Search

The objective of this paper is to optimize the energy consumption and performance of fuel cell electric bicycles (FCEBs) under specific key input parameters. The paper applied an integrated method that includes an artificial neural network (ANN) and genetic algorithm (GA) to forecast and identify an optimal performance and energy consumption of FCEBs. The simulation model of FCEBs is established and simulated in MATLAB-Simulink environment to generate 1000 data points, that are used for training, validating, testing artificial neural network, the ANN architecture containing five input neurons, two hidden neurons and two output neurons, respectively. Furthermore, the GA is integrated to find the maximum performance and energy consumption once the ANN has exactly been trained. The study found that the FCEB configuration can achieve an effective performance at 30.3 km/h with required power of 210.4 W under speed level_5, radius of wheel 0.39 m, frontal area 0.423 m2, slope grade 0%. In order to validate and verify the simulated results, the experimental approach method was conducted in the same condition. The experimental results fit well with the simulated results in the same initial input parameters. • A method integrating the artificial neural network with genetic algorithm is applied. • The consumed power and fuel cell voltage are optimized by combining artificial neural network and genetic algorithm (ANN-GA). • The effective performance area of fuel cell electric bicycle under various a structure and operating parameters is identified. • Deep-learning results reflect major physical mechanism of fuel cell electric bicycle operation. [ABSTRACT FROM AUTHOR]

Considering the dynamic nature of market conditions, this paper introduces a state-dependent futures hedging optimization model and methodology. This approach dynamically adjusts the traditional model-driven hedging strategy, effectively balancing the pursuit of returns with the imperative of risk mitigation. Empirical evidence shows that integrating Hidden Markov Model (HMM) with machine learning techniques, as demonstrated in this study, improves the accuracy of market state forecasts. Compared to the traditional model-driven hedging strategy, the innovative state-dependent hedging strategy introduced here significantly enhances the return-to-risk ratio, and revenue, without increasing hedging risks. Moreover, the hedging portfolio developed under this strategy achieves an average hedging efficiency of 0.76, highlighting the effectiveness of the proposed methodology. Additional robustness tests indicate that this market state-dependent hedging optimization strategy is promising under various conditions, including different position adjustment ratios, volatility benchmarks, evaluation periods, types of crude oil, and transaction costs. The research conducted in this paper not only contributes to and expands traditional hedging theories but also provides a practical risk management solution for market participants. • A state-dependent hedging strategy is proposed. • A state prediction model combining HMM and machine learning is established. • The hedging model effectively balances the pursuit of returns with the imperative of risk mitigation. [ABSTRACT FROM AUTHOR]

Increasing integration of distributed solar photovoltaic (PV) into distribution networks could result in adverse effects on grid operation. Traditional model-based control algorithms require accurate model information that is difficult to acquire and thus are challenging to implement in practice. This paper proposes a surrogate model-enabled grid visibility scheme to empower deep reinforcement learning (DRL) approach for distribution network voltage regulation using PV inverters with minimal system knowledge. In contrast to existing DRL methods, this paper presents and corroborates the adverse impact of missing load information on DRL performance and, based on this finding, proposes a surrogate model methodology to impute load information utilizing observable data. Additionally, a multi-fidelity neural network is utilized to construct the DRL training environment, chosen for its efficient data utilization and enhanced robustness to data uncertainty. The feasibility and effectiveness of the proposed algorithm are assessed by considering DRL testing across varying degrees of observable load information and diverse training environments on a realistic power system. • A visibility-enhanced surrogate has been introduced, empirically demonstrating its effectiveness in enhancing deep reinforcement learning (DRL) performances. • A multi-fidelity neural network is employed as a surrogate model within the training environment, improving data utilization efficiency while ensuring the effectiveness of DRL training. • Realistic test feeders have been used for testing, further validating the feasibility of the proposed method in practical scenarios. [ABSTRACT FROM AUTHOR]

In the last decade, with the introduction of smart meters to smart grids, demand response programs (DRPs) have been widely adopted to establish a generation and consumption balance. DRPs provide many benefits for efficient grid management. However, these programs are conducive to higher levels of dissatisfaction by changing grid customers' consumption patterns. This paper aims to investigate the effects of DRPs on social welfare (SW). To this end, the paper presents a mathematical model for SW during the implementation of DRPs. In the proposed model, the level of customer satisfaction is assumed the main factor contributing to SW. This mathematical model considers different types of DRPs in terms of their impacts on SW. The paper also seeks to obtain linear and nonlinear models of DRPs and the coefficient of participation (CoP). CoP as an indicator shows the percentage of customers who actively participate in each DRP and plays a significant role in the assessment of the SW level. Moreover, owing to the sparsity and variety of distribution network customers, load patterns are classified into different clusters to take the load types into account. As a matter of fact, this process aims to identify similar patterns and thus, the same level of satisfaction for each separate cluster. The classification process is performed by using a machine learning-based clustering method known as the Affinity Propagation (AP) algorithm. Then, the model calculates the level of SW for the clusters based on the usage of electrical equipment and the time of day when they are turned on. The obtained levels of SW help operators select the best programs for every cluster in terms of customer satisfaction, and achieve the highest performance of DRPs. Lastly, the model is evaluated using real data of a distribution network to ensure the effectiveness and accuracy of the model. [Display omitted] • Development of a mathematical model for assessing Social Welfare during DRPs execution, considering customer satisfaction. • Integration of Coefficient of Participation (CoP) as an important factor in evaluating SW during DRP implementation. • Development of Affinity Propagation algorithm to cluster load patterns, enabling tailored DRPs to maximize effectiveness. [ABSTRACT FROM AUTHOR]

Accurate estimation of State of Health (SOH) is crucial to ensure optimal performance and safe operation of lithium-ion battery. This paper proposes a Stacking ensemble learning paradigm for SOH estimation. The Stacking ensemble learning increases adaptability to different features by using base learners with different structures, reducing the risk of overfitting. The model utilizes random vector functional link (RVFL) and active state tracking long-short-term memory network (AST-LSTM) as base learners, where AST-LSTM actively tracks long-term information of lithium-ion battery, and RVFL acts as the meta-learner for stacking. The random vector functional link network helps to avoid the problem of gradient vanishing that is commonly encountered in neural networks due to the gradient descent principle. To further improve estimation accuracy, Singer initialization method and dimension learning method are employed to enhance the Heap-based optimization (HBO) algorithm. In this study, the IHBO algorithm is used to optimize the hyperparameters of the model. Comparing with other methods, the hybrid model proposed in this paper demonstrates superior estimation performance under different operating conditions: at a temperature of 24 °C with a discharge current of 1 A, at a temperature of 4 °C with a discharge current of 1 A, and at a temperature of 4 °C with a discharge current of 2 A. The highest RMSE of the proposed method for the three working conditions are 0.006, 0.01, and 0.017, respectively. Therefore, the proposed Stacking ensemble learning is feasible for SOH estimation of lithium-ion battery and can better adapt to lithium-ion battery data under different operating conditions. • A Stacking ensmeble learning strategy is proposed for SOH estimation. • The improved HBO algorithm is introduced to optimize the Stacking model. • SOH for lithium batteries under three different operating conditions is estimated. • Six benchmark models are used to verify the performance of the proposed model. [ABSTRACT FROM AUTHOR]

To mitigate the impact of market uncertainty on trading investments, this paper proposes a forecasting and investing framework for crude oil market by integrating interval models and machine learning models. Firstly, natural language processing technique is employed to analyze text information from social and news media, enabling the capture of market and societal sentiment. Subsequently, deep learning models are integrated to combine sentiment data with other economic variables for more accurate predictions of crude oil prices. Furthermore, this paper introduces a trading strategy with interval constraints based on interval prediction models to reduce trading risk arising from the uncertainty of point forecasts in investments. Through trading simulations, it is discovered that employing the interval constrained strategy is more effective in reducing trading risk and enhancing investment returns compared to point-based trading strategies. This interval-based strategy offers a novel approach to mitigating investment risk in the crude oil market. • A novel prediction framework is proposed to integrate text data and interval data. • Combining point-valued and interval-valued deep learning models. • An interval-based trading strategy is developed. • The proposed trading strategy can reduce risk while improving investment returns. [ABSTRACT FROM AUTHOR]

Wind power efficiency is an essential factor affecting wind power development, and efficient wind power control methods are the key to improving wind power efficiency. Previous wind power control methods require high internal system expertise in internal systems and struggled to balance the accuracy of the maximum power control and the output stability under high wind speed. Therefore, this paper proposes a reward-adaptive control method for wind power tracking based on Deep Deterministic Policy Gradient (DDPG). The method can use one controller to simultaneously control the generator torque and pitch angle in various operating conditions following the system state and the designed flag of operating conditions, thus enabling efficient tracking of the maximum power generation, and stabilizing the output power under high wind speed. Moreover, this paper proposes an internal and external reward algorithm that integrates the Intrinsic Curiosity Module (ICM) and the Actor–Critic (AC) architecture, which can adaptively calculate the reward of DDPG, thereby effectively resolving the sparse reward problem, accelerating the convergence of neural network, and improving the learning effect. From the simulation, the control method proposed in this paper can effectively improve the power generation efficiency under turbulent wind speed, reduce the pitch angle variation by about 30%, and improve the power tracking accuracy by more than 70%. • A reward-adaptive wind power control method based on DDPG is studied. • The method uses one controller to control the power in different operation. • A reward algorithm integrates the ICM and the Actor–Critic is proposed. • The results show that the method improves the power tracking performance. [ABSTRACT FROM AUTHOR]

