Estimation of Lithium Battery Soc Based on the Gwo-Cnn-Gru Hybrid Model

Accurate prediction of the State of Charge (SOC) of lithium batteries is crucial for enhancing the driving range of electric vehicles and establishing Battery Management Systems (BMS). To improve prediction accuracy, this paper proposes a hybrid prediction model based on the Grey Wolf Optimizer (GWO), Convolutional Neural Networks (CNN), and Gated Recurrent Unit (GRU) neural networks. Initially, several grey wolf individuals are randomly initialized, each corresponding to a set of GRU hyperparameter configurations. The fitness of each individual is assessed to identify the outstanding performers among all individuals, and their positions are updated according to the strategies corresponding to the “Alpha, Beta, Delta“ roles. This process is iterated until the optimal parameters are reached. CNN is used to extract multi-channel charging characteristics, while GRU is used to extract historical capacity data features related to time-dependent discharge characteristics, thereby establishing a GWO-CNN-GRU hybrid model for prediction. The results show that the GWO-CNN-GRU hybrid model has good predictive effects on the SOC of lithium batteries, with residual values within 0.1.