基于数据驱动与机理模型的锂离子电池健康状态估计与拐点识别

Abstract Accurate state-of-health (SOH) estimation and knee points identification are crucial for optimizing battery performance and lifecycle management. An SOH estimation method combining an improved Newton-Raphson-based optimizer algorithm for optimizing support vector regression and an adaptive boosting algorithm (INRBO-SVR-AdaBoost) is proposed, as well as a knee point identification method considering failure thresholds based on the maximum vertical distance method. Firstly, three improvements are introduced to enhance the global search ability and convergence speed of the standard NRBO algorithm, enabling the SVR method to obtain optimal parameters. Then, the AdaBoost algorithm is applied to integrate the INRBO-SVR method, improving SOH estimation accuracy. Experimental results show that the INRBO-SVR-AdaBoost method provides higher SOH estimation accuracy than other methods, with root mean square error and mean absolute error both below 0.89 % and 0.75 %, respectively. Secondly, based on the accurate SOH estimation, an empirical model combining a double-exponential and a second-order polynomial ( SOH EM ) is constructed, and the maximum vertical distance ( VD max,LCD ) between SOH EM and the linear SOH degradation curve is calculated for different failure thresholds. By computing the maximum vertical distance ( VD max,LKC ) between VD max,LCD and the linear knee point curve for different failure thresholds, the final knee point is identified. Experimental results show that the identified knee points have an error within 46 cycles, with the identification accuracy of the knee points reaching at least 90 %, demonstrating strong flexibility and precision. The proposed high-precision SOH estimation method and flexible knee point identification method have significant guiding implications for battery life prediction and retirement management.