Charging Pile Fault Prediction Model Based on GRU Network and WOA

Aihua Yang; Tianke Fang; Elcid A. Serrano; Bin Liu; Fucai Liu; Zhenxiang Chen

doi:10.37965/jait.2024.0507

Charging Pile Fault Prediction Model Based on GRU Network and WOA

Authors

Aihua Yang School of Information Technology, Mapua University, Manila, Philippines & School of Information Media, Zhangzhou Vocational College of Science and Technology, Zhangzhou, China https://orcid.org/0009-0006-4223-3891
Tianke Fang School of Information Technology, Mapua University, Manila, Philippines & School of Computing and Information Engineering, Xiamen University of Technology, Xiamen, China https://orcid.org/0009-0009-2518-2502
Elcid A. Serrano School of Information Technology, Mapua University, Manila, Philippines https://orcid.org/0000-0001-7786-9829
Bin Liu College of Software Engineering, Xiamen University of Technology, Xiamen, China
Fucai Liu Zhangzhou Vocational College of Science and Technology, Zhangzhou, China
Zhenxiang Chen School of Information Media, Zhangzhou Vocational College of Science and Technology, Zhangzhou, China

DOI:

https://doi.org/10.37965/jait.2024.0507

Keywords:

charging pile, fault prediction, GRU network, performance analysis, whale optimization algorithm

Abstract

The global energy structure is transforming, and new energy vehicles (NEVs) are becoming the future of the automobile industry. However, the development of charging piles (CPs) and related facilities has not kept pace with the growth of NEVs. This study uses the gated recurrent unit (GRU) network and the whale algorithm to construct a high-performance CP fault prediction model. The proposed model, which utilizes the whale algorithm to prevent the GRU network from falling into local optima, demonstrates improved predictive information extraction and prediction ability. The experimentally verified results indicate that the proposed model achieved 92.02% prediction accuracy, 85.66% recall, and 93.87% F1 value. Additionally, the proposed model demonstrates excellent computational ability with an average running time of under 5 minutes on both datasets. This result is a substantial reduction from the control model’s running time. The experimental findings show that the study’s suggested model has a good ability to anticipate fault data. Its sophistication is verified by comparative tests, which can provide a reference for subsequent research.