A combination model of wavelet analysis and neural network for predicting oil and gas exploration accidents

Abstract

Oil and gas exploration (OGE) accidents are usually observed in severe fatalities. How to predicting risks of accidents at worksite in a targeted way is still a challenge, but few researchers have studied this issue due to lack of accident data from the worksite. To address this problem, the accident cases for the past 50 years in a globally operated OGE company are collected as a basis, and a wavelet neural network (WNN) accident prediction model which combines the wavelet analysis and the traditional BP-neural network (BPNN) to predict the OGE accident has been developed in this paper. The wavelet denoising processing is used to effectively remove high-frequency noise of the data series, as well as to conform with the original data trend simultaneously. The denoised data is imported to BPNN for WNN model training. The trained WNN is used for OGE accident prediction, and a comparative study has been conducted between the prediction results from the WNN model and the traditional BPNN. The comparison results indicate that the WNN model is more precise. An application study of this prediction model has been discussed. This method can be used to provide targeted pre-warning information and corresponding prevention strategies before conducting OGE fieldwork. This is expected to improve the safety of OGE field operations.