Search

Chromatographic separation processes are described by nonlinear partial differential and algebraic equations, which may result in high computational cost, hindering further online applications. To decrease the computational burden, different data‐driven modeling approaches can be implemented. In this work, we investigate different strategies of data‐driven modeling for chromatographic processes, using artificial neural networks to predict pseudo‐dynamic elution profiles, without the use of explicit temporal information. We assess the performance of the surrogates trained on different dataset sizes, achieving good predictions with a minimum of 3400 data points. Different activation functions are used and evaluated against the original high‐fidelity model, using accuracy, interpolation, and simulation time as performance metrics. Based on these metrics, the best performing data‐driven models are implemented in a process optimization framework. The results indicate that data‐driven models can capture the nonlinear profile of the process and that can be considered as reliable surrogates used to aid process development. [ABSTRACT FROM AUTHOR]

With the goal of forecasting streamflow time series with sufficient lead time, we evaluate the efficiency and accuracy of data-based models ranging from relatively simple to complex. Based on this, we systematically explain the model construction and selection process according to lead time, type and amount of data, and optimization method. This analysis involved optimizing the inputs and hyperparameters of four unique data-driven models: Autoregressive Integrated Moving Average (ARIMA), Artificial Neural Network (ANN), Long Short-Term Memory (LSTM), and Transformer (TRANS), which were applied to the Soyang watershed, South Korea. The type and amount of model inputs are determined through a fine-tuning process that samples based on a correlation threshold, correlation to predictand, and autocorrelation to historical data and evaluates the simulated objective function. Hyperparameters are simultaneously optimized using three conventional optimization methods: Bayesian optimization (BO), particle swarm optimization (PSO), and gray wolf optimization (GWO). The experimental results provide insight into the role of input predictors, data preparations (e.g., wavelet transform), hyperparameter optimization, and model structures. From this, we can provide guidelines for model selection. Relatively simple models can be used when the dataset is small or there are few input variables, when only the near future is predicted, or when the selection of optimization methods is limited. However, a more complex model should be selected if the type and amount of data are sufficient, various optimization methods can be applied, or it is necessary to secure more lead time. More parameters, more complex model structures, and more training materials make this possible. [ABSTRACT FROM AUTHOR]

Landslides due to rainfall are among most destructive natural disasters that cause property damages, huge financial losses and human deaths in different parts of the World. To plan for mitigation and resilience, the prediction of the volume of rainfall-induced landslides is essential to understand the relationship between the volume of soil materials debris and their associated predictors. Objectives of this research are to construct a model by utilizing advanced data-driven algorithms (i.e., ordinary least square or Linear regression (OLS), random forest (RF), support vector machine (SVM), extreme gradient boosting (EGB), generalized linear model (GLM), decision tree (DT), and deep neural network (DNN), K-nearest neighbor (KNN) and Ridge regression (RR)) for the prediction of the volume of landslides due to rainfall considering geological, geomorphological, and environmental conditions. Models were tested on the Korean landslide dataset to observe the best-performing model, and among tested algorithms, the extreme gradient boosting ranked high with the coefficient of determination (R2 = 0.85) and mean absolute error (MAE = 150.421 m3). The volume of landslides was strongly influenced by slope length, drainage status, slope angle, aspect, and age of trees. The anticipated volume of landslide can be important for land use allocation and efficient landslide risk management. [ABSTRACT FROM AUTHOR]

Deformable linear objects (DLOs), such as rods, cables, and ropes, play important roles in daily life. However, manipulation of DLOs is challenging as large geometrically nonlinear deformations may occur during the manipulation process. This problem is made even more difficult as the different deformation modes (e.g., stretching, bending, and twisting) may result in elastic instabilities during manipulation. In this paper, we formulate a physics-guided data-driven method to solve a challenging manipulation task—accurately deploying a DLO (an elastic rod) onto a rigid substrate along various prescribed patterns. Our framework combines machine learning, scaling analysis, and physical simulations to develop a physics-based neural controller for deployment. We explore the complex interplay between the gravitational and elastic energies of the manipulated DLO and obtain a control method for DLO deployment that is robust against friction and material properties. Out of the numerous geometrical and material properties of the rod and substrate, we show that only three non-dimensional parameters are needed to describe the deployment process with physical analysis. Therefore, the essence of the controlling law for the manipulation task can be constructed with a low-dimensional model, drastically increasing the computation speed. The effectiveness of our optimal control scheme is shown through a comprehensive robotic case study comparing against a heuristic control method for deploying rods for a wide variety of patterns. In addition to this, we also showcase the practicality of our control scheme by having a robot accomplish challenging high-level tasks such as mimicking human handwriting, cable placement, and tying knots. [ABSTRACT FROM AUTHOR]

Higher Himalayan catchments are often poorly monitored for hydrological activities involving flood flow prediction for the safety of riverside communities and the successful operation of hydropower projects. This study aimed to estimate the comparative performance of artificial neural network (ANN) based flow prediction models using 10 years of daily river flow data of Kaligandaki catchment at Kotagaun, Nepal, which is a snow-fed catchment in the Himalayan region. The flow prediction models were trained and tested at a hydrological station using the previous 3 days' river flow data to predict the 1-day ahead flow data. Eight different training functions were employed in an ANN model for comprehensive statistical assessment of accuracy and precision of each training function. The most significant and validated result obtained in this study is the comprehensive comparison of various training functions' performance, and identification of the most efficient training function for the study case. Among the training functions investigated, the Levenberg-Marquardt backpropagation function exhibits the best performance for the model having Nash-Sutcliffe efficiency, root mean square error and mean absolute error values of 0.866, 209.578 and 75.422, respectively. This study provides a fundamental basis for accurate flow prediction of topographically challenged catchments where hydrological monitoring and data collection may be limited. In particular, this model will help to improve early warning system, hydrological planning, and the safety of riverside communities in the Himalayan region. [ABSTRACT FROM AUTHOR]

Batteries have been considered a key element in several applications, ranging from grid-scale storage systems through electric vehicles to daily-use small-scale electronic devices. However, excessive charging and discharging will impair their capabilities and could cause their applications to fail catastrophically. Among several diagnostic indices, state-of-charge estimation is essential for evaluating a battery's capabilities. Various approaches have been introduced to reach this target, including white, gray, and black box or data-driven battery models. The main objective of this work is to provide an extensive comparison of currently highly utilized machine learning-based estimation techniques. The paper thoroughly investigates these models' architectures, computational burdens, advantages, drawbacks, and robustness validation. The evaluation's main criteria were based on measurements recorded under various operating conditions at the Energy Systems Research Laboratory (ESRL) at FIU for the eFlex 52.8 V/5.4 kWh lithium iron phosphate battery pack. The primary outcome of this research is that, while the random forest regression (RFR) model emerges as the most effective tool for SoC estimation in lithium-ion batteries, there is potential to enhance the performance of simpler models through strategic adjustments and optimizations. Additionally, the choice of model ultimately depends on the specific requirements of the task at hand, balancing the need for accuracy with the complexity and computational resources available and how it can be merged with other SoC estimation approaches to achieve high precision. [ABSTRACT FROM AUTHOR]

Eutrophication is a major environmental issue with many negative consequences, such as hypoxia and harmful cyanotoxin production. Monitoring coastal eutrophication is crucial, especially for island countries like the Republic of Cyprus, which are economically dependent on the tourist sector. Additionally, the open-sea aquaculture industry in Cyprus has been exhibiting an increase in recent decades and environmental monitoring to identify possible signs of eutrophication is mandatory according to the legislation. Therefore, in this modeling study, two different types of artificial neural networks (ANNs) are developed based on in situ data collected from stations located in the coastal waters of Cyprus. These ANNs aim to model the eutrophication phenomenon based on two different data-driven modeling procedures. Firstly, the self-organizing map (SOM) ANN examines several water quality parameters' (specifically water temperature, salinity, nitrogen species, ortho-phosphates, dissolved oxygen, and electrical conductivity) interactions with the Chlorophyll-a (Chl-a) parameter. The SOM model enables us to visualize the monitored parameters' relationships and to comprehend complex biological mechanisms related to Chl-a production. A second feed-forward ANN model is also developed for predicting the Chl-a levels. The feed-forward ANN managed to predict the Chl-a levels with great accuracy (MAE = 0.0124; R = 0.97). The sensitivity analysis results revealed that salinity and water temperature are the most influential parameters on Chl-a production. Moreover, the sensitivity analysis results of the feed-forward ANN captured the winter upwelling phenomenon that is observed in Cypriot coastal waters. Regarding the SOM results, the clustering verified the oligotrophic nature of Cypriot coastal waters and the good water quality status (only 1.4% of the data samples were classified as not good). The created ANNs allowed us to comprehend the mechanisms related to eutrophication regarding the coastal waters of Cyprus and can act as useful management tools regarding eutrophication control. [ABSTRACT FROM AUTHOR]

