DL-RAP: Deep-Learning based Real-time Accident Diffusion Prediction

With the rapid development of the industrial field, the frequency of accidents caused by accidental leakage and diffusion of dangerous gases is also increasing. Habitat and environmental monitoring and prediction are therefore of particular importance. Computer simulation-based analysis, as one of the current mainstream approaches, is generally too time-consuming, limiting its effectiveness in leakage-prediction-based applications. In this paper, we propose Deep-Learning based Real-time Accident diffusion Prediction model (DL-RAP) to predict the trend of gas concentration distributions over both temporal and spatial dimensions. Firstly, the COMSOL multiphysics software and the theory of fluid mechanics are utilized to produce reliable data for typical industrial environments. Then, an improved transformer neural network gas concentration diffusion prediction model is proposed, which learns the historical data and predicts the concentration distribution of gas in the future. Experiment results confirmed that the DL-RAP can predict leakage distribution both timely and accurately.