[Display omitted] • Estimated relationship between nighttime stable light data and carbon emissions. • Reference to shared social economy and representative concentration to set scenarios. • Classification of cities based on industry output and GDP. • The carbon emission pathway of the city from 2021 to 2060 was dynamically simulated. • Service cities are more likely to have autonomous carbon peaks. As cities are critical actors in mitigating climate change and achieving the "3060″ target, multi-scenario studies on urban carbon emissions can provide a scientific basis for formulating urban carbon peaking action plans. To remedy the problems of missing regional statistics, inconsistent caliber, and lack of city-scale studies in carbon emission research, this paper uses the sparrow optimization neural network algorithm to fit carbon emission data with nighttime stable light for training. Carbon emission data were obtained for 281 cities in China during 2000–2020. The rates of change of influencing factors are set based on shared socioeconomic pathways (SSPs) and representative concentration pathways (RCPs) for different periods and different scenarios. The carbon emission and carbon peaking evolution paths of service, industrial and comprehensive cities from 2021 to 2060 are dynamically simulated. The results show that (1) service cities are significantly higher than industrial and comprehensive cities in population, GDP, secondary industry output, and energy consumption. (2) The economic development effect, as the primary driver of carbon emission growth, increases and then decreases in all five categories of cities, with 2010 as the inflection point. Industrial structure improvement has an increasingly strong offsetting effect on carbon emissions and is one of the critical directions for future carbon emission reduction. (3) Service cities such as Beijing and Shanghai are already at the completion stage of urban transformation and are more likely to reach the carbon peak on their own than other types of cities. In the low carbon following scenario, comprehensive cities such as Kaifeng, Rizhao, and Jilin can achieve their carbon peaking targets efficiently. The findings of this paper can provide valid theoretical support for carbon peaking action programs in China and other countries. [ABSTRACT FROM AUTHOR]

• The AES power system intelligent energy scheduling environment is built. • The BI-LSTM-Att action network is propose by combining BI-LSTM with attention mechanism. • The DQN-CE algorithm is proposed by improving the DQN algorithm. • The experiments on an all-electric ferry is conducted using a variety of methods. At present, Ship energy scheduling based on deep reinforcement learning (DRL) is an important research direction in which Deep Q learning algorithm (DQN) has been successfully applied in the field of energy scheduling. In terms of the shortcomings of DQN in all electric ships(AES) energy scheduling, such that, the insufficient performance of using multiplayer perceptron(MLP) as action network and the degradation of convergence performance due to over estimation of Q value, in this paper, we propose a DQN cross entropy(DQN-CE) energy scheduling algorithm combing bi-directional LSTM and attention mechanism(BI-LSTM-Att). Instead of using MLP, firstly, we use BI-LSTM-Att as action network of DQN, and then an improved DRL algorithm called DQN-CE is proposed, that is, we add cross-entropy loss of predicted action of the next time for the target network and the action network to DQN. Finally, we conduct simulation experiments on an all-electric ferry, and the experimental results show that the energy scheduling algorithm proposed in this paper reduces the economic consumption by 4.11%, increases the utilization rate of energy storage system by 24.4%, and reduces the exploration time of agent training by 31.3% compared with the original DQN energy scheduling algorithm. Finally, the effectiveness and superiority of the algorithm were further verified in the new case study. [ABSTRACT FROM AUTHOR]

Integrating enormous quantities of wind energy into the electrical system requires precise planning and forecasting. This paper presents a novel framework for wind power forecasting, which establishes a new standard for accuracy and reliability. It uses advanced deep neural networks and a stacking ensemble mechanism to make probabilistic forecasts. These forecasts address wind speed volatility and variability, which continue to be key issues in producing consistently accurate wind power forecasts despite advances in deep learning. Base learners generate forecasts that include lower and upper bounds, as well as median prediction intervals. The Huber regressor, also known as the Meta-learner, combines projections from many models to reduce susceptibility to extreme values. Expanding window cross-validation is used in performance evaluation, with a focus on Mean Absolute Error, Mean Squared Error (MSE), Symmetric Mean Absolute Percentage Error, Prediction Interval Coverage Probability, and Average Interval Width for one, two, and three step ahead predictions. The model's stability is measured by the standard deviation of MSE throughout each validation window. Additionally, the Diebold–Mariano test is used to validate the Meta-learner's predictive accuracy in contrast to other advanced models and the seasonal naïve benchmark. This study found that the meta model outperformed all other models in eight of the nine forecasts tested, indicating its effectiveness in medium-range forecasting. In Germany, it exceeded Temporal Fusion Transformer by two and three steps ahead, with 22% and 34% improvements, respectively. Indeed, this study not only provides a reliable prediction tool, but also lays foundations for efficient and effective energy management and policy planning in the renewable energy industry. • Deep neural network architectures are applied to probabilistic wind power forecasting. • Transformer and artificial neural network approaches combined with conformal prediction intervals is proposed. • A window cross-validation scheme on wind power forecasting is evaluated. • Wind power data from Germany, the UK, and the Netherlands are used for forecasting purposes. • SHAP and LIME analyses perform the explanatory analysis of the variables in the meta-learner's predictions. [ABSTRACT FROM AUTHOR]

Power grid damage and blackouts are increasing with climate change. Load forecasting methods that integrate climate resilience are therefore essential to facilitate timely and accurate network reconfiguration during periods of extreme stress. Our paper proposes a generalised Wildfire Resilient Load Forecasting Model (WRLFM) to predict electricity load based on operational data of a Distribution Network (DN) in Australia during wildfire seasons in 2015–2020. We demonstrate that load forecasting during wildfire seasons is more challenging than during non-wildfire seasons, motivating an imperative need to improve forecast performance during wildfire seasons. To develop the robust WRLFM, comprehensive comparative analyses were conducted to determine proper Machine Learning (ML) forecast structures and methods for incorporating multiple factors. Bi-directional Gated Recurrent Unit (Bi-GRU) and Vision Transformer (ViT) were selected as they performed the best among all 13 recently trending ML methods. Multi-factors were incorporated to contribute to forecast performance, including input sequence structures, calendar information, flexible correlation-based temperature conditions, and categorical Fire Weather Index (FWI). High-resolution categorical FWI was used to build a forecasting model with climate resilience for the first time, significantly enhancing the average stability of forecast performances by 42%. A sensitivity analysis compared data set patterns and model performances during wildfire and non-wildfire seasons. The improvement rate of load forecasting performance during wildfire seasons was more than two times greater than in non-wildfire seasons. This indicates the significance and effectiveness of applying the WRLFM to improve forecast accuracy under extreme weather risks. Overall, the WRLFM reduces the Mean Absolute Percentage Error (MAPE) of the forecast by 14.37% and 20.86% for Bi-GRU and ViT-based models, respectively, achieving an average forecast MAPE of around 3%. • Novel climate-resilient load forecast model for use during extreme wildfire seasons. • Bi-GRU and ViT outperform 11 recently trending machine learning models. • Categorical FWI with high resolution enhances forecast stability by 42%. • Optimised calendar, temperature and FWI input reduce errors by 17.6%. • Forecast improvement during wildfires is twice that in non-wildfire seasons. [ABSTRACT FROM AUTHOR]

The increasing adoption of electric vehicles (EVs) enhances charging stations' role in balancing power grid demands through dynamically pricing and adjusting charging power strategies. However, unpredictable nature of EV flow and the necessity for rapid response to real-time peak shaving present complex challenges. To tackle this, a novel dual-center control scheme and learning optimization methods are proposed in this paper. It adopts a station operation model based on queuing theory, considering charging pile capacity constraints. The proposed control scheme, integrating a service price maker (SPM) and charging power controller (CPC), adapts service prices and charging power in response to electricity prices and peak-shaving directives. Employing a model-free Deep Hierarchical Reinforcement Learning (DHRL) approach, the SPM targets economic maximization, while the CPC aims to balance peak-shaving efficacy and user satisfaction. Furthermore, to improve the offline learning efficiency of the algorithm, feature fusion (FF) technology is used to condense high-dimensional data into essential low-dimensional features, improving the algorithm's learning efficiency. The effectiveness of the proposed FF-DHRL was verified in a simulated charging station system. • A dual-center control scheme based on price guidance and charging power control is proposed. • A station operation model considering dynamic charging demand and grid peak shaving is formulated under the proposed dual-center control scheme. • The bi-layer collaborative optimization method based on DHRL is proposed. • A novel fused state is designed to improve the training performance of agents. [ABSTRACT FROM AUTHOR]

Establishing building energy models (BEMs) for building design and analysis poses significant challenges due to demanding modeling efforts, expertise to use simulation software, and building science knowledge in practice. These make building modeling labor-intensive, hindering its widespread adoptions in building development. Therefore, to overcome these challenges in building modeling with enhanced automation in modeling practice, this paper proposes Eplus-LLM (EnergyPlus-Large Language Model) as the auto-building modeling platform, building on a fine-tuned large language model (LLM) to directly translate natural language description of buildings to established building models of various geometries, occupancy scenarios, and equipment loads. Through fine-tuning, the LLM (i.e., T5) is customized to digest natural language and simulation demands from users and convert human descriptions into EnergyPlus modeling files. Then, the Eplus-LLM platform realizes the automated building modeling through invoking the API of simulation software (i.e., the EnergyPlus engine) to simulate the auto-generated model files and output simulation results of interest. The validation process, involving four different types of prompts, demonstrates that Eplus-LLM reduces over 95% modeling efforts and achieves 100% accuracy in establishing BEMs while being robust to interference in usage, including but not limited to different tones, misspells, omissions, and redundancies. Overall, this research serves as the pioneering effort to customize LLM for auto-modeling purpose (directly build-up building models from natural language), aiming to provide a user-friendly human-AI interface that significantly reduces building modeling efforts. This work also further facilitates large-scale building model efforts, e.g., urban building energy modeling (UBEM), in modeling practice. • This study represents the pioneering effort in customizing LLM for auto-modeling. • A large language model is used to digest natural language descriptions for modeling. • Integration of four prompts enhances the flexibility and versatility of our platform. • Our platform provides a human-AI interface and reduces building modeling efforts. [ABSTRACT FROM AUTHOR]

