Study On A Power Grid Fault Diagnosis Method Based On Rough Set Theory And Naive Bayesian Networks

As the enlarging of power grid scale and increasing of power grid automatization level, more and more comprehensive alarm information is transmitted to control centers of all grades. When power grid is in fault, if the fault information is imperfect and indeterminate or even the key information is lost, it may result in the condition that correct conclusion could not be given by fault diagnosis.To settle this problem the paper proposes a new method to diagnose faults in local power grid in which the Rough Set theory is integrated with Naive Bayesian networks (RSNB). At first, the protections and circuit breakers are taken as conditional attributes and faulty region as decision-making attribute, various faults are investigated and decision table is established; then, by use of attribute reducing method based on cognizable matrix and information entropy, the optimal attribute reduction combination is extracted; finally, by means of the reduction decision table formed by optimal attribute reduction combination, the naive Bayesian networks model is built and the nodal probability is trained.On the basis of above, considering the fault information being both localized and connected, this paper attempts the idea of distributed faults diagnosis. First, a method is proposed to partition the large-scale power grid. Then, at the sub-grids, the RSNB method is used to diagnose faults. After an in-depth study of partitioning method for power grid, this paper attempts the method of graph partitioning integrated with papilionaceous partitioning. This method can partition the large-scale power grid into a desired number of connected sub-grids with balanced working burdens in performing fault diagnosis. The proposed method consists of three basic steps: forming the weighted depth-first-search tree of the studied power grid; partitioning the power grid into a desired number of sub-grids with balanced working burdens use graph partitioning method; using papilionaceous partitioning method to assure connection capability of sub-grid at the basis of the minimal redundancy.The RSNB distributed fault diagnosis software is programmed by VC++ programming language. Results of calculation examples show that the proposed method is correct and effective, and that the grid partitioning method is very fast and effective for large-scale power grid, and that the RSNB method can improve the fault tolerance capability of the fault diagnosis system while the kernel attribute is lost, so this method is available.