Complex systems, as interwoven miscellaneous interacting entities that emerge and evolve through self-organization in a myriad of spiraling contexts, exhibit subtleties on global scale besides steering the way to understand complexity which has been under evolutionary processes with unfolding cumulative nature wherein order is viewed as the unifying framework. Indicating the striking feature of non-separability in components, a complex system cannot be understood in terms of the individual isolated constituents' properties per se, it can rather be comprehended as a way to multilevel approach systems behavior with systems whose emergent behavior and pattern transcend the characteristics of ubiquitous units composing the system itself. This observation specifies a change of scientific paradigm, presenting that a reductionist perspective does not by any means imply a constructionist view; and in that vein, complex systems science, associated with multiscale problems, is regarded as ascendancy of emergence over reductionism and level of mechanistic insight evolving into complex system. While evolvability being related to the species and humans owing their existence to their ancestors' capability with regards to adapting, emerging and evolving besides the relation between complexity of models, designs, visualization and optimality, a horizon that can take into account the subtleties making their own means of solutions applicable is to be entailed by complexity. Such views attach their germane importance to the future science of complexity which may probably be best regarded as a minimal history congruent with observable variations, namely the most parallelizable or symmetric process which can turn random inputs into regular outputs. Interestingly enough, chaos and nonlinear systems come into this picture as cousins of complexity which with tons of its components are involved in a hectic interaction with one another in a nonlinear fashion amongst the other related systems and fields. Relation, in mathematics, is a way of connecting two or more things, which is to say numbers, sets or other mathematical objects, and it is a relation that describes the way the things are interrelated to facilitate making sense of complex mathematical systems. Accordingly, mathematical modeling and scientific computing are proven principal tools toward the solution of problems arising in complex systems' exploration with sound, stimulating and innovative aspects attributed to data science as a tailored-made discipline to enable making sense out of voluminous (-big) data. Regarding the computation of the complexity of any mathematical model, conducting the analyses over the run time is related to the sort of data determined and employed along with the methods. This enables the possibility of examining the data applied in the study, which is dependent on the capacity of the computer at work. Besides these, varying capacities of the computers have impact on the results; nevertheless, the application of the method on the code step by step must be taken into consideration. In this sense, the definition of complexity evaluated over different data lends a broader applicability range with more realism and convenience since the process is dependent on concrete mathematical foundations. All of these indicate that the methods need to be investigated based on their mathematical foundation together with the methods. In that way, it can become foreseeable what level of complexity will emerge for any data desired to be employed. With relation to fractals, fractal theory and analysis are geared toward assessing the fractal characteristics of data, several methods being at stake to assign fractal dimensions to the datasets, and within that perspective, fractal analysis provides expansion of knowledge regarding the functions and structures of complex systems while acting as a potential means to evaluate the novel areas of research and to capture the roughness of objects, their nonlinearity, randomness, and so on. The idea of fractional-order integration and differentiation as well as the inverse relationship between them lends fractional calculus applications in various fields spanning across science, medicine and engineering, amongst the others. The approach of fractional calculus, within mathematics-informed frameworks employed to enable reliable comprehension into complex processes which encompass an array of temporal and spatial scales notably provides the novel applicable models through fractional-order calculus to optimization methods. Computational science and modeling, notwithstanding, are oriented toward the simulation and investigation of complex systems through the use of computers by making use of domains ranging from mathematics to physics as well as computer science. A computational model consisting of numerous variables that characterize the system under consideration allows the performing of many simulated experiments via computerized means. Furthermore, Artificial Intelligence (AI) techniques whether combined or not with fractal, fractional analysis as well as mathematical models have enabled various applications including the prediction of mechanisms ranging extensively from living organisms to other interactions across incredible spectra besides providing solutions to real-world complex problems both on local and global scale. While enabling model accuracy maximization, AI can also ensure the minimization of functions such as computational burden. Relatedly, level of complexity, often employed in computer science for decision-making and problem-solving processes, aims to evaluate the difficulty of algorithms, and by so doing, it helps to determine the number of required resources and time for task completion. Computational (-algorithmic) complexity, referring to the measure of the amount of computing resources (memory and storage) which a specific algorithm consumes when it is run, essentially signifies the complexity of an algorithm, yielding an approximate sense of the volume of computing resources and seeking to prove the input data with different values and sizes. Computational complexity, with search algorithms and solution landscapes, eventually points toward reductions vis à vis universality to explore varying degrees of problems with different ranges of predictability. Taken together, this line of sophisticated and computer-assisted proof approach can fulfill the requirements of accuracy, interpretability, predictability and reliance on mathematical sciences with the assistance of AI and machine learning being at the plinth of and at the intersection with different domains among many other related points in line with the concurrent technical analyses, computing processes, computational foundations and mathematical modeling. Consequently, as distinctive from the other ones, our special issue series provides a novel direction for stimulating, refreshing and innovative interdisciplinary, multidisciplinary and transdisciplinary understanding and research in model-based, data-driven modes to be able to obtain feasible accurate solutions, designed simulations, optimization processes, among many more. Hence, we address the theoretical reflections on how all these processes are modeled, merging all together the advanced methods, mathematical analyses, computational technologies, quantum means elaborating and exhibiting the implications of applicable approaches in real-world systems and other related domains. [ABSTRACT FROM AUTHOR]

In-cylinder pressure is a key parameter for assessing marine engines health; therefore, its measurement or prediction is paramount for these engines' diagnosis. Thermodynamic models are typically employed for predicting the in-cylinder pressure, which, however, face challenges pertinent to their calibration and computational time requirements. Recent advances in the field of machine learning have leveraged the development of data-driven models. This study aims to compare two approaches for input features and six regression techniques to select the most effective combination for developing data-driven models to predict the in-cylinder pressure of marine four-stroke engines. Two approaches with different input and output features are initially compared. The first employs regression to directly predict the in-cylinder pressure signal, whereas the second predicts the harmonics coefficients by regression and subsequently estimates the in-cylinder pressure by using a Fourier series function. Typical regression techniques, including linear, elastic, and polynomial regression, support vector machines (SVM), decision trees (DT), and artificial neural networks (ANN), are employed to develop data-driven models based on the second approach. The required datasets for training and testing are derived by using a physical digital twin for the investigated marine engine, which is calibrated against the shop trials and acquired shipboard measurements. The accuracy of the data-driven models are estimated based on the root mean square error considering the testing datasets. For the data-driven model based on the second approach and the ANN regression, a sensitivity study is carried out considering the training datasets and the harmonics number to derive recommendations for these parameters' values. The results demonstrate that the second approach provides higher accuracy, whereas the ANN regression is the most effective technique for developing data-driven models to estimate the in-cylinder pressure, as the exhibited root mean square error is retained within ± 0.2 bar for the ANN trained with 20 samples. This study supports the development and use of data-driven models for marine engines health diagnosis. [ABSTRACT FROM AUTHOR]

Estimation of total sediment loads is a significant topic in river management as direct measurement is costly and time-consuming. This study aims not only to use the eXtreme Gradient Boosting (XGBoost) model but also to compare its performance with that of other empirical equations and ML models, including Artificial Neural Networks (ANN), AdaBoost, Gradient Boost Regressor, Random Forest Regressor, and Gaussian Process. 543 data points from the United States Geological Survey were used to train and test different models. The results showed that XGBoost outperformed other methods considering six performance metrics. To be more specific, the root mean square errors and determination coefficient were 216 and 0.95, respectively, whereas the corresponding metrics for ANN were 316.23 and 0.87, respectively. To interpret the sediment predictions and delineate the importance of each feature, XGBoost feature importance and SHapley Additive exPlanations (SHAP) were utilized. According to the feature importance analysis, estimations of the XGBoost model was mostly (72%) affected by the water surface width. Moreover, SHAP analysis verified the importance of water surface width on the final predictions. Finally, based on the results achieved in this study, further applications of XGBoost in water resources management are postulated. [ABSTRACT FROM AUTHOR]

