Recent advances in artificial intelligence and deep learning have spurred a paradigm shift in gas dispersion modeling, transitioning from traditional methods such as Frozen Cloud Analysis and Response Surface Methodology to machine learning and deep learning-based approaches. The existing Bayesian Regularization Artificial Neural Network gas dispersion prediction model is more accurate than traditional models, but is limited to single gas scenarios and static predictions. This study integrates a new Convolutional Neural Network combined with Transformer model by combining CNN extraction with Transformer attention to predict gas dispersion from offshore platforms using a dataset generated from Flame Acceleration Simulator simulations. Unlike earlier models, the CNN_Transformer model captures the entire dynamic process of gas dispersion by predicting temporal variations in dispersion and managing multiple gas components—including methane, ethane, propane, and gas mixtures—while considering critical environmental factors such as wind speed, wind direction, and internal pressure. Through comparative analysis of BRANN, CNN and CNN_Transformer, the CNN_Transformer model achieved a higher prediction accuracy, and the error between the predicted value and the FLACS simulation data was within 30 %. Moreover, while FLACS simulations—used as a benchmark—typically require hours to compute, the CNN_Transformer model delivers predictions in seconds, making it a promising tool for real-time explosion risk assessment and intelligent platform deployment.
扫码关注我们
求助内容:
应助结果提醒方式:
