Search

The decision-making process in real-world implementations has been affected by a growing reliance on data-driven models. Recognizing the limitations of isolated methodologies - namely, the lack of domain understanding in data-driven models, the subjective nature of empirical knowledge, and the idealized assumptions in first-principles simulations, we explore their synergetic integration. We showed the potential risk of biased results when using data-driven models without causal analysis. Through a case study on energy consumption in building design, we demonstrate how causal analysis significantly enhances the modeling process, mitigating biases and spurious correlations. We concluded that: (a) Sole data-driven models' accuracy assessment or domain knowledge screening may not rule out biased and spurious results; (b) Data-driven models' feature selection should involve careful consideration of causal relationships, especially colliders; (c) Integrating causal analysis results aid to first-principles simulation design and parameter checking to avoid cognitive biases. We advocate for the routine integration of causal inference within data-driven models in engineering practices, emphasizing its critical role in ensuring the models' reliability and real-world applicability.

The building sector in Sweden is responsible for 40 % of the final energy use, and space heating largely contributes to the total energy demand. Minimum energy performance standards and building codes are becoming more strict and comprehensive globally. In the European Union there are substantial goals of achieving a higher energy efficiency in the residential building sector. Energy Performance Certificates (EPCs) are a tool that can be used to track and assess building energy in a standardized way, to follow up on the energy efficiency of buildings. Privately owned single-family houses do not need to have a valid EPC, it is only mandatory to perform an assessment when the house is subject to sale, meaning that only about 20 % of single-family houses in Sweden have an EPC. The objective of the study is therefore to develop a data driven model based on existing EPC data, to estimate the annual space heating energy demand of Swedish single family houses that do not have an EPC. Sub-objectives include testing different data-driven machine learning models to identify which algorithm achieves the highest accuracy. Input features such as floor area, construction year, geographical location and heating source were used to make predictions. The results show that out of the investigated models, the Histogram-GBM performed the best when predicting the space heating energy of an individual house, with a RMSE of 24.33 kWh/m2 /year, MAE of 17.50 kWh/m2 /year, MAPE of 24.62 % and R2 -value of 0.736. When evaluating the results of predictions for the entire single-family housing stock, the errors of each individual house are hidden within the entire housing stock, minimizing the error of the total prediction. The best performing model, Histogram GBM, achieved a MBE of 0.59 kWh/prediction, equivalent to an error smaller than 0.005 %., I Sverige står byggnadssektorn för 40 % av den slutliga energianvändningen, och uppvärmning av byggnader är en stor andel av den totala energianvändningen. Minimikrav för energiprestanda och byggregler blir globalt allt mer strikta och omfattande. EU har betydande mål att uppnå högre energieffektivitet för alla typer av bostadshus. Energideklarationer (EPC) är ett verktyg som kan användas för att samla in och bedömma energianvändning i byggnader på ett standardiserat sätt, för att kunna följa upp energieffektivisering av byggnader. Till skillnad från andra byggnadstyper behöver privat ägda småhus inte ha en giltig energideklaration, de är endast obligatoriska att utfärda när huset säljs, vilket betyder att bara 20 % av småhus i Sverige har en energideklaration. Syftet med studien är därför att utveckla en datadriven modell baserat på data från existerande energideklarationer, för att estimera den årliga värmeenergianvändningen i svenska småhus som saknar en energideklaration. Syftet är också att testa olika data-drivna maskininlärningsmodeller för att identifiera vilken som uppnår det lägsta felet. I modellerna har indata såsom golvyta, byggår, geografisk plats och värmekälla använts för att göra estimeringar. Resultaten visar att av de undersökta modellerna var Histogram-GBM den bästa för att estimera värmeanvändning, med ett RMSE på 24.33 kWh/m2 /år, MAE på 17.50 kWh/m2 /år, MAPE på 24.62 % och ett R2 -värde på 0.736. Resultaten visar också att när värmeanvändning estimeras för hela beståndet av småhus, minimeras det totala felet när alla kvarvarande fel summeras. Den bästa modellen, Histogram-GBM uppnådde ett MBE på 0.59 kWh/estimering, vilket motsvarar ett fel mindre än 0.005 %.

Traffic analysis is vital for enhancing the performance of transportation systems, where continuous evaluation of traffic states helps responsible road authorities detect and address issues. High-quality traffic data is key to analysis, as it aids in planning and investments. Traditionally, traffic data collection has been costly and limited. Nowadays, connected vehicles and mobile phones have transformed this process, enabling traffic data collection across large geographic regions without the need for dedicated measurement devices. The availability of large-scale and detailed traffic data allows for in-depth analysis using mathematical models. This thesis develops models to utilize available traffic data for transportation system improvements, aiming to enhance traffic conditions and road user experience. It utilizes data from link flows and travel times, applying models over large geographic areas. The thesis addresses transportation engineering issues through data-driven methods. The thesis proposes two methods for allocating electric vehicle charging stations using optimization and route sampling techniques. It introduces a new index for assessing travel time reliability. It shows how clustering analysis of descriptive travel time statistics can be used to detect different traffic states. Furthermore, this thesis presents a statistical model to estimate link flow propagation using measured link flow data, analyzing traffic influence across surrounding areas. The thesis also uses traffic simulation, focusing on combining speed cameras and probe vehicles for data collection and developing a model to identify probable routes based on hourly link flows. The thesis results highlight the importance of data-driven models in optimizing transportation systems and improving road user travel experiences.

Blending of partially cracked ammonia with biosyngas is an attractive strategy for improving NH3 combustion. In practice, products of biomass gasification and those of thermo-catalytic cracking of NH3 are subject to some compositional uncertainties. Despite their practical importance, so far, the effects of such uncertainties on combustion systems remained largely unexplored. Hence, this paper quantifies the effects of small compositional uncertainties of reactants upon combustion of partially cracked NH3/syngas/air mixtures. An uncertainty quantification method, based on polynomial chaos expansion and a data-driven model, is utilised to investigate the effects of uncertainty in fuel composition on the laminar flame speed (SL) and adiabatic flame temperature (Tad) at different inlet pressures (Pi). The analysis is then extended to the power output of a combined-cycle gas turbine fuelled by the reactants. It is found that 1.5% fuel compositional uncertainty can cause 12–21% of SL uncertainty depending on the inlet pressure. Furthermore, the effect of compositional uncertainty on Tad increases at higher ratios of H2 to NH3. Sensitivity analysis reveals that the uncertainty of CO contribution to SL uncertainty is higher than that of NH3, while the trend is reversed for the Tad uncertainty. In addition, the power output from the combined-cycle gas turbine system varies between 4 and 6% with 1.5% of fuel compositional uncertainty. This become more noticeable at elevated Pi [5–10 atm], particularly when the fuel mixture contains high H2 which is the main contributor to Tad variability., QC 20240520

Traffic analysis is vital for enhancing the performance of transportation systems, where continuous evaluation of traffic state helps responsible road authorities detect and address issues. High-quality traffic data aids in planning and investments. Traditionally, traffic data collection was costly and limited. Nowadays, connected vehicles and mobile phones have transformed this process, enabling traffic data collection across large geographic regions without the need for dedicated measurement devices. The availability of large-scale and detailed traffic data allows for in-depth analysis using mathematical models. This thesis develops models to utilize available traffic data for system improvements, aiming to enhance performance and road user experience. It explores data sources such as link flow data and travel time data, applying models over large geographic areas. The thesis addresses transportation engineering issues through data-driven methods. The thesis proposes two methods for allocating electric vehicle charging stations using optimization and route sampling techniques. It introduces a new index for assessing travel time reliability. It shows how clustering analysis of descriptive travel time statistics can be used to detect different traffic states. Furthermore, this thesis presents a statistical model to estimate link flow propagation using measured link flow data, analyzing traffic influence across surrounding areas. The final contributions of the thesis focus on traffic simulation, combining speed cameras and probe vehicles for data collection, and developing a model to identify probable routes based on hourly link flows. The results of the thesis highlight the importance of data-driven models in optimizing transportation systems and improving road user travel experiences.