Due to the dynamic and complexity of leachate percolation within municipal solid waste (MSW), planning and operation of solid waste management systems are challenging for decision-makers. In this regard, data-driven methods can be considered robust approaches to modeling this problem. In this paper, three black-box data-driven models, including artificial neural networks (ANN), adaptive neuro-fuzzy inference systems (ANFIS), and support vector regression (SVR), and also three white-box data-driven models, including the M5 model tree (M5MT), classification and regression trees (CART), and group method of data handling (GMDH), were developed for modeling and predicting landfill leachate permeability (k ). Based on a previous study conducted by Ghasemi et al. (2021), k can be formulated as a function of impermeable sheets ( IS ) and copper pipes ( CP ). Hence, in the present study, IS and CP were adopted as input variables for the prediction of k and evaluated for the performance of the suggested black-box and white-box data-driven models. Scatter plots and statistical indices such as coefficient of determination (R2), root mean square error (RMSE), and mean absolute error (MAE) were used for qualitative and quantitative evaluations of the effectiveness of the suggested methods. The outcomes indicated all of the provided models successfully predicted k . However, ANN and GMDH had higher accuracy between the proposed black-box and white-box data-driven models. ANN with R2 = 0.939, RMSE = 0.056, and MAE = 0.017 was marginally better than GMDH with R2 = 0.857, RMSE = 0.064, and MAE = 0.026 in the testing stage. Nevertheless, an explicit mathematical expression provided by GMDH to predict k was easier and more understandable than ANN. [ABSTRACT FROM AUTHOR]

Liquid loading is the most common operational problem influencing gas well productivity for the petroleum operator. Liquid loading is defined as an operational constraint that is associated with gas wells where the major driving mechanism for hydrocarbon production is by the associated gas-driven mechanisms. Liquid loading occurs when liquid accumulated in the tubing or casing results in the gas velocity lower than the critical value (the minimum velocity required for gas to push the liquid out of the gas well), which overtime leads to a hydrostatic back pressure greater than the formation pressure of the well, thereby limiting the flow of gas into the well. The continuous build-up of pressure from liquid loading eventually minimizes well productivity and expensive work over operations. However, current mathematical models to predict liquid loading are flawed with varying inaccuracies and depending on the models deployed will ultimately lead to loss of production time and well productivity. In our work we present prediction of liquid loading using a software-based model incorporating the particle swarm optimization algorithm, genetic algorithm, and artificial neural network and Bayesian neural network algorithms applications. The results of our research findings show that artificial neural network software-based model with a simulated accuracy of 93% and 92% for test and trained data, respectively, outperformed the particle swarm optimization data-driven model with a simulated sensitivity accuracy of 92% and 83%, and genetic algorithm data-driven models with a simulated accuracy of 89% and 83%. The Bayesian neural network was postulated as a robust model because of its simplicity shown to have simulated accuracy of 77% and 73% for train and test data, respectively. Thus software-based code environment and data-driven model developed and presented in this paper may resolve many of current deficiencies and gaps in the current technical literature to predict liquid loading with high precision offering saving in millions of dollars to the operators. [ABSTRACT FROM AUTHOR]

Redundant consumption of energy in buildings is an important issue that causes increasing problems of climate change and global warming in the world. Therefore, it is necessary to develop efficient energy management approaches in buildings. Accurate prediction of energy consumption plays an important role to obtain energy-efficient buildings. Data-driven methods gained attention for estimation of energy consumption in buildings which would provide more accurate prediction results. In this study, hourly energy consumption prediction is performed on a university office building to increase energy efficiency in the building using machine learning algorithms. A new parameter is proposed, air conditioning demand, to improve accuracy of the algorithms. Moreover, temporal parameters, i.e. day of week, month of year, and hour of day, were used along with meteorological parameters to improve prediction performance of the algorithms. Experimental results show that hourly energy consumption of the building could be predicted using machine learning algorithms with high performance. When the results were analysed, Deep Neural Network (DNN) achieved better performance among other alternative algorithms. The average values of R2, RMSE and MAPE for DNN were 0.959, 4.796 kWh, and 5.738 %, respectively. Also, the addition of proposed air conditioning demand parameter provided improved performance to the algorithms. • Long-term building energy consumption prediction (BECP) is performed. • Five machine learning algorithms are utilized for BECP. • A new parameter is proposed, named as air conditioning demand. • Machine learning algorithms produced satisfying experimental results. • Efficiency of proposed air conditioning demand parameter is also justified. [ABSTRACT FROM AUTHOR]

The integration of Photovoltaic (PV) systems into grid has a detrimental effect on grid stability, dependability, reliability, efficiency, economy, planning and scheduling. Thus, a reliable PV output prediction is necessary for grid stability. This paper presents a detailed review on PV power forecasting technique. A detailed evaluation of forecasting techniques reveals that solar irradiance is highly correlated with PV output which makes it mandatory to classify the weather as well as analyze the cloudy motion. The researchers in literature dominantly preferred very short term and short-term horizon for PV forecasting. Different PV forecasting models categorized into physical, persistence, statistical, machine learning and hybrid model used in literature are studied and evaluated using various performance metrics such as root mean square error and mean absolute error for accuracy. A thorough comparison of various forecasting techniques suggests that deep neural networks with ensemble technique or hybrid techniques supersedes the traditional approaches of PV power forecasting in terms of efficiency and accuracy.Graphical abstract: The integration of Photovoltaic (PV) systems into grid has a detrimental effect on grid stability, dependability, reliability, efficiency, economy, planning and scheduling. Thus, a reliable PV output prediction is necessary for grid stability. This paper presents a detailed review on PV power forecasting technique. A detailed evaluation of forecasting techniques reveals that solar irradiance is highly correlated with PV output which makes it mandatory to classify the weather as well as analyze the cloudy motion. The researchers in literature dominantly preferred very short term and short-term horizon for PV forecasting. Different PV forecasting models categorized into physical, persistence, statistical, machine learning and hybrid model used in literature are studied and evaluated using various performance metrics such as root mean square error and mean absolute error for accuracy. A thorough comparison of various forecasting techniques suggests that deep neural networks with ensemble technique or hybrid techniques supersedes the traditional approaches of PV power forecasting in terms of efficiency and accuracy. [ABSTRACT FROM AUTHOR]

Innovation capability (IC) is a fundamental firms’ strategic asset to sustain competitive advantage. In this article, relying on patent data, patents forward citations are used as proxy of IC and the main patents numerical and categorical variables are considered as proxy of IC determinants. The main purpose of this article is to understand which patents features are relevant to predict IC, i.e., which are the determinants of IC within patents. Three different algorithms of machine learning, widely applied to model real-world phenomena, regularized least squares, deep neural networks, and random forest are applied for this investigation. Results show that the most important patent features useful to predict IC refer to internal determinants, such as the technological scope of the company (technological domains and International Patent Classification classes), the backward citations, and technical concepts. Some external variables are also relevant, such as the family size, a time-related variable, and the timespan between the youngest and oldest patent of the family. Implications from this article concern two perspectives. From a methodological perspective, the study shows the usefulness of machine learning approaches in simplifying the decision-making process as they reduce the number of variables to be considered to investigate the company’s IC and they provide accurate and easily interpretable result. From a managerial standpoint, this article points out the few and relevant patent variables to be considered when dealing with patents and IC. [ABSTRACT FROM AUTHOR]

Accurate rainfall-runoff modelling is particularly challenging due to complex nonlinear relationships between various factors such as rainfall characteristics, soil properties, land use, and temporal lags. Recently, with improvements to computation systems and resources, data-driven models have shown good performances for runoff forecasting. However, the relative performance of common data-driven models using small temporal resolutions is still unclear. This study presents an application of data-driven models using artificial neural network, support vector regression and long-short term memory approaches and distributed forcing data for runoff predictions between 2010 and 2019 in the Russian River basin, California, USA. These models were used to predict hourly runoff with 1–6 h of lead time using precipitation, soil moisture, baseflow and land surface temperature datasets provided from the North American Land Data Assimilation System. The predicted results were evaluated in terms of seasonal and event-based performance using various statistical metrics. The results showed that the long-short term memory and support vector regression models outperforms artificial neural network model for hourly runoff forecasting, and the predictive performance of the models was greater during the wet seasons compared to the dry seasons. In addition, a comparison of the data-driven model results with the National Water Model, a fully distributed physical-based hydrologic model, showed that the long-short term memory and support vector regression models provide comparable performance. The results demonstrate that data-driven models for hourly runoff forecasting are sufficiently predictive and useful in areas where observation systems are not available. [ABSTRACT FROM AUTHOR]

