基于BiLSTM-STW神经网络的锂电池剩余容量预测  

Remaining capacity prediction of lithium battery based on BiLSTMSTWneural network

作　　者：姚俊荣 唐学用 李庆生 YAO Junrong;TANG Xueyong;LI Qingsheng(College of Electrical Engineering,Guizhou University,Guiyang Guizhou 550025,China;Power Grid Planning&Research Center Guizhou Power Grid Co.,Ltd.,Guiyang Guizhou 550003,China)

机构地区：[1]贵州大学电气工程学院,贵州贵阳550025 [2]贵州电网有限责任公司电网规划研究中心,贵州贵阳550003

出　　处：《电源技术》2023年第7期889-893,共5页Chinese Journal of Power Sources

摘　　要：针对锂电池剩余容量预测精度无法满足当前工程应用的问题,结合双向长短时记忆网络(bi-directional long short-term memory,BiLSTM)与滑动时间窗口(sliding time window,STW)算法的优点,提出一种电池剩余容量预测方法。首先分析BILSTM神经网络和STW算法原理,构建了BiLSTM-STW神经网络模型,采用自适应矩优化算法(adaptive moment estimation,Adam)对模型超参数进行优化,实现模型修正;然后选取美国国家航空航天局(National Aeronautics Space and Administration,NASA)埃姆斯研究中心锂电池数据,对数据进行处理并选取容量衰减特征数据作为神经网络的预测输入量;最后利用构建的神经网络对NASA锂电池数据集进行剩余容量预测实验。实验结果表明,所构建的神经网络模型能够精确预测锂电池的剩余容量,相比LSTM神经网络模型有更好的精确度。To address the problem that the remaining capacity prediction accuracy of lithium batteries cannot meet the current engineering applications,a battery remaining capacity prediction method was proposed by combining the advantages of bi-directional long short-term memory(BiLSTM)and sliding time window(STW)algorithms method.Firstly,the principle of BILSTM neural network and STW algorithm was analyzed,a BiLSTM-STW neural network model was constructed,and adaptive moment estimation(Adam)was used to optimize the hyperparameters of the model to achieve model correction.Then,the lithium battery data from the National Aeronautics and Space Administration(NASA)Ames Research Center were selected,and the capacity decay characteristics were selected as the prediction input of the neural network.The experimental results show that the neural network model constructed can accurately predict the remaining capacity of lithium batteries with better accuracy than the LSTM neural network model.

关 键 词：锂电池 双向长短时记忆网络 滑动时间窗口 剩余容量预测 神经网络 

分 类 号：TM912[电气工程—电力电子与电力传动]