Developing predictive models that can accurately simulate future positions in sports could revolutionise current modelling practices. This thesis endeavours to create a model capable of simulating football games two seconds into the future. By utilising tracking data, in combination with a player database from Football Manager, the goal of the project is to create sophisticated deep learning models with accurate predictions. The data-driven approach learns the movement of football players using neural networks (NN), and particularly networks with the long short-term memory (LSTM) architecture. By combining these complex models with features such as velocity, distance to ball, and tiredness, we achieve the desired results. The most complex model is an LSTM model, averaging a prediction error of 1.59 meters for the two-second time-frame. The results vary based on player position, with the movement of central midfielders and defensive midfielders being the hardest to predict accurately, while the movement of goalkeepers, forwards, and centre-backs tends to be more predictable. The findings of this thesis demonstrate the potential of machine learning in sport analytics, with the predictive modelling offering valuable insights in the realm of analytics. Future work could further enhance the predictive accuracy of the models, by the integration of event-data and the utilisation of team-based modelling.

The building sector in Sweden is responsible for 40 % of the final energy use, and space heating largely contributes to the total energy demand. Minimum energy performance standards and building codes are becoming more strict and comprehensive globally. In the European Union there are substantial goals of achieving a higher energy efficiency in the residential building sector. Energy Performance Certificates (EPCs) are a tool that can be used to track and assess building energy in a standardized way, to follow up on the energy efficiency of buildings. Privately owned single-family houses do not need to have a valid EPC, it is only mandatory to perform an assessment when the house is subject to sale, meaning that only about 20 % of single-family houses in Sweden have an EPC. The objective of the study is therefore to develop a data driven model based on existing EPC data, to estimate the annual space heating energy demand of Swedish single family houses that do not have an EPC. Sub-objectives include testing different data-driven machine learning models to identify which algorithm achieves the highest accuracy. Input features such as floor area, construction year, geographical location and heating source were used to make predictions. The results show that out of the investigated models, the Histogram-GBM performed the best when predicting the space heating energy of an individual house, with a RMSE of 24.33 kWh/m2 /year, MAE of 17.50 kWh/m2 /year, MAPE of 24.62 % and R2 -value of 0.736. When evaluating the results of predictions for the entire single-family housing stock, the errors of each individual house are hidden within the entire housing stock, minimizing the error of the total prediction. The best performing model, Histogram GBM, achieved a MBE of 0.59 kWh/prediction, equivalent to an error smaller than 0.005 %., I Sverige står byggnadssektorn för 40 % av den slutliga energianvändningen, och uppvärmning av byggnader är en stor andel av den totala energianvändningen. Minimikrav för energiprestanda och byggregler blir globalt allt mer strikta och omfattande. EU har betydande mål att uppnå högre energieffektivitet för alla typer av bostadshus. Energideklarationer (EPC) är ett verktyg som kan användas för att samla in och bedömma energianvändning i byggnader på ett standardiserat sätt, för att kunna följa upp energieffektivisering av byggnader. Till skillnad från andra byggnadstyper behöver privat ägda småhus inte ha en giltig energideklaration, de är endast obligatoriska att utfärda när huset säljs, vilket betyder att bara 20 % av småhus i Sverige har en energideklaration. Syftet med studien är därför att utveckla en datadriven modell baserat på data från existerande energideklarationer, för att estimera den årliga värmeenergianvändningen i svenska småhus som saknar en energideklaration. Syftet är också att testa olika data-drivna maskininlärningsmodeller för att identifiera vilken som uppnår det lägsta felet. I modellerna har indata såsom golvyta, byggår, geografisk plats och värmekälla använts för att göra estimeringar. Resultaten visar att av de undersökta modellerna var Histogram-GBM den bästa för att estimera värmeanvändning, med ett RMSE på 24.33 kWh/m2 /år, MAE på 17.50 kWh/m2 /år, MAPE på 24.62 % och ett R2 -värde på 0.736. Resultaten visar också att när värmeanvändning estimeras för hela beståndet av småhus, minimeras det totala felet när alla kvarvarande fel summeras. Den bästa modellen, Histogram-GBM uppnådde ett MBE på 0.59 kWh/estimering, vilket motsvarar ett fel mindre än 0.005 %.

Developing predictive models that can accurately simulate future positions in sports could revolutionise current modelling practices. This thesis endeavours to create a model capable of simulating football games two seconds into the future. By utilising tracking data, in combination with a player database from Football Manager, the goal of the project is to create sophisticated deep learning models with accurate predictions. The data-driven approach learns the movement of football players using neural networks (NN), and particularly networks with the long short-term memory (LSTM) architecture. By combining these complex models with features such as velocity, distance to ball, and tiredness, we achieve the desired results. The most complex model is an LSTM model, averaging a prediction error of 1.59 meters for the two-second time-frame. The results vary based on player position, with the movement of central midfielders and defensive midfielders being the hardest to predict accurately, while the movement of goalkeepers, forwards, and centre-backs tends to be more predictable. The findings of this thesis demonstrate the potential of machine learning in sport analytics, with the predictive modelling offering valuable insights in the realm of analytics. Future work could further enhance the predictive accuracy of the models, by the integration of event-data and the utilisation of team-based modelling.

Current scientific workflow systems do not typically integrate simulation-centric and data-centric aspects due to their very different software/infrastructure requirements. A transparent integration of such components into a single end-to-end workflow would lead to a more efficient and automated way for generating insights from large simulation data. This work presents a complex case study related to extreme events analysis of future climate data that integrates in the same workflow numerical simulations, Big Data analytics and Machine Learning models. The case study is being implemented in the context of the eFlows4HPC project using the project’s software stack for deployment and orchestration of the workflow. The solution implemented in the project has shown to simplify the development and execution of end-to-end climate workflows with heterogeneous software requirements. Moreover, such an approach can, in the long term, increase the reuse of workflows by scientists and their portability over different HPC infrastructures., Thiswork has received funding from the European High-Performance Computing Joint Undertaking (JU) under grant agreement No 955558. The JU receives support from the European Union’s Horizon 2020 research and innovation programme and Spain, Germany, France, Italy, Poland, Switzerland and Norway. In Spain, it has received complementary funding from MCIN/AEI/10.13039/501100011033, Spain and the European Union NextGenerationEU/PRTR (contracts PCI2021-121957, PCI2021-121931, PCI2021-121944, and PCI2021- 121927). In Italy, it has been preliminary approved for complimentary funding by Ministero dello Sviluppo Economico (MiSE) (ref. project prop. 2659). The authors also acknowledge financial support by MCIN/AEI /10.13039/501100011033, Spain through the "Severo Ochoa Programme for Centres of Excellence in R&D" under Grant CEX2021-001148-S, the Spanish Government (contract PID2019- 107255 GB) and by Generalitat de Catalunya (contract 2021-SGR-00412)., Peer Reviewed, Postprint (author's final draft)

Snow avalanches are defined as a mass of snow that rapidly flows down a sloping surface, such as a hill or a mountain. Depending on their velocity and volume they can have highly destructive force and pose major threat to people, property, infrastructure, and ecosystems. Andorra is a small, mountainous country located in the Axial Zone of the central Pyrenees mountain range in southwestern Europe, enclosed by France and Spain. It is exposed to several natural hazards including snow avalanches. The latter occur in the country several times per year and the popularity of winter tourism in the area results in high risk levels. Due to the frequency of the events, Andorra maintains a long avalanche inventory since the 1980’s. Although the authorities have developed advanced regulations in terms of managing risk, have put in place a series of monitoring systems, as well as protection measures for avalanches, currently, there is not a forecasting tool available, and the issued warnings are based on weather observations and experience. The aim of this master thesis has been to evaluate the suitability of data-driven models to assess susceptibility and map initiation areas of snow avalanches in Andorra. This aim was formulated as a classification task and the machine learning algorithms that were brought to the test included the Decision Tree, Random Forest, Adaptive Boosting, Gradient Boosting, Extreme Gradient Boosting, Logistic Regression, Support Vector Classifier, and Neural Network. The models received as input topographical information of the study area, derived by GIS analysis of a 5x5 m DEM and a land cover map, interpolated weather station data, snow cover information from Landsat 8 and the susceptibility to shallow landslides acquired by Shalstab. The obtained results were satisfactory, with achieved accuracies of 80% or higher for all the models.

