Pub Date : 2025-04-04DOI: 10.1016/j.egyai.2025.100510
Joan Tomàs Villalonga Palou , Javier Serrano González , Jesús Manuel Riquelme Santos , Juan Manuel Roldán Fernández
The emergence of new resources and services in the electricity system implies that more and more agents need to obtain more accurate forecasts to optimize their operations. It is common for these agents to have different sources of forecasts (from specialized consultants or meteorological services, among others).
The proposed approach aims to obtain more accurate predictions by optimally combining a set of predictions obtained by different techniques. In this way it is possible to obtain a resulting prediction that improves the error and uncertainty associated with each of the individual forecasts. The objective is achieved by the analytical minimization of the errors obtained by each of the individual predictors. This allows to obtain dynamically the optimized weights assigned to each of the algorithms so that the combination outperforms the individual behaviour of each of them. The proposed ensemble approach has been successfully tested on a real time series of electric vehicle charging.
Likewise, the results obtained have been compared exhaustively with other ensemble techniques consolidated in the literature based on different methods, including dynamic ensembles as machine learning approaches. The results obtained show an appreciable improvement of the errors obtained in the predictions using the proposed techniques.
{"title":"A novel weight-based ensemble method for emerging energy players: an application to electric vehicle load prediction","authors":"Joan Tomàs Villalonga Palou , Javier Serrano González , Jesús Manuel Riquelme Santos , Juan Manuel Roldán Fernández","doi":"10.1016/j.egyai.2025.100510","DOIUrl":"10.1016/j.egyai.2025.100510","url":null,"abstract":"<div><div>The emergence of new resources and services in the electricity system implies that more and more agents need to obtain more accurate forecasts to optimize their operations. It is common for these agents to have different sources of forecasts (from specialized consultants or meteorological services, among others).</div><div>The proposed approach aims to obtain more accurate predictions by optimally combining a set of predictions obtained by different techniques. In this way it is possible to obtain a resulting prediction that improves the error and uncertainty associated with each of the individual forecasts. The objective is achieved by the analytical minimization of the errors obtained by each of the individual predictors. This allows to obtain dynamically the optimized weights assigned to each of the algorithms so that the combination outperforms the individual behaviour of each of them. The proposed ensemble approach has been successfully tested on a real time series of electric vehicle charging.</div><div>Likewise, the results obtained have been compared exhaustively with other ensemble techniques consolidated in the literature based on different methods, including dynamic ensembles as machine learning approaches. The results obtained show an appreciable improvement of the errors obtained in the predictions using the proposed techniques.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"20 ","pages":"Article 100510"},"PeriodicalIF":9.6,"publicationDate":"2025-04-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143816082","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Early detection of faults in photovoltaic (PV) arrays has always been the center of attention to maintain system efficiency and reliability. However, conventional protection devices have shown various deficiencies, especially when dealing with less severe faults. Hence, artificial intelligence (AI) models, specifically machine learning (ML) have complemented the conventional protection devices to compensate for their limitations. Despite their obvious advantages, ML models have also shown several shortcomings, such as (i) most of them relied on a massive amount of training dataset to provide a fairly satisfying accuracy, (ii) not many of them were able to detect less severe faults, and (iii) those which were able to detect less severe faults could not produce high accuracy. To this end, the present paper proposes a state-of-the-art deep reinforcement learning (DRL) model based on deep Q-network (DQN) to overcome all the existing challenges in previous ML models for PV arrays fault detection and diagnosis. The model carries out a two-stage process employing two DQN-based agents which is not only able to accurately detect and classify (first stage) various faults in PV arrays, but it is also able to assess the severity of line-to-line (LL) and line-to-ground (LG) faults (second stage) in PV arrays using only a small training dataset. The training and testing datasets include several voltage and current values on PV array current-voltage (I-V) characteristic curve which is extracted using the variable load technique for PV array I-V curve extraction. The model has been implemented on an experimental standalone PV array and the results show outstanding accuracies of 98.61 % and 100 % when it is verified by testing datasets in the first and the second stage, respectively.