Deploying renewable energy and implementing smart energy management strategies are crucial for decarbonizing Building Energy Systems (BES). Despite recent advancements in data-driven Deep Reinforcement Learning (DRL) for BES optimization, significant challenges still exist, such as the time-consuming and data-intensive nature of training DRL controllers and the complexity of environment dynamics in Multi-Agent Reinforcement Learning (MARL). Consequently, these obstacles impede the synchronization and coordination of multiple agent control, leading to slow DRL convergence performance. To address these issues. This paper proposes a novel approach to optimize hybrid building energy systems. We introduce an integrated system combining a multi-stage Proximal Policy Optimization (PPO) on-policy framework with Imitation Learning (IL), interacting with the model environment. To improve scalability and robustness of Multi-agent Systems (MAS), this approach is designed to enhance training efficiency with centralized training and decentralized execution. Simulation results of case studies demonstrate the effectiveness of the Multi-agent Deep Reinforcement Learning (MADRL) model in optimizing the operations of hybrid building energy systems in terms of indoor thermal comfort and energy efficiency. Results show the proposed framework significantly improve performance in achieving convergence in just 50 episodes for dynamic decision-making. The scalability and robustness of the proposed model have been validated across various scenarios. Compared with the baseline during cold and warm weeks, the proposed control approach achieved improvements of 34.86% and 46.10% in energy self-sufficiency ratio, respectively. Additionally, the developed MADRL effectively improved solar photovoltaic (PV) self-consumption and reduced household energy costs. Notably, it increased the average indoor temperature closer to the desired set-point by 1.33 °C, and improved the self-consumption ratio by 15.78% in the colder week and 18.47% in the warmer week, compared to baseline measurements. These findings highlight the advantages of the multi-stage PPO on-policy framework, enabling faster learning and reduced training time, resulting in cost-effective solutions and enhanced solar PV self-consumption. • Integrated a multi-stage RL algorithm with imitation learning in the MAPPO framework to optimize system energy performances. • Proposed model achieved faster MAPPO convergence under resource constraints • Introduced a shared reward function in MAPPO to align agent incentives with collective energy optimization. • Improved solar PV self-consumption and reduced energy costs while maintaining indoor thermal comfort with MAPPO. • Validated the scalability and adaptability of MAPPO across various scenarios. [ABSTRACT FROM AUTHOR]

Non-intrusive load monitoring is a prominent part of demand-side energy management that provides visibility of flexible loads to support real-time electricity market pricing strategies and intelligent demand response programs. Compared with household-level load disaggregation, substation-level load disaggregation can significantly preserve residential privacy and reduce facility costs while providing sufficient information of flexible loads for intelligent demand-side energy management from the area scale. Especially, among various flexible loads, thermostatically controlled loads are highlighted due to their large proportion and high demand response elasticity. However, due to the variation and complexity of residential routines on a large scale, disaggregation of flexible loads from the substation level remains unsolved. To this end, focusing on thermostatically controlled loads, this paper proposes a contrastive sequence-to-point learning algorithm for substation-level flexible load disaggregation to fill the research gap. In the first stage, the theory of the effect of load aggregation and thermal inertia effect is introduced, and significant impact factors on flexible loads are summarized. Secondly, a substation-level flexible load disaggregation algorithm based on contrastive sequence-to-point learning is proposed, where pair-wise comparison and residual mechanism are combined in a semi-supervised structure to extract deep features and track fluctuations in flexible loads. Then, SHapley Additive exPlanations are utilized to ensure the optimization and interpretability of the algorithm. The proposed algorithm is tested and verified on public datasets under low frequency, reducing the disaggregation Mean Absolute Percentage Error of thermostatically controlled loads to as low as 8.78% and 11.26% for bi-directional and unidirectional structures separately. Additionally, it is generalizable to disaggregate other flexible loads, including photovoltaic and electric vehicles, demonstrating satisfactory performance. The algorithm has proved to be robust to data sparsity problems and practical for substation-level demand response potential estimation. • A novel contrastive-based sequence-to-point learning architecture is proposed for substation-level flexible load disaggregation. • Effects of load aggregation and thermal inertia of TCLs are analyzed as theory support to determine crucial impact factors. • Residual connections and pair-wise comparison mechanisms are employed to support the in-depth feature extraction and overcome the data sparsity challenge. • Shapley Additive exPlanations are utilized to ensure the optimization and interpretability of the proposed algorithm. • Verified on large-scale public datasets to prove the generalization for other flexible loads and the practicability for potential estimation of demand response [ABSTRACT FROM AUTHOR]

Emerging data-driven approaches, such as deep reinforcement learning (DRL), aim at on-the-field learning of powertrain control policies that optimize fuel economy and other performance metrics. Indeed, they have shown great potential in this regard for individual vehicles on specific routes/drive cycles. However, for fleets of vehicles that must service a distribution of routes, DRL approaches struggle with learning stability issues that result in high variances and challenge their practical deployment. In this paper, we present a novel framework for shared learning among a fleet of vehicles through the use of a distilled group policy as the knowledge sharing mechanism for the policy learning computations at each vehicle. We detail the mathematical formulation that makes this possible. Several scenarios are considered to analyze the framework's functionality, performance, and computational scalability with fleet size. Comparisons of the cumulative performance of fleets using our proposed shared learning approach with a baseline of individual learning agents and another state-of-the-art approach with a centralized learner show clear advantages to our approach. For example, we find a fleet average asymptotic improvement of 8.5% in fuel economy compared to the baseline while also improving on the metrics of acceleration error and shifting frequency for fleets serving a distribution of suburban routes. Furthermore, we include demonstrative results that show how the framework reduces variance within a fleet and also how it helps individual agents adapt better to new routes. • A shared learning algorithm for a fleet of vehicles serving a distribution of routes. • A characterization of the scalability and computational complexity of the algorithm. • Demonstration of the benefits of shared learning for fleets in varying scenarios. [ABSTRACT FROM AUTHOR]

• A two-stage method is proposed for electricity price forecasting. • The method diversifies the data resources improve the forecasting accuracy. • The method is tested using the real-world data of Texas electricity market. • Causal relationship combined with physical models guarantee better results. • Deep neural networks show a great performance in electricity price forecasting. Over the years, the growing penetration of renewable energy into the electricity market has resulted in a significant change in the electricity market price. This change makes the existing forecasting method prone to error, decreasing the economic benefits. Hence, more precise forecasting methods need to be developed. This paper starts with a survey and benchmark of existing machine learning approaches for forecasting the real-time market (RTM) price. While these methods provide sufficient modeling capability via supervised learning, their accuracy is still limited due to the single data source, e.g., historical price information only. In this paper, a novel two-stage supervised learning approach is proposed by diversifying the data sources such as highly correlated power data. This idea is inspired by the recent load forecasting methods that have shown extremely well performances. Specifically, the proposed two-stage method, namely the rerouted method, learns two types of mapping rules. The first one is the mapping between the historical wind power and the historical price. The second is the forecasting rule for wind generation. Based on the two rules, we forecast the price via the forecasted generation and the first learned mapping between power and price. Additionally, we observed that it is not the more training data the better, leading to our validation steps to quantify the best training intervals for different datasets. We conduct comparisons of numerical results between existing methods and the proposed methods based on datasets from the Electric Reliability Council of Texas (ERCOT). For each machine learning step, we examine different learning methods, such as polynomial regression, support vector regression, neural network, and deep neural network. The results show that the proposed method is significantly better than existing approaches when renewables are involved. [ABSTRACT FROM AUTHOR]

Highlights • Multiple optimization targets are considered for a generalized energy Internet. • Power supply-demand balance is realized in the entire energy Internet system. • A model-free approach is applied with real-world power data. • Advanced deep reinforcement learning approach is applied. Abstract This paper investigates the energy management problem in the field of energy Internet (EI) with interdisciplinary techniques. The concept of EI has been proposed for a while. However, there still exist many fundamental and technical issues that have not been fully investigated. In this paper, a new energy regulation issue is considered based on the operational principles of EI. Multiple targets are considered along with constraints. Then, the practical energy management problem is formulated as a constrained optimal control problem. Notably, no explicit mathematical model for power of renewable power generation devices and loads is utilized. Due to the complexity of this problem, conventional methods appear to be inapplicable. To obtain the desired control scheme, a model-free deep reinforcement learning algorithm is applied. A practical solution is obtained, and the feasibility as well as the performance of the proposed method are evaluated with numerical simulations. [ABSTRACT FROM AUTHOR]

