An On-line SOH estimation method for power battery under low sampling rate

In the process of online state monitoring of electric vehicle power battery, the higher sampling rate can improve the prediction accuracy of the regression model to some extent, but it will lead to an increase in storage and computation costs. How to further improve the prediction accuracy of data-driven SOH estimation algorithm with low sampling rate is the key problem in the engineering application of this kind of algorithm. To solve this problem, a feature reconstruction algorithm of power battery condition monitoring data stream based on double sliding windows has been proposed. In this algorithm, several adjacent historical discharge cycles with similar SOH are defined as discharge cycle window, several adjacent voltage intervals and their corresponding average accumulated capacity are defined as input feature window. The algorithm applies data stream mining technology to these two windows, and reconstructs the voltage, current, time and other continuous input information detected in a single discharge cycle into a new input feature consisting of ”voltage interval - cumulative capacity” of several adjacent discharge cycles. The experimental results show that the prediction model trained with reconstructed input features can achieve higher prediction accuracy than the prediction model trained with original input features at lower sampling rates and lower hardware resource consumption for SOH prediction using short-duration discharge data.