{"title":"Two-stage deep Q-network reinforcement learning based ultra-efficient fault diagnosis and severity assessment scheme for photovoltaic protection","authors":"Sherko Salehpour , Aref Eskandari , Amir Nedaei , Mohammadreza Aghaei","doi":"10.1016/j.egyai.2025.100512","DOIUrl":"10.1016/j.egyai.2025.100512","url":null,"abstract":"<div><div>Early detection of faults in photovoltaic (PV) arrays has always been the center of attention to maintain system efficiency and reliability. However, conventional protection devices have shown various deficiencies, especially when dealing with less severe faults. Hence, artificial intelligence (AI) models, specifically machine learning (ML) have complemented the conventional protection devices to compensate for their limitations. Despite their obvious advantages, ML models have also shown several shortcomings, such as (i) most of them relied on a massive amount of training dataset to provide a fairly satisfying accuracy, (ii) not many of them were able to detect less severe faults, and (iii) those which were able to detect less severe faults could not produce high accuracy. To this end, the present paper proposes a state-of-the-art deep reinforcement learning (DRL) model based on deep Q-network (DQN) to overcome all the existing challenges in previous ML models for PV arrays fault detection and diagnosis. The model carries out a two-stage process employing two DQN-based agents which is not only able to accurately detect and classify (first stage) various faults in PV arrays, but it is also able to assess the severity of line-to-line (LL) and line-to-ground (LG) faults (second stage) in PV arrays using only a small training dataset. The training and testing datasets include several voltage and current values on PV array current-voltage (I-V) characteristic curve which is extracted using the variable load technique for PV array I-V curve extraction. The model has been implemented on an experimental standalone PV array and the results show outstanding accuracies of 98.61 % and 100 % when it is verified by testing datasets in the first and the second stage, respectively.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"20 ","pages":"Article 100512"},"PeriodicalIF":9.6,"publicationDate":"2025-04-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143816083","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-04-01DOI: 10.1016/j.egyai.2025.100504
Nicholas Majeske , Shreyas Sunil Vaidya , Ryan Roy , Abdul Rehman , Hamed Sohrabpoor , Tyson Miller , Wenhui Li , C.R. Fiddyment , Alexander Gumennik , Raj Acharya , Vikram Jadhao , Prateek Sharma , Ariful Azad
We develop a comprehensive framework for storing, analyzing, forecasting, and visualizing industrial energy systems consisting of multiple devices and sensors. Our framework models complex energy systems as a dynamic knowledge graph, utilizes a novel machine learning (ML) model for energy forecasting, and visualizes continuous predictions through an interactive dashboard. At the core of this framework is A-RNN, a simple yet efficient model that uses dynamic attention mechanisms for automated feature selection. We validate the model using datasets from two manufacturers and one university testbed containing hundreds of sensors. Our results show that A-RNN forecasts energy usage within 5% of observed values. These enhanced predictions are as much as 50% more accurate than those produced by standard RNN models that rely on individual features and devices. Additionally, A-RNN identifies key features that impact forecasting accuracy, providing interpretability for model forecasts. Our analytics platform is computationally and memory efficient, making it suitable for deployment on edge devices and in manufacturing plants.
