Research on Power System Fault Diagnosis Method Based on Bayesian Networks
|School||Southwest Jiaotong University|
|Course||Proceedings of the|
|Keywords||Power system Alarm processing Wide-area backup protection Fault Diagnosis Bayesian network Time causal|
Power system operation, often subjected to a variety of natural and man-made interference, difficult to avoid failures. When the power grid fails, there will be a lot of information dealt with in the influx control centers, to dispatch personnel accident caused great difficulties, an automated alarm handling system (ie fault diagnosis system) to provide decision support for scheduling staff and accordingly system for rapid recovery after the accident, and reduce economic losses. The alarm processing early start in many ways, mainly in how to deal with the uncertainty on the alert information space and how to model the large-scale grid, less the consideration of the timing characteristics of alarm information and the lack of a uniform model to deal with the uncertainty on the time and space of the alarm information. Large grid interconnection improve the optimal allocation of resources, brought huge economic power also makes the system more vulnerable to large-scale blackout against incorrect action of the study showed that 75% of the electric power system large interference relay protection related. With the rapid development of wide-area communications network, wide-area backup protection (Wide-area Backup Protection) concept, in order to provide better selectivity, real-time and reliability. To achieve as soon as possible, and as much as possible small-scale removal of faulty components to complete a better backup protection functions, the primary task is to quickly and accurately identify the faulty components. Wide-area backup protection, mainly in system architecture, collaboration mechanism, the back-up protection fault diagnosis method in artificial intelligence is still rare. This paper first introduces the research status of the alarm processing, fault diagnosis method of the status quo, as well as wide-area backup protection algorithm fault diagnosis method. Then introduces some basic concepts of Bayesian network research on the importance of conditional independence in Bayesian network, and Leaky Noisy-Or model through its independent causal mechanism to reduce the complexity of modeling, and Bayesian network inference problem and its reasoning algorithm is introduced. Wide-area distributed power grid fault diagnosis method. Refuse to move and misoperation priori probability of component failure, protect access issues discussed, an event-sampling of a priori probability calculation method. As a diagnostic evidence of wide-area backup protection requirements of real-time and accuracy of fault diagnosis, use of protective measurement signal, a distributed power grid fault diagnosis method of dynamic modeling based on Bayesian network to reduce diagnostic modeling and complexity of reasoning computed. Time causal Bayesian network model to solve a class of alarm the fault diagnose problems with timing characteristics. Discussion of the time prisoners causality uncertainty, fuzzy set theory to express the the time causal relationship between the fault and alarm, the two most common node model - \and the method of calculation of the conditional probability of the node, and the probability of the fault hypothesis, last time causal Bayesian network model of fault diagnosis reasoning calculation process demonstrated by an example, compare the accuracy of the model example sex. SCADA and SOE information grid failure (for scheduling control center alerts at buried). Discussed the importance of the SOE (protection and circuit breaker operation timing information), the addition of the power grid fault diagnosis accuracy, and proposes the use of causal Bayesian network model grid fault diagnosis. The introduction of a new element - the switch in time causal Bayesian network model, then the concept of dynamic association path is proposed to improve the method based on the breadth-first search to find the associated path to determine the switch state. Finally, an example was given for validation.