【摘要】 针对传统BP神经网络在线估算锂离子电池健康状态(state of health,SOH)容易使权值陷入局部最优解,导致SOH预测不精确。结合模拟退火(simulate anneal,SA)算法能有效收敛于全局最优的特点,提出一种基于SA算法优化BP神经网络的锂离子电池SOH在线预测方法。以锂离子电池为研究对象,分析了微分电压、欧姆内阻、循环次数与电池SOH的关系,并以此作为电池的健康状态因子(health indicator,HI)输入至BP神经网络。利用SA算法优化BP神经网络的权值,使预测模型得到最优解。实验结果表明:利用优化算法对电池SOH进行预测,其最大误差仅为1.98%,平均误差为1.09%。相较于传统BP神经网络,优化算法预测最大误差降低了5.62%,平均误差降低2.33%。从而验证了基于SA算法优化BP神经网络能够获取全局最优值并提高电池SOH估算精度是有效的。更多还原

【Abstract】 Estimating the state of health(SOH) of lithium ion batteries for BP neural networks tends to bring the weights fall into local optimal solutions,leading to the problem of inaccurate SOH prediction.Combined with the simulated annealing(SA) algorithm,it can effectively converge to the global optimal characteristics.A method based on simulated annealing optimized BP neural network for lithium ion battery SOH prediction is proposed.Taking lithium ion battery as the research object,the relationship between differential voltage,ohmic internal resistance,cycle number and battery SOH was analyzed,and the health indicator(HI) of the battery was used as the input of BP neural network.The SA-BP algorithm is used to optimize the weight of the neural network,so that the prediction model obtains the optimal solution.The experimental results show that the optimal algorithm is used to predict the battery SOH,and the maximum error is only 1.98%,and the average error is 1.09%.Compared with the traditional BP neural network,the optimization algorithm predicts that the maximum error is reduced by 5.62% and the average error is reduced by 2.33%.It is verified that the SA optimized BP neural network can obtain global optimal values and improve the accuracy of battery SOH estimation.更多还原