{"title":"Industrial energy forecasting using dynamic attention neural networks","authors":"Nicholas Majeske , Shreyas Sunil Vaidya , Ryan Roy , Abdul Rehman , Hamed Sohrabpoor , Tyson Miller , Wenhui Li , C.R. Fiddyment , Alexander Gumennik , Raj Acharya , Vikram Jadhao , Prateek Sharma , Ariful Azad","doi":"10.1016/j.egyai.2025.100504","DOIUrl":"10.1016/j.egyai.2025.100504","url":null,"abstract":"<div><div>We develop a comprehensive framework for storing, analyzing, forecasting, and visualizing industrial energy systems consisting of multiple devices and sensors. Our framework models complex energy systems as a dynamic knowledge graph, utilizes a novel machine learning (ML) model for energy forecasting, and visualizes continuous predictions through an interactive dashboard. At the core of this framework is A-RNN, a simple yet efficient model that uses dynamic attention mechanisms for automated feature selection. We validate the model using datasets from two manufacturers and one university testbed containing hundreds of sensors. Our results show that A-RNN forecasts energy usage within 5% of observed values. These enhanced predictions are as much as 50% more accurate than those produced by standard RNN models that rely on individual features and devices. Additionally, A-RNN identifies key features that impact forecasting accuracy, providing interpretability for model forecasts. Our analytics platform is computationally and memory efficient, making it suitable for deployment on edge devices and in manufacturing plants.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"20 ","pages":"Article 100504"},"PeriodicalIF":9.6,"publicationDate":"2025-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143808118","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-03-27DOI: 10.1016/j.egyai.2025.100506
Sara Ghane , Stef Jacobs , Furkan Elmaz , Thomas Huybrechts , Ivan Verhaert , Siegfried Mercelis
This paper introduces a novel privacy-aware Federated Proximal Policy Optimization (FPPO) method combined with action masking. As a Federated Reinforcement Learning (FRL) approach, the proposed method is used for optimizing the reloading of Domestic Hot Water (DHW) storage tanks, with a focus on energy savings and DHW thermal comfort in collective heating systems. The proposed approach combines FedProx as the Federated Learning (FL) method and Proximal Policy Optimization (PPO) as the Deep Reinforcement Learning (DRL) technique to address the challenges of distributed control while ensuring data privacy. Key contributions include: (1) employing action masking to guarantee compliance with comfort level, (2) designing a global reward function to align agents actions toward collective energy savings, (3) implementing a privacy-aware design where only model parameters are shared with a global aggregator, avoiding raw data transmission, and (4) optimizing PPO’s loss function for improved performance.
PPO was benchmarked using a common FL method (FedAvg) alongside two other DRL methods, where PPO outperformed both in scalability and energy savings, especially in larger systems. Then, PPO-based FRL was refined into FPPO by integrating a proximal term with coefficient into the loss function to enhance the performance. Experiments were conducted with both fixed and dynamically adjusted , with the latter demonstrating better energy savings and comfort. Results show that FPPO achieves up to 10.08% energy savings while maintaining DHW discomfort below 8.72% in systems with at least 20 dwellings. These findings highlight FPPO as a scalable, privacy-aware, and energy-efficient solution for distributed control in collective heating systems.
{"title":"Federated proximal policy optimization with action masking: Application in collective heating systems","authors":"Sara Ghane , Stef Jacobs , Furkan Elmaz , Thomas Huybrechts , Ivan Verhaert , Siegfried Mercelis","doi":"10.1016/j.egyai.2025.100506","DOIUrl":"10.1016/j.egyai.2025.100506","url":null,"abstract":"<div><div>This paper introduces a novel privacy-aware Federated Proximal Policy Optimization (FPPO) method combined with action masking. As a Federated Reinforcement Learning (FRL) approach, the proposed method is used for optimizing the reloading of Domestic Hot Water (DHW) storage tanks, with a focus on energy savings and DHW thermal comfort in collective heating systems. The proposed approach combines FedProx as the Federated Learning (FL) method and Proximal Policy Optimization (PPO) as the Deep Reinforcement Learning (DRL) technique to address the challenges of distributed control while ensuring data privacy. Key contributions include: (1) employing action masking to guarantee compliance with comfort level, (2) designing a global reward function to align agents actions toward collective energy savings, (3) implementing a privacy-aware design where only model parameters are shared with a global aggregator, avoiding raw data transmission, and (4) optimizing PPO’s loss function for improved performance.