Uncertainty is a common problem in spatial modeling and geographical information systems (GIS). Furthermore, urban gain modeling (UGM) contains various dimensions and components of uncertainties. Data sampling is important in UGM, and may cause the results of the models to contain many uncertainties as well as affects their precision and accuracy. A poorly sampled or biased dataset can lead to inaccurate predictions and decreased performance of the models. This paper aims to present and develop novel strategies for sampling and building training datasets that can enhance the performance of data-driven models. In other words, the present study used maximum entropy (ME) and ecological niche factor analysis (ENFA) models to select pure non-change samples with minimal uncertainty for training datasets in UGM of Isfahan and Tabriz cities in Iran. The urban gain of two time intervals of 1992–2002 and 2002–2012 were used for Tabriz City and two time intervals of 1994–2004 and 2004–2014 for Isfahan City. Nine and 14 urban gain drivers were used in the UGM of Isfahan and Tabriz cities, respectively. After the ME and ENFA models produced a training dataset with change and non-change samples with the lowest uncertainty, three well-known models, namely random forest (RF), artificial neural network (ANN), and support vector machine (SVM) were used for the modeling. Moreover, the ME and ENFA models that were used to investigate the uncertainty of the sampling procedure were used as the one-class prediction models. Compared to extant studies, the proposed ME – based sampling strategy increased the area under the receiver operating characteristic curve (AUROC), figure of merit, producer’s accuracy, and overall accuracy by 5.5%, 5%, 5%, and 3%, respectively, in the validation phase of Isfahan City and by 5%, 6%, 14%, and 17%, respectively, for Tabriz City. For Isfahan, the accuracies of ME (AUROC = 0.649) and ENFA (AUROC = 0.661) one – class models were closer to that of the ANN – M, Validerad;2023;Nivå 2;2023-06-14 (joosat);Licens fulltext: CC BY License

Uncertainty is a common problem in spatial modeling and geographical information systems (GIS). Furthermore, urban gain modeling (UGM) contains various dimensions and components of uncertainties. Data sampling is important in UGM, and may cause the results of the models to contain many uncertainties as well as affects their precision and accuracy. A poorly sampled or biased dataset can lead to inaccurate predictions and decreased performance of the models. This paper aims to present and develop novel strategies for sampling and building training datasets that can enhance the performance of data-driven models. In other words, the present study used maximum entropy (ME) and ecological niche factor analysis (ENFA) models to select pure non-change samples with minimal uncertainty for training datasets in UGM of Isfahan and Tabriz cities in Iran. The urban gain of two time intervals of 1992–2002 and 2002–2012 were used for Tabriz City and two time intervals of 1994–2004 and 2004–2014 for Isfahan City. Nine and 14 urban gain drivers were used in the UGM of Isfahan and Tabriz cities, respectively. After the ME and ENFA models produced a training dataset with change and non-change samples with the lowest uncertainty, three well-known models, namely random forest (RF), artificial neural network (ANN), and support vector machine (SVM) were used for the modeling. Moreover, the ME and ENFA models that were used to investigate the uncertainty of the sampling procedure were used as the one-class prediction models. Compared to extant studies, the proposed ME – based sampling strategy increased the area under the receiver operating characteristic curve (AUROC), figure of merit, producer’s accuracy, and overall accuracy by 5.5%, 5%, 5%, and 3%, respectively, in the validation phase of Isfahan City and by 5%, 6%, 14%, and 17%, respectively, for Tabriz City. For Isfahan, the accuracies of ME (AUROC = 0.649) and ENFA (AUROC = 0.661) one – class models were closer to that of the ANN – M, Validerad;2023;Nivå 2;2023-06-14 (joosat);Licens fulltext: CC BY License

The prediction of ship motions and quiescent periods, is of paramount importance for the maritime industry. The capability to predict these events sufficiently in advance has the potential to improve the safety and efficiency of several marine operations, such as landing and take-off on aircraft carriers, transfer of cargo, and mating operations between ships. Several models have been proposed in the literature for the prediction of ship motions and quiescent period. This work will review them by first grouping them into three main categories (i.e., physical, data-driven, and hybrid models) and then by detailing the most recent and relevant ones describing the advantages and disadvantages of each approach. Review concludes with the open problems and future perspectives of this important field of research., Ship Design, Production and Operations

In this work, a purely data-driven approach to mapping out the state of a dynamical system over a set of chosen parameters is presented and demonstrated along a case study using real-world experimental data from a friction brake system. Complex engineering systems often exhibit a rich bifurcation behavior with respect to one or several parameters, which is difficult to grasp using experimental approaches or numerical simulations. At the same time, the growing need for energy-efficient machines that can operate under varying or extreme environmental conditions also calls for a better understanding of these systems to avoid critical transitions. The proposed method combines machine learning techniques with synthetic data augmentation to create a complete state map for a dynamical system. First, a machine learning model is trained on experimental data, picking up hidden mechanisms and complex parametric relations of the underlying dynamical system. The model is then exploited to assess the state of the system for a set of synthetically generated data to obtain a state map over the complete space spanned by the chosen parameters. In addition, an extension of the concept to a probability state map is introduced. The results indicate that the proposed method can uncover hidden variables which drive dynamical transitions between different states of a system that were previously inaccessible., Federal Ministry of Education and Research (BMBF), Ministère de l'enseignement supérieur de la recherche et de l'innovation (MESR)

In conjunction with tides, storm surge is one major driver of coastal flooding associated with storm events. Because local inundation is strongly modulated by the local shape of the coastline and the bathymetric slope, accurate storm surge predictions using traditional numerical models require the use of very fine grids and are hence resource intensive. Therefore, the performance of a live prediction system based on such methods will likely be subject to a trade-off between prediction accuracy, prediction speed and cost., This study explores the use of data driven methods as an alternative to numerical models to reconstruct the daily storm surge maximum levels along the entire coast of New Zealand. Firstly, several atmospheric predictors are utilized that incorporate different variables, time lags and spatial domains, using 3 statistical models, in a selected number of locations in New Zealand, to find the combination that optimizes the reconstruction. Finally, the storm surge daily maxima are reconstructed with the different statistical models along the entire coast, using the best performing predictor., Results show very good performance for the best atmospheric predictor and statistical model, providing average values of 0.88 for the Pearson correlation coefficient and 4.3 cm for the root mean squared error metric (RMSE) (the average value for the RMSE in the 99% percentile is 8.2 cm). For the Kling–Gupta Efficiency (KGE; incorporating 3 sub-metrics: correlation, bias term and variability term), which is the metric used to rank the models, the average value is 0.82., Our results highlight the suitability of data driven models to simulate storm surge maximum levels, and prove the methodology is appropriate for finding a well performing atmospheric predictor that is able for reconstruct these values. Moreover, this methodology can be also applied to new variables, regions and problems, as there are no physical restrictions on the used predictors nor predictands.

