Modern Methods of Data Preprocessing to Increase the Accuracy of AGV Battery Discharge Forecast

To minimize its idle time automated guided vehicle (AGV) is an important task of predicting its battery discharge. This is important for increasing the efficiency of resource use and ensuring business continuity in production environments where people work intensively. To improve the quality of AGV battery discharge data, modern processing methods are used: data cleaning, normalization, dimensionality reduction, interpolation, extrapolation, noise filtering and anomaly detection. Various methods were investigated to supplement the missing instantaneous energy consumption data of the AGV using the Formica 1 example. Data processing methods were applied to improve the prediction accuracy, including simple approaches such as averaging and filtering, as well as more complex methods such as Kalman and Wiener filters. In addition, machine learning algorithms such as RNNs and autoencoders have been used to recover lost data. The results showed that neural network-based methods are the most accurate for complex nonlinear data, while simpler methods have the advantage of speed and ease of use. The MSE values highlight that RNNs have the highest accuracy (0.468), while methods such as Wiener filter (0.501) and autoencoders (0.504) trade speed for slightly lower accuracy. The Kalman filter (0.494) provides good balance for linear systems, especially where noise suppression is key.