Research on vehicle battery data cleaning method based on OOA-VMD-ATGRU-GAN

Abstract

Battery health monitoring is influenced by environmental and human factors, resulting in the presence of abnormal and missing values in the detection data. These issues compromise the accuracy of subsequent life prediction and fault diagnosis. To address this problem, we propose a deep learning-based method for cleaning battery anomalies and imputing missing data. Initially, we optimize the Variational Modal Decomposition method using the Osprey Optimization Algorithm to minimize the influence of continuous discharge processes on local anomaly detection. This process allows us to obtain the trend of the time series, and subsequently, we determine the anomalies by using the interquartile range after removing the trend components. The identified anomalies are then converted into missing values for further processing. Next, we fill in these missing values by constructing a Generative Adversarial Network. The generator structure of the network combines the attention mechanism with the Gated Recurrent Unit. We validate our approach using a real vehicle dataset and subsequently perform prediction on the cleaned dataset. Our results demonstrate that the subsequent Long Short-term Memory Networks and Gated Recurrent Unit prediction model improves the RMSE value by approximately 35% and the MAPE value by roughly 42%. Thus, our method effectively enhances the quality of the original data.