Monitoring and evaluating the biofouling state and its effects on the vessel's hull and propeller performance is a crucial problem that attracts the attention of both academy and industry. Effective and reliable tools to address this would allow a timely cleaning procedure able to trade off costs, efficiency, and environmental impacts. In this paper, the authors carry out a critical review, accompanied with summary tables, of the biofouling problem with a particular focus on the shipping industry and the state-of-the-art techniques for monitoring and evaluating the biofouling state and its effects on the vessel's hull and propeller performance. In particular, different techniques are grouped according to the three main families of numerical models that have been designed and exploited in the literature: Physical Models (i.e., models relying on the mechanistic knowledge of the phenomena), Data-Driven Models (i.e., models relying on historical data about the phenomena together with Artificial Intelligence), and Hybrid Models (i.e., a hybridisation between Physical and Data-Driven Models). A conclusion from the performed review, open problems, and future direction of this field of research is detailed at the end of the review., Ship Design, Production and Operations

Reactivity controlled compression ignition (RCCI) is a highly efficient and clean combustion concept, which enables the use of a wide range of renewable fuels. Consequently, this promising dual fuel combustion concept is of great interest for realizing climate neutral future transport. RCCI is very sensitive for operating conditions and requires advanced control strategies to guarantee stable and safe operation. For real-world RCCI implementation, we face control challenges related to transients and varying ambient conditions. Currently, a multivariable air–fuel path controller that can guarantee robust RCCI engine operation is lacking. In this work, we present a RCCI engine controller, which combines static decoupling and a diagonal MIMO feedback controller. For control design, a frequency domain-based approach is presented, which explicitly deals with cylinder-to-cylinder variations using data-driven, cylinder-individual combustion models. This approach enables a systematic trade-off between fast and robust performance and gives clear design criteria for stable operation. The performance of the developed multivariable engine controller is demonstrated on a six-cylinder diesel-E85 RCCI engine. From experimental results, it is concluded that the RCCI engine controller accurately tracks the five desired combustion and air path parameters, simultaneously. For the studied transient cycle, this results in 12.8% reduction in NOx emissions and peak in-cylinder pressure rise rates are reduced by 3.8 bar/deg CA. Compared to open-loop control, the stable and safe operating range is increased from 25 ∘C up to 35 ∘C intake manifold temperature and maximal load range is increased by 14.7% up to BMEP = 14.8 bar

Only a subset of degrees of freedom are typically accessible or measurable in real-world systems. As a consequence, the proper setting for empirical modeling is that of partially-observed systems. Notably, data-driven models consistently outperform physics-based models for systems with few observable degrees of freedom; e.g. hydrological systems. Here, we provide an operator-theoretic explanation for this empirical success. To predict a partially-observed system’s future behavior with physics-based models, the missing degrees of freedom must be explicitly accounted for using data assimilation and model parametrization. Data-driven models, in contrast, employ delay-coordinate embeddings and their evolution under the Koopman operator to implicitly model the effects of the missing degrees of freedom. We describe in detail the statistical physics of partial observations underlying data-driven models using novel maximum entropy and maximum caliber measures. The resulting nonequilibrium Wiener projections applied to the Mori-Zwanzig formalism reveal how data-driven models may converge to the true dynamics of the observable degrees of freedom. Additionally, this framework shows how data-driven models infer the effects of unobserved degrees of freedom implicitly, in much the same way that physics models infer the effects explicitly. This provides a unified implicit-explicit modeling framework for predicting partially-observed systems, with hybrid physics-informed machine learning methods combining both implicit and explicit aspects.

Risk modelling is an active area of research in computational finance, which involves design of models for quantifying and forecasting the uncertainty associated with investments. Unlike traditional financial assets, the risk modelling for blockchain powered digital assets is rarely studied. The main contributions of this thesis lie in design and application of the novel data-driven models to risk forecasting of the price and algorithmic returns of cryptocurrency assets. In our first study, we proposed data-driven and neuro-volatility risk forecasting models for cryptocurrencies. Our results are quantified with two widely used risk metrics - Value at Risk (VaR) and Expected Shortfall (ES). Both the data-driven and neuro-volatility estimates demonstrated a significantly higher risk for cryptocurrencies as compared to traditional technology stocks. We observed that the data-driven models produced better forecasts for cryptocurrencies, while better forecasts resulted for the regular stocks and indexes with the neuro-volatility model. Also, the data-driven models are more efficient in terms of computation time. We did another study with the algorithmic returns for simple moving average crossover strategy using Sharpe ratio as risk quantifier, which we extended to develop a novel algorithmic trading strategy for cryptocurrencies. As a case study, we applied this knowledge to decentralized finance (DeFi) ecosystem. The main findings from the studies indicate superiority of the data-driven approach proposed in this research for the risk forecasting problems. The data-driven approach performed equivalently well or better compared to neuro volatility models in different studies reported in this thesis and also that the data-driven approach was significantly efficient in terms of computation time in each of these studies.

Reactivity controlled compression ignition (RCCI) is a highly efficient and clean combustion concept, which enables the use of a wide range of renewable fuels. Consequently, this promising dual fuel combustion concept is of great interest for realizing climate neutral future transport. RCCI is very sensitive for operating conditions and requires advanced control strategies to guarantee stable and safe operation. For real-world RCCI implementation, we face control challenges related to transients and varying ambient conditions. Currently, a multivariable air–fuel path controller that can guarantee robust RCCI engine operation is lacking. In this work, we present a RCCI engine controller, which combines static decoupling and a diagonal MIMO feedback controller. For control design, a frequency domain-based approach is presented, which explicitly deals with cylinder-to-cylinder variations using data-driven, cylinder-individual combustion models. This approach enables a systematic trade-off between fast and robust performance and gives clear design criteria for stable operation. The performance of the developed multivariable engine controller is demonstrated on a six-cylinder diesel-E85 RCCI engine. From experimental results, it is concluded that the RCCI engine controller accurately tracks the five desired combustion and air path parameters, simultaneously. For the studied transient cycle, this results in 12.8% reduction in NOx emissions and peak in-cylinder pressure rise rates are reduced by 3.8 bar/deg CA. Compared to open-loop control, the stable and safe operating range is increased from 25 ∘C up to 35 ∘C intake manifold temperature and maximal load range is increased by 14.7% up to BMEP = 14.8 bar

Accurate, reliable, and computationally inexpensive models of the dynamic state of combustion engines are a fundamental tool to investigate new engine designs, develop optimal control strategies, and monitor their performance. The use of those models would allow to improve the engine cost-efficiency trade-off, operational robustness, and environmental impact. To address this challenge, two state-of-the-art alternatives in literature exist. The first one is to develop high fidelity physical models (e.g., mean value engine, zero-dimensional, and one-dimensional models) exploiting the physical principles that regulate engine behaviour. The second one is to exploit historical data produced by the modern engine control and automation systems or by high-fidelity simulators to feed data-driven models (e.g., shallow and deep machine learning models) able to learn an accurate digital twin of the system without any prior knowledge. The main issues of the former approach are its complexity and the high (in some case prohibitive) computational requirements. While the main issues of the latter approach are the unpredictability of their behaviour (guarantees can be proved only for their average behaviour) and the need for large amount of historical data. In this work, following a recent promising line of research, we describe a modelling framework that is able to hybridise physical and data driven models, delivering a solution able to take the best of the two approaches, resulting in accurate, reliable, and computationally inexpensive models. In particular, we will focus on modelling the dynamic state of a four-stroke diesel engine testing the performance (both in terms of accuracy, reliability, and computational requirements) of this solution against state-of-the-art physical modelling approaches on real-world operational data., Green Open Access added to TU Delft Institutional Repository 'You share, we take care!' - Taverne project https://www.openaccess.nl/en/you-share-we-take-care Otherwise as indicated in the copyright section: the publisher is the copyright holder of this work and the author uses the Dutch legislation to make this work public., Ship Design, Production and Operations