Highlights • Review of the application of reinforcement learning (RL) for demand response (DR) • DR is relevant for integrating renewable energy source into the smart grid. • Considering RL/DR from energy generation and storage, to supply and user satisfaction. • Typical algorithms and open research directions are discussed. • Most articles focus on single-agent systems in stationary environments. Abstract Buildings account for about 40% of the global energy consumption. Renewable energy resources are one possibility to mitigate the dependence of residential buildings on the electrical grid. However, their integration into the existing grid infrastructure must be done carefully to avoid instability, and guarantee availability and security of supply. Demand response, or demand-side management, improves grid stability by increasing demand flexibility, and shifts peak demand towards periods of peak renewable energy generation by providing consumers with economic incentives. This paper reviews the use of reinforcement learning, a machine learning algorithm, for demand response applications in the smart grid. Reinforcement learning has been utilized to control diverse energy systems such as electric vehicles, heating ventilation and air conditioning (HVAC) systems, smart appliances, or batteries. The future of demand response greatly depends on its ability to prevent consumer discomfort and integrate human feedback into the control loop. Reinforcement learning is a potentially model-free algorithm that can adapt to its environment, as well as to human preferences by directly integrating user feedback into its control logic. Our review shows that, although many papers consider human comfort and satisfaction, most of them focus on single-agent systems with demand-independent electricity prices and a stationary environment. However, when electricity prices are modelled as demand-dependent variables, there is a risk of shifting the peak demand rather than shaving it. We identify a need to further explore reinforcement learning to coordinate multi-agent systems that can participate in demand response programs under demand-dependent electricity prices. Finally, we discuss directions for future research, e.g., quantifying how RL could adapt to changing urban conditions such as building refurbishment and urban or population growth. [ABSTRACT FROM AUTHOR]

Highlights • A deep recurrent neural network (RNN) model is used for medium-term thermal load prediction. • The deep RNN model outperforms a simple multilayer perceptron for time series forecasting. • An optimization framework is proposed for sizing a thermal storage tank to meet thermal loads. • The optimization method can be used in selecting a thermal storage tank for use with a building. • The combined predictions and optimization allow for estimating performance of the tank. Abstract This paper evaluates the performance of deep recurrent neural networks in predicting heating demand for a commercial building over a medium-to-long term time horizon (⩾ 1 week), and proposes a modeling framework to demonstrate how these longer-term predictions can be used to aid design of a stratified thermal storage tank. The building sector contributes significantly to primary energy consumption in the US, and as such, there is a need to predict heating demand in buildings over longer time horizons, and to develop methods that can facilitate installation, planning and management of distributed generation and thermal storage to meet these heating demands. Key objectives of this paper are: (a) Investigate how a deep recurrent neural network model performs in predicting heating demand in campus buildings at University of Utah over multiple weeks, and (b) develop an optimization framework that which can provide definitive guidelines on sizing a stratified thermal storage tank without requiring high performance computing resources. The results showed that the predictions by the deep RNN are comparatively more accurate than those by a 3-layer MLP, and that these deep RNN predictions can adequately serve as proxy for future demand while considering sizing in the design of a complementary stratified thermal storage tank. [ABSTRACT FROM AUTHOR]

The pulp and paper industry is a fundamental sector of the economy of many countries. However, this sector requires real collaboration and initiatives from stakeholders to reduce its significant consumption of energy and emission of greenhouse gases. Heat exchangers are examples of equipment in pulp mills that are subjected to undesirable and complex phenomena such as evolution of fouling over time, which leads to inefficiency in terms of energy consumption and unplanned shutdowns, resulting in ineffective maintenance strategies and production costs. Therefore, there is a clear need to develop an accurate predictive maintenance tool that helps mill operators avoid such situations. It is necessary for that tool to effectively track the fouling evolution level and, based on it, deploy a reliable prognostics approach to estimate more accurately the time-to-clean of this equipment. This study presents a new hybrid prognostics approach for fouling prediction in heat exchangers. The proposed approach relies on the fusion of information of different prediction horizons to estimate the time-to-clean. Employing long short-term memory, it allows adaptation of long-term predictions by accurate short-term predictions using multiple non-linear auto-regressive exogenous models. This fusion not only captures the changes in degradation trend over time, but also ensures a good accuracy of prognostics results in both the short- and long-term horizons for planning maintenance actions. The effectiveness of the proposed approach was successfully proven on real industrial data collected from a pulp mill heat exchanger located in Canada. • New adaptive prognostic approach based on fusion of short and long-term predictors. • Selection of relevant physical parameters for monitoring of fouling evolution. • Accurate long-term fouling prediction by using short-term predictors. • High accuracy of fouling time-to-clean in a real heat exchanger plant. [ABSTRACT FROM AUTHOR]

• A commuting-pattern-oriented representation of random driving conditions is developed. • Integrated stochastic optimization of powertrain design and control is engineered. • Machine-learning solvers for stochastic optimization are developed. • The stochastic optimization improves the expected energy economy by up to 12%. • Topology modification to EV powertrains improves energy economy by over 17%. The performance of an electric vehicle (EV) powertrain is greatly influenced by the design scheme, control strategy, and driving conditions. Owing to uncertainties in driving conditions and the mutual correlation between design scheme and control strategy, identifying the optimal topology and component parameters of EV powertrains for improving the energy efficiency of EVs remains a challenging task. Oriented to the commuting patterns of individuals, this paper proposes a stochastic optimization method for identifying the superior powertrain design that optimizes the expected energy economy of the EV powertrain and reduces the energy economy variability in random driving conditions. This paper also introduces a machine-learning solution to the complex mutual restrictions among design variables, control strategies, and driving conditions, which determines the feasible design space and powertrain performance. Suggested by massive validations, the stochastic optimization method reduces the expectation and deviation of the energy consumption of EV powertrains up to 12.0% and 5.6%, compared with deterministic optimization methods. Additionally, by comparative evaluation among stochastic optimal design results of different topologies, contributions of topology modifications to the energy efficiency of EV powertrains are clarified, and the superior powertrain topology is identified, which improves 17.1% expected energy economy compared with the current popular topology. [ABSTRACT FROM AUTHOR]

• An RF based model is proposed for crown snow load outage risk estimation of OHLs. • Shapley additive explanations values are used to estimate variable impacts. • Accurate outage risk estimates are achieved with open-source explanatory variables. In the northern hemisphere, snow accumulating on trees and overhead lines causes widespread outages in the electricity distribution networks. Accurate outage risk models are an essential element in improving the resilience of modern distribution networks. In this paper, a Random Forest-based model for estimating the susceptibility of overhead lines to outages caused by tree crown snow loads is proposed. The model uses a novel combination of an aerial inspection outage risk dataset, an advanced forest crown snow load risk map, a canopy height model, and forest characteristics data. All predictor variables used in the study are available as open data. As a result, outage risk probability in 50 m overhead line sections for a distribution network was generated. Cross-validation of the model showed a good predictive performance with a receiver operating characteristic area under curve (ROC AUC) of 0.75 and an accuracy of 0.74. The impact of the predictor variables was investigated by using Shapley additive explanations (SHAP) values. The most impactful variables were the forest crown snow load risk, the number of nearby canopy height model pixels, and the birch tree volume. The outage risk probability model developed in this paper could be similarly applied to assess the crown snow load risk in other distribution networks or even in other types of networks, such as roads and railways. [ABSTRACT FROM AUTHOR]

Accurate crude oil price forecasting is crucial for maintaining energy economy stability and enhancing investment and operational decision-making. Nowadays, the link prediction model for crude oil prices based on a complex network model has become an emerging and promising approach. However, existing link prediction methods are unable to extract complex information about interval-valued crude oil prices, and fail to consider the impact of multi-source information. Therefore, this paper proposes the link prediction model for interval-valued crude oil prices using complex network and multi-source information, which converts crude oil price interval series into complex networks. Through link prediction, it takes into account both the impact of past interval-crude oil prices on future interval time series and the interference of external real-time information on interval-valued prices. First, news headlines and geopolitical risk indices related to crude oil prices are captured, and search index data and news sentiment scores associated with interval data are analyzed. Subsequently, the interval sequence data and related influencing factors are decomposed and reconstructed respectively. The reconstructed interval center and radius sequence are then converted into a directed network, and node similarity in the network is calculated to generate similar nodes of the interval subsequence as feature values. Finally, multiple machine learning models are employed to acquire the optimal weighted combination prediction with interval prediction values. According to the experimental data, this approach outperforms other benchmark methods in terms of prediction accuracy. • Predictive modeling to transform interval-valued into complex networks is proposed. • A link prediction model based on multi-source information is constructed. • Combining AI, link prediction, and complex networks forms a multidisciplinary mothed. • The proposed model is well validated in predicting interval WTI crude oil prices. • The constructed model can effectively extract multi-scale data features. [ABSTRACT FROM AUTHOR]

Lithium-ion batteries have found their way into myriad sectors of industry to drive electrification, decarbonization, and sustainability. A crucial aspect in ensuring their safe and optimal performance is monitoring their energy levels. In this paper, we present the first study on predicting the remaining energy of a battery cell undergoing discharge over wide current ranges from low to high C-rates. The complexity of the challenge arises from the cell's C-rate-dependent energy availability as well as its intricate electro-thermal dynamics especially at high C-rates. To address this, we introduce a new definition of remaining discharge energy and then undertake a systematic effort in harnessing the power of machine learning to enable its prediction. Our effort includes two parts in cascade. First, we develop an accurate dynamic model based on integration of physics with machine learning to capture a battery's voltage and temperature behaviors. Second, based on the model, we propose a machine learning approach to predict the remaining discharge energy under arbitrary C-rates and pre-specified cut-off limits in voltage and temperature. The experimental validation shows that the proposed approach can predict the remaining discharge energy with a relative error of less than 3% when the current varies between 0 ∼ 8 C for an NCA cell and 0 ∼ 15 C for an LFP cell. The approach, by design, is amenable to training and computation. • A new C-rate-dependent definition of remaining discharge energy for lithium-ion batteries. • A new machine learning approach to predict the remaining discharge energy. • A novel way to train the machine learning approach using synthetic data generated by hybrid physics+machine learning models. • Experiments and simulations show the approach is highly accurate across broad C-rate ranges. [ABSTRACT FROM AUTHOR]

