Abstract: As the core of electric vehicles, the power battery's state of health (SOH) is an important indicator to characterize whether the battery can work normally, indicating the current service life and reliability of the battery, and directly affecting the performance of the battery. Accurately estimating the SOH of the battery can predict the overall life of the lithium-ion battery and improve the charging and discharging strategy to avoid the occurrence of battery abuse and other failures. In order to accurately predict the health state of power battery, this paper selects the characteristic parameters that have a strong correlation with the health state of battery as the health factor of health state prediction, and designs and trains the NARX nonlinear autoregressive neural network. By establishing different training sets and control group of input characteristic parameters to analyze and compare the influence of training sets and input parameters on the prediction results, obtaining accurate battery health status value can improve the power performance of electric vehicles.

Key words: Electric vehicles; Battery health state prediction; Fault diagnosis; NARX neural network

摘要： 动力电池作为电动汽车的核心，其健康状态（SOH）为表征电池能否正常工作的重要 指标，表示电池当前的使用寿命及其可靠性，并直接影响电池的性能。准确估计电池的 SOH 能够预知锂离子电池的整体寿命，完善充放电策略，以避免电池滥用等故障的发生。为确保 对动力电池的健康状态进行准确预测，文章选择与电池健康状态具备极强相关性的特征参数 作为健康状态预测的健康因子，设计并训练 NARX 非线性自回归神经网络，通过建立不同的 训练集和输入特征参数的对照组去分析对比训练集和输入参数带给预测结果的影响，获取精 确的电池健康状态值，能够提高电动汽车的动力性。

关键词: 电动汽车；电池健康状态预测；故障诊断；NARX 神经网络