An improved parameter identification and radial basis correction-differential support vector machine strategies for state-of-charge estimation of urban-transportation-electric-vehicle lithium-ion batteries

Abstract

The State estimation and determination of time-varying model parameters are crucial for ensuring the safe management of lithium-ion batteries. This paper designs a limited memory recursive least square algorithm to improve the accuracy of online parameter identification. An adaptive radial basis correction-differential support vector machine model is constructed to correct the state of charge value by considering the dynamic characteristic parameters. It greatly reduces estimation error and noise, while monitoring the critical conditions for safe and reliable online battery operation. The estimation effects of the proposed model are verified under hybrid pulse power characterization and dynamic stress test working conditions. The maximum error values obtained are 0.037 % and 0.336 %, respectively, thus achieving high-precision estimation. The proposed method is adaptive to real-time battery management applications, laying a foundation for robust state estimation of lithium-ion batteries used in urban transportation electric vehicles.