Spring discharge always illustrates the groundwater-flux and aquifer storage oscillations. Because of inherent heterogeneity in karst environments, it is essential to mimic karst spring flows to acquire a superior understanding of hydrological processes and provide sustainable management and protection of karst waters. The framework of karst media is nonlinear and complex, which can be demonstrated by data-driven models. In this study, the performance of Support Vector Regression (SVR), the Adaptive Neuro-Fuzzy Inference System (ANFIS), and Artificial Neural Networks (ANN) was assessed to predict spring discharge 1-, 3-, 7-, 10- and 14-day ahead. A hybrid Gamma Test-Genetic Algorithm was performed to establish an optimal input combination. SVR, ANFIS, and ANN performances were analyzed via four residuals: Correlation Coefficient, Mean Absolute Error, Root Mean Squared Error (RMSE), Nash–Sutcliffe Efficiency, and Developed Discrepancy Ratio. According to the RMSE values (of 0.08, 0.18, 0.64 and 0.86 using ANN; 0.19, 0.22, 0.83, and 0.61 using ANFIS; and 0.15, 0.26, 0.78 and 0.59 using SVR for Lordegan, Deime, Dehcheshmeh, and Dehghara springs, respectively), the results demonstrated that ANN was highly accurate for the discharge prediction of the Lordegan, Deime, and Dehcheshme springs whereas it had the least accuracy for the discharge prediction of the Dehghara spring up to 14-day ahead. However, SVR performed better than the other models for all prediction steps in the Dehghara spring, having a more complex and heterogeneous flow system compared to the others. For all the springs, the models' accuracy decreased as the time ahead increased. [ABSTRACT FROM AUTHOR]

Copyright of Journal of Beijing University of Technology is the property of Journal of Beijing University of Technology, Editorial Department and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Information and communication technologies combined with in-situ sensors are increasingly being used in the management of urban drainage systems. The large amount of data collected in these systems can be used to train a data-driven soft sensor, which can supplement the physical sensor. Artificial Neural Networks have long been used for time series forecasting given their ability to recognize patterns in the data. Long Short-Term Memory (LSTM) neural networks are equipped with memory gates to help them learn time dependencies in a data series and have been proven to outperform other type of networks in predicting water levels in urban drainage systems. When used for soft sensing, neural networks typically receive antecedent observations as input, as these are good predictors of the current value. However, the antecedent observations may be missing due to transmission errors or deemed anomalous due to errors that are not easily explained. This study quantifies and compares the predictive accuracy of LSTM networks in scenarios of limited or missing antecedent observations. We applied these scenarios to an 11-month observation series from a combined sewer overflow chamber in Copenhagen, Denmark. We observed that i) LSTM predictions generally displayed large variability across training runs, which may be reduced by improving the selection of hyperparameters (non-trainable parameters); ii) when the most recent observations were known, adding information on the past did not improve the prediction accuracy; iii) when gaps were introduced in the antecedent water depth observations, LSTM networks were capable of compensating for the missing information with the other available input features (time of the day and rainfall intensity); iv) LSTM networks trained without antecedent water depth observations yielded larger prediction errors, but still comparable with other scenarios and captured both dry and wet weather behaviors. Therefore, we concluded that LSTM neural network may be trained to act as soft sensors in urban drainage systems even when observations from the physical sensors are missing. [ABSTRACT FROM AUTHOR]

The influence of weather conditions on sea state, and in particular on the dynamic evolution of waves, is an important issue that affects several areas, including maritime traffic and the planning of coastal works. To collect relevant data, buoys are used to set up distributed sensor networks along coastal areas. However, unfavourable weather conditions can lead to downtime, which can be extended due to maintenance issues. The ability to improve the robustness of these sensor systems using predictive models, i.e. digital twins, to interpolate and extrapolate missing data is an important and growing area of research. To accomplish such a task, models must be found that can account for both the spatial and temporal dynamics of the input data to correctly estimate the variables of interest. In this work, a deep learning architecture is proposed to realize a digital twin for the monitoring buoy for significant wave height estimation using spatial and temporal information about the wind field in the area of interest. The proposed methodology was applied to a case study using wave height data from an Italian Sea Monitoring Network buoy installed near the coast of Sicily and wind field data from the Copernicus Climate Change Service ERA5 reanalysis. The reported results show that the use of a multi-block hybrid deep neural network consisting of convolutional layers for spatial feature extraction and short-term memory layers for modelling the involved dynamics, which takes into account the buoy surrounding area, outperforms other empirical, numerical, machine learning and deep learning methods used in the literature. • The wind field and the significant wave height relation are modelled with a hybrid neural network. • Single and multiple CNN-LSTM layers are compared. • Peak estimation is addressed through dataset balancing strategy. • Specific performance analysis on extreme events has been reported [ABSTRACT FROM AUTHOR]

Prediction of nonlinear and dynamic systems is a challenging task, however with the aid of machine learning techniques, particularly neural networks, is now possible to accomplish this objective. Most common neural networks used are the multilayer perceptron (MLP) and recurrent neural networks (RNN) using long-short term memory units (LSTM-RNN). In recent years, deep learning neural network models have become more relevant due the improved results they show for various tasks. In this paper the authors compare these neural network models with deep learning neural network models such as long-short term memory deep recurrent neural network (LSTM-DRNN) and gate recurrent unit deep recurrent neural network (GRU-DRNN) when presented with the task of predicting three different chaotic systems such as the Lorenz system, Rabinovich–Fabrikant and the Rossler System. The results obtained show that the deep learning neural network model GRU-DRNN has better results when predicting these three chaotic systems in terms of loss and accuracy than the two other models using less neurons and layers. These results can be very helpful to solve much more complex problems such as the control and synchronization of these chaotic systems. [ABSTRACT FROM AUTHOR]

Weather data are major input for building energy usage predictions. However, solar-radiation-related historical and real-time weather data are unavailable or incomplete in many locations. Therefore, many models, which use more available weather parameters to predict solar radiation's components, were developed in the last 30 years. An experimental evaluation of these models is needed since measurement devices and satellite techniques are improved, and weather files are consequently updated. In this study, we review, calibrate, and validate the accuracy of global and diffuse irradiation prediction models and efficacy models using experimental data over an 18-month data collection period in Taipei, Taiwan, and Kingsville, Texas. The evaluation also covers data-driven models such as neural networks. The results show that the SUNY model provides good solar irradiance estimations; Perez and Muneer efficacy models provide good daylight illuminance estimations; and Erbs, Muneer, Reindl, and Perez models have similar accuracy but different error trends when separating direct and diffuse irradiance. The results can be used as a guideline when filling in solar-radiation-related fields. [ABSTRACT FROM AUTHOR]

Conventional numerical models of fluid film bearings (FFB) typically require significant amounts of computations to achieve acceptable calculation accuracy. This may cause many inconveniences, especially for computationally expensive rotor dynamics tasks. The study addresses general issues in building synthetic data-driven dynamic FFB models using machine learning methods, focusing on artificial neural networks (ANNs). In the study, surrogate models predicting journal bearing forces were built using a variety of approaches, including single- and multi-component ANN-based models, as well as so-called physics-informed neural networks (PINN). The work presents a comparative analysis of the considered approaches, primarily in the context of rotor dynamics calculations. It includes an assessment of the accuracy and performance of considered models, as well as the time spent on their creation. The results show that all tested solutions outperform numerical methods by order of magnitude or more, but it is difficult to talk about the overwhelming advantage of any one approach over others. When choosing a methodology for building surrogate FFB models, one should proceed from the requirements for the model, the benefits and shortcomings of certain methods. The work gives an idea of their relationships, including the influence of key hyperparameters of the methods on the properties of the resulting models. The results can be useful both for traditional calculations in the area of rotor dynamics, and for relatively new problems in the rotor systems, such as predictive analytics and optimal design systems, predictive controllers of active components of rotary machines. • Various machine learning methods tested for prediction the journal fluid film bearing forces • One-component, multi-component and physics-informed ANN-based bearing models are compared • Surrogate models achieved good accuracy and excellent performance compared to numerical models • Data-driven models tested under extrapolation conditions and when applying reduced dataset for training • Relations between the method' hyperparameters and models' properties are described [ABSTRACT FROM AUTHOR]

