Models And Methods For Power