State of Charge and State of Energy Co-estimation for Lithium-ion Battery Based on PatchTST Model

To optimize battery energy management, enhance battery safety performance, and improve its utilization efficiency, it is crucial to accurately estimate the State of Charge (SOC) and State of Energy (SOE) of lithium-ion batteries. Since the Transformer network is designed for natural language processing problems and has some drawbacks in long time series prediction tasks, this paper adopts a Transformer-based multivariate time series prediction model called Patch Time Series Transformer (PatchTST) to co-estimate battery SOC and SOE. The model uses Patch operations to extract local semantics from time series data, which makes its training faster and capable of handling longer input sequences. Compared with other Transformer-based models at room temperature, the proposed model improves SOC and SOE estimation by 36.8% and 41.1% respectively on the RMSE metric. The PatchTST was compared with several other algorithms, The algorithm demonstrates increased estimation accuracy for both SOC and SOE across various environmental temperature conditions.