A critical challenge in the data-driven modeling of dynamical systems is producing methods robust to measurement error, particularly when data is limited. Many leading methods either rely on denoising prior to learning or on access to large volumes of data to average over the effect of noise. We propose a novel paradigm for data-driven modeling that simultaneously learns the dynamics and estimates the measurement noise at each observation. By constraining our learning algorithm, our method explicitly accounts for measurement error in the map between observations, treating both the measurement error and the dynamics as unknowns to be identified, rather than assuming idealized noiseless trajectories. We model the unknown vector field using a deep neural network, imposing a Runge-Kutta integrator structure to isolate this vector field, even when the data has a non-uniform timestep, thus constraining and focusing the modeling effort. We demonstrate the ability of this framework to form predictive models on a variety of canonical test problems of increasing complexity and show that it is robust to substantial amounts of measurement error. We also discuss issues with the generalizability of neural network models for dynamical systems and provide open-source code for all examples. • Presenting a novel method to simultaneously learn measurement noise and dynamics. • Embedding network in a numerical integrator for learning from unevenly spaced data. • A discussion of apparent pitfalls with using neural networks. [ABSTRACT FROM AUTHOR]

A novel neural network algorithm for indirect measurement of the polished rod load of the beam-pumping unit is proposed. The dynamometer card is a two-dimension graph of polished rod load versus horse head position, and it is widely used for working state monitoring of pumping unit. The drift problem of load sensor makes the direct measurement of polished rod load not economical and practical, so indirect measurement of polished rod load is crucial. Through the analysis of the mechanical characteristics of the pumping unit, a neural network with physical meaning (physical network) is designed. The experiment shows that the physical network can simulate the energy transfer process properly with only 18 training data items, which is about 1% of other data-driven models, and maintains good performance under different down-hole conditions. [ABSTRACT FROM AUTHOR]

The aim of this paper is to improve semiseasonal forecast of groundwater availability in response to climate variables, surface water availability, groundwater level variations, and human water management using a two‐step data‐driven modeling approach. First, we implement an ensemble of artificial neural networks (ANNs) for the 300 wells across the High Plains aquifer (USA). The modeling framework includes a method to choose the most relevant input variables and time lags; an assessment of the effect of exogenous variables on the predictive capabilities of models; and the estimation of the forecast skill based on the Nash‐Sutcliffe efficiency (NSE) index, the normalized root mean square error, and the coefficient of determination (R2). Then, for the ANNs with low‐ accuracy, a MultiModel Combination (MuMoC) based on a hybrid of ANN and an instance‐based learning method is applied. MuMoC uses forecasts from neighboring wells to improve the accuracy of ANNs. An exhaustive‐search optimization algorithm is employed to select the best neighboring wells based on the cross correlation and predictive accuracy criteria. The results show high average ANN forecasting skills across the aquifer (average NSE > 0.9). Spatially distributed metrics of performance showed also higher error in areas of strong interaction between hydrometeorological forcings, irrigation intensity, and the aquifer. In those areas, the integration of the spatial information into MuMoC leads to an improvement of the model accuracy (NSE increased by 0.12), with peaks higher than 0.3 when the optimization objectives for selecting the neighbors were maximized.tT Key Points: Artificial neural networks can accurately forecast semiseasonal groundwater level changesMuMoC improved the groundwater well‐level forecasting skill for 1‐ to 4‐month lead times with respect to a single ANN model by 25% in NSEThe implementation of MuMoC is recommended in case of densely gauged areas [ABSTRACT FROM AUTHOR]

Highlights • Coarse-mesh CFD snapshots are used for Model reduction of CVD reactor. • CPU time reduction of 60% for fine-mesh computations with complex chemistry. • Candidate models for deposition of Al from DMEAA evaluated. Abstract A computational methodology is introduced for the development of a Reduced Order Model (ROM) using data from low-fidelity, in terms of the spatial discretization, Computational Fluid Dynamics (CFD) simulations. The methodology is applied in the context of investigating efficiently, new chemistry pathways in Chemical Vapor Deposition (CVD) processes by comparing experimental findings with simulation results. The proposed approach involves building a very coarse, yet three-dimensional CFD model of the process that does not include chemical reactions, which provides "snapshots", i.e. time instances, of the dynamic behavior of the system. This bundle of low-fidelity data is used for generating a reduced order model by implementing Proper Orthogonal Decomposition (POD), for the time-invariant subspace approximation of the data and Artificial Neural Networks (ANN) for the time-dependent coefficients. This last step circumvents the need for Galerkin projection of the equations on the subspace, rendering therefore the approach equation-free and purely data-driven. The ultimate goal addressed in this work is to produce fast approximations of the temperature, pressure and flow field in a CVD reactor, which will then be used as starting points of high-fidelity, in terms of the computational mesh, CFD simulations, that include multiple surface and gas phase reactions. Given the quality of approximations, the model corrects for the species distributions after only a few time-steps, providing thus quick and accurate deposition-rate and uniformity results. This allows for efficient and easy fine-tuning adjustments of the chemistry model diminishing the need to solve the high-fidelity model many times. [ABSTRACT FROM AUTHOR]

Highlights • A workflow is presented for data-driven equation-free Reduced Order Model. • Principal Component Analysis/Artificial Neural Networks are combined for the ROM. • Only data from simulations are used to derive the ROM. • Data is first classified with the Support Vector Machine algorithm. • Method produces fast and accurate results on nonlinear CVD process model. Abstract This paper presents an equation-free, data-driven approach for reduced order modeling of a Chemical Vapor Deposition (CVD) process. The proposed approach is based on process information provided by detailed, high-fidelity models, but can also use spatio-temporal measurements. The Reduced Order Model (ROM) is built using the method-of-snapshots variant of the Proper Orthogonal Decomposition (POD) method and Artificial Neural Networks (ANN) for the identification of the time-dependent coefficients. The derivation of the model is completely equation-free as it circumvents the projection of the actual equations onto the POD basis. Prior to building the model, the Support Vector Machine (SVM) supervised classification algorithm is used in order to identify clusters of data corresponding to (physically) different states that may develop at the same operating conditions due to the inherent nonlinearity of the process. The different clusters are then used for ANN training and subsequent development of the ROM. The results indicate that the ROM is successful at predicting the dynamic behavior of the system in windows of operating parameters where steady states are not unique. [ABSTRACT FROM AUTHOR]

• A novel definition of the best machine model for N 2 O soft sensors was established. • Multivariate outlier detection considered special characteristics of wastewater. • Feature reduction balanced data acquisition, computation time, and model accuracy. • Boosted decision trees showed better overall performance than deep neural network. • Interpretable ML is a step towards guided decisions for N 2 O emissions mitigation. Nitrous oxide (N 2 O) emissions may account for up to 80 % of a wastewater treatment plant's (WWTP) total carbon footprint. Given the complexity of the pathways involved, estimating N 2 O emissions through mechanistic models still often fails to precisely depict process dynamics. Alternatively, data-driven methods for predicting N 2 O emissions hold substantial potential. However, so far, a comprehensive approach is still overlooked, impeding the advancement of full-scale application. Therefore, this study develops a comprehensive approach for using machine learning to perform online process modeling of N 2 O emissions. The approach is tested on a long-term N 2 O emission dataset from a full-scale WWTP. Uniquely, the proposed approach emphasizes not just model accuracy, but it also considers model complexity, computational speed, and interpretability, equipping operators with the insights needed for informed corrective actions. Algorithms with varying levels of complexity and interpretability including k-Nearest Neighbors (kNN), decision trees, ensemble learning models, and deep neural networks (DNN) were considered. Furthermore, a parametric multivariate outlier removal method was adjusted to account for data statistical distributions, significantly reducing data loss. By employing an effective feature selection methodology, a trade-off between data acquisition, model performance, and complexity was found, reducing the number of features by 40 % and decreasing data collection cost, model complexity and computational burden without significant effect on modeling accuracy. The best performing models are kNN (R2 = 0.88), AdaBoost (R2 = 0.94), and DNN (R2 = 0.90). Feature importance of models was analyzed and compared with process knowledge to test interpretability, guiding N 2 O mitigation decisions. [Display omitted] [ABSTRACT FROM AUTHOR]

