Lithium Battery SOC Estimation Method Study Based On Principal Component Analysis

In recent years, environment protection and energy consumption has becomean urgent problem for further development, which has accelerated the prosperityof electric vehicle (EV) and hybrid electric vehicle (HEV) industry, focusing onenergy conservation and emission reduction. And Li-ion battery, as the basicenergy source of newly emerging vehicle, has obtained close attention everbefore.In the common research and maintenance, prediction of state of charge(SOC), is acted as an essential parameter for EV and HEV while generatingcontrol strategy for the whole vehicle. The precision and method for theprediction has also been the topic of researchers all over the world. However,there has been no other prediction method, which can obtain a more accurateprediction result. In addition, principle component analysis (PCA) has alwaysacted as multi-component analysis method in extracting essential factors basedon simplification and compression of data. The factors of which can be applied toconduct advanced analysis and prediction. Therefore, PCA method can be appliedto establish SOC prediction model with good theoretical and practicalsignificance.Corresponding experiments aimed at the performance parameters affectingSOC are conducted to obtain the essential factors. Feasibility for SOC predictionwith PCA is proved based on the theoretical analysis of PCA, and in advance anSOC prediction model based on PCA along with least square regression (LSR).Corresponding simulation and experiments are conducted to verify the proposedmodel. PCA cannot manage to extract non-linear factors in the parameters, whichmay induce prediction error based on the analysis result.To amend the prediction error, kernel principle component analysis (KPCA) is applied to establish the prediction model. Corresponding simulation andexperiments are conducted to verify the proposed model. The results havedescribed that the amendment model should also improve the variety intemperature and deterioration degree of battery based on actual workingcondition. The revised model has obtained better prediction results than purePCA model based on corresponding simulation and experiments.The simulationresults have shown that the revised model can be applied to various workingcondition with superior real-time ability, reliability and improved precision. Theaverage error of the prediction results can be reduced to1.46%in comparisonwith Ah measurement.