Power System Fault Diagnosis Based On DS Evidence Theory And Mutual Information Network

As the size of modern power system gets larger and larger, more and more automatic equipment is applied to power system. With the process of digitalization and informatization, the fault information obtained is more and more complete, whereas there is still uncertainty in the fault information, confronted with the mass fault information, the staff is always at a loss. Therefore there is a strong need for power system fault diagnosis, which is to analyze the reason of fault quickly, reappear the process of fault, and evaluate the performance of device and equipment. Studying and developing the power system fault diagnosis have important meaning.DS evidence theory has special advantage on dealing with uncertainty. There is uncertainty, such as refuse and misoperation, misinformation and missing, etc. in power system fault diagnosis, hence, a power system fault diagnosis method based on DS evidence theory is proposed. In terms of rationale of protective relay, diffusion and information theories, make the probable fault elements in closed or approximately closed area bounded by operated circuit breakers be frame of discernment; based on fault diagnosis Bayesian inference, a posterior probability formula for fault element is deducted; according to the fault information got from SCADA, RMS systems, prior fault probability of element, refuse and misoperation probabilities of protective relay and circuit breaker, which are calculated in terms of expert knowledge, testing data, historical information, statistical data, etc., are assigned to posterior fault probability formula, then we can obtain fault posterior probability of every element under known fault sequence; process these probabilities, setup basic probability assignment for the elements to be recognized, form evidences; combine these evidences in terms of Dempster combining rule; and obtain a judgment of fault possibility to every element in frame of discernment.Mutual information network is a feature selection and data classification method based on information theory. Mutual information is used to test the relating degree between input attributes and objective attributes, deduce the relating rules between input attributes and objective attributes, and realize the pattern classification of data sequence. Power system fault diagnosis estimates the fault elements in terms of the fault information, therefore it can be described as a pattern classification problem. On this basis, the features of fault information of transmission network are analyzed in this thesis, and a power system fault diagnosis model based on mutual information network is proposed. In the problem of transmission network fault diagnosis, a fault diagnosis model based on mutual information network is proposed. The information got from SCADA, EMS, and SOE (Sequence of Events) is utilized thoroughly. The information of protective relay and circuit breaker is taken as candidate input attributes, and the fault element is taken as objective attribute. In terms of the fault history and operation rules of protective relay and circuit breaker, the sample set is built, the key features are selected and the redundant features are eliminated by using the mutual information network algorithm. Fault diagnosis result can be got by putting online fault information into the mutual information network. The simulation of a 28 elements power grid shows that, when there are lost, distorted information, and refuse, misoperation of protective relay and circuit breaker, it is still able to get diagnosis result of fault elements, showed the good fault tolerance ability of this method. For large and complex power system, it can be divided into several small parts, and then build mutual information network model for every sub network. Therefore the algorithm is more flexible, and the calculation speed is promoted.The programming is based on MATLAB 7.0 in this thesis. The testing results demonstrate that these methods can be applied into power system fault diagnosis successfully, and have important theoretical value and application prospect.