Journal of Engineering, Project, and Production Management, 2026, 16(3), 2026-0020
Substation Surroundings Hazard Detection and Change Monitoring Using Multi-Source Satellite Remote Sensing and Deep Learning
1 Assistant Engineer, Power Transformation Department,
Guiyang Power Supply Bureau, Guiyang, Guizhou,550004, China, E-mail:
yangz0912@gzgy.csg.cn
2
Senior Engineer, Power Transformation Department, Guiyang Power Supply
Bureau, Guiyang, Guizhou,550004, China, E-mail: dqhp1794@outlook.com
(corresponding author).
3
Engineer, Production Technology Department, Guiyang, Guizhou,550004,
China, E-mail: 765142348@qq.com
4
Senior Engineer, Production Technology Department, Guiyang,
Guizhou,551417, China, E-mail: 78180764@qq.com
5
Senior Engineer, Production Technology Department, Guiyang,
Guizhou,563000, China, E-mail: caolei@gzdy.csg.cn
Project Management
Received Apr 22, 2026; revised May 14, 2026; accepted May 21, 2026
Available online May 29, 2026
Abstract: Improving high-voltage substation safety and reliability requires continuous monitoring of spatially heterogeneous and temporally persistent environmental hazards. Existing approaches rely on single-source satellite imagery or isolated Unmanned Aerial Vehicle (UAV) inspections and typically decouple continuous hazard estimation from categorical risk assessment. It remains unclear how to design a unified spatio-temporal framework that simultaneously reconstructs hazard intensity fields and generates operationally consistent risk levels under multi-source sensing uncertainty. This study develops a hybrid deep spatio-temporal architecture that integrates multispectral satellite data, Synthetic Aperture Radar (SAR) observations, UAV-derived anomaly cues, and environmental forcing variables within a shared regression–classification backbone for substation-surroundings hazard monitoring. The framework was trained on 2.8 million spatial samples from 2021 to 2023 and optimized via regularized stochastic gradient updates with temporal encoding and multi-source fusion. On independent tests, the model achieves a root mean square error of 0.113, a mean absolute error of 0.075, and a coefficient of determination of 0.905, explaining over 90 percent of spatial hazard variance. Risk-level classification attains 91.2 percent accuracy, a Critical-class F1-score of 0.94, and a false negative rate below 6 percent. Relative to the strongest baseline, hazard root mean square error decreases by 23 percent, and cumulative operational risk cost decreases by 12 to 16 percent compared with deep reinforcement learning alternatives. The framework provides a unified, deployable solution for proactive hazard detection, enabling early warning with lead times exceeding 30 hours before confirmed field events.
Keywords: Multi-source remote sensing data fusion, spatio-temporal hazard prediction, hybrid regression classification learning, infrastructure risk assessment, substation environmental monitoring.
Copyright © Journal of Engineering, Project, and Production Management (EPPM-Journal).
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivs 3.0 Unported License.
Requests for reprints and permissions at eppm.journal@gmail.com.
Citation: Yang, Z., Cao, J., Wang, Y., Tao, L., and Cao, L. (2026). Substation Surroundings Hazard Detection and Change Monitoring Using Multi-Source Satellite Remote Sensing and Deep Learning. Journal of Engineering, Project, and Production Management, 16(3), 2026-0020.
