Bayesian Networks for Fault Diagnosis of a Large Power Station and its Transmission Lines

Abstract

This article proposes a simplified fault-diagnosis system based on Bayesian networks with noisy-OR/AND nodes to estimate the faulty item/section(s) of a large power station and its transmission lines. The proposed method utilizes the final information of protective relays and corresponding circuit breakers to construct the Bayesian fault diagnosis model for each section. The learning algorithm for Bayesian network parameters takes the sum of the mean-squared error between the expected values and the computed values of certain target variables as the minimizing optimization function to adjust the network parameters continuously. By comparing the result beliefs of possible faulty sections, the faulty item/section(s) becomes a candidate. In order to test the validity and feasibility of that method, a computer simulation of the High Dam power station and its 500-kV transmission lines is used. It is shown that the proposed diagnosis method has many merits, such as rapid reasoning, less storage memory and processing time, easy correctness of diagnosing results, flexibility, and application into a large power station and its transmission lines for real-time fault diagnosis. Finally, it assists and supports the operator of the control room to make the right decision, especially in case of communication loss.