A real-time early warning classification method for natural gas leakage based on random forest

Serious natural gas leakage explosion accidents have brought seriously threatening to people's lives and properties. Efficient warning classifications is of great significance to make rapid response, thus reducing the losses caused by accidents. This paper describes a novel early warning classification method for natural gas leakage based on a multi-classification random forest (RF) model, which allows evaluating the level of early warning of gas accidents timely and accurately, assisting monitoring department and gas company in timely rapid decision and scientific disposal. Fully considering the laws of natural gas leakage and the change of comprehensive risks in underground spaces adjacent to natural gas pipeline, an early warning classification index system was established, and multiple warning factors features were extracted from recorded warning events of natural gas leakage. Then the early-warning level labels of the warning events was gained by K-mean clustering and experts scoring methods. The extracted warning features and the associated early-warning level labels were used to train and validate the proposed model. The effectiveness and feasibility of this model is further verified by comparing with other popular approaches. Furthermore, the verified model is loaded into real time module, which can achieve the real time warning classification. The research results demonstrated that the proposed method can timely and accurately classify the levels of the early warning events. The prediction accuracy of the natural gas leakage early warning classification model based on the RF algorithm is 88.02 %. For real time warning events, rapid decision can be made according to the characteristics of early-warning grades, and the emergency disposal can be guided more effectively based on the warning classification results.