【摘要】 针对一般的电池峰值功率状态估算只考虑单个因素的影响,或使用的电池模型相对简单等问题,提出一种新的估算方法.以三元锂动力电池作为研究对象,综合考虑了电池的温度、荷电状态和欧姆内阻等对电池放电峰值功率的影响,运用Matlab中的神经网络工具箱与Matlab语言编程,建立了基于数据统计和机器学习的模拟退火(simulated annealing, SA)+BP神经网络混合算法的神经网络电池模型.采用混合脉冲功率特性测试方法进行试验,共得到245组有效的试验数据,其中200组数据作为训练样本,45组数据作为测试样本.对比了单一BP算法和SA+BP混合算法训练模型的仿真结果,证明SA+BP混合算法训练的电池模型估算精度更高,能更加准确地描述电池的功率特性.更多还原

【Abstract】 To solve the problems that only single factor was considered for battery peak power estimation and the battery model was simple, a new estimation method was proposed. Taking the ternary lithium power lithium batteries as research object, considering the comprehensive effects of battery temperature, charging state and ohm resistance on the power state, a neural network battery model on simulated annealing and back propagation(SA+BP)hybrid algorithm was eatablished based on data statistics and machine learning by neural network toolbox and Matlab programming. Hybrid pulse power characteristic(HPPC) method was used to conduct the experiments. 245 groups valid experimental data were obtained with selected 200 groups experimental data as training samples, and the rest experimental data were used as test samples. The simulation results of the training model by single BP algorithm were compared with those by SA+BP hybrid algorithm. The results show that the trained model based on SA+BP hybrid algorithm has better estimation accuracy, which can more accurately describe the peak power of battery.更多还原