Understanding user-level energy demand is pivotal for sustainable development and smart grid implementation, as it facilitates optimal resource allocation and energy conservation. Traditional machine learning models, however, often ignored the potential connections between users, limiting comprehension of user-level energy demands. Capturing behavioral correlations among users in high-dimensional temporal data remains challenging. In this paper, a novel scheme is proposed for revealing the implicit energy behavior correlation, which serves as prior knowledge to enhance predictive model performance. In addition, a spatio-temporal feature extraction framework is introduced to fuse spatial and temporal information, capturing the coherence of energy consumption data. This approach captures the correlation between the energy consumption patterns of different users, achieving highly accurate demand forecasting at the user level. In fine-grained demand forecasting experiments, the prediction accuracy of the proposed approach was improved by 14% compared with the baseline model. Based on fine-grained prediction results, an innovative visual analytical interface is also developed to characterize the migration of energy demand over time in both physical and topological space, offering valuable insights into demand-side energy management. In the empirical study, we found that there is an obvious mental inertia in the energy behavior of urban residents, which leads to energy waste. Our research provides critical insights for policymakers and planners in addressing the sustainability challenges of urban energy systems. • Mapped high-dimensional energy data in 2D space, identified 6 basic energy patterns. • Developed a user-level energy forecasting model with spatio-temporal correlations. • Created a visual tool to track demand migration, aided demand-side management. [ABSTRACT FROM AUTHOR]

With the surge of electric vehicles, accurate estimation of their energy consumption becomes increasingly critical. Data-driven models have been widely used for estimating the energy consumption of electric vehicles; however, their applications often face limitations due to inadequate training data, resulting in over-fitting and poor generalizability. In this paper, a physically rational data augmentation approach is proposed to expand the driving trip dataset. By incorporating physical coherence into the augmentation process, new driving trips are synthesized with rational physical context. The effectiveness of the proposed approach is validated by applying it to three data-driven models for estimating the energy consumption of electric vehicles across different validation scenarios. Compared with two baseline data augmentation approaches, our proposed approach demonstrates superior model training performance with less data synthesized. In the best case, the proposed approach achieved a 34% accuracy improvement over the raw data and an 11% improvement over the best-performing baseline. This proposed approach shows considerable promise in facilitating the effective adoption of advanced machine learning algorithms in industrial applications by significantly reducing the data collection requirements. [Display omitted] • The requirements of small-scale data for BEV data-driven models are discussed. • An approach that synthesizes physically rational BEV trip samples is proposed. • Experiments show that conventional data augmentation may produce invalid samples. • The proposed approach effectively improves the performance of data-driven models. [ABSTRACT FROM AUTHOR]

Energy conservation is a fundamental requirement in achieving sustainable development, enhancing environmental protection and improving economic and social wellbeing by reducing energy consumption. Carbon emissions are a significant manifestation of energy consumption. Real-time monitoring and forecasting of air quality plays a crucial role in effectively controlling carbon emissions and improving energy conservation measures. Complex Event Processing (CEP) is a technical framework for streaming data processing based on predefined rules and is widely used for real-time detection. Yet as the diversity of data increases, it becomes exceptionally difficult for experts to formulate the CEP rules. Rule-based algorithms can address this issue to a certain extent. However, the CEP rules extracted through these methods are obtained from static data and are often simple and independent of each other. Here, we regard such rules as direct rules, which ignore possible connections between events that usually imply deeper rules. In contrast, we refer to rules that reflect interconnections between events and reveal higher-level principles as indirect rules. In this paper, we propose a methodology fusing interpretable machine learning and decision mining to extract direct and indirect multi-level CEP rules from air pollution datasets for real-time detection, prediction, and early warning. First, we utilize the decision tree algorithm to extract direct rules to detect air quality in real-time. Next, we utilize the process mining algorithm to model the changes in air states and extract the Petri net model from it that reflects the changes in air quality. Then, we utilize our proposed decision mining algorithm to extract indirect CEP rules. Indirect rules can help to understand how pollutants in the air change over time. In addition, they can reveal underlying trends and patterns of air quality change for prediction and early warning. Finally, we validate the accuracy and effectiveness of our approach using a real air pollution dataset. This work contributes to the field of air pollution detection and can help promote the optimization of energy conservation, while it can also support the development of regulatory strategies. • A multi-level CEP rules extraction method is proposed for air quality detection. • A novel method combining machine learning and process mining to extract CEP rules. • A novel way to analyze decision points in Petri nets to monitor air states changes. [ABSTRACT FROM AUTHOR]

Accurately predicting effective thermal conductivity is of great importance for the design and performance evaluation of emerging composites. In this paper, an efficient and implementation-friendly lattice Boltzmann (LB) scheme for predicting the effective thermal conductivity of 3D complex structures is proposed. The key innovation is that the optimum convergence parameter of the 3D thermal LB method is found, which enables the LB equation to converge to steady heat conduction equation with the fastest speed and without losing any accuracy. To deal with the thermal contact resistance between different components, an interface treatment scheme is derived. In comparison with the existing schemes, the present scheme enjoys several hundred times higher computational efficiency. By virtue of this LB scheme, the effective thermal conductivity of the reinforced composites with different dimensional fillers are systematically calculated, and a comprehensive machine learning model is developed. This work provides a powerful numerical tool for high-throughput simulations of the 3D representative volume elements with high thermal conductivity ratios and large grid numbers. It may facilitate the application of data-driven techniques in study of the thermal transport properties of emerging composite materials and structures. • A high-efficiency LB scheme for 3D thermal transfer simulations is proposed. • A treatment scheme of interface thermal resistance to TRT model is derived. • A random method is established to reconstruct RVE of composites with hybrid fillers. • A k eff machine learning model for composites with 0D, 1D and 2D fillers is developed. [ABSTRACT FROM AUTHOR]

This paper presents a recurrent neural network model to make medium-to-long term predictions, i.e. time horizon of ⩾ 1 week, of electricity consumption profiles in commercial and residential buildings at one-hour resolution. Residential and commercial buildings are responsible for a significant fraction of the overall energy consumption in the U.S. With advances in sensors and smart technologies, there is a need for medium to long-term prediction of electricity consumption in residential and commercial buildings at hourly intervals to support decision making pertaining to operations, demand response strategies, and installation of distributed generation systems. The modeler may have limited access to information about building’s schedules and equipment, making data-driven machine learning models attractive. The energy consumption data that is available may also contain blocks of missing data, making time-series predictions difficult. Thus, the main objectives of this paper are: (a) Develop and optimize novel deep recurrent neural network (RNN) models aimed at medium to long term electric load prediction at one-hour resolution; (b) Analyze the relative performance of the model for different types of electricity consumption patterns; and (c) Use the deep NN to perform imputation on an electricity consumption dataset containing segments of missing values. The proposed models were used to predict hourly electricity consumption for the Public Safety Building in Salt Lake City, Utah, and for aggregated hourly electricity consumption in residential buildings in Austin, Texas. For predicting the commercial building’s load profiles, the proposed RNN sequence-to-sequence models generally correspond to lower relative error when compared with the conventional multi-layered perceptron neural network. For predicting aggregate electricity consumption in residential buildings, the proposed model generally does not provide gains in accuracy compared to the multi-layered perceptron model. [ABSTRACT FROM AUTHOR]

Time-series modeling of PV output for solar panels can help solar panel owners understand the power systems’ time-varying behavior and be prepared for the load demand. The time-series forecast/prediction can become challenging due to many missing observations or a lack of historical records that are not sufficient to establish statistical models. Increasing PV measurement frequency over a longer period increases the cost in the detection of the PV fluctuation. This paper proposes an efficient approach to iterative multi-task learning for time series (MTL-GP-TS) that improves prediction of the PV output without increasing measurement efforts by sharing the information among PV data from multiple similar solar panels. The proposed iterative MTL-GP-TS model learns/imputes unobserved or missing values in a dataset of time series associated with the solar panel of interest to predict the PV trend. Additionally, the method improves and generalizes the traditional multi-task learning for Gaussian Process to the learning of both global trend and local irregular components in time series. A real-world case study demonstrated that the proposed method could result in substantial improvement of predictions over conventional approaches. The paper also discusses the selection of parameters and data sources when implementing the proposed algorithm. [ABSTRACT FROM AUTHOR]

