A Deep Learning Model Based on the Introduction of Attention Mechanism is Used to Predict Lithium-Ion Battery SOC

Abstract

In order to enhance the accuracy of state of charge (SOC) prediction for lithium-ion batteries, this paper developed a deep learning model optimized by slime mold algorithm (SMA). This model combines convolutional neural networks, bidirectional gated recurrent units, and attention mechanisms. Through SMA optimization of critical parameters in the model, the predictive performance has been significantly improved. In the experimental phase, the paper collected discharge data from an 18650 battery pack under 8 different operating conditions, totaling 10596 sets of data. These data were used to fully train and validate the model. The test results show that the new model demonstrates exceptional accuracy in SOC prediction, with average absolute error, root mean square error, and mean absolute percentage error reaching 0.46462%, 0.56406%, and 6.8028%, respectively. Moreover, the decision coefficient R reaches 0.962. This result not only surpasses single models and unoptimized models but also provides important technical support for improving the battery life and driving range of electric vehicles.