Abstract Presence-absence methods are widely-used data-driven models for landslide susceptibility mapping. Landslide absence data included in the training data of presence-absence methods is usually not available and has to be generated. In consideration of low availability and uncertain quality of landslide absence data, many presence-only methods which simply use landslide presence as training data were proposed to map landslide susceptibility. However, whether the presence-only methods can circumvent the influence of the shortcomings inherent to landslide absence data and perform better than presence-absence methods are worth studying. Moreover, the effect of landslide absence data in data-driven models for landslide susceptibility mapping can be discussed. In this study, two presence-only methods including one-class support vector machine (one-class SVM), kernel density estimation (KDE), and two presence-absence methods including artificial neural networks (ANN) and two-class support vector machine (two-class SVM) are developed and compared to evaluate their respective performance in mapping landslide susceptibility. The AUC values are 0.705, 0.720, 0.929, and 0.951 for one-class SVM, KDE, ANN, and two-class SVM, respectively. From the comparison of the four methods, two-class SVM has the best performance in landslide susceptibility mapping among the four methods, while one-class SVM has the worst. Two presence-absence methods can constrain the over-prediction of susceptibility value better and have better performance than the two presence-only methods since they classify less percentage of areas to be susceptible with more landslide occurrences located inside. The landslide absence data is proven to constrain the over-prediction of models, which makes it necessary in landslide susceptibility mapping. Highlights • Presence-only methods would over-predict the potential distribution of landslide. • Two-class SVM methods shows the most promising results. • The landslide absence data is proven to constrain the over-prediction of models. [ABSTRACT FROM AUTHOR]

In this study, an expert knowledge-based model, a logistic regression model, and an artificial neural network model were compared for their accuracy and portability in landslide susceptibility mapping. Two study areas (the Kaixian and the Three Gorges areas in China) were selected for this comparison based on their well-known, high landslide hazard. To evaluate the performance of these models and to minimize the impact of the particularity of a study area on model performance, cross-applications of three models between the two study areas were conducted. When the Kaixian area was used as a model development area, prediction accuracy for the expert knowledge-based model, the logistic regression model, and the artificial neural network model were 71.5%, 81.0% and 100.0%, respectively. The high prediction accuracy of the two data-driven models were expected because the observed landslide occurrence used in training the models were also used to validate their respective performance, while the expert knowledge-based model did not use these observations for training. The perfect accuracy for the neural network model can also be attributed to its over-prediction of the susceptibility. When breaking the susceptibility into four classes: very low susceptibility (0–0.25), low susceptibility (0.25–0.5), high susceptibility (0.5–0.75), and very high susceptibility (0.75–1), the observed landslide density at the very high susceptibility level is 0.303/km 2 , 0.212/km 2 , and 0.195/km 2 for the expert knowledge-based model, the logistic regression model, and the artificial neural network model, respectively. This suggests that the expert knowledge-based model was much better than the other two data-driven models at evaluating landslide occurrence in very high susceptibility areas. When the three models developed in the Kaixian area were applied in the Three Gorges area without any changes, their prediction accuracy dropped to 44.8% for the logistic regression model and 81.6% for the artificial neural network model, while the expert knowledge-based model maintained its initial accuracy level of 82.8%. The landslide density for the very high susceptibility areas in the Three Georges area was 0.275/km 2 , 0.082/km 2 , and 0.060/km 2 for the expert knowledge-based model, the logistic model, and the artificial neural network model, respectively. These results indicate that the expert knowledge-based model is more effective at predicting areas with very high susceptibility. When the Three Gorges area was used as a model development area and the Kaixian area was used as the model application area, similar results were obtained. Results from the two experiments show that the performance of the logistic regression model and artificial neural network model is not as stable as the expert knowledge-based model when transferred to a new area. This suggests that the expert knowledge-based model is more suitable for landslide susceptibility mapping over large areas. [ABSTRACT FROM AUTHOR]

Artificial Neural Networks (ANNs) are becoming more popular to model ocean engineering problems. With the development of Artificial Intelligence, data-driven models are showing better behaviour and performance than the traditional models used in ocean engineering. However, the main limitation of ANNs models is the uncertainty associated to them and their "black box" nature. ANNs models present final results without any uncertainty or explanation and this limits enormously their applicability, especially in decision-making tasks. This research paper tries to deal with this problem. Given the exponential growth that artificial intelligence models are currently experiencing, it is necessary to address these limitations so that the field of ocean engineering does not fall behind. This is the final aim of this review research paper. A review of how ocean engineering studies have dealt with ANNs uncertainties has been carried out with the final objective of proposing new methods to deal with these uncertainties to make the application of this tool less constrained by its "black-box nature" and facilitate the expansion of these models in the field. [Display omitted] • Model uncertainty and the black-box nature of ANNs is the main limitation of this methodology. • Not considering uncertainty in ANNs can lead to errors in the order of tenths of a percent. • Bootstrap methodology is the most appropriate to deal with input uncertainty in ANNs. • Bayesian techniques are recommended to deal with input and model uncertainty for small datasets. • Monte Carlo simulations are recommended to deal with uncertainty in maritime structures studies. [ABSTRACT FROM AUTHOR]

Occupancy information that can be used to determine heating, ventilation and lighting requirements is one of the important parameters for the control of energy efficient buildings. In this paper, we propose a fusion framework for building occupancy estimation with environmental parameters. Based on the environmental sensor data, a coarse estimation of building occupancy can be achieved using data-driven models that include extreme learning machine (ELM), support vector machine (SVM), artificial neural network (ANN), K-nearest neighbors (KNN), linear discriminant analysis (LDA) and classification and regression tree (CART). Due to the extremely fast learning speed of the ELM algorithm, we apply an ELM-based wrapper method to select the best feature set of environmental parameters. To further improve the estimation accuracy of building occupancy, taking occupancy dynamics into consideration, we fuse the results of data-driven models with well developed occupancy models by using a particle filter algorithm. Real experiments have shown that our proposed fusion framework can achieve improvements of around 5–14% and 3–12% for the estimation accuracy and the detection accuracy (presence/absence) respectively among the different methodologies. [ABSTRACT FROM AUTHOR]

This work investigates the uncertainty associated to the presence of multiple subsets of predictors yielding data-driven models with the same, or similar, predictive accuracy. To handle this uncertainty effectively, we introduce a novel input variable selection algorithm, called Wrapper for Quasi Equally Informative Subset Selection (W-QEISS), specifically conceived to identify all alternate subsets of predictors in a given dataset. The search process is based on a four-objective optimization problem that minimizes the number of selected predictors, maximizes the predictive accuracy of a data-driven model and optimizes two information theoretic metrics of relevance and redundancy, which guarantee that the selected subsets are highly informative and with little intra-subset similarity. The algorithm is first tested on two synthetic test problems and then demonstrated on a real-world streamflow prediction problem in the Yampa River catchment (US). Results show that complex hydro-meteorological datasets are characterized by a large number of alternate subsets of predictors, which provides useful insights on the underlying physical processes. Furthermore, the presence of multiple subsets of predictors—and associated models—helps find a better trade-off between different measures of predictive accuracy commonly adopted for hydrological modelling problems. [ABSTRACT FROM AUTHOR]

Water resources management is a complex task. It requires accurate prediction of inflow to reservoirs for the optimal management of surface resources, especially in arid and semi-arid regions. It is in particular complicated by droughts. Markov chain models have provided valuable information on drought or moisture conditions. A complementary method, however, is required that can both evaluate the accuracy of the Markov chain models for predicted drought conditions, and forecast the values for ensuing months. To that end, this study draws on Artificial Neural Networks (ANNs) as a data-driven model. The employed ANNs were trained and tested by means of a statistically-based input selection procedure to accurately predict reservoir inflow and consequently drought conditions. Thirty three years' data of inflow volume on a monthly time resolution were selected to enable calculation of the standardized streamflow index (SSI) for the Markov chain model. Availability of hydro-climatic data from the Doroodzan reservoir in the Fars province, Iran, allowed us to develop a reservoir specific ANN model. Results demonstrated that both models accurately predicted drought conditions, by employing a randomization procedure that facilitated the selection of the required data for the ANN to forecast reservoir inflow close to the observed values over a validation period. The results confirmed that combining the two models improved short-term prediction reliability. This was in contrast to single model applications that resulted into substantial uncertainty. This research emphasized the importance of the correct selection of data or data mining, prior to entering a specific modeling routine. [ABSTRACT FROM AUTHOR]