Heating, ventilation, and air-conditioning (HVAC) are among the major energy demand in the buildings sector globally. Improving the energy efficiency of such systems is a critical objective for mitigating greenhouse gas emissions and transitioning towards renewable sources of energy supply. The interest of this paper is to explore means to increase the efficiency of HVAC systems in accommodating occupants’ behavior in real time. For instance, rooms in office buildings are not always occupied by occupants during scheduled HVAC service periods. This offers an opportunity to reduce unnecessary energy demands of HVAC systems associated with occupants’ behavior. An in-depth analysis of occupants’ stochastic behavior within an office building is conducted in this paper. A demand-driven control strategy is proposed that automatically responds to occupants’ energy-related behavior for reducing energy consumption and maintains room temperature for occupants with similar performances as a static cooling. In this control strategy, two types of machine learning methods – unsupervised and supervised learning – are applied to learn occupants’ behavior in two learning processes. The occupancy-related information learned by the algorithms is used by a set of specified rules to infer real-time room setpoints for controlling the office's space cooling system. This learning-based approach intends to reduce the need for human intervention in the cooling system’s control. The proposed strategy was applied to control the cooling system of the office building under real-world conditions. Eleven case study office spaces were selected, representing three typical office uses: single person offices, multi-person offices, and meeting rooms. The experimental results report between 7% and 52% energy savings as compared to the conventionally-scheduled cooling systems. [ABSTRACT FROM AUTHOR]

This paper presents data-driven models for photovoltaic module temperature prediction and analyzes the relation and effects of ambient conditions to module temperature. A total of 12 different machine learning and regression algorithms are implemented, with a large experimental dataset of 345,600 × 7. Prior to implementing those algorithms, data preprocessing is performed to prepare the datasets and determine the informative attributes for the models. Using PCA with module temperature as target to predict, the selected features for models' inputs were determined to be ambient temperature, solar radiation, wind speed, and relative humidity, and each algorithm is cross-validated and tuned with optimal performance parameters. Results show that while relative humidity is more likely to introduce less information to the model, other aforementioned features are the important parameters to predict the module temperature. While for linear modeling, LASSO algorithm provided the best performance, the ANN model demonstrated the best overall results as it produced the most accurate predictions with lowest errors. A similar performance is attained by the proposed non-linear model, KRR and Gradient Boosting algorithm, with a slight advantage to the KRR model. Furthermore, in comparison to experimental data, the ANN model and the proposed non-linear model provided an R2 values of 0.986 and 0.981, with a MAE of 0.982 and 1.476, and MSE of 2.181and 3.464, respectively. Moreover, the proposed model supplied accurate results when compared to models from literature in an out-of-sample testing, it also proven to be robust and accurate when used to predict the PV power output. • Prediction of photovoltaic module temperature considering ambient weather conditions. • Predictive models have been developed using twelve different machine learning and regression algorithms. • Ambient temperature and solar radiation are key parameters and important variables to predict the PV module temperature. • The performance of developed models was evaluated and compared with other models in the literature. [ABSTRACT FROM AUTHOR]

Load shapes derived from smart meter data are frequently employed to analyze daily energy consumption patterns, particularly in the context of applications like Demand Response (DR). Nevertheless, one of the most important challenges to this endeavor lies in identifying the most suitable consumer clusters with similar consumption behaviors. In this paper, we present a novel machine learning based framework in order to achieve optimal load profiling through a real case study, utilizing data from almost 5000 households in London. Four widely used clustering algorithms are applied specifically K-means, K-medoids, Hierarchical Agglomerative Clustering and Density-based Spatial Clustering. An empirical analysis as well as multiple evaluation metrics are leveraged to assess those algorithms. Following that, we redefine the problem as a probabilistic classification one, with the classifier emulating the behavior of a clustering algorithm, leveraging Explainable AI (xAI) to enhance the interpretability of our solution. According to the clustering algorithm analysis the optimal number of clusters for this case is seven. Despite that, our methodology shows that two of the clusters, almost 10% of the dataset, exhibit significant internal dissimilarity. As a result, these clusters have been excluded from consideration for DR programs. The scalability and versatility of our solution makes it an ideal choice for power utility companies aiming to segment their users for creating more targeted DR programs. • Assessment of four ML clustering algorithms of different nature. • Re-definition of the problem using probabilistic classification and xAI. • Optimization of demand-side management clusters combining ML and empirical knowledge. • Experimental application on historical data from 4438 households in London. [ABSTRACT FROM AUTHOR]

As the significance of participation in the Day-Ahead Market (DAM) for stakeholders managing the charging of Electric Vehicle (EV) fleets increases, the necessity for precise EV Load Curve (EVLC) forecasting emerges as crucial. This paper presents an extensive investigation of nine diverse EVLC forecasting methodologies, encompassing statistical, machine learning, and deep learning techniques. These methodologies are evaluated on four public, real-world EV datasets, keeping in line with the specific forecasting horizon required for DAM. The study incorporates models with and without online historical data, ensuring broad applicability across varied data availability scenarios. An exploration of seasonal variations in forecasting performance is conducted via one year-long rolling simulations, providing deep insight on seasonal patterns. Furthermore, a detailed methodology for constructing EVLCs from session-based, tabular EV datasets is presented. The conducted research establishes a first-of-its-kind comprehensive comparison of EVLC forecasting methodologies for the real-world challenge of DAM participation. The research results provide indispensable guidance to wholesale electricity market participants, namely suppliers and electric vehicle aggregators, advancing the understanding in the field and significantly contributing to optimization of DAM participation for EV fleets. • Evaluation of forecasting in specific horizon required for DAM participation. • Nine diverse EV Load Curves forecasting methods. • Forecasts are evaluated on four public, real-world EV datasets. • Evaluation for a year-long period and diverse characteristics of EV charging data. • Methodology for constructing EV Load Curve from session-based tabular EV dataset. [ABSTRACT FROM AUTHOR]

Energy demand from the built environment is among the most important contributors to greenhouse gas emissions. One promising way to curtail these emissions is through innovative energy management systems (EMS's). These systems often rely on access to real-world demand data, which remains elusive in practice. Even when available, energy demand data typically suffers from missing data as well as many irregularities and anomalies. This precludes the application of many off-the-shelf machine learning algorithms for time series analysis and modelling, necessary for downstream energy management. Transforming energy demand time series to low dimensional feature matrices has been shown to work well in determining similar buildings and predicting meta-data, both of which can be used to create better forecast algorithms used as input in EMS's. These studies are, however, often marred by the limited size of datasets, as well as the non-interpretable nature of extracted features. This paper addresses these concerns and makes several important contributions: (1) it collates several open-source datasets to create a large meta-analysis dataset containing energy demand data for over 13,000 buildings; (2) it investigates the use of different interpretable feature extraction methods on this collated dataset; and (3) it shows that this feature matrix can be used more generally to determine similar buildings and predict building properties such as missing meta-data. The large feature matrix resulting from the work is open-sourced as part of a web-based dashboard to enable the community to reproduce and further develop our results. • A large building energy demand dataset containing over 13,000 buildings is created. • Different interpretable time series feature extraction methods are investigated. • Features show considerable discriminative power in detecting similar buildings. • Extracted features can be used to predict missing meta-data in building datasets. [ABSTRACT FROM AUTHOR]

Accurately capturing data on the external loads that large structural systems endure is crucial for improving the performance of energy equipment. This paper introduces a novel hybrid model-data-driven framework for the dynamic load identification of interval structures, which seamlessly combines finite-element modeling with machine learning techniques. To address potential ill-posed issues in model-driven methods and the interpretability limitations of data-driven methods, we propose a physics-informed neural network. This neural network effectively inverts uncertain modal responses with low data requirements and high predictive performance high by integrating the underlying modal transformation equation into the loss function of a fully connected neural network. To identify the modal loads using predicted modal displacement/acceleration responses, we introduce a pioneering dynamics inversion method. This method modifies the traditional Kalman filter with an assumption of unknown inputs to reduce the sensitivity of load identification process to different noises. In addition, our approach incorporates a subinterval Chebyshev expansion method to adaptively determine the interval boundaries of external loads. The efficiency of the proposed method is assessed through two numerical examples and validated through comparative research against baseline methods. Our findings suggest that this approach enhances precision, robustness, and generalization in dynamic load identification, even when facing challenges such as limited training data, significant noise interference, and non-zero initial conditions. • A novel hybrid model-data-driven dynamic load identification method for interval structures is proposed. • A physics-informed neural network is established to inverse the uncertain modal responses effectively. • Two improved Kalman filter algorithms are developed to identify the modal loads at specified parameters. [ABSTRACT FROM AUTHOR]

This paper investigates the scheduling problem in smart distribution networks equipped with distributed photovoltaic energy storage systems (PV-ESS) to address excessive power losses, economic revenue, and over-voltage issues. Accurately modeling the grid structure and ensuring adequate sensor coverage pose significant challenges in network settings of this nature, and therefore, we put forth a novel approach known as Principal Component Analysis-based incomplete data equivalence (PIDE) for constructing a data-driven power flow model under incomplete data. Moreover, the presence of distributed PV-ESS, coupled with the lack of data sharing, introduces a hybrid cooperation-competition dynamic, resulting in suboptimal solutions and local optima. To address this challenge, we approach the scheduling problem by formulating it as a multi-agent reinforcement learning task. Meanwhile, we present Counterfactual Multi-agent Soft Actor–Critic (COSAC), which incorporates stochastic policy learning to enhance exploration and facilitates credit assignment in the continuous action space, so as to accurately determine the individual contributions of agents involved in the task. Simulation results conducted on the IEEE 33 and 123 bus systems demonstrate the effectiveness of the proposed method. Specifically, we find that PIDE achieves a substantial reduction in the necessary data sampling coverage, and COSAC outperforms state-of-the-art multi-agent reinforcement learning methods by at least 4.14%. [Display omitted] • A non-convex multi-objective distributed PV-ESS scheduling problem is addressed. • Data-driven power flow model reduces training time and improves accuracy. • A novel machine learning method is proposed to reduce sampling coverage. • MARL realizes exploration-enhanced credit assignment in continuous action space. • Decentralized control policy is executed without communication among PV-ESSs. [ABSTRACT FROM AUTHOR]

