摘要
介绍了几种常用的电池等效模型,通过试验选择了适用于磷酸铁锂电池的Thevenin模型并辨识了模型参数;分析了EKF算法和BP神经网络原理,提出了BP-EKF算法,使用BP神经网络的自学习能力和逼近能力,优化和补偿EKF算法的非线性误差,同时降低了等效模型的精度要求;使用UDDS电流模拟汽车行驶电流设计了仿真试验,同时使用BP-EKF算法和EKF算法对数据进行处理,结果表明,当SOC初值误差较大时,BP-EKF算法可在300 s内接近理论值,且其收敛精度比EKF算法提高了70%以上。
Several frequently-used battery equivalent models were introduced, the Thevenin model which was suitable for LFP battery was selected through test, and parameters were identified. Principle of EKF algorithm and BP neural network were analyzed, and BP-EKF algorithm was proposed, which used the self-learning ability and approximation capability of BP neural network to optimize and compensate non-linear error of EKF, lower accuracy requirement of equivalent model. The simulation experiment was designed using UDDS current to simulate vehicle driving current, and BPEKF and EKF were applied to process simulation data. The result shows that, when initial error of SOC is greater, BP-EKF algorithm can close to theoretical within 300 s, and its convergence precision increases by 70% compared with EKF algorithm.
出处
《汽车技术》
CSCD
北大核心
2018年第2期19-23,共5页
Automobile Technology
基金
院级自然科学重点项目(YJKJ2016-05)