Deterministic models based on the laws of physics, as well as data-driven models, are often used to assess the current state of vessels and their systems, as well as predict their future behaviour. Predictive maintenance methodologies (i.e., Condition Based Maintenance) and advanced control strategies (i.e., Model Predictive Control) are built upon the use of such numerical tools to identify ensuing performance shifts. In fact, forecasting near-future performance can substantially contribute to enhancing operational efficiency and enabling advanced system control. Data from modern sensor technology, which has become more readily available, combined with automatic control systems capable of prescribing optimal control strategies, can improve vessel operation and reduce energy consumption. A data-driven model that relies on recent advances in Artificial Intelligence, Machine Learning, and Data Mining, leveraging historical observations is employed to forecast a vessel’s onboard power generation trends as a function of the past, present, and future behaviour of a ship and its systems. To prove the framework, the proposed methodology is tested on real data collected from the Integrated Platform Management System of an Oceangoing Patrol Vessel of the Royal Netherlands Navy. The developed data-driven model is achieves high forecasting accuracy in the near-term. The authors foresee that the proposed methodology could be used as part of an electric energy control strategy, within a more integrated and intelligent mission planning framework., Ship Design, Production and Operations

Monitoring and evaluating the biofouling state and its effects on the vessel's hull and propeller performance is a crucial problem that attracts the attention of both academy and industry. Effective and reliable tools to address this would allow a timely cleaning procedure able to trade off costs, efficiency, and environmental impacts. In this paper, the authors carry out a critical review, accompanied with summary tables, of the biofouling problem with a particular focus on the shipping industry and the state-of-the-art techniques for monitoring and evaluating the biofouling state and its effects on the vessel's hull and propeller performance. In particular, different techniques are grouped according to the three main families of numerical models that have been designed and exploited in the literature: Physical Models (i.e., models relying on the mechanistic knowledge of the phenomena), Data-Driven Models (i.e., models relying on historical data about the phenomena together with Artificial Intelligence), and Hybrid Models (i.e., a hybridisation between Physical and Data-Driven Models). A conclusion from the performed review, open problems, and future direction of this field of research is detailed at the end of the review., Ship Design, Production and Operations

Risk modelling is an active area of research in computational finance, which involves design of models for quantifying and forecasting the uncertainty associated with investments. Unlike traditional financial assets, the risk modelling for blockchain powered digital assets is rarely studied. The main contributions of this thesis lie in design and application of the novel data-driven models to risk forecasting of the price and algorithmic returns of cryptocurrency assets. In our first study, we proposed data-driven and neuro-volatility risk forecasting models for cryptocurrencies. Our results are quantified with two widely used risk metrics - Value at Risk (VaR) and Expected Shortfall (ES). Both the data-driven and neuro-volatility estimates demonstrated a significantly higher risk for cryptocurrencies as compared to traditional technology stocks. We observed that the data-driven models produced better forecasts for cryptocurrencies, while better forecasts resulted for the regular stocks and indexes with the neuro-volatility model. Also, the data-driven models are more efficient in terms of computation time. We did another study with the algorithmic returns for simple moving average crossover strategy using Sharpe ratio as risk quantifier, which we extended to develop a novel algorithmic trading strategy for cryptocurrencies. As a case study, we applied this knowledge to decentralized finance (DeFi) ecosystem. The main findings from the studies indicate superiority of the data-driven approach proposed in this research for the risk forecasting problems. The data-driven approach performed equivalently well or better compared to neuro volatility models in different studies reported in this thesis and also that the data-driven approach was significantly efficient in terms of computation time in each of these studies.

With the embrace of the performance-based seismic design as the state-of-the-art design method, recent emphasis has been placed on eliminating its drawbacks and facilitating its application in practice. This study aims to propose an alternative design method: performance-based analytics-driven seismic design, which is applied to steel moment resisting frame buildings. First, the seismic performance of self-centering (with post-tensioned connections) and conventional moment resisting frames (with reduced-beam section connection) is comparatively assessed. The comparison indicates that the economic benefit for adopting the post-tensioned connection is not significant. Then, an end-to-end computational platform, which automates the seismic design, nonlinear structural model construction, and response simulation (static and dynamic) of steel moment resisting frames is developed. Using this platform, a comprehensive database is developed, which includes 621 special steel moment resisting frames designed in accordance with modern codes and standards and their corresponding nonlinear structural models and seismic responses (i.e., peak story drifts, peak floor accelerations, and residual story drifts). Using this database, the efficacy of mechanics-based, data-driven, and hybrid (combination of mechanics-based and data driven) approaches to estimating the seismic drift demand are evaluated. The evaluation results reveal that the hybrid approach has the best performance whereas the mechanics-based model has the lowest performance. Next, a set of non-parametric and parametric surrogate models are developed for estimating the engineering demand parameter distributions. A comparative assessment of the proposed surrogate models and the simplified analysis method proposed by FEMA P-58 is conducted to demonstrate the superior predictive performance of the former. Finally, the effect of various design variables on the collapse performance of steel moment resisting frames are evalua

The number of installed Floating Offshore Wind Turbines (FOWTs) has doubled since 2017, quadrupling the total installed capacity, and is expected to increase significantly over the next decade. Consequently, there is a growing consideration towards the main challenges for FOWT projects: monitoring the system's integrity, extending the lifespan of the components, and maintaining FOWTs safely at scale. Effectively and efficiently addressing these challenges would unlock the wide-scale deployment of FOWTs. In this work, we focus on one of the most critical components of the FOWTs, the Mooring Lines (MoLs), which are responsible for fixing the structure to the seabed. The primary mechanical failure mechanisms in MoLs are extreme load and fatigue, both of which are functions of the axial tension. An effective solution to detect long term drifts in the mechanical response of the MoLs is to develop a Digital Twin (DT) able to accurately predict the behaviour of the healthy system to compare with the actual one. Authors will leverage operational data collected from the world's first commercial floating wind farm (Hywind Pilot Park1) in 2018, to investigate the effectiveness of the DT for the prediction of the MoL axial tension. The DT will be developed using state-of-the-art data-driven methods, and results based on real operational data will support our proposal., Accepted Author Manuscript, Ship Design, Production and Operations

A data-driven scheme for modeling electrical impedance in biosensors is presented by a subspace method working with the singular value decomposition of structured voltage and current data. Contrary to the classical electrical impedance spectroscopy (EIS) methods, our scheme uses simple instrumentation, works in time-domain, provides fast results, and does not require semi-empirical assumptions to retrieve structured models from data. We show how data-driven models exhibit a close relationship with lumped-element circuits, encoding dielectric and conductive properties detected by the sensor in the range from 10 kHz up to 10 MHz. Performance results are shown for calibration networks and two case studies: (i) a buffer solution, and (ii) a biological cell suspension. Finally, the viability of the scheme is discussed when compared with the classical EIS method.

