A progressive decomposition time series forecasting method for monitoring the voltage of energy storage batteries

Accurately predicting voltage is crucial for ensuring the safety monitoring of energy storage battery systems in energy storage stations. However, the battery system, as a highly nonlinear and dynamically changing complex system, faces the issue of cumulative errors in long-term forecasting. To address the poor accuracy and stability of voltage prediction in long-term scenarios for energy storage batteries, this paper proposes a Long-Term Series Forecasting model based on the progressive sequence decomposition concept and Transformer. The model consists of an encoder-decoder structure based on the Transformer network, incorporating a progressive sequence decomposition module to extract high-dimensional features from multiple semantic perspectives. This approach enables the model to capture long-term stable trends in unknown correlated feature dimensions, thereby improving the final prediction accuracy. Additionally, an improved auto-correlation mechanism network block is designed in the network to enhance the model's perception of the similarity of time series at different time spans, thereby enhancing the stability of the model in long-term series forecasting scenarios. Furthermore, this paper optimizes features for lithium-ion energy storage battery voltage data, encoding sequence features, battery parameter features, and trend features in separate domains. By effectively aggregating the correlations of various parameters, the prediction performance of the lithium-ion energy storage battery voltage forecasting model is significantly improved. Experimental results demonstrate that the proposed method exhibits good prediction accuracy and stability for voltage changes in the next 24 hours, with an average absolute error of only 0.013 compared to the actual voltage values. Through extensive comparative experiments, our method surpasses other time series models in terms of providing state-of- the-art accuracy, effectively enhancing the stability of long-term voltage forecasting.