Accurate cluster photovoltaic power prediction (CPPP) is crucial for the operation and control of renewable energy grid-connected power systems. The traditional modeling strategies for CPPP such as direct aggregation (DA) and statistical upscaling (SU) have limitations such as error accumulation and upscaling factor uncertainty. To address these issues, this paper proposed a novel hybrid approach for CPPP by combining machine learning models with an improved SU technique. Firstly, a robust broad learning system (BLS) model, in which the Generalized Maximum Correntropy Criterion (GMCC) is used to replace the original mean square error (MSE) loss in BLS, is proposed to solve the problem of multiple outliers affecting the prediction accuracy of regional cluster stations, and it is called GBLS. Then, the Relevance Vector Machine (RVM) as an effective nonlinear regression model is further utilized to compensate for the prediction errors obtained by the GBLS to form the hybrid prediction model, namely GBLS-RVM. Moreover, to mitigate the uncertainty associated with scaling factors in traditional SU strategy, a new SU strategy is developed to refine the relationship between the reference station and the cluster sub-region, enabling direct modeling for regional power prediction. Finally, data from two PV clusters in different regions of China are used to validate the effectiveness of the proposed model, and the results show that under the improved SU strategy, the GBLS-RVM model, reduced RMSE by approximately 33.7% compared to the traditional BLS model, and the RMSE decreased by 12.89% and 30.2% when compared to traditional DA and traditional SU strategies. • A novel hybrid method for CPPP is developed by combining ML and improved SU technique. • A BLS with GMCC (GMCC-BLS) as a prediction model is developed to handle the outliers in measurement data. • An innovative error correction strategy based on RVM is proposed to improve prediction reliability. • A new SU technique is used to establish direct relationships between reference stations and subregions. [ABSTRACT FROM AUTHOR]

The use of machine learning in building technology has become increasingly important in recent years. One of the applications is heating load prediction, which enables demand-side flexibility. Most studies consider the heating load prediction without sufficient context with the existing building characteristics. For an accurate load prediction, suitable features have to be selected according to their importance, the feature importance (FI). The scope of this paper is to investigate whether there is a relationship between the building characteristics and the FI and if so, how strong this relationship is. Additionally, an analysis has been conducted to determine which building characteristic have the most significant impact on FI. For this purpose, a full factorial design of a room with six different building characteristics is carried out. In total, the heating load is calculated for 15 552 room variants. The thermal balance, correlation, random forest FI, permutation FI and SHapley Additive exPlanations (SHAP) values are calculated for these different rooms. The local SHAP values were used to explain the model. These values also provide insight into the interaction of individual features with the heating load. For most variants, the outdoor temperature had the highest FI. It is investigated which building characteristics have the greatest influence on the thermal balance, correlation, FI and SHAP values. A relationship was found between the proportion of thermal balance, the correlation between the features and the label as well as the FI. The greatest association with the thermal balance characteristics was found for the SHAP values. The study shows the systematic relationship between building characteristics and FI. Therefore, FI should always be considered in the context of building characteristics. [Display omitted] • SHAP values provide an excellent way of showing how features affect heating load. • Comparison of various feature selection methods for 15 552 different rooms. • Characteristics of the building are reflected in the feature importance. • Air exchange rate has the greatest influence on the feature importance. [ABSTRACT FROM AUTHOR]

Smart home energy management systems help the distribution grid operate more efficiently and reliably, and enable effective penetration of distributed renewable energy sources. These systems rely on robust forecasting, optimization, and control/scheduling algorithms that can handle the uncertain nature of demand and renewable generation. This paper proposes an advanced ML algorithm, called Recurrent Trend Predictive Neural Network based Forecast Embedded Scheduling (rTPNN-FES), to provide efficient residential demand control. rTPNN-FES is a novel neural network architecture that simultaneously forecasts renewable energy generation and schedules household appliances. By its embedded structure, rTPNN-FES eliminates the utilization of separate algorithms for forecasting and scheduling and generates a schedule that is robust against forecasting errors. This paper also evaluates the performance of the proposed algorithm for an IoT-enabled smart home. The evaluation results reveal that rTPNN-FES provides near-optimal scheduling 37.5 times faster than the optimization while outperforming state-of-the-art forecasting techniques. • A novel rTPNN-based Forecast Embedded Scheduling (rTPNN-FES) is presented. • rTPNN-FES simultaneously forecasts and schedules household loads. • Significant improvement is achieved in forecasting compared to the state-of-the-art. • Near optimal scheduling is achieved with much lower computation time. [ABSTRACT FROM AUTHOR]

With the wide adoption of building automation system, and the advancement of data, sensing, and machine learning techniques, data-driven fault detection and diagnostics (FDD) for building heating, ventilation, and air conditioning systems has gained increasing attention. In this paper, data-driven FDD is defined as those that are built or trained from data via machine learning or multivariate statistical analysis methods. Following this definition, this paper reviews and summarizes the literature on data-driven FDD from three aspects: process, systems studied (including the systems being investigated, the faults being identified, and the associated data sources), and evaluation metrics. A data-driven FDD process is further divided into the following steps: data collection, data cleansing, data preprocessing, baseline establishment, fault detection, fault diagnostics, and potential fault prognostics. Literature reported data-driven methods used in each step of an FDD process are firstly discussed. Applications of data-driven FDD in various HVAC systems/components and commonly used data source for FDD development are reviewed secondly, followed by a summary of typical metrics for evaluating FDD methods. Finally, this literature review concludes that despite the promising performance reported in the literature, data-driven FDD methods still face many challenges, such as real-building deployment, performance evaluation and benchmarking, scalability and transferability, interpretability, cyber security and data privacy, user experience, etc. Addressing these challenges is critical for a broad real-building adoption of data-driven FDD. • This paper reviews data-driven fault detection and diagnostics studies for HVAC systems. • Papers are collected by Sub-keyword Synonym Subtopics Searching method. • Papers are reviewed regarding FDD process, systems studied, and evaluation metrics. • Ongoing efforts, gaps, and challenges are identified. [ABSTRACT FROM AUTHOR]

This paper presents a laboratory study of Offline-trained Reinforcement Learning (RL) control of a Heating Ventilation and Air-Conditioning (HVAC) system. We conducted the experiments on a radiant floor heating system consisting of two temperature zones located in Denmark. The buildings are subjected to real-world weather. A previous paper describes the algorithm we tested, which we summarize in this paper. First, we present a benchmarking test which we conducted during spring 2021 and winter 2021/2022. This data is used in the Offline RL framework to train and deploy the RL policy, which we then tested during winter 2021/2022 and spring 2022. An analysis of the data shows that the RL policy showed predictive control-like behavior, and reduced the oscillations of the system by a minimum of 40%. Additionally, we show that the RL policy is minimum 14% more cost-effective than the traditional control policy used in the benchmarking test. • Real-world data from laboratory test with underfloor heating. • Offline Multi-Agent Reinforcement Learning can eliminate poor behavior during training while converging. • Long Short Term Memory layers are an effective method for obtaining training models for Reinforcement Learning. • Heating costs are reduced by approximately 15% and comfort is maintained. [ABSTRACT FROM AUTHOR]

• Cloud-based technologies adoption in real-time microgrid management. • Boosting the microgrid data forecasting by machine-learning techniques. • Impact of the internet of things usage in data acquisition for a real-life energy management system. Under a hierarchical structure perspective, the integration of tools like the internet of things, cloud computing, and machine learning into a microgrid real-time energy management system involves superior capabilities like an autonomous and scalable design, massive storage capabilities, real-time information analysis and processing, and security issues control, to mention a few. This paper evaluates most of them using a cloud-based real-time energy management system integrated into a real-life hardware-in-the-loop testbed. The proposed cloud-based system is tested by solving an economic power dispatch problem including the equalization of the multiple battery-based energy storage systems interacting within a multi-microgrid environment. The test assessment combined reviewing and running microgrid models, incorporating these models into a real-life power-hardware-in-the-loop unit, linking the testbed to a cloud server, and merging the energy management system with on-demand computing tools, primarily machine learning and the internet of things. As established by the experimental evidence, this paper cites the benefits of combining machine learning techniques and internet of things tools with a scalable and autonomous real-life cloud-based energy management system architecture to improve the framework's functionality, enhance the energy forecasting for generation and usage, and cut down the price paid to the service provider. [ABSTRACT FROM AUTHOR]

• A BIM-supported energy performance analysis for green building design is proposed. • This systematic analytical framework integrates explainable machine learning and multi-objective optimization. • It can serve as a data-driven decision tool to guide green building development even under uncertainty. • The top important features that greatly affect the building energy performance can be quantitatively identified. • Building energy performance can be improved by 13.43% using the obtained optimal strategy. Supported by the combination of the advanced BIM technique with intelligent algorithms, this paper develops a systematic framework using explainable machine learning and multi-objective optimization to realize the automatic prediction and optimization of building energy performance towards the sustainable development goal. There are three critical parts incorporated, including the DesignBuilder simulation, BO-LGBM (Bayesian optimization-LightGBM) and an explainable method SHAP (SHapley Additive explanation)-based prediction and explanation of building energy performance, and AGE-MOEA algorithm-based multi-objective optimization (MOO) under sources of uncertainty. It has been verified in a case study for green building design. Results show that: (1) The predictive BO-LGBM model can make a highly precise prediction in agreement with the simulation data, reaching up the R2 larger than 93.4% and MAPE smaller than 2.13%. From the SHAP calculation, features related to the HAVC (Heating Ventilation and Air Conditioning) system tend to contribute more to affecting the prediction results. (2) The AGE-MOEA-based optimization can identify a set of Pareto optimal solutions in simultaneously minimizing the building energy consumption, CO 2 emission, and indoor thermal discomfort degree, arriving at the highest optimization rate of 13.43% under proper adjustment of certain features. (3) In the MOO task, the consideration of model and data uncertainty by prediction intervals and Monte-Carlo simulation can further increase the optimization rate by around 4.0% than the deterministic scenario, resulting in more desired strategies in optimizing the green building performance. In short, this paper enriches the area of green building development. For one thing, it raises the transparency and interpretability of machine learning to make the prediction more convincing. For another, it efficiently determines the passive and active design solutions along with the detailed profile of influential factors for green building design. [ABSTRACT FROM AUTHOR]