The number of installed floating offshore wind turbines (FOWTs) has doubled since 2017, quadrupling the total installed capacity, and is expected to increase significantly over the next decade. Consequently, there is a growing consideration towards the main challenges for FOWT projects: monitoring the system’s integrity, extending the lifespan of the components, and maintaining FOWTs safely at scale. Effectively and efficiently addressing these challenges would unlock the wide-scale deployment of FOWTs. In this work, we focus on one of the most critical components of the FOWTs, the Mooring Lines (MoLs), which are responsible for fixing the structure to the seabed. The primary mechanical failure mechanisms in MoLs are extreme load and fatigue, both of which are functions of the axial tension. An effective solution to detect long-term drifts in the mechanical response of the MoLs is to develop a Digital Twin (DT) able to accurately predict the behaviour of the healthy system to compare with the actual one. Moreover, we will develop another DT able to accurately predict the near future axial tension as an effective tool to improve the lifespan of the MoLs and the safety of FOWT maintenance operations. In fact, by changing the FOWT operational settings, according to the DT prediction, operators can increase the lifespan of the MoLs by reducing the stress and, additionally, in the case where FOWT operational maintenance is in progress, the prediction from the DT can serve as early safety warning to operators. Authors will leverage operational data collected from the world’s first commercial floating-wind farm [the Hywind Pilot Park (https://www.equinor.com/en/what-we-do/floating-wind/hywind-scotland.html.)] in 2018, to investigate the effectiveness of DTs for the prediction of the MoL axial tension for the two scenarios depicted above. The DTs will be developed using state-of-the-art data-driven methods, and results based on real operational data will support our prop, Ship Design, Production and Operations

Evaluating an adequate computer model representing the characteristics of a subsurface reservoir is of great importance in the oil and gas industry. With recent data acquisition developments, high-resolution geological models can be generated for the subsurface reservoirs; however, uncertainties of geological data and too many details in the model have brought up remarkable limitations for application of high-fidelity geo-models in reservoir management. To generate efficient low-resolution models, which are relatively representative of the reservoir dynamic behaviour, data-driven models are proposed. The data-driven models are trained principally based on the production history of the reservoir with primary knowledge of structural information. The Data-driven Discrete Well Affinity (DiWA) model is designed to generate efficient proxy models that require basic geological information and full production history of the reservoir. Since it benefits the computational power of the Delft Advanced Terra Simulator (DARTS) and adjoint gradient-based optimization, the DiWA model has become an efficient framework in terms of accuracy and computational costs. This research shows that the DiWA model can relatively reconstruct the reference reservoir parameters out of an ensemble of priors, and the final results can be improved utilizing closer prior knowledge to the true response. Additionally, it is demonstrated that the performance of the DiWA model in a complex, realistic oil field can be enhanced using a comprehensive objective function, applying appropriate well controls, understanding the underlying driving mechanisms, using sufficient degrees of freedom, and applying consistent boundary conditions., Civil Engineering

The increasing amount of data on biofilter treatment performance over the past decade has made it possible to use data-driven approaches to explore the relationships between biofilter performance and a range of input variables. The knowledge gap lies in lack of models to predict the biofilter performance considering both design and operational variables, especially for heavy metals. In this study, we tested three machine learning (ML) approaches, namely multilinear regression (MLR), artificial neural network (NN), and random forest (RF), to predict biofilter outflow concentrations of heavy metals (Cd, Cr, Cu, Fe, Ni, Pb and Zn) using a range of design and operational factors as input variables. The results show that RF performed relatively better than other two models, with median Nash-Sutcliffe Efficiency (NSE) values of 0.995, 0.317, 0.762, 0.636, 0.726, 0.896 and 0.656 for Cd, Cr, Cu, Fe, Ni, Pb and Zn, respectively during model training. However, all the models were less accurate during model validation, with the better performance found for Cd (average NSE=0.964), Zn (0.530) and Ni (0.393) and poorer performance observed for Cu (0.219), Pb (0.058), Fe (-0.054) and Cr (-0.062). Infiltration rate (IR) and inflow concentration (Cin) were sensitive to all pollutants’ removal in biofilters. The ratio of system size to catchment size was also found to be important for Zn, Ni and Cd, while ponding depth was an important variable for Cd. Based on thousands of hypothetical design and operational scenarios (generated using raw data), the best ML models were used to predict the biofilter outflow concentrations and estimate the risk quotient (RQ) values with regards to reuse of treated stormwater for various purposes. Results suggest that biofilters were able to reduce health risks associated with heavy metals in stormwater and therefore produce reliable water fit for reuses such as irrigation, swimming, and toilet flushing. Modelling results showed that biofilt

Railways are currently experiencing higher demands on safety, infrastructure performance, network capacity and service quality, etc. As a result, a high level of track availability, service quality and infrastructure robustness against unexpected events is required, at reduced cost, to meet the demands. The track geometry is one of the critical parameters influencing the performance of the infrastructure and the capacity and quality of service of the railway network. The track geometry continuously degrades and deviates from the designed level over time and under the influence of various factors. A poor track geometry negatively affects the ride quality and passenger comfort and increases the risk of train derailment. To ensure a safe and reliable track for the trains, the track geometry should be kept in an acceptable condition. This necessitates the development of an applicable and effective tamping regime, as the main maintenance action for track geometry rejuvenation. An effective tamping regime enables enhancement of the availability and safety performance of the track, to control the track degradation and to restore the damaged track to an operational state, at the lowest possible cost. The purpose of the research conducted for this thesis was to develop data-driven methodologies and optimization approaches for the development of a plan for predictive railway track geometry maintenance. First, a study was carried out to investigate the application of artificial neural networks to estimation of the track geometry degradation rate in spatial space by considering the operational and constructional covariates. In addition, the relative importance of the observed covariates for the track geometry degradation was also explored. The results showed that the maintenance history, the degradation value after tamping and the frequency of the trains passing along the track were the most important covariates affecting the track geometry degradation rate. Second, a case stu

With the embrace of the performance-based seismic design as the state-of-the-art design method, recent emphasis has been placed on eliminating its drawbacks and facilitating its application in practice. This study aims to propose an alternative design method: performance-based analytics-driven seismic design, which is applied to steel moment resisting frame buildings. First, the seismic performance of self-centering (with post-tensioned connections) and conventional moment resisting frames (with reduced-beam section connection) is comparatively assessed. The comparison indicates that the economic benefit for adopting the post-tensioned connection is not significant. Then, an end-to-end computational platform, which automates the seismic design, nonlinear structural model construction, and response simulation (static and dynamic) of steel moment resisting frames is developed. Using this platform, a comprehensive database is developed, which includes 621 special steel moment resisting frames designed in accordance with modern codes and standards and their corresponding nonlinear structural models and seismic responses (i.e., peak story drifts, peak floor accelerations, and residual story drifts). Using this database, the efficacy of mechanics-based, data-driven, and hybrid (combination of mechanics-based and data driven) approaches to estimating the seismic drift demand are evaluated. The evaluation results reveal that the hybrid approach has the best performance whereas the mechanics-based model has the lowest performance. Next, a set of non-parametric and parametric surrogate models are developed for estimating the engineering demand parameter distributions. A comparative assessment of the proposed surrogate models and the simplified analysis method proposed by FEMA P-58 is conducted to demonstrate the superior predictive performance of the former. Finally, the effect of various design variables on the collapse performance of steel moment resisting frames are evalua

En este proyecto de investigación, se utilizaron un modelo físico (FSLAM) y cuatro modelos basados en datos (regresión logística, SVC, árbol de clasificación y bosque aleatorio) para mapear la susceptibilidad a los deslizamientos para un área de estudio ubicada en los Pirineos Catalanes. Seguidamente, se compararon los resultados de todos los modelos para determinar cuál funcionó mejor y en qué condiciones. También se discutieron las ventajas y desventajas de cada modelo, así como las limitaciones de sus productos finales., In this research project, a physically-based (FSLAM) and four data-driven models (logistic regression, SVC, classification tree and random forest) were used to map landslide susceptibility for a case study area located in the Catalan Pyrenees. The results for all models were then compared in order to determine which performed best and under which conditions. The advantages and disadvantages of each model were also discussed as well as the limitations of their end products.

A data-driven scheme for modeling electrical impedance in biosensors is presented by a subspace method working with the singular value decomposition of structured voltage and current data. Contrary to the classical electrical impedance spectroscopy (EIS) methods, our scheme uses simple instrumentation, works in time-domain, provides fast results, and does not require semi-empirical assumptions to retrieve structured models from data. We show how data-driven models exhibit a close relationship with lumped-element circuits, encoding dielectric and conductive properties detected by the sensor in the range from 10 kHz up to 10 MHz. Performance results are shown for calibration networks and two case studies: (i) a buffer solution, and (ii) a biological cell suspension. Finally, the viability of the scheme is discussed when compared with the classical EIS method.