</div><div>PPO was benchmarked using a common FL method (FedAvg) alongside two other DRL methods, where PPO outperformed both in scalability and energy savings, especially in larger systems. Then, PPO-based FRL was refined into FPPO by integrating a proximal term with coefficient <span><math><mi>μ</mi></math></span> into the loss function to enhance the performance. Experiments were conducted with both fixed and dynamically adjusted <span><math><mi>μ</mi></math></span>, with the latter demonstrating better energy savings and comfort. Results show that FPPO achieves up to 10.08% energy savings while maintaining DHW discomfort below 8.72% in systems with at least 20 dwellings. These findings highlight FPPO as a scalable, privacy-aware, and energy-efficient solution for distributed control in collective heating systems.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"20 ","pages":"Article 100506"},"PeriodicalIF":9.6,"publicationDate":"2025-03-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143738722","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-03-27DOI: 10.1016/j.egyai.2025.100508
Yuekuan Zhou
Circular Economy and Sustainability in Industry 4.0 Era are promoters for carbon neutrality transformation, while their interconnected nexus and specific roles in low-carbon transition have not been clearly revealed. Furthermore, an integrated circular economy framework with buildings, PVs, battery and EVs, with overlaps in renewable-driven operational stages has not been considered in lifecycle decarbonization. This study is to reveal the nexus between Circular Economy and Sustainability in Industry 4.0 Era. Operational modes and mechanism of Circular Economy in PVs, buildings, electric vehicle industries and batteries are specifically analysed, together with energy and carbon flow analysis and optimization. Roles of Circular Economy in Sustainability have been provided, through an integrated circular economy framework with buildings, PVs, battery and electric vehicles (EVs), considering the overlap in renewable-energy driven operational stages in lifecycle decarbonization. Last but not the least, waste material recovery and waste-to-energy conversion have been analysed within the close-in-loop cycle for sustainability transition. Advanced digital technology in future Circular Economy is formulated with data-driven circular economy and internet-of-thing (IoT)-based waste-to-energy framework. Research results indicate that circular economy plays significant roles in sustainability, including cascade reuse paradigm, reverse supply chain with the recovery of end-of-life batteries, EV lifetime extension via repair and reuse, low carbon with refurbishing and remanufacturing, and less new primary materials via recycling materials, waste material recovery and waste-to-energy conversion. The renewable-driven battery-building-transportation-waste circular economy chain with the cross overlap in clean energy utilization can partially offset carbon emissions during the raw materials mining, manufacturing and recycling stages. This study can promote the waste to energy and advanced machine learning techniques with Circular Economy and Sustainability in Industry 4.0 Era.
{"title":"AI-driven digital circular economy with material and energy sustainability for industry 4.0","authors":"Yuekuan Zhou","doi":"10.1016/j.egyai.2025.100508","DOIUrl":"10.1016/j.egyai.2025.100508","url":null,"abstract":"<div><div>Circular Economy and Sustainability in Industry 4.0 Era are promoters for carbon neutrality transformation, while their interconnected nexus and specific roles in low-carbon transition have not been clearly revealed. Furthermore, an integrated circular economy framework with buildings, PVs, battery and EVs, with overlaps in renewable-driven operational stages has not been considered in lifecycle decarbonization. This study is to reveal the nexus between Circular Economy and Sustainability in Industry 4.0 Era. Operational modes and mechanism of Circular Economy in PVs, buildings, electric vehicle industries and batteries are specifically analysed, together with energy and carbon flow analysis and optimization. Roles of Circular Economy in Sustainability have been provided, through an integrated circular economy framework with buildings, PVs, battery and electric vehicles (EVs), considering the overlap in renewable-energy driven operational stages in lifecycle decarbonization. Last but not the least, waste material recovery and waste-to-energy conversion have been analysed within the close-in-loop cycle for sustainability transition. Advanced digital technology in future Circular Economy is formulated with data-driven circular economy and internet-of-thing (IoT)-based waste-to-energy framework. Research results indicate that circular economy plays significant roles in sustainability, including cascade reuse paradigm, reverse supply chain with the recovery of end-of-life batteries, EV lifetime extension via repair and reuse, low carbon with refurbishing and remanufacturing, and less new primary materials via recycling materials, waste material recovery and waste-to-energy conversion. The renewable-driven battery-building-transportation-waste circular economy chain with the cross overlap in clean energy utilization can partially offset carbon emissions during the raw materials mining, manufacturing and recycling stages. This study can promote the waste to energy and advanced machine learning techniques with Circular Economy and Sustainability in Industry 4.0 Era.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"20 ","pages":"Article 100508"},"PeriodicalIF":9.6,"publicationDate":"2025-03-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143768708","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-03-24DOI: 10.1016/j.egyai.2025.100503
