Forecasting of non-stationary battery manufacturing capacity based on Q-learning principle

The forecasting problem of lithium battery manufacturing capability is crucial for developing production plans and can be regarded as a non-stationarity time series forecasting problem. This paper proposes a combined forecasting method for battery manufacturing based on a Q-learning algorithm. First, to solve the problem of poor decomposition caused by large subjectivity in the parameter selection process of the variational mode decomposition (VMD) algorithm, this study proposes a Q-learning decision optimization algorithm to optimize the decomposition layer number and penalty factor; thus, the optimal sequence decomposition effect is obtained. Then, the number of hidden layer nodes of the long short-term memory (LSTM) was optimized. Finally, by considering the manufacturing data of a listed Chinese battery company as the experimental data, the proposed algorithm was compared with the LSTM and VMD-LSTM methods. The results illustrate that the proposed algorithm can reduce the disturbance caused by nonstationary data, optimize the model, and significantly improve the forecast accuracy.