A novel time series hybrid model for online prediction of electric vehicles battery pack capacity with real charging data

Abstract

Accurate on-road vehicle capacity degradation path prediction model is of great significance to guide vehicle maintenance and performance optimization and ensure safety performance. This paper proposes a novel hybrid model, which mainly focus on the characteristics of the data set and combines the random forest (RF), particle swarm optimization (PSO), variational mode decomposition (VMD), and convolutional long short-term memory (ConvLSTM) algorithm to online forecast the capacity degradation path of electric vehicles (EVs) battery pack. First, battery capacity mean values during a week are regarded as the real capacity. Then, the PSO method is mainly used to obtain the parameters of the VMD offline and online to reduce the noise of the real capacity trajectory. In addition, the importance of the extracted battery pack features is evaluated through the RF method, and features with high importance are screened out. Finally, ConvLSTM is designed to extract the spatial relationship between feature sets to obtain better prediction performance. The charging data of 20 EVs are used to verify the performance of the proposed hybrid model. The experiment results show that the performance of proposed model outperforms other baseline models and can be used to accuracy predict the battery pack capacity degradation path.