Power System Fault Diagnosis Based On Bayesian Network And Dempeter-Shafer Evidencd Theory

With the increasing scale and more comoplex structure of power system,when the power system failure, especially the emergence of complex fault (including cascade tripping) conditions, a large number of fault information go into the dispatch control center. This situation has brought some difficulties in scheduling staff. At the same time, further requests have been advanced to power system fault diganosis. The traditional centralized power system fault diagnosis has to meet the demand, therefore, a distributed diagnostic model has great practical significance.This paper introduced the Bayesian network and Dempster-Shafer evidence theory. Firstly, it introduced the basic theory of Bayesian network, network structure and network inference, then introduced the basic theory of Dempster-Shafer evidence theory, including basic concepts of Dempster-Shafer evidence theory, theory of evidence combination rule and the weight assignment and explained by example.It became the basis of distributed power system fault diagnosis based on Bayesian network and Dempster-Shafer evidence theory.Some improvement has been made on fault diagnosis of the current centralized Bayesian network model. Firstly, a real-time analysis method was used to determine wiring fault diagnosis to narrow the scope of fault, zones;secondly, the threshold criterion has been used to determine the component failure, the failure of the backward reasoning component probabilities were reasonable amendments to effectively solve the system of information leading to the miscarriage of justice due to interference, improving the diagnosis of fault tolerance; the five numerical examples were made to demonstrate the feasiblity and effectiveness of above approaches.Finally, the paper focused on distributed network fault diagnosis model based on the Bayesian network and Dempster-Shafer evidence theory. It also used the method of regional fault recognition, respectively separated by butterfly and leaf segmentation method to reasonably divide the power system.For the failure region of overlap with the subnet components, paper respectively used coincidenced degree method and the geometric mean method to calculate the subnet body weights of evidence, and then mix the subnet diagnosis to gain the final result. In order to facilitate comparison with the centralized diagnosis, this paper get the same five examples and another reference to an actual power, two partitioning methods and two Dempster-Shafer evidence combination methods were used for simulation. The simulation examples presented the model based on Bayesian network and Dempster-Shafer evidence theory of distributed network fault model was effective, and compared to the traditional centralized diagnostic model, whose timesliness and precision in diagnosis have been greatly improved. the method of butterfly segmentation and geometric mean value to the distributed fault diagnosis was the most accurate and reasonable.