Engineering Implementation of SOH Estimation and RUL Prediction for Power Battery Based on Sparse Sampled Data

In this study, the SOH estimation is first obtained using the inverse ampere-hour integral method on sparsely sampled battery operation data with a 30-second cycle in the new energy vehicle big data platform. Next, the Random Forest metric feature variables are used, and the driving mileage and the driving temperature with high importance are selected as the feature variables. The mileage prediction is made using the SARIMA fitting method, and the temperature prediction is made using the Random Forest metric feature variables. The approach of multiple linear regression is used to determine the final SOH prediction. In order to obtain the relevant RUL forecast value, 85% of the SOH is used as the failure threshold, and the advance prediction time is set to be at least 150 days. This approach may achieve the SOH prediction RMSE error within 5% and the RUL prediction error within 500 km for the measured cars by comparing the deviation between the predicted values and the actual estimated values. The prediction method used in this article is more robust than using previous SOH estimating methods directly because of the generally consistent driving behavior and temperature cycle in this method. A big data platform for new energy cars can use this technology to monitor a huge number of vehicles because it just requires a modest amount of compute.