Sharun Parayil Shaji , Wolfgang Tress
Perovskite solar cells are promising photovoltaic devices because of the high defect tolerance and desirable optoelectronic properties of the metal-halide perovskite absorber materials. The transition from lab to industry is still an open problem, which is mainly limited by upscaling and stability. In this study we try to use tools from data science namely Pearson correlation and random forest regressor applied to the data from the open-source platform “Perovskite Database” to understand the correlations with material choice, fabrication techniques, and current-voltage key features to the stability and hysteresis index. We find that the cell stack as a whole plays a crucial role in hysteresis and not a single layer. We statistically confirm that p-i-n and higher-efficient solar cells generally show reduced hysteresis. We identify certain cross correlations, which would lead to wrong conclusions e.g. claiming an open-circuit voltage not correlated with the hysteresis or some apparent correlations with material parameters, which originate from the historical development. Regarding stability, we are not able to obtain good performance from the machine learning model. Reasons are non-standardized measurements and lack of sufficient data.
{"title":"Data-driven analysis of hysteresis and stability in perovskite solar cells using machine learning","authors":"Sharun Parayil Shaji , Wolfgang Tress","doi":"10.1016/j.egyai.2025.100503","DOIUrl":"10.1016/j.egyai.2025.100503","url":null,"abstract":"<div><div>Perovskite solar cells are promising photovoltaic devices because of the high defect tolerance and desirable optoelectronic properties of the metal-halide perovskite absorber materials. The transition from lab to industry is still an open problem, which is mainly limited by upscaling and stability. In this study we try to use tools from data science namely Pearson correlation and random forest regressor applied to the data from the open-source platform “Perovskite Database” to understand the correlations with material choice, fabrication techniques, and current-voltage key features to the stability and hysteresis index. We find that the cell stack as a whole plays a crucial role in hysteresis and not a single layer. We statistically confirm that p-i-n and higher-efficient solar cells generally show reduced hysteresis. We identify certain cross correlations, which would lead to wrong conclusions e.g. claiming an open-circuit voltage not correlated with the hysteresis or some apparent correlations with material parameters, which originate from the historical development. Regarding stability, we are not able to obtain good performance from the machine learning model. Reasons are non-standardized measurements and lack of sufficient data.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"20 ","pages":"Article 100503"},"PeriodicalIF":9.6,"publicationDate":"2025-03-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143724741","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-03-22DOI: 10.1016/j.egyai.2025.100500
Zhian Ye , Dawei Qiu , Shuangqi Li , Zhong Fan , Goran Strbac
The rapid development of distributed energy resources has led to an increasing number of prosumers enhancing their energy utilization, thereby raising the demands on energy management technologies. As a result, the development of future smart grids is becoming increasingly important, with a particular emphasis on integrating demand-side flexibility into electricity market. To facilitate distributed interaction among prosumers, the double-side auction market enables peer-to-peer (P2P) energy trading, maximizing the social welfare within the dynamic local electricity market. In this setup, prosumers can set their own bidding prices and optimize their operations and trading strategies. However, trading in double-side auction market faces limitations due to the complexity of the market clearing algorithm and the difficulty of predicting other participants’ bidding behaviors. To address these challenges, this paper models the P2P energy trading problem in the double-side auction market as a multi-agent reinforcement learning (MARL) task. The concept of federated learning is introduced to enhance scalability among market participants while protecting the private information of individual prosumers. Additionally, the parameter-sharing framework is proposed to accelerate the learning process. To further improve the stability of MARL training, the global information of P2P energy trading price is integrated into the critic network. The proposed federated MARL algorithm is evaluated using a real-world open-source dataset from an European residential community of 250 households with a 15-minute resolution. The evaluation assesses both the training performance of the algorithm as well as the economic and operational benefits of the P2P energy trading market compared to a traditional electricity retail market.
