An approach towards AGV battery cell voltage prediction using DL and data pre-processing

Modern methods for solving the AGV battery cell voltage prediction problem include a symbiosis of probabilistic and machine learning methods. In this study, we propose to use two RNN-based approaches with preliminary data analysis and processing. In the first approach, we make predictions based on smoothed data. Three filters were chosen for smoothing: exponential smoothing; moving average; weighted moving average. The moving average method showed the best results, but according to the Dickey-Fuller test, the predicted signal was still not stationary. The second approach was based on a centered stationary signal obtained by subtracting the value of the moving average from the original signal.The forecast accuracy was evaluated using the following errors: MAPE, RMSE, MSPE, RMSPE, MFE, SEBIAS, ME. Adequacy of the proposed neural network model based on the coefficient of determination R2. As can be seen from Table 2, a significant improvement in prediction accuracy is observed for the centered signal. The maximum error was reduced by 5.6 times R2 increasing by 0.0596. At the same time, both models are adequate.