[Display omitted] In recent years deep neural networks have been proposed as a lightweight data-driven model to capture high-dimensional, nonlinear physical processes to predict building thermal responses. However, the need of a large amount of data for the training process of deep neural networks clashes with the potential limited data availability in most existing or new buildings. Transfer learning aims to enhance the performance of a target learner exploiting knowledge from related and similar environments. This study conducted a suite of experiments that leveraged 250 data-driven models based on a synthetic dataset of a building archetype to study the influence of data availability, energy efficiency level, occupancy and climate for the transfer process of thermal dynamics. The performance of the transfer learning process was compared against a classical machine learning approach. The results suggest that building thermal dynamics can be effectively transferred under the same climatic conditions, increasing performance when dealing with different occupancy schedules, efficiency levels and low data availability. Furthermore, the paper compares the performance of both transfer learning and machine learning approaches in an online fashion, to support the implementation in real-world deployment. [ABSTRACT FROM AUTHOR]

Drilling optimization is one of the most important management and engineering objectives in the upstream oil and gas industry, which has been the subject of numerous studies during the last two decades. Although the role of geomechanical parameters has rarely been considered in these studies. Therefore, this study aimed to optimize the controllable drilling parameters using a novel geomechanics-based workflow. For this purpose, data for 4229 m drilling from 4 wells located in one of the oil fields in southwestern Iran were studied. A 1D geomechanical model was developed for each well following the data pre-processing. In the feature selection stage, using the non-dominated sorting genetic algorithm-II (NSGA-II) integrated by the artificial neural network with multilayer perceptron (ANN-MLP), bit diameter (BS), equivalent circulating density (ECD), weight on bit (WOB), revolution per minute (RPM), flow pump rate (FPR), depth (Depth), formation resistivity (RT), static Young's modulus (E), and minimum horizontal stress (Sh) were selected to develop intelligent prediction models for drilling rate of penetration (ROP) and torque (TRQ), so two models ROP and TRQ were then developed using ANN-MLP. Subsequently, the coupling of ROP, TRQ, and mechanical specific energy (MSE) models and the NSGA-II multi-objective optimization algorithm were performed to achieve the best controllable parameters including WOB, RPM, and FPR at any depth (real-time), so that ROP was maximized and TRQ and MSE were minimized. Then, the optimal parameters for each formation thickness, 50 m and 100 m intervals were presented. In addition, the scenario of the set of optimal parameters for each geomechanical unit (GMU) was provided for the first time. Finally, the drilling performance results obtained from the two optimization techniques based on geomechanics and best practices were compared to each other. According to the results, the application of optimal parameters obtained from the best practices significantly leads to a doubling of mechanical specific energy. However, the application of geomechanics-based optimization technique with different scenarios successfully improved the performance of drilling operations; so that in the real-time scenario, the drilling time by 62%, 82%, 56%, and 68% and the MSE by averagely 7%, 29%, 11%, and 15% were reduced for wells A, B, C, and D, respectively. Furthermore, the application of scenarios of GMUs, formation thickness, and 50 m and 100 m intervals had the best performance, respectively. Therefore, the application of geomechanics-based drilling optimization for carbonate formations within the studied parameter ranges is strongly recommended according to the information range in this study. • Using 1D MEM results (Rock strength and Geo stress) as inputs for Artificial Intelligent-based drilling models. • Distinguishing Geomechanical Units (GMU) using an optimum clustering method. • Using Geomechanical Units (GMU) as interval for developing and designing optimum controllable drilling parameters. • Multi-objective optimization of controllable drilling parameters with maximizing ROP and minimizing TRQ and MSE. • Successfully implementations of the results of Geomechanical-based optimizationin a simulation-based experiment. [ABSTRACT FROM AUTHOR]

Input variable selection (IVS) is vital in the development of data-driven models. Among different IVS methods, partial mutual information (PMI) has shown significant promise, although its performance has been found to deteriorate for non-Gaussian and non-linear data. In this paper, the effectiveness of different approaches to improving PMI performance is investigated, focussing on boundary issues associated with bandwidth estimation. Boundary issues, associated with kernel-based density and residual computations within PMI, arise from the extension of symmetrical kernels beyond the feasible bounds of potential inputs, and result in an underestimation of kernel-based marginal and joint probability distribution functions in the PMI. In total, the effectiveness of 16 different approaches is tested on synthetically generated data and the results are used to develop preliminary guidelines for PMI IVS. By using the proposed guidelines, the correct inputs can be identified in 100% of trials, even if the data are highly non-linear or non-Gaussian. [ABSTRACT FROM AUTHOR]

Advances in industrial digital technologies have led to an increasing volume of data generated from industrial bioprocesses, which can be utilised within data-driven models (DDM). However, data volume and variability complications make developing models that captures the underlying biological nature of the bioprocesses challenging. In this study, a framework for developing data-driven models of bioprocesses is proposed and evaluated by modelling an industrial bioprocess, which treats industrial or agrifood wastewaters whilst simultaneously generating bioenergy. Six models were developed to predict the reduction in chemical oxygen demand from the wastewater by the bioprocess and statistically evaluated using both testing data (randomly partitioned data from the model development) and unseen data (new data not used during the model development). The statistical error metrics employed were the coefficient of determination (R2), root mean square error (RMSE), mean absolute error (MAE) and mean absolute percentage error (MAPE). The stacked neural network model was best able to model the bioprocess, having the highest accuracy on the testing data (R2: 0.98; RMSE: 1.29; MAE: 2.27; MAPE: 4.08) and the unseen data (R2: 0.82; RMSE: 2.57; MAE: 1.75; MAPE: 3.68). Data visualisation is used to observe (or confirm) whether new data points are within the model boundaries, helping to increase confidence in the model's predictions on future data. • Modelling bioprocesses face data volume and variability challenges. • Data-driven modelling method is proposed and applied to an industrial bioprocess. • The stacked neural network had the highest accuracy when interpolating. • However, the model accuracy reduced when extrapolating. • Visualisation used to communicate when model is interpolating/extrapolating. [ABSTRACT FROM AUTHOR]

• We formulated a methodology framework to establish data-driven models for effective groundwater management. • We verified the applicability of data-driven models across tempo-spatial scale. • We demonstrated the tempo-spatial distribution of changes in groundwater level by using O 2 S 2 method. The overexploitation of groundwater resource and its delicacy management has gained increasing attentions in recent years worldwide because of causing a series of serious environmental and geological problems. Currently, accurately predicting the groundwater level (GWL) is an important issue in effective groundwater management across scales. In the present study, three popularly-used data-driven models, which are an autoregressive integrated moving average (ARIMA), a back-propagation artificial neural network (BP-ANN) and long short-term memory (LSTM), were established in five zones with different hydrogeological properties to explore the model's accuracy in predicting the GWL at monthly and daily scales in a Northern Plain in China. The developed models were evaluated by both the Nash-Sutcliffe efficiency coefficient (NSE) and root mean square error (RMSE). The results indicate that the performance of the LSTM model is best at monthly time scales with the NSEs greater than 0.76 and RMSEs smaller than 1.15 m in each zone during the training period and demonstrate a good performance at daily time scales with the NSEs greater than 0.9 and the RMSEs smaller than 0.55 m at a local area. Meanwhile, the tempo-spatial distribution of the probability of drawdowns from the LSTM model was estimated by using the object-oriented spatial statistical (O 2 S 2) method. The results show that cumulative drawdowns greater than 10 m are mainly concentrated in water source areas, with probabilities over 0.7 from 2003 to 2010 and declining to less than 0.3 from 2011 to 2014. The GWL rose generally in the study area from 2015 to 2018, but the probability of a drawdown with more than 5 m exceeded 0.8 in Zone V because of continuing groundwater exploitation. This study formulates a framework on developing effective data-driven models for predicting the GWL across scales which have the potential to aid groundwater management. [ABSTRACT FROM AUTHOR]