{"title":"Federated Reinforcement Learning for decentralized peer-to-peer energy trading","authors":"Zhian Ye , Dawei Qiu , Shuangqi Li , Zhong Fan , Goran Strbac","doi":"10.1016/j.egyai.2025.100500","DOIUrl":"10.1016/j.egyai.2025.100500","url":null,"abstract":"<div><div>The rapid development of distributed energy resources has led to an increasing number of prosumers enhancing their energy utilization, thereby raising the demands on energy management technologies. As a result, the development of future smart grids is becoming increasingly important, with a particular emphasis on integrating demand-side flexibility into electricity market. To facilitate distributed interaction among prosumers, the double-side auction market enables peer-to-peer (P2P) energy trading, maximizing the social welfare within the dynamic local electricity market. In this setup, prosumers can set their own bidding prices and optimize their operations and trading strategies. However, trading in double-side auction market faces limitations due to the complexity of the market clearing algorithm and the difficulty of predicting other participants’ bidding behaviors. To address these challenges, this paper models the P2P energy trading problem in the double-side auction market as a multi-agent reinforcement learning (MARL) task. The concept of federated learning is introduced to enhance scalability among market participants while protecting the private information of individual prosumers. Additionally, the parameter-sharing framework is proposed to accelerate the learning process. To further improve the stability of MARL training, the global information of P2P energy trading price is integrated into the critic network. The proposed federated MARL algorithm is evaluated using a real-world open-source dataset from an European residential community of 250 households with a 15-minute resolution. The evaluation assesses both the training performance of the algorithm as well as the economic and operational benefits of the P2P energy trading market compared to a traditional electricity retail market.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"20 ","pages":"Article 100500"},"PeriodicalIF":9.6,"publicationDate":"2025-03-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143687633","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-03-22DOI: 10.1016/j.egyai.2025.100499
Dapeng Wang, Zhaojian Liang, Ziqi Zhang, Mengying Li
Convective cooling by wind is crucial for large-scale photovoltaic (PV) systems, as power generation inversely correlates with panel temperature. Therefore, accurately determining the convective heat transfer coefficient for PV arrays with various geometric configurations is essential to optimize array design. Traditional methods to quantify the effects of configuration utilize either Computational Fluid Dynamics (CFD) simulations or empirical methods. These approaches often face challenges due to high computational demands or limited accuracy, particularly with complex array configurations. Machine learning approaches, especially hybrid learning models, have emerged as effective tools to address challenges in heat transfer design optimization. This study introduces a method that combines Physics-Informed Machine Learning with a Deep Convolutional Neural Network (PIML-DCNN) to predict convective heat transfer rates with high accuracy and computational efficiency. Additionally, an innovative loss function, termed the ”Pocket Loss”, is developed to enhance the interpretability and robustness of the PIML-DCNN model. The proposed model achieves relative estimation errors of 2.5% and 2.7% on the validation and test datasets, respectively, when benchmarked against comprehensive CFD simulations. These results highlight the potential of the proposed model to efficiently guide the configuration design of PV arrays, thereby enhancing power generation in real-world operations.