The number of installed Floating Offshore Wind Turbines (FOWTs) has doubled since 2017, quadrupling the total installed capacity, and is expected to increase significantly over the next decade. Consequently, there is a growing consideration towards the main challenges for FOWT projects: monitoring the system's integrity, extending the lifespan of the components, and maintaining FOWTs safely at scale. Effectively and efficiently addressing these challenges would unlock the wide-scale deployment of FOWTs. In this work, we focus on one of the most critical components of the FOWTs, the Mooring Lines (MoLs), which are responsible for fixing the structure to the seabed. The primary mechanical failure mechanisms in MoLs are extreme load and fatigue, both of which are functions of the axial tension. An effective solution to detect long term drifts in the mechanical response of the MoLs is to develop a Digital Twin (DT) able to accurately predict the behaviour of the healthy system to compare with the actual one. Authors will leverage operational data collected from the world's first commercial floating wind farm (Hywind Pilot Park1) in 2018, to investigate the effectiveness of the DT for the prediction of the MoL axial tension. The DT will be developed using state-of-the-art data-driven methods, and results based on real operational data will support our proposal., Accepted Author Manuscript, Ship Design, Production and Operations

The number of installed floating offshore wind turbines (FOWTs) has doubled since 2017, quadrupling the total installed capacity, and is expected to increase significantly over the next decade. Consequently, there is a growing consideration towards the main challenges for FOWT projects: monitoring the system’s integrity, extending the lifespan of the components, and maintaining FOWTs safely at scale. Effectively and efficiently addressing these challenges would unlock the wide-scale deployment of FOWTs. In this work, we focus on one of the most critical components of the FOWTs, the Mooring Lines (MoLs), which are responsible for fixing the structure to the seabed. The primary mechanical failure mechanisms in MoLs are extreme load and fatigue, both of which are functions of the axial tension. An effective solution to detect long-term drifts in the mechanical response of the MoLs is to develop a Digital Twin (DT) able to accurately predict the behaviour of the healthy system to compare with the actual one. Moreover, we will develop another DT able to accurately predict the near future axial tension as an effective tool to improve the lifespan of the MoLs and the safety of FOWT maintenance operations. In fact, by changing the FOWT operational settings, according to the DT prediction, operators can increase the lifespan of the MoLs by reducing the stress and, additionally, in the case where FOWT operational maintenance is in progress, the prediction from the DT can serve as early safety warning to operators. Authors will leverage operational data collected from the world’s first commercial floating-wind farm [the Hywind Pilot Park (https://www.equinor.com/en/what-we-do/floating-wind/hywind-scotland.html.)] in 2018, to investigate the effectiveness of DTs for the prediction of the MoL axial tension for the two scenarios depicted above. The DTs will be developed using state-of-the-art data-driven methods, and results based on real operational data will support our prop, Ship Design, Production and Operations

Evaluating an adequate computer model representing the characteristics of a subsurface reservoir is of great importance in the oil and gas industry. With recent data acquisition developments, high-resolution geological models can be generated for the subsurface reservoirs; however, uncertainties of geological data and too many details in the model have brought up remarkable limitations for application of high-fidelity geo-models in reservoir management. To generate efficient low-resolution models, which are relatively representative of the reservoir dynamic behaviour, data-driven models are proposed. The data-driven models are trained principally based on the production history of the reservoir with primary knowledge of structural information. The Data-driven Discrete Well Affinity (DiWA) model is designed to generate efficient proxy models that require basic geological information and full production history of the reservoir. Since it benefits the computational power of the Delft Advanced Terra Simulator (DARTS) and adjoint gradient-based optimization, the DiWA model has become an efficient framework in terms of accuracy and computational costs. This research shows that the DiWA model can relatively reconstruct the reference reservoir parameters out of an ensemble of priors, and the final results can be improved utilizing closer prior knowledge to the true response. Additionally, it is demonstrated that the performance of the DiWA model in a complex, realistic oil field can be enhanced using a comprehensive objective function, applying appropriate well controls, understanding the underlying driving mechanisms, using sufficient degrees of freedom, and applying consistent boundary conditions., Civil Engineering

Railways are currently experiencing higher demands on safety, infrastructure performance, network capacity and service quality, etc. As a result, a high level of track availability, service quality and infrastructure robustness against unexpected events is required, at reduced cost, to meet the demands. The track geometry is one of the critical parameters influencing the performance of the infrastructure and the capacity and quality of service of the railway network. The track geometry continuously degrades and deviates from the designed level over time and under the influence of various factors. A poor track geometry negatively affects the ride quality and passenger comfort and increases the risk of train derailment. To ensure a safe and reliable track for the trains, the track geometry should be kept in an acceptable condition. This necessitates the development of an applicable and effective tamping regime, as the main maintenance action for track geometry rejuvenation. An effective tamping regime enables enhancement of the availability and safety performance of the track, to control the track degradation and to restore the damaged track to an operational state, at the lowest possible cost. The purpose of the research conducted for this thesis was to develop data-driven methodologies and optimization approaches for the development of a plan for predictive railway track geometry maintenance. First, a study was carried out to investigate the application of artificial neural networks to estimation of the track geometry degradation rate in spatial space by considering the operational and constructional covariates. In addition, the relative importance of the observed covariates for the track geometry degradation was also explored. The results showed that the maintenance history, the degradation value after tamping and the frequency of the trains passing along the track were the most important covariates affecting the track geometry degradation rate. Second, a case stu

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 ma

The past decades, the increasing availability of data has paved the way for a new, data-driven generation of models. This research proposes a non-parametric Bayesian network (NPBN) to model hydrologic processes. The Bayesian network (BN) is a directed, acyclic graph in which the variables are represented by the nodes, and the conditional probability distribution between variable pairs is represented by the arcs. NPBNs are computationally less expensive than many conceptual hydrologic models and are sufficiently flexible to be able to handle different continuous data sources. The goal of this thesis is to make an NPBN for a lowland catchment and test its performance. The case study concerns the catchment of the Vledder, Wapserveense and Steenwijker Aa. This catchment makes this research the first one in which an NPBN is comprehensively implemented for (1.) a single catchment in which the catchment processes are modelled, (2.) a Dutch catchment, and (3.) a lowland, partially managed, catchment. For the BN model, seven hydro-meteorological variables have been selected for the model, complemented by the target variable, which is the monthly maximum daily average discharge (MMDAD). The aim of the BN is to be able to accurately predict the MMDAD, and the Kling-Gupta efficiency (KGE) acts as a performance indicator by which to optimize the BN’s parameters. For this thesis, the Gaussian copula was selected to be implemented for all variable combinations in the BN, because this type allows for the use of the multivariate normal distribution to calculate a conditioned network. The fit of the Gaussian copula to the data is tested in this thesis. This method is far more convenient than the alternative called the vine-copula method and most likely gives a better fit than the other alternatives. Three distributions are compared to model the marginal distributions, of which the Gaussian mixture model has been selected. This function extrapolated too little, so a novel alteration f

The high substrate variability and complexity of fermentation media derived from lignocellulosic feedstocks affect the concentration profiles and the length of the fermentations. Not accounting for such variability raises operational and scheduling issues and affects the overall performance of these processes. In this work, a hybrid soft sensor was developed to monitor and forecast the evolution of cellulose-to-ethanol fermentations. The soft sensor consisted of two modules (a data-driven and a kinetic model) connected sequentially. The data-driven module used a partial-least-squares model to estimate the current state of glucose from spectroscopic data. The kinetic model was recursively fit to the known concentrations of glucose to update the long-horizon predictions of glucose, xylose and ethanol. This combination of real-time data update from an actual fermentation process to a high fidelity kinetic model constitutes the basis of the digital twin concepts and allows for the better real-time understanding of complex inhibition phenomena caused by different inhibitors commonly found in lignocellulosic feedstocks. The soft sensor was experimentally validated with three different cellulose-to-ethanol fermentations and the results suggested that this method is suitable to monitor and forecast fermentations when the measurements provide reasonably good estimates of the real states of the system. These results would allow increasing the flexibility of the operation of cellulosic processes and adapting the scheduling to the inherent variability of such substrates.