Abstract: Species–environment relationships are key information for the development of planning and management strategies for conservation or restoration of ecosystems. Artificial neural networks (ANNs) are one widely applied type of species distribution model (SDM). Fuzzy neural networks (FNNs), that is, fuzzified ANNs, have been introduced to take into account the uncertainties inherent in fish behaviour and errors in input data. Despite their high predictive ability in modelling complex systems, FNNs cannot describe habitat preference curves (HPCs), although these are the basis for habitat quality assessment. The present study therefore aimed to evaluate the applicability of FNNs for modelling habitat preference and spatial distributions of Japanese medaka (Oryzias latipes), one of the most common freshwater fish in Japan. Three independent data sets were collected during a series of field surveys and used for model development and evaluation of FNNs. A weight decay backpropagation algorithm was additionally introduced, and its effects on the FNNs were evaluated on the basis of model performance and habitat preference information retrieved from the field observation data. Modified sensitivity analysis was applied to derive HPCs of the target fish. Application of weight decay backpropagation markedly reduced the variability of the model structures, improved the generalization ability of the FNNs, and resulted in well-converged and consistent HPCs that were similar to those evaluated by fuzzy habitat preference models. These results support the applicability of FNNs to habitat preference modelling, which can provide useful information on the habitat use by the target fish. Further study should focus on the effects of sources of uncertainty, such as zero abundance, on the SDMs and the resulting habitat preference evaluation. [Copyright &y& Elsevier]

Machine learning approaches, onboard measurements, and widely available wave forecast and hindcast data present an opportunity to develop predictive models for vessel motion forecasting. Detailed vessel motion forecasts would support underway and deployment decisions for safer and more efficient vessel operation. To demonstrate this application, ridge regression and neural network models for heave, pitch, and roll prediction were trained and tested using time-and-place specific, multidirectional wave model parameters as input. Additionally, the performance benefits of providing these predictive models with computationally efficient, physics-based model predictions (PBMPs) of heave, pitch, and roll as additional inputs were examined. Data from approximately 13,500 30-minute windows, measured aboard an operational research vessel, were used to train and test the data-driven models. Data from over 2,500 additional 30-minute windows, measured aboard a sister vessel, were also used to test the versatility of the trained models. The results of this study showed effective reduction of motion amplitude mean-squared error (MSE) values on multiple test datasets relative to the PBMPs alone. The results also showed that inclusion of PBMPs as input to the data-driven models was typically beneficial in terms of MSE reduction, stressing the importance of retaining physics-based information in data-driven models. [ABSTRACT FROM AUTHOR]

Abstract: Along with existing numerical process models describing the wind–wave interaction, the relatively recent development in the area of machine learning make the so-called data-driven models more and more popular. This paper presents a number of data-driven models for wind–wave process at the Caspian Sea. The problem associated with these models is to forecast significant wave heights for several hours ahead using buoy measurements. Models are based on artificial neural network (ANN) and instance-based learning (IBL) .To capture the wind–wave relationship at measurement sites, these models use the existing past time data describing the phenomenon in question. Three feed-forward ANN models have been built for time horizon of 1, 3 and 6h with different inputs. The relevant inputs are selected by analyzing the average mutual information (AMI). The inputs consist of priori knowledge of wind and significant wave height. The other six models are based on IBL method for the same forecast horizons. Weighted k-nearest neighbors (k-NN) and locally weighted regression (LWR) with Gaussian kernel were used. In IBL-based models, forecast is made directly by combining instances from the training data that are close (in the input space) to the new incoming input vector. These methods are applied to two sets of data at the Caspian Sea. Experiments show that the ANNs yield slightly better agreement with the measured data than IBL. ANNs can also predict extreme wave conditions better than the other existing methods. [Copyright &y& Elsevier]

The article presents the study which examines the efficacy of a hybrid-data model in determing the annual streamflows of rivers that exhibit complex dependence. The hydrid model is a combination of two non-linear data-driven models, an artificial neural network (ANN) and a moving block bootstrap (MBB). The model has been applied to three annual streamflow records with variable record length selected from different geographic regions, Africa, U.S., and Russia. The performance of the proposed ANN-based non-linear hybrid model has been compared with that of the linear parametric hybrid model. The study concludes that the proposed ANN-based hybrid model (ANNHM) is more effective than the linear parametric-based hybrid model (LPHM) in modelling annual streamflows of rivers.

The article discusses the use of hydrologic models such as instance-based learning with machine learning algorithms to forecast floods. The authors compare instance-based learning methods with artificial neural networks, M5 model trees, and conceptual hydrologic models. They conclude that instance-based learning methods can produce better results if used with the Gaussian kernel function compared to M5 model trees and artificial neural networks.

Abstract: Harmful algal blooms, which are considered a serious environmental problem nowadays, occur in coastal waters in many parts of the world. They cause acute ecological damage and ensuing economic losses, due to fish kills and shellfish poisoning as well as public health threats posed by toxic blooms. Recently, data-driven models including machine-learning (ML) techniques have been employed to mimic dynamics of algal blooms. One of the most important steps in the application of a ML technique is the selection of significant model input variables. In the present paper, we use two extensively used ML techniques, artificial neural networks (ANN) and genetic programming (GP) for selecting the significant input variables. The efficacy of these techniques is first demonstrated on a test problem with known dependence and then they are applied to a real-world case study of water quality data from Tolo Harbour, Hong Kong. These ML techniques overcome some of the limitations of the currently used techniques for input variable selection, a review of which is also presented. The interpretation of the weights of the trained ANN and the GP evolved equations demonstrate their ability to identify the ecologically significant variables precisely. The significant variables suggested by the ML techniques also indicate chlorophyll-a (Chl-a) itself to be the most significant input in predicting the algal blooms, suggesting an auto-regressive nature or persistence in the algal bloom dynamics, which may be related to the long flushing time in the semi-enclosed coastal waters. The study also confirms the previous understanding that the algal blooms in coastal waters of Hong Kong often occur with a life cycle of the order of 1–2 weeks. [Copyright &y& Elsevier]

Quantifying and analyzing the geological uncertainties associated with the geostatistical models is critical to design a reliable production strategy. A large number, hundreds, even thousands, of model realizations should be run in a reservoir simulator to quantify the uncertainties in terms of a target flow response. However, this becomes a time consuming and computationally demanding process. In practice, different methods are used to select only a few representative models for further analyses. This study aims to take a fresh look at uncertainty quantification solutions by means of a proxy model based on artificial neural networks (ANN). The idea is replacing the flow simulator with an ANN-based proxy model by which the target responses of all realizations are calculated much faster and with a lower computational demand. The ability of computing the combined effect of several uncertain parameters along with successful experiences of utilizing ANNs for reservoir simulation, motivated us to propose our ANN-based methodology as a solution for uncertainty quantification problem. We apply our methodology on a realistic 3D synthetic model which is a channelized model with uncertainties in porosity, permeability, and facies properties. The results are compared with those of two widely used methods, i.e., the traditional ranking and distance-based clustering techniques. It is shown that the ANN-based proxy is able to accurately quantify the geological uncertainties in a much shorter time. Hence, it can be used as a reliable tool for uncertainty quantification and representative models selection. • A new solution for quantifying the geological uncertainties of reservoir models is presented. • A data-Driven proxy model based on Artificial Neural Networks is built to replace the comprehensive reservoir simulator. • The Data-Driven proxy model calculates the desired target response much faster. • The proposed solution outperforms the widely-used uncertainty quantification techniques. [ABSTRACT FROM AUTHOR]

Accurate prediction of soil moisture is important yet challenging in various disciplines, such as agricultural systems, hydrology studies, and ecosystems studies. However, many data-driven models are being used to simulate and predict soil moisture at only a single depth. To predict soil moisture at various soil depths with depths of 100, 200, 500, and 1000 mm from the surface, based on the weather and soil characteristic data, this study designed two data-driven models: artificial neural networks and long short-term memory models. The developed models are applied to predict daily soil moisture up to 6 days ahead at four depths in the Eagle Lake Observatory in California, USA. The overall results showed that the long short-term memory model provides better predictive performance than the artificial neural network model for all depths. The root mean square error of the predicted soil moisture from both models is lower than 2.0, and the correlation coefficient is 0.80–0.97 for the artificial neural network model and 0.90–0.98 for the long short-term memory model. In addition, monthly based evaluation results showed that soil moisture predicted from the data-driven models is highly useful for analyzing the effects on the water cycle during the wet season as well as dry seasons. The prediction results can be used as basic data for numerous fields such as hydrological study, agricultural study, and environment, respectively. [ABSTRACT FROM AUTHOR]