Research On Power System Fault Diagnosis Method Based On Bayesian Networks

The power system often suffers natural and artificial disturbances during operation, so malfunction is inevitab. When a malfunction occurs, lots of information will emerge into the dispatching control center and make it difficult for the scheduling person to deal with the malfunction. An automatic alarm processing system, namely fault diagnosis system, can give the scheduling person to make decisions and help the system to recover quickly. Studies on alarm processing began early and many methods have been found on how to deal with the alarm information's uncertainty in space. However, temporal information is not given enough attention and we lack a uniform model which can handle spatial and temporal uncertainty at the same time.Large-scale interconnection of power grids improve the optimization of resources and bring huge economic benefits. Howerver, it makes the power system more accessible to balckouts. A study by NERC shows that about 75 percent of the power system's major disturbances are caused by the incorrrect activities of the relay protection. As the communication network develops quickly, a concept of Wide-area Backup Protection is put forward to offer better effectiveness and reliability. The primary task of communication network is to find out the fault component quickly. Studies on Wide-area Backup Protection focus on system structure and protective strategy and collaboration, not on fault diagnosis.In this paper, the current situation of fault diagnosis studies connected with alarm processing as well as Wide-area Backup Protection is introduced.Then the basic concepts on Bayesian Network studies are presented and the importance of conditional independence is elaborated. The paper also expounds on how Leaky Noisy-Or model to lower the complexity of building models through causal independence, as well as the reasoning of Bayesian Network.Regarding fault diagosis on Wide-area Backup Protection, we discuss on component breakdown, failure operation protection and Prior probability of False operation and put forward a Prior probability calculating method of event sampling. According to Wide-area Backup Protection's requirement for real-time and accuracy, we adopt relay protection measuring signal as diagnosing evidence and bring forward a Bayesian Network-based distributed power grid's fault diagnosing method to reduce the complexity of inference calculation.As to fault diagnosis on temporal information alarm, we bring forward Temporal Causal Bayesian Network Model in this paper, discuss time uncertainty in causality and adopt fuzzy set theory to express the temporal causality between fault and alarm. We also set up two common node models and give calculating methods on the node's conditional probability and fault hypothesis.We study power system's fault method based on SCADA and SOE information, discuss the importance of SOE information for enhancing fault diagnosis accuracy of power grids and put forward Temporal Causal Bayesian Network Model for fault diagnosis. A new element-switch, is introduced to Temporal Causal Bayesian Network Model and the concept of dynamic realted path is improved. A mothod based on breadth-first search is adopted to look for related path to determine the status of the swith.