Safety Analysis Improvement in Fire Risk Assessment Model and Optimized Risk Indexing using Deep Learning Approach

Abstract

Gas pipeline risk assessment is crucial due to many dangers and financial losses. The findings are more accurate and reliable when deep learning is used. This study's objective is to assess pipeline fire, explosion, and hazardous gas release risks using an Efficient Deep Learning Model (EDLM)-optimized risk indexing procedure. Deep Recurrent Neural Network (DRNN) and the Ebola Optimization Algorithm are combined to create the proposed EDLM (EOA). EOA is used in the DRNN to facilitate the weight update process. Utilizing MATLAB software, the evaluation of the fire risk was finished.  Using EDLM process analysis, the weight of each fire risk indexing was added up, and the fire risk level was determined using a five-state criterion system that comprised highly unpleasant, greatly desirable, favourable, moderate, and unfavourable. The ultimate risk score for a fire, explosion, and poisonous gas leak in a gas pipeline in Korea is in a favourable range. The fact that there are many different variables that might cause a fire and that they don't happen very often is proof positive of this result. To improve the accuracy of fire incidence, an efficient deep learning model is given. Furthermore, the fire risk indexing model is built upon the fire prediction model. The proposed method is put into practise and contrasted with traditional approaches like recurrent neural networks (RNN), artificial neural networks (ANN), and support vector machines (SVM), in that order. Consequently, it is expected that safety managers will find the results useful in making decisions on the risk management of gas pipelines.