The past decades, the increasing availability of data has paved the way for a new, data-driven generation of models. This research proposes a non-parametric Bayesian network (NPBN) to model hydrologic processes. The Bayesian network (BN) is a directed, acyclic graph in which the variables are represented by the nodes, and the conditional probability distribution between variable pairs is represented by the arcs. NPBNs are computationally less expensive than many conceptual hydrologic models and are sufficiently flexible to be able to handle different continuous data sources. The goal of this thesis is to make an NPBN for a lowland catchment and test its performance. The case study concerns the catchment of the Vledder, Wapserveense and Steenwijker Aa. This catchment makes this research the first one in which an NPBN is comprehensively implemented for (1.) a single catchment in which the catchment processes are modelled, (2.) a Dutch catchment, and (3.) a lowland, partially managed, catchment. For the BN model, seven hydro-meteorological variables have been selected for the model, complemented by the target variable, which is the monthly maximum daily average discharge (MMDAD). The aim of the BN is to be able to accurately predict the MMDAD, and the Kling-Gupta efficiency (KGE) acts as a performance indicator by which to optimize the BN’s parameters. For this thesis, the Gaussian copula was selected to be implemented for all variable combinations in the BN, because this type allows for the use of the multivariate normal distribution to calculate a conditioned network. The fit of the Gaussian copula to the data is tested in this thesis. This method is far more convenient than the alternative called the vine-copula method and most likely gives a better fit than the other alternatives. Three distributions are compared to model the marginal distributions, of which the Gaussian mixture model has been selected. This function extrapolated too little, so a novel alteration f

The Dynamic OD Matrix Estimation (DODME) is a hard problem since no direct full observations are available, and therefore one should resort to indirect estimation approaches. This formulation solves at the upper level a nonlinear optimization that minimizes some distance measures between observed and estimated link flow counts at certain counting stations located in a subset of links in the network, and at the lower level a traffic assignment that estimates these link flow counts assigning the current estimated matrix. Since these estimations are based on a traffic assignment at the lower level, these analytical approaches, although numerically efficient, imply a high computational cost. The advent of ICT applications has made available new sets of traffic related measurements enabling new approaches. This research report explores how to extract such information from the recorded data., Dynamic traffic models require dynamic inputs, and one of the main inputs are the Dynamic Origin-Destinations (OD) matrices describing the variability over time of the trip patterns across the network. The Dynamic OD Matrix Estimation (DODME) is a hard problem since no direct full observations are available, and therefore one should resort to indirect estimation approaches. Among the most efficient approaches, the one that formulates the problem in terms of a bilevel optimization problem has been widely used. This formulation solves at the upper level a nonlinear optimization that minimizes some distance measures between observed and estimated link flow counts at certain counting stations located in a subset of links in the network, and at the lower level a traffic assignment that estimates these link flow counts assigning the current estimated matrix. The variants of this formulation differ in the analytical approaches that estimate the link flows in terms of the assignment and their time dependencies. Since these estimations are based on a traffic assignment at the lower level, these analytical approaches, although numerically efficient, imply a high computational cost. The advent of ICT applications has made available new sets of traffic related measurements enabling new approaches; under certain conditions, the data collected on used paths could be interpreted as an, de facto, estimated assignment observed . This allows extracting empirically the same information provided by an assignment that is used in the analytical approaches. This research report explores how to extract such information from the recorded data., This research was funded by PTV Group, by TRA2016-76914-C3-1-P Spanish R+D Programs and by Secretaria d’Universitats-i-Recerca-Generalitat de Catalunya- 2017- SGR- 1749 and Industrial PhD Program 2017 DI 041., Preprint

This case study uses a long short-term memory (LSTM)-based model to predict short-term urban water demands for the Hefei City of China. The performance of the LSTM-based model is compared with the autoregressive integrated moving average (ARIMA) model, the support vector regression (SVR) model, and the random forests (RF) model based on data with time resolutions ranging from 15 min to 24 h. Additionally, this paper investigates the performance of the LSTM-based model in predicting multiple successive data points. Results show that the LSTM-based model can offer predictions with improved accuracy than the other models when dealing with data with high time resolutions, data points with abrupt changes, and data with a relatively high uncertainty level. It is also observed that the LSTM-based model exhibits the best performance in predicting multiple successive water demands with high time resolutions. In addition, the inclusion of external parameters (e.g., temperature) cannot enhance the performance of the LSTM-based model, but it can improve ARIMAX's prediction ability (ARIMAX is the ARIMA with variables). These observations provide additional and improved evaluations regarding the LSTM-based models used for short-term urban water demand forecasting, thereby enabling their wider adoption in practical applications., Accepted Author Manuscript, Sanitary Engineering

Railways are a vital and effective means of mass transportation and play a vital role in modern transportation and social development. The benefits of the railway compared to other transportation modes are a high capacity, high efficiency and low pollution, and owing to these advantages, railways are nowadays experiencing a higher demand for the transportation of passengers and goods. This is in turn imposing higher demands on the railway capacity and service quality. As a result, infrastructure managers are being driven to develop new strategies and plans to fulfil new requirements, which include a higher level of resilience against failure, a more robust and available infrastructure, and cost reduction. This can be achieved by making efficient and effective maintenance decisions by applying RAMS (reliability, availability, maintainability, and safety) analysis and LCC (life cycle cost) assessment. A major part of the railway maintenance burden is related to track geometry maintenance. Due to the forces induced on the track by traffic, the railway degrades over time, causing deviations from the designed vertical and horizontal alignment. When the track geometry degrades to an unacceptable level, this can cause catastrophic consequences, such as derailment. Maintenance actions are used to control the degradation of the track and restore the geometry condition of the track sections to an acceptable state. With the current advancements in the field of technologies for railway track geometry measurement, a large amount of event data and condition monitoring data is available. Such technologies, along with advances in predictive analytics, are providing the possibility of predicting the track geometry condition in support of a predictive maintenance strategy. The aim of the research conducted for this thesis has been to develop methodologies and tools for the prediction of railway track geometry degradation, in order to facilitate and enhance the capability of making ef

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., Water Resources

There has been significant growth in online social science experiments in order to understand behavior at-scale, with finer-grained data collection. Considerable work is required to perform data analytics for custom experiments. In this dissertation, we design and build composable and extensible automated software pipelines for evaluating social phenomena through iterative experiments and modeling. To reason about experiments and models, we design a formal data model. This combined approach of experiments and models has been done in some studies without automation, or purely conceptually. We are motivated by a particular social behavior, namely collective identity (CI). Group or CI is an individual's cognitive, moral, and emotional connection with a broader community, category, practice, or institution. Extensive experimental research shows that CI influences human decision-making. Because of this, there is interest in modeling situations that promote the creation of CI in order to learn more from the process and to predict human behavior in real life situations. One of our goals in this dissertation is to understand whether a cooperative anagram game can produce CI within a group. With all of the experimental work on anagram games, it is surprising that very little work has been done in modeling these games. Also, abduction is an inference approach that uses data and observations to identify plausibly (and preferably, best) explanations for phenomena. Abduction has broad application in robotics, genetics, automated systems, and image understanding, but have largely been devoid of human behavior. We use these pipelines to understand intra-group cooperation and its effect on fostering CI. We devise and execute an iterative abductive analysis process that is driven by the social sciences. In a group anagrams web-based networked game setting, we formalize an abductive loop, implement it computationally, and exercise it; we build and evaluate three agent-based models (A

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., Water Resources