Probabilistic Predictive Model for Electric Vehicle’s Battery’s State of Health

Electric vehicles (EVs) are in high demand due to environmental benefits, but conventional lithium-ion batteries (LIBs) can degrade, posing safety risks and maintenance costs. Early prediction of battery failure using multiple parameters is crucial and requires the development of an appropriate predictive algorithm for LIB state of health (SOH). There are two main categories of methods for estimating the SOH of LIB, i.e., physicsbased models and data-driven methods. Physics-based models rely on complex mathematical equations and physical knowledge of the system’s degradation behaviors, which is troublesome. Data-driven methods do not require accurate physical modelling and instead develop their own decision criteria based on training datasets. Therefore, long-short-term memory neural network (LSTMNN) model, which is a data-driven method, is a good choice for estimation of battery SOH because of its ability to learn long-term relationships between inputs and outputs. This study proposed a LSTMNN optimized by Genetic algorithm (LSTMNNGA) to overcome the issue of poor hyperparameter selection in neural networks (NNs). The robustness of the LSTMNNGA is evaluated by comparing it with the LSTMNN and an artificial neural network (ANN) model. The proposed model outperforms the approaches mentioned in the comparative study with RMSE, MAE, and MAPE of 0.0795,0.0664, and 45.31% respectively. To address the unique behavior of the test battery compared to other batteries in the dataset, potential future work could include incorporating charger voltage and impedance data, as well as experimenting with different NN architectures.