Article: MultiParamNet: a multi-parameter fusion approach for anomaly detection in DC power supply systems Journal: International Journal of Wireless and Mobile Computing (IJWMC) 2026 Vol.31 No.1 pp.42 - 52 Abstract: To address the limitations of traditional abnormal state detection methods for Direct Current (DC) power systems, this paper proposes an intelligent anomaly detection approach based on multi-parameter fusion. The proposed method fully exploits the temporal dependencies and cross-domain correlations among multiple heterogeneous parameters, including voltage, current, temperature, battery state, communication signals and control logic, thereby constructing a unified high-dimensional feature representation. A deep temporal modelling network is developed by integrating Gated Recurrent Units (GRU) with a multi-scale attention mechanism, enabling accurate perception of dynamic operating states and early identification of abnormal behaviours. Furthermore, a contrastive learning-based self-supervised pre-training strategy is introduced to enhance the model's generalisation capability and feature discrimination under limited labelled data. An isolation forest algorithm is then employed for graded classification and interpretability analysis of multiple types of anomalies. Experiments conducted on real-world data sets from a DC power supply system demonstrate that the proposed method achieves a high anomaly detection accuracy of 95.2%, significantly outperforming traditional statistical models and state-of-the-art deep learning approaches. Inderscience Publishers - linking academia, business and industry through research

Title: MultiParamNet: a multi-parameter fusion approach for anomaly detection in DC power supply systems

Authors: Yu Zhang; Chuanqi Shen; Jinlong Zhang; Lutong Zhang; Luansong Yue; Mingyue Fan; Wei Liu; Qing Lei; Shengnan Cui

Addresses: State Grid Jilin Electric Power Co., Ltd., Songyuan Power Supply Company, Songyuan, China ' State Grid Jilin Electric Power Co., Ltd., Songyuan Power Supply Company, Songyuan, China ' State Grid Jilin Electric Power Co., Ltd., Songyuan Power Supply Company, Songyuan, China ' State Grid Jilin Electric Power Co., Ltd., Songyuan Power Supply Company, Songyuan, China ' State Grid Jilin Electric Power Co., Ltd., Songyuan Power Supply Company, Songyuan, China ' State Grid Jilin Electric Power Co., Ltd., Songyuan Power Supply Company, Songyuan, China ' State Grid Jilin Electric Power Co., Ltd., Songyuan Power Supply Company, Songyuan, China ' State Grid Jilin Electric Power Co., Ltd., Songyuan Power Supply Company, Songyuan, China ' State Grid Jilin Electric Power Co., Ltd., Songyuan Power Supply Company, Songyuan, China

Abstract: To address the limitations of traditional abnormal state detection methods for Direct Current (DC) power systems, this paper proposes an intelligent anomaly detection approach based on multi-parameter fusion. The proposed method fully exploits the temporal dependencies and cross-domain correlations among multiple heterogeneous parameters, including voltage, current, temperature, battery state, communication signals and control logic, thereby constructing a unified high-dimensional feature representation. A deep temporal modelling network is developed by integrating Gated Recurrent Units (GRU) with a multi-scale attention mechanism, enabling accurate perception of dynamic operating states and early identification of abnormal behaviours. Furthermore, a contrastive learning-based self-supervised pre-training strategy is introduced to enhance the model's generalisation capability and feature discrimination under limited labelled data. An isolation forest algorithm is then employed for graded classification and interpretability analysis of multiple types of anomalies. Experiments conducted on real-world data sets from a DC power supply system demonstrate that the proposed method achieves a high anomaly detection accuracy of 95.2%, significantly outperforming traditional statistical models and state-of-the-art deep learning approaches.

Keywords: DC power system; multi-parameter fusion; anomaly detection; deep temporal modelling; self-supervised learning; operational state assessment.

DOI: 10.1504/IJWMC.2026.155329

International Journal of Wireless and Mobile Computing, 2026 Vol.31 No.1, pp.42 - 52

Received: 15 Aug 2025
Accepted: 07 Nov 2025
Published online: 30 Jul 2026 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article

Title: MultiParamNet: a multi-parameter fusion approach for anomaly detection in DC power supply systems

Keep up-to-date