Periodic Reporting for period 1 - ULICBat (User behaviour informed learning and intelligent control for charging of vehicle battery packs)
Reporting period: 2022-08-01 to 2024-09-30
Electric vehicles (EVs) have recently experienced rapid development as an ideal alternative to traditional fossil-fuel-powered vehicles. By transitioning away from fossil fuels to renewable energy sources, EVs significantly reduce carbon emissions and harmful pollutants, improving public health and environmental quality. Moreover, EV technology promotes energy diversification, decreasing dependence on non-renewable resources and enhancing energy affordability and stability. As one of the most expensive and perhaps the least understood components of EVs, the lithium-ion (Li-ion) battery pack necessitates advanced management for safe and efficient use since it is often associated with risks and uncertainty, exemplified by ageing and accidents.
Charging is a critical process for Li-ion batteries to replenish energy. The widely adopted fast charging strategies can reduce the charging time, alleviating the mileage anxiety of EV users to a limited extent. However, the high electrical current rates utilised in these strategies inevitably lead to extensive battery capacity degradation. Hence, how to intelligently optimise the charging process according to different user demands instead of blindly fast charging to relieve battery degradation is an urgent and valuable research problem. The lack of accurate battery prediction models for practical use forms one of the biggest technical challenges in the field of advanced charging management. This fact has strongly motivated the design of a model-free charging control strategy that can still achieve multi-objective charging optimization. Last but not least, cell imbalances caused by inhomogeneous conditions and manufacturing variations result in insufficient energy use, accelerated capacity degradation, and even safety hazards of the entire battery pack. How to ensure that the battery cells' states in a battery pack are consistent at the end of the charging process is another major challenge that this project focuses on. The objective of this project is to create intelligent and health-aware charging strategies for Li-ion battery packs, in which the user’s short-term demands and long-term behaviours will be fully considered.
Charging is a critical process for Li-ion batteries to replenish energy. The widely adopted fast charging strategies can reduce the charging time, alleviating the mileage anxiety of EV users to a limited extent. However, the high electrical current rates utilised in these strategies inevitably lead to extensive battery capacity degradation. Hence, how to intelligently optimise the charging process according to different user demands instead of blindly fast charging to relieve battery degradation is an urgent and valuable research problem. The lack of accurate battery prediction models for practical use forms one of the biggest technical challenges in the field of advanced charging management. This fact has strongly motivated the design of a model-free charging control strategy that can still achieve multi-objective charging optimization. Last but not least, cell imbalances caused by inhomogeneous conditions and manufacturing variations result in insufficient energy use, accelerated capacity degradation, and even safety hazards of the entire battery pack. How to ensure that the battery cells' states in a battery pack are consistent at the end of the charging process is another major challenge that this project focuses on. The objective of this project is to create intelligent and health-aware charging strategies for Li-ion battery packs, in which the user’s short-term demands and long-term behaviours will be fully considered.
To achieve a reliable and high-performance charging control for Li-ion batteries, the following three tasks were accomplished.
Firstly, we have proposed an enhanced hybrid stacking-boosting ensemble model for EV charging duration prediction. The time index features, EV information features, and historical statistical features are selected as the prediction model input. Then, an ensemble model is designed by combining multiple linear regression, support vector regression, random forest, XGBoost, and artificial neural network. Experimental results show that better prediction performance (less root mean square of prediction error) can be achieved compared with the commonly used single machine learning model. At last, an optimal charging algorithm is proposed for EV battery charging to meet the EV user charging demand that can extend the lifetime of the battery.
Secondly, we have conducted a comprehensive review of the state of charge estimation techniques. Over the last decade, numerous attempts have been made to effectively analyze and compare the state of charge estimation methods for commercial Li-ion batteries. However, they seldom reflect on the state of charge estimation techniques based on a control-oriented viewpoint for a Li-ion battery system. To fill this gap, we have reviewed the most up-to-date battery state of charge estimation methods applied to lithium-ion battery systems. They are broadly classified as open-loop-based, closed-loop-based, and hybrid approaches. Finally, we provide an analysis of the positive and negative aspects of the reviewed techniques and some suggestions for future research.
