Online parameter identification based on split battery model

The battery state of charge (SOC) estimation is still one of the key technologies to be solved in the battery management system. The estimation accuracy plays a decisive role in other key issues such as vehicle safety management, residual power monitoring, and battery pack equalisation control. The second-order RC model is a common battery model for SOC estimation. However, there are mutual interferences between the state variables of the traditional second-order RC model, which causes unnecessary oscillations in the initial SOC estimation and affects the stability and accuracy of estimation. Although the split battery model (SBM) can effectively solve this problem, the parameter identification of the model still uses offline parameter identification, which cannot effectively track the model parameters affected by temperature, resulting in low identification accuracy and poor tracking performance. To solve this problem, this paper first analyses how the model solves the problem of mutual interference between state variables; then, through the measurement data and curve simulation obtained by the battery detection system, an OCV-SOC relationship curve is established; then a online parameter identification method, Recursive Least-Squares with Forgetting Factor(FFRLS) based on split battery model is proposed, to make up for the problem that the original offline parameter identification cannot solve the model parameters influenced by temperature, so as to achieve rapid mode parameters identify and improve the recognition accuracy on-line; finally the Unscented Kalman Filter(UKF) algorithm is used to estimate the SOC and verify the effect of the parameter recognition. This article takes a 3.3V 20Ah lithium iron phosphate battery as the research object, through parameter identification comparison experiments and SOC estimation and verification experiments, verifying that FFRLS can effectively improve the estimation accuracy of the model parameters and improve the SOC estimation effect.