Improved Harris Hawk-Optimized Attentive Deep Belief Network for Reliable Lithium-Ion EV Battery SOC–SOH Prediction

Resumo

Lithium-ion batteries are an integral part of modern electrification systems, which provide power for Electric Vehicles (EVs), and many portable smart devices. Therefore, in order to maintain an accurate battery management system, accurate prediction of a lithium-ion battery’s State of Charge (SoC) and its State of Health (SoH) is necessary. Since these batteries degrade slowly over time, it results in deterioration and decreased performance. This work describes a sophisticated data-driven Deep Learning (DL) model that accurately estimates the prediction of SoC and SoH. The model’s creation begins with a thorough data preprocessing step, where all missing values, anomalies, and redundancies are addressed to maintain the integrity of the sensor data used to build the model. A complete exploratory analysis of the data aids in identifying the degradation-sensitive features of the data. A combination of two layers, a Deep Belief Network and an Attention Mechanism (DBN-AM), is used in the prediction stage to formulate a model that demonstrates non-linear dynamics of deterioration and significance assignment of the temporal dynamics to represent long-range temporal dependencies. The hyperparameters in this model are adjusted by means of Improved Harris Hawk Optimisation (IHHO) to enhance the proficiency and convergence stability of the prediction stage. This framework is a reliable and intelligent solution for predictive management of batteries for usage in real world applications and the results of simulation experiments carried out in python and achieves the lowest error. For SoC, MSE of 0.001, RMSE of 0.023 and MAE of 0.017, provides an accurate estimation of the battery charge and for SoH, MSE of 0.001, RMSE of 0.032 and MAE of 0.019 provides an accurate estimation of health of the battery.