Research On Power System Complex Initial Fault Diagnosis Method Based On Adaptive Fingerprint Recognition

This paper analyzes the evolution of typical power system faults, such as "8.14" blackout etc., draws the conclusion that each failure of the power system can be divided into five stages, including slow succession breaking stage, rapid succession breaking stage, short oscillations stage, avalanches stage, and the last restoration stage. So, this paper focuses the research objective on achieving fast and accurate diagnosis of the complex initial faults, and hopes to design a real-time power system fault diagnosis method, to be able to self-adapt the change of the power grid, and quickly analyze the fault nodes in the early stages of failure, thus effectively to assist the operator in fault processing, and prevent the faults further expand.Through the analysis to the current domestic and international power system fault diagnosis methods, there are two common problems:1) Because of the frequent changes of the power grid structure and operation mode, the modeling, rules, algorithms of those fault diagnosis methods are difficult to adapt to, and thus affecting the practical use of these methods; 2) The uncertainty of the switch protection signal, leads to a higher diagnosis error rate of those fault diagnosis methods and also affecting the use of these methodsIn order to solve the influence of power grid operation mode to the fault diagnosis, the paper introduces the theory of complex adaptive system (CAS), studies the concept, the core ideas and features of CAS, and then considering for the specific needs of the power system fault diagnosis, this paper proposes an adaptive analysis modeling mothed for all of the power system devices, which can automatically adapt to the changes of the grid operation mode. And on this basis, a kind of adaptive searching analysis method for the contingency set of the complex initial fault. The construction of the contingency set is the basis of the adaptive fingerprint recognition fault diagnosis mothed (AFRFD).As for the question of the uncertainty of the switch and protection signal, the new idea of compositing the information of action signal and power grid branch flow is proposed, which use the change of the branch power flow as the "flow fingerprint" of the grid failure, and use the action logical relationship of the switch and protection signal as the "action signal fingerprint" of the grid failure, then learn from the mature experience of the real fingerprint recognition system, diagnose the grid fault by means of the pattern recongnition method. In order to realize this idea, this paper put emphasis on researching the extracting method for those two kinds of fingerprint:1) Through exploring the switch-protection action information of the power grid fault, an action signal fingerprint adaptive extraction method is proposed, which extract the action informantion, the temporal constraints among these actions, the topological constraints, and the relative measurement constraints etc. to form the action signal fingerprint. On the one hand, this method has inherited the available diagnosis knowledge of expert system knowledge base, and on the other hand, through the quantitative processing of transforming the discrete signals into the continuous variables, the expert system inference error problems can be avoided to a large extent, and simultaneously the quantitative results can be achieved in combination with other fault diagnosis method conveniently, such as the flow fingerprint fault diagnosis as follows.2) The extracting method for power flow fingerprint eigenvector based on the principal component analysis is proposed:The principal component analysis method can achieve dimensionality reduction to these high-dimensional variables at the same time preserving the mainly original data. According to the minimum mean square error principle, carried out an orthogonal linear transformation to the original fault power flow fingerprint feature space which was comprised of all branch active power, and received a set of orthogonal vectors which formed a new principal component space. By the PCA cumulative contribution rate, extracted some of the most important orthogonal vectors, and obtained a low-dimensional sub-space of grid fault power flow. The fault power flow fingerprint feature extracted by using this method, can significantly improve the computational efficiency, and has higher recognition accuracy compared with the original feature vector.On this basis, according to the characteristics of the power flow fingerprint and the action signal fingerprint, the corresponding fault identification strategies are designed. And on the analysis of the diagnosis blind area of these two strategies, this paper further puts forward a fusion fault diagnosis strategy based on the action signal fingerprint and power flow fingerprint. This strategy is based on the evidence combination rules of the D-S evidence theory, combined the power flow fingerprint information and action signal fingerprint information, largely solved the problem of the above two effects on fault diagnosis, and improved the correctness and practicabiliy of the AFRFD method.Finally, the article explains the implementation of the AFRFD method, with a specific regional power grid fault example, which validates the practical value of this mothed.