基于DESSA-DESN和NCA的锂离子电池剩余寿命预测

Remaining useful life prediction of lithium-ion batteries based on the DESSA-DESN model and the NCA algorithm

锂离子电池的剩余使用寿命(RUL)对于锂离子电池在设备中的管理、使用至关重要,为了提高RUL的预测精度,本工作提出一种基于混合差分进化-麻雀搜索算法(DESSA)优化的深度回声状态网络(DESN)和邻域成分分析法(NCA)的锂离子电池RUL预测方法。首先,对锂离子电池的容量衰减特性进行分析,对于多种能够描述电池老化状态的间接健康指标,利用NCA算法降维处理,得到4个高相关度的健康因子作为模型的输入;其次,将差分进化算法(DE)和麻雀搜索算法(SSA)相结合,将突变、交叉、筛选等操作引入SSA算法的种群更新过程中,提出混合差分进化-麻雀搜索算法(DESSA)算法,利用DESSA算法对DESN网络的参数进行寻优,建立DESSA-DESN预测模型。最后,利用NASA数据集和CALCE数据集对所提模型的有效性和泛化性能进行验证,并与SSA-DESN、GPR等现有方法进行比较,结果表明本工作提出的DESSA-DESN模型能够更加准确追踪锂离子电池的退化状态,具有更小的预测误差,对RUL预测结果的均方根误差(RSME)能够保持在1.5%以内,平均绝对误差(MAE)保持在1%以下。

The remaining useful life(RUL) of lithium-ion batteries is crucial in managing and using energy storage devices.To improve the prediction accuracy,this study proposes a RUL prediction method based on improved deep echo state network(DESN) and neighborhood component analysis(NCA),in which the DESN is optimized by a hybrid differential evolution(DE)-sparrow search algorithm(SSA).First,various health indicators(HIs) are selected to describe the battery aging mechanism by analyzing the capacity decay characteristics of lithium-ion batteries.The NCA is used to reduce the HI dimensionality.Four high-correlation health factors are then obtained as the model input.Next,the DE algorithm(DE) and the SSA are combined to construct the DESSA algorithm,which is used to optimize the DESN network parameters.As a result,the DESSA-DESN prediction model is established.Finally,the validity and the generalization performance of the proposed model are verified using datasets from the National Aeronautics and Space Administration and the Center for Advanced Life Cycle Engineering.The results show that,compared with existing methods(e.g.,SSA-DESN and ground penetrating radar),the proposed method more accurately tracks the degradation state of lithium-ion batteries with smaller prediction errors.The root-mean-squared error of the prediction results remains within 1.5%,while the mean absolute error remains within 1%.