{"title":"Efficient estimation of convective cooling of photovoltaic arrays: A physics-informed machine learning approach","authors":"Dapeng Wang, Zhaojian Liang, Ziqi Zhang, Mengying Li","doi":"10.1016/j.egyai.2025.100499","DOIUrl":"10.1016/j.egyai.2025.100499","url":null,"abstract":"<div><div>Convective cooling by wind is crucial for large-scale photovoltaic (PV) systems, as power generation inversely correlates with panel temperature. Therefore, accurately determining the convective heat transfer coefficient for PV arrays with various geometric configurations is essential to optimize array design. Traditional methods to quantify the effects of configuration utilize either Computational Fluid Dynamics (CFD) simulations or empirical methods. These approaches often face challenges due to high computational demands or limited accuracy, particularly with complex array configurations. Machine learning approaches, especially hybrid learning models, have emerged as effective tools to address challenges in heat transfer design optimization. This study introduces a method that combines Physics-Informed Machine Learning with a Deep Convolutional Neural Network (PIML-DCNN) to predict convective heat transfer rates with high accuracy and computational efficiency. Additionally, an innovative loss function, termed the ”Pocket Loss”, is developed to enhance the interpretability and robustness of the PIML-DCNN model. The proposed model achieves relative estimation errors of 2.5% and 2.7% on the validation and test datasets, respectively, when benchmarked against comprehensive CFD simulations. These results highlight the potential of the proposed model to efficiently guide the configuration design of PV arrays, thereby enhancing power generation in real-world operations.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"20 ","pages":"Article 100499"},"PeriodicalIF":9.6,"publicationDate":"2025-03-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143704169","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-03-19DOI: 10.1016/j.egyai.2025.100507
Gwiman Bak , Youngchul Bae
This study introduces the positive and negative convolution cross-connect neural network (PNCCN), a novel deep learning framework designed for accurately predicting the remaining useful life (RUL) of lithium-ion batteries (LIBs). The model leverages the positive and negative convolution (PNC) and nonlinear cross-connect (NCC) architectures to effectively capture complex nonlinear interactions and degradation patterns in battery data. The PNCCN model was developed and evaluated using a comprehensive dataset comprising 118 battery cells, processed at 10 s intervals during charge-discharge cycles. The training, validation, and test datasets were divided in a 60:20:20 ratio to ensure robust performance evaluation across diverse operational conditions. By excluding internal resistance (IR) data, the model simplifies data acquisition, reduces dependency on costly sensors, and improves the practicality of battery management system (BMS) integration. The PNCCN model achieved an average root mean square errors (RMSEs) of 9.47 and 93.58 cycles for training and test datasets, respectively, with mean absolute percentage errors (MAPEs) of 1.03 % and 8.28 %. Comparative analysis demonstrates that the PNCCN model outperforms existing methods, offering a reliable and scalable solution for LIB RUL prediction. These results highlight the model's potential for real-world applications, emphasizing its effectiveness in reducing system complexity and enhancing predictive accuracy without relying on IR data.
{"title":"Positive and negative convolution cross-connect neural network for predicting the remaining useful life of lithium-ion batteries","authors":"Gwiman Bak , Youngchul Bae","doi":"10.1016/j.egyai.2025.100507","DOIUrl":"10.1016/j.egyai.2025.100507","url":null,"abstract":"<div><div>This study introduces the positive and negative convolution cross-connect neural network (PNCCN), a novel deep learning framework designed for accurately predicting the remaining useful life (RUL) of lithium-ion batteries (LIBs). The model leverages the positive and negative convolution (PNC) and nonlinear cross-connect (NCC) architectures to effectively capture complex nonlinear interactions and degradation patterns in battery data. The PNCCN model was developed and evaluated using a comprehensive dataset comprising 118 battery cells, processed at 10 s intervals during charge-discharge cycles. The training, validation, and test datasets were divided in a 60:20:20 ratio to ensure robust performance evaluation across diverse operational conditions. By excluding internal resistance (IR) data, the model simplifies data acquisition, reduces dependency on costly sensors, and improves the practicality of battery management system (BMS) integration. The PNCCN model achieved an average root mean square errors (RMSEs) of 9.47 and 93.58 cycles for training and test datasets, respectively, with mean absolute percentage errors (MAPEs) of 1.03 % and 8.28 %. Comparative analysis demonstrates that the PNCCN model outperforms existing methods, offering a reliable and scalable solution for LIB RUL prediction. These results highlight the model's potential for real-world applications, emphasizing its effectiveness in reducing system complexity and enhancing predictive accuracy without relying on IR data.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"20 ","pages":"Article 100507"},"PeriodicalIF":9.6,"publicationDate":"2025-03-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143697135","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-03-19DOI: 10.1016/j.egyai.2025.100505
