Remaining Useful Life Prediction of Batteries Using Long Short-Term Memory Networks Based on Variable-Order Fractional Gradient Descent Method

With the popularity of the battery industry in daily life, in order to avoid accidents and economic losses caused by internal aging of batteries during use, it is extremely important to effectively and accurately predict the remaining useful life (RUL) of the battery. In order to improve the accuracy of RUL prediction of batteries, this paper proposes a new long short-term memory (LSTM) neural network prediction method based on the variable order fractional gradient descent method, combining the advantages of the fractional gradient descent method and the long short-term memory neural network. First, a multi-layer LSTM model is designed to capture long-term dependencies in battery sequence data. Secondly, in the reverse parameter update process, the variable-order fractional gradient descent algorithm is introduced as the training optimizer of the model. Finally, the trained neural network model is used to predict the RUL in the future based on existing battery indicators. Experimental results show that the LSTM model based on variable-order fractional gradient descent has higher prediction accuracy and stability than the traditional prediction model, and effectively improves various performances of battery life prediction.