Weather-Related Failure Risk Prediction of Overhead Contact Lines Based on Deep Gaussian Process

Due to highly complicated working conditions of overhead contact lines, it is inevitable to suffer from the dynamic external weather conditions and environmental factors, and further trigger a variety of risk events, even causing a series of serious consequences. To prevent the weather-related risks in advance, this paper proposes a weather-related failure risk prediction approach based on deep gaussian process (DGP), with its superior performance of nonlinear processing and uncertainty quantification. After preprocessing the weather data and the associated failure records, the weather-related failure risk prediction dataset is established for the studied issue of this paper, that is predictive classification problem. To simultaneously predict the lighting-related trip-out, wind-related floater intrusion, and fog-haze-related pollution flashover risk, a multi-task learning framework in DGP is formulated to capture the complex dependencies between weather parameters and OCL failure risk. The extensive experiments investigated on the constructed dataset reflect the effectiveness and superior of the proposed approach, with capacity of uncertainty quantification and giving trustworthy prediction results.