Application of GAN–CNN in Risk Assessment of Pipeline Failures in Multiphase Pipeline With Image Information Encoding Approach

Abstract

In this paper, a novel approach for risk assessment of multiphase pipeline failures applying a generative adversarial network with a convolutional neural network (GAN–CNN) approach is proposed. To enhance the risk prediction ability of a machine learning model, generating cases and expanding the data set of multiphase pipeline data groups are ideal ways. The Image Information Encoding approach is employed to store the key features information in the grayscale images and reserve the relationships between different features. The GAN–CNN algorithm will generate the data group in the form of grayscale images, considering the relationship between key features of the data groups; meanwhile, the categories of the synthetic data types can be switched to adapt to the requirements of the practical application. The model performance of risk assessment will be compared based on different data sets, including the unexpanded one and the expanded one, and the relevant analysis will be presented.