• Algorithm can be sensitive to algorithm setting, affecting optimality and robustness. • The discrepancy between the digitized and initial objectives cannot be ignored. • The system dynamics that can affect potential performance improvement is revealed. • Algorithms can obtain energy savings and violation reductions in some scenarios. • Actor-critic, off-policy, and model-based algorithms exhibit better performance. Deep Reinforcement Learning has been increasingly researched for Dynamic Thermal Management in Data Centers. However, existing works typically evaluate the performance of algorithms on a specific task, utilizing models or data trajectories without discussing in detail their implementation feasibility and their ability to deal with diverse work scenarios. The lack of these works limits the real-world deployment of Deep Reinforcement Learning. To this end, this paper comprehensively evaluates the strengths and limitations of state-of-the-art algorithms by conducting analytical and numerical studies. The analysis is conducted in four dimensions: algorithms, tasks, system dynamics, and knowledge transfer. As an inherent property, the sensitivity to algorithms settings is first evaluated in a simulated data center model. The ability to deal with various tasks and the sensitivity to reward functions are subsequently studied. The trade-off between constraints and power savings is identified by conducting ablation experiments. Next, the performance under different work scenarios is investigated, including various equipment, workload schedules, locations, and power densities. Finally, the transferability of algorithms across tasks and scenarios is also evaluated. The results show that actor-critic, off-policy, and model-based algorithms outperform others in optimality, robustness, and transferability. They can reduce violations and achieve around 8.84% power savings in some scenarios compared to the default controller. However, deploying these algorithms in real-world systems is challenging since they are sensitive to specific hyperparameters, reward functions, and work scenarios. Constraint violations and sample efficiency are some aspects that are still unsatisfactory. This paper presents our well-structured investigations, new findings, and challenges when deploying deep reinforcement learning in Data Centers. [ABSTRACT FROM AUTHOR]

• Applications of Reinforcement Learning (RL) in energy market operation are reviewed. • Fundamentals of RL including basic concepts and algorithms are summarized. • Critical discussions on applicability and obstacles of RL deployment are provided. • Advanced RL techniques developed recently are recommended as future perspectives. The increasing penetration of renewable generations, along with the deregulation and marketization of power industry, promotes the transformation of energy market operation paradigms. The optimal bidding strategy and dispatching methodologies under these new paradigms are prioritized concerns for both market participants and power system operators. In contrast with conventional solution methodologies, the Reinforcement Learning (RL), as an emerging machine learning technique that exhibits a more favorable computational performance, is playing an increasingly significant role in both academia and industry. This paper presents a comprehensive review of RL applications in deregulated energy market operation including bidding and dispatching strategy optimization, based on more than 150 carefully selected papers. For each application, apart from a paradigmatic summary of generalized methodology, in-depth discussions of applicability and obstacles while deploying RL techniques are also provided. Finally, some RL techniques that have great potentiality to be deployed in bidding and dispatching problems are recommended and discussed. [ABSTRACT FROM AUTHOR]

• A co-simulation platform for Hybrid Electric Vehicles is developed for virtual assessment of Energy Management, performance, CO 2 & pollutant emissions. • CO 2 & pollutant emissions models are based on convolutional neural network for accurate transient state prediction. • A continuous deep reinforcement learning based EMS of HEV is proposed. • DDQL is trained with random generated driving cycles based on Markov Chain approach from real recordings. • Cumulative d error for predicted pollutants emissions is below 8.5% while HEV simulation is running up to 10 times faster than real time in a standard laptop. The Energy Management Strategy (EMS) in an HEV is the key for improving fuel economy and simultaneously reducing pollutant emissions. This paper presents a methodology for developing hybrid models that enable EMS testing as well as the evaluation of fuel consumption, CO 2 and pollutant emissions (CO, NO x and THC). In this context, pollutant emissions are hard to quantify with static models such as the well-known map-based approach which is mainly due to the pronounced impact of transient effects. The novelty of this paper primarily comes from the characterization of pollutant emissions through Convolutional Neural Networks (CNN), providing high accuracy for both, instantaneous and cumulative values. The input parameters are classical Internal Combustion Engine (ICE) measurements such as engine speed, air mass flow, torque and exhaust temperature. The proposed CNNs are reduced to a minimum for low complexity and fast computability. These models are developed with experimental data from chassis dyno testing of a conventional turbo-charged gasoline engine vehicle. The pollutant emission models are used in conjunction with physical models of the remaining powertrain allowing for real time simulations of the complete HEV vehicle. The Double Deep-Q learning algorithm is proposed for the EMS and compared to the Dynamic programming (DP) solution. The introduced methodology is developed in a co-simulation framework between MATLAB-Simulink and AMESIM. The resulting model runs between 8 and 10 times faster than real time in an off-the-shelf PC. This provides the capability for developing models suitable for HIL (hardware-in-the-loop) and SIL (software-in-the-loop) applications. The final error in predicted CO 2 remains below 2.5% while the final cumulative error for pollutants is below 8.5% in the case of CO and HC emissions. [ABSTRACT FROM AUTHOR]

This research paper presents a data-driven framework for design optimization of renewable energy communities (RECs) in the residential sector, considering both techno-economic challenges and environmental impact. The study's focus is to determine suitable sizes for photovoltaic systems, wind turbines, and battery electrical energy systems by evaluating energy, economic, and environmental criteria. To achieve this, we develop a data-driven model that incorporates Homer Pro and an in-house tool developed in Python programming language that integrates a machine learning algorithm, life cycle cost (LCC), life cycle assessment (LCA) calculations of the REC model. Furthermore, a multi-objective optimization model is established to minimize the LCC and LCA parameters while maximizing green energy use. Moreover, a multi-criteria decision-making approach based on Weighted Sum Model (WSM) is proposed to help the stakeholders to see beyond the selection criteria based on LCC and LCA to choose the most appropriate scenario optimal solution for the desired energy community and interpret the effect of various economic parameters on the sustainable performance of REC. The framework application is illustrated through a case study for the optimal design of REC for a residential community in Tarragona, Spain, consisting of 100 buildings. The results revealed a substantial improvement in economic and environmental benefits for designing REC, the optimal minimum cost solution with a levelized cost of energy (LCOE = 0.044 $/kWh) and a payback period of 7.1 years with an LCOE reduction of 85.04% compared to the base case. The minimum impact with an LCOE = 0.220 $/kWh and a payback period of 12.5 years with a reduction in environmental impact of 54.59% compared to the base case. Overall, the developed data-driven provides policy decision-making with an evaluation of REC in the residential sector. [Display omitted] • A data-driven framework for optimizing the residential renewable energy community is framed. • The framework is applied to a real case study residential community in Tarragona, Spain. • Multi-objective optimization is applied to minimize LCC and LCA parameters while maximizing green energy use. • A multi-criteria decision-making approach is carried out to design optimized configurations for energy communities. [ABSTRACT FROM AUTHOR]

Vehicle modularization has become an emerging trend in the automotive industry, leading to research on modular configuration, composition, and related control strategies. In this paper, we propose a modular power system with a hydraulic free piston engine (HFPE) as the power unit and develop a power distribution control strategy to enhance the overall efficiency of the system. Firstly, we determine the configuration scheme of the modular power system and establish a simulation model of the HFPE using MATLAB/Simulink. We conduct principle verification of the simulation model. Secondly, based on the simulation model of HFPE, we research the power unit control strategy using the machine learning regression prediction algorithm, enabling dynamic working condition switching of the power unit. Next, we propose a power distribution optimization algorithm which is named as the Rule Based Double Iterative Optimization Algorithm (RBDI) and compare it with several mature optimization algorithms under the framework of model predictive control, considering related constraints. Finally, we validate the performance of the proposed power distribution control strategy using a hardware-in-loop system. The results demonstrate that the output power of the modular power system can be effectively ensured. Compared with the average distribution algorithm (AVE), the genetic algorithm (GA), and the ameliorated particle swarm optimization algorithm (APSO), the overall working efficiency of the modular power system using the proposed control strategy is increased by 6.57%, 6.13%, and 5.59%, respectively, under the three test driving cycles. • The HFPE simulation model and a modular power architecture (MPA) composed of HFPE are proposed. • A dynamic switching strategy for HFPE working conditions based on machine learning regression prediction is proposed. • An optimal power distribution control algorithm based on model predictive control for MPA is proposed. • Results show that the control effect of the RBDI in improving work efficiency is better than AVE, GA and APSO. • The control effect has been verified in a hardware-in-loop (HIL) system. [ABSTRACT FROM AUTHOR]