Precise SOC Estimation for Li-ion Batteries based on Dual-Channel Neural Network

Accurate estimation of state of charge (SOC) is crucial to ensure the safe and efficient operation of li-ion battery. Data-driven SOC estimation is increasingly demonstrating its advantages as it can automatically extract features from the original signal to predict SOC. In this paper, we propose a SOC estimation method for li-ion batteries based on dual-channel neural network architecture. Firstly, in order to address the challenge of capturing the local feature instability of time-variant battery data, we have innovatively employed a CNN network embedded by Koopman operator to capture the local feature difference within a finite neighborhood of the dynamic system distribution. Additionally, given the complexity in mining the global nonlinear coupling relationship between battery parameters, we have utilized a Transformer module to analyze the time-invariant data of the battery at a global dimension and capture both long-term trends and coupling correlation characteristics within the battery sequence. Finally, the dual-channel architecture is employed to integrate the distribution characteristics of the local dynamic system and the global coupling characteristics in order to achieve precise estimation of SOC. The efficacy of our proposed approach is substantiated by experimental findings derived from the battery dataset.