A Data-driven Model for Prediction of Lithium Battery State of Health
To address the problems of low accuracy and poor robustness of traditional data-driven battery state of health prediction model,this paper develops a time series prediction model fusing both nonlinear and linear branches.The nonlinear time series prediction branch is formed by a one-dimensional convolutional neural network with a multi-size parallel structure in series with a bidirectional gated recurrent neural network,and the linear branch is constructed by an autore-gressive model.Two branches in parallel output prediction results through a fully connected lay-er.The above prediction model has the generalization ability of the nonlinear part and the memo-ry ability of the linear part,and is more sensitive to the change of input amplitude.The whale optimization algorithm is used to effectively search the optimal model hyper-parameters.The ef-fectiveness and superiority of the linear nonlinear fusion prediction model proposed in this paper are verified by comparing existing models and ablation experiments.
battery state of healthbidirectional gated recurrent neural networkautoregressive modelwhale optimization algorithm
万方数据
维普
