Research on SOC of new energy electric vehicles based on THRO-CNN-Bi LSTM-AM model

Accurate prediction of battery state of charge (SOC) is essential for optimizing the battery utilization of new energy electric vehicles. This paper proposes a CNN-BiLSTM-AM model based on the Tian Ji Horse Racing Optimization (THRO) algorithm to predict SOC at road forks. The EVDATA dataset is preprocessed to handle outliers and missing values. A two-layer convolutional neural network (CNN) extracts spatial features from battery voltage, temperature, and other parameters, while a bidirectional long short-term memory network (BiLSTM) captures long-term dependencies. The THRO algorithm optimizes the model's hyperparameters, and the attention mechanism (AM) dynamically weights key feature dimensions. Experimental results show that the proposed THRO-CNN-BiLSTM-AM model effectively predicts the SOC values with high accuracy. For instance, the predicted SOC values for the next 10 time points are as follows: 46.82%, 46.67%, 46.93%, 47.45%, 47.96%, 48.48%, 48.99%, 49.48%, 49.94%, and 50.36%. The model demonstrates significant improvement over traditional models, achieving a higher prediction accuracy and better generalization to dynamic charging and discharging processes.