Dacheng Li , Songshan Guo , Jihong Wang , Yongliang Li , Chenggong Sun , Geng Qiao , Chaomurilige , Yulong Ding
Local energy systems are undergoing significant transformation by integrating more solar photovoltaics (PVs) and battery energy storage systems (BESS) to achieve net-zero targets in the energy sector. To ensure an affordable and sustainable decarbonisation process, optimising both system design and operation together is crucial for maximising system profitability and encouraging broader stakeholder participation in the energy transition. However, the complex interdependent influence on the system economic flows, along with the nonlinear characteristics of the system, make the economic optimisation extremely challenging. To address this, we developed a new framework based on advanced artificial intelligence to exploit a wider arbitrage margin under various trading mechanisms, including net metering, day-ahead, and dynamic frequency. We conducted optimisation study on a local energy system operating at University of Warwick using real data from demonstrated BESS and solar PVs, and the effectiveness of the proposed intelligent approach was validated, and the necessity of interdependent optimisation was highlighted. Results showed that, compared to the original campus system (20 MW-level), a carbon reduction rate of up to 61.4 % was achieved through net metering trading, while a maximum annual profit increase of 251 % was realised with dynamic frequency trading. The proposed intelligent framework can be applied to any energy systems with integrated solar PVs and BESS, where the adopted trading mechanism are associated with the system design and operation. The findings offer a practical tool for academics, investors, and policy makers to collaborate in the deployment of renewable energy and energy storage to accelerate the decarbonisation of energy supply.
{"title":"Interdependent design and operation of solar photovoltaics and battery energy storage for economically viable decarbonisation of local energy systems","authors":"Dacheng Li , Songshan Guo , Jihong Wang , Yongliang Li , Chenggong Sun , Geng Qiao , Chaomurilige , Yulong Ding","doi":"10.1016/j.egyai.2025.100505","DOIUrl":"10.1016/j.egyai.2025.100505","url":null,"abstract":"<div><div>Local energy systems are undergoing significant transformation by integrating more solar photovoltaics (PVs) and battery energy storage systems (BESS) to achieve net-zero targets in the energy sector. To ensure an affordable and sustainable decarbonisation process, optimising both system design and operation together is crucial for maximising system profitability and encouraging broader stakeholder participation in the energy transition. However, the complex interdependent influence on the system economic flows, along with the nonlinear characteristics of the system, make the economic optimisation extremely challenging. To address this, we developed a new framework based on advanced artificial intelligence to exploit a wider arbitrage margin under various trading mechanisms, including net metering, day-ahead, and dynamic frequency. We conducted optimisation study on a local energy system operating at University of Warwick using real data from demonstrated BESS and solar PVs, and the effectiveness of the proposed intelligent approach was validated, and the necessity of interdependent optimisation was highlighted. Results showed that, compared to the original campus system (20 MW-level), a carbon reduction rate of up to 61.4 % was achieved through net metering trading, while a maximum annual profit increase of 251 % was realised with dynamic frequency trading. The proposed intelligent framework can be applied to any energy systems with integrated solar PVs and BESS, where the adopted trading mechanism are associated with the system design and operation. The findings offer a practical tool for academics, investors, and policy makers to collaborate in the deployment of renewable energy and energy storage to accelerate the decarbonisation of energy supply.</div></div>","PeriodicalId":34138,"journal":{"name":"Energy and AI","volume":"20 ","pages":"Article 100505"},"PeriodicalIF":9.6,"publicationDate":"2025-03-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143734811","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号