Deep reinforcement learning based state of charge estimation and management of electric vehicle batteries

• Published in: IET smart grid 2023-08, Vol.6 (4), p.422-431
• Main Authors: Saba, Irum, Tariq, Muhammad, Ullah, Mukhtar, Poor, H. Vincent
• Format: Article
• Language: English
• Subjects: Algorithms; Alternative energy sources; battery powered vehicles; Climate change; Controllers; Deep learning; deep reinforcement learning; Electric charge; electric vehicle charging; Electric vehicles; Energy storage; Environmental management; Fuzzy logic; Lithium; Methods; Parameter estimation; Peak load; Power; Product safety; Rechargeable batteries; Smart grid; smart grid devices; State estimation; State of charge; Vehicle-to-grid
• ISSN: 2515-2947; 2515-2947
• DOI: 10.1049/stg2.12110

In vehicle‐to‐grid (V2G) networks, electric vehicle (EV) batteries have significant potential as storage elements to smooth out variations produced by renewable and alternative energy sources and to address peak demand catering to smart grids. State estimation and management are crucial for assessing the performance of EV batteries. Existing approaches to these tasks typically do not include the effect of various parameters like route type, environmental conditions, current, and torque to estimate the state of charge (SoC) of EV batteries. In experiments, it is observed that the overall driving cost is affected by these parameters. A new method based on deep reinforcement learning is proposed to estimate and manage the SoC of nickel‐metal hybrid batteries, with an emphasis on the realisation of the parameters that affect a battery's health. The proposed deep deterministic policy gradient‐based SoC estimation and management for EV batteries, under the effect of battery parameters, are compared with the existing state‐of‐the‐art models to validate their usefulness in terms of overall battery life, thermal safety, and performance. The proposed method demonstrates an accuracy of up to 98.8% in SoC estimation and overall driving cost with less convergence time as compared to the state‐of‐the‐art models for EV batteries. In vehicle‐to‐grid (V2G) networks, electric vehicle (EV) batteries have significant potential as storage elements to smooth out variations produced by renewable and alternative energy sources and to address peak demand catering in smart grids. State estimation and management are crucial for assessing the performance of EV batteries. Existing approaches to these tasks typically do not include the effect of various parameters like route type, environmental conditions, current, and torque to estimate the state of charge (SoC) of EV batteries. In experiments, it is observed that the overall driving cost is affected by these parameters. A new method based on deep reinforcement learning is proposed to estimate and manage the SoC of nickel‐metal hybrid batteries, with an emphasized realisation of the parameters that affect a battery’s health. We proposed deep deterministic policy gradient‐based SoC estimation and management for EV batteries, in which under the effect of battery parameters we compared the results with the existing state‐of‐the‐art models to validate its usefulness in terms of overall battery life, thermal safety, and performance. We pro