Reliable Field Prediction for Industrial Safety: a GPR-MRF-K Approach to Emergency Response and Risk Assessment

Accurate and efficient field prediction is crucial for real-time environmental monitoring and emergency response systems. Traditional spatiotemporal methods suffer from high computational complexity, while single time-series models inadequately capture spatial correlations. This paper presents GPR-MRF-K, a novel hybrid approach integrating Gaussian Process Regression (GPR) with Markov Random Field-enhanced Kriging (MRF-K) for rapid field prediction. GPR handles temporal dependencies and uncertainty quantification, while MRF-K enables efficient spatial interpolation through neighborhood structure optimization, reducing computational complexity from O(N3) to O(n3). Experimental validation using a 12 -sensor ammonia leakage model demonstrates that GPR-MRF-K achieves 67 % lower mean square error compared to LSTM-based methods and traditional Kriging, with an average prediction time of 37 seconds. The approach shows significant potential for environmental monitoring, industrial safety, and emergency response applications requiring both accuracy and real-time performance.