Real-Time Spontaneous Combustion Risk Assessment of Coal Based on Edge Computing and Federated Learning

Spontaneous combustion of coal poses a significant threat to coal mine safety, and its real-time, accurate risk assessment is crucial for fire prevention. To address the limitations of traditional centralized monitoring systems, insufficient privacy protection, and delayed model updates—this study proposes a real-time risk assessment framework for coal spontaneous combustion by integrating edge computing and federated learning. The framework deploys lightweight risk assessment models at the edge nodes within coal mines, enabling localized, real-time analysis and early warning of key parameters such as temperature, CO concentration, and O₂ concentration, thereby significantly reducing response latency. Meanwhile, a federated learning mechanism is employed to collaboratively train a global risk prediction model across multiple edge nodes without sharing raw data, effectively preserving data privacy and security. Periodic parameter aggregation and model updates enhance the model’s generalization and adaptability. Experimental results demonstrate that the proposed method achieves high-accuracy, low-latency real-time risk assessment while ensuring data privacy. Compared to traditional centralized models, the proposed approach improves early warning accuracy by 18.7% and reduces average response time by 62.3%. This study provides a feasible technical solution for intelligent, secure, and collaborative early warning of coal spontaneous combustion in complex underground environments.