Thirdly, we introduce a novel, hypergraph-based approach to establish the first unified model for various active battery equalization systems. This model reveals the intrinsic relationship between battery cells and equalizers by representing them as the vertices and hyperedges of hypergraphs, respectively. With the developed model, we identify the necessary conditions for all equalization systems to achieve balance through controllability analysis, offering valuable insights for selecting the number of equalizers. It can be used to solve the cell imbalance problem in the battery pack charging process.
Firstly, we have proposed an enhanced hybrid stacking-boosting ensemble model for EV charging duration prediction. The time index features, EV information features, and historical statistical features are selected as the prediction model input. Then, an ensemble model is designed by combining multiple linear regression, support vector regression, random forest, XGBoost, and artificial neural network. Experimental results show that better prediction performance (less root mean square of prediction error) can be achieved compared with the commonly used single machine learning model. At last, an optimal charging algorithm is proposed for EV battery charging to meet the EV user charging demand that can extend the lifetime of the battery.
Secondly, we have conducted a comprehensive review of the state of charge estimation techniques. Over the last decade, numerous attempts have been made to effectively analyze and compare the state of charge estimation methods for commercial Li-ion batteries. However, they seldom reflect on the state of charge estimation techniques based on a control-oriented viewpoint for a Li-ion battery system. To fill this gap, we have reviewed the most up-to-date battery state of charge estimation methods applied to lithium-ion battery systems. They are broadly classified as open-loop-based, closed-loop-based, and hybrid approaches. Finally, we provide an analysis of the positive and negative aspects of the reviewed techniques and some suggestions for future research.
Thirdly, we introduce a novel, hypergraph-based approach to establish the first unified model for various active battery equalization systems. This model reveals the intrinsic relationship between battery cells and equalizers by representing them as the vertices and hyperedges of hypergraphs, respectively. With the developed model, we identify the necessary conditions for all equalization systems to achieve balance through controllability analysis, offering valuable insights for selecting the number of equalizers. It can be used to solve the cell imbalance problem in the battery pack charging process.
The research results contribute to advancements in the field of electric transportation by introducing an EV charging demand prediction and a healthy optimized charging strategy.
Compared to the state-of-the-art charging methods, such as fast charging, the research results dynamically adapts charging based on real-time battery conditions and user behavior, offering superior performance, efficiency, or longevity. Moreover, the implementation of developed balancing techniques can ensure uniform states of charge across all cells in the battery pack, addressing a critical challenge in enhancing their usable capacity and lifetime.
These results have the potential to revolutionize the way batteries are charged, benefiting end-users, manufacturers, and environmental sustainability. The impacts include: 1) Extended battery lifespan, i.e. reducing battery degradation contributes to lower replacement rates and waste, supporting sustainability goals. 2) Cost savings for users, i.e. healthier batteries require fewer replacements, reducing long-term costs. 3) Scalability across industries, i.e. this technology can be adapted to various applications, including electric vehicles, consumer electronics, and renewable energy storage systems.
Compared to the state-of-the-art charging methods, such as fast charging, the research results dynamically adapts charging based on real-time battery conditions and user behavior, offering superior performance, efficiency, or longevity. Moreover, the implementation of developed balancing techniques can ensure uniform states of charge across all cells in the battery pack, addressing a critical challenge in enhancing their usable capacity and lifetime.
These results have the potential to revolutionize the way batteries are charged, benefiting end-users, manufacturers, and environmental sustainability. The impacts include: 1) Extended battery lifespan, i.e. reducing battery degradation contributes to lower replacement rates and waste, supporting sustainability goals. 2) Cost savings for users, i.e. healthier batteries require fewer replacements, reducing long-term costs. 3) Scalability across industries, i.e. this technology can be adapted to various applications, including electric vehicles, consumer electronics, and renewable energy storage systems.