Synthetic Time-Series Data Generation with 3D Convolution for EV Systems
Xudong Hu, Guihai Zhang, Biplab Sikdar · 2024
As electric vehicles (EVs) gain widespread acceptance for sustainable transportation, robust testing and validation for related technologies are becoming difficult due to challenges in acquiring real-world data due to limited availability, high costs, and privacy concerns. To address this issue, this paper introduces the 3D-time-series Generative Adversarial Network (3DTS GAN) to generate high-resolution, multivariate synthetic driving data for EV systems. Integrating Auto-encoder and GAN structures, the proposed method addresses the shortcomings of existing data generation methods, offering a more comprehensive representation of driving data. Evaluation results show that this method is able to generate synthetic data that is similar to original driving data with higher similarity scores than those attained using existing methods. Moreover, a functional check is done to demonstrate that there is no significant difference between using the original driving data and the synthetic data to perform further tasks such as energy consumption prediction.