Charge and discharge energy prediction model of lithium-ion battery based on ECA-PatchTST

Lithium-ion battery energy is affected by multidimensional charge and discharge parameters and cycle life, resulting in insufficient energy measurement accuracy during charge and discharge process, which affects battery management and distribution of energy storage power station. In this paper, an Efficient Channel Attention-Patch time series Transformer (ECA-PatchTST) method for energy estimation of Li-ion batteries is proposed. By introducing an independent channel input mechanism, the ECA-PatchTST model inputs the time series data of different rounds separately, and calculates the attention between channels independently for each sequence. The Patch mechanism is introduced in the model to fragment each channel sequence, which increases the feature extraction ability of the model for long time series data. By introducing onedimensional adaptive convolution to capture inter channel interaction attention, the mechanism further enhances useful information and supress non-important information. Experimental results show that the accuracy of ECA-PatchTST proposed in this paper can reach $\mathbf{9 9. 1 7 \%}$ under the condition of battery energy attenuation, which proves that ECA-PatchTST can realize battery energy measurement under complex environment with high accuracy, and has effective engineering application value.