Machine learning-based optimization of battery charging speed with health aware constraints

The energy storage systems are under pressure in terms of finding smart charging systems that are safe and fast, there is more demand of the high-performance and long life of the system. Constant-current constant-voltage (CC-CV) charging, though popular, is known to encourage high temperature and extreme rapidity in the losses of capacity in highly dynamic charging. To address this problem, this paper will introduce a Machine Learning (ML) Based Optimization Framework of adaptive battery charging that will maximize the charging rate and the battery will not be damaged. This framework has been used to combine deep learning and Reinforcement Learning (RL) to actively regulate the charging current and voltage, based on the current values of the battery state, e.g., State of Charge (SoC), State of Health (SoH), temperature, and internal resistance. A multi-objective rewarding process has been created to optimalize the working time, thermal stability and degradation rate as a combination. Some comparative analysis was conducted among different methods including Conventional CC -CV, Rule-Based fast Charging, LSTM-RL and the developed Transformer-RL model. The results indicate that Transformer-RL controller allowed decreasing the total time of charge by about a quarter of 39 percent, top temperature by less than 41 o C, and cycle life by an average of 35 percent comparing to the traditional methods. The convergence of rewards analysis supported the stable learning and good trade-off between the performance and the safety. The findings report the potential to expand Transformer-RR optimization on building health-conscious, intelligent, and self-adaptive charging protocols of the next-generation electric vehicles and renewable energy storage systems. It relies on the approach of real-time launching of smart charging systems which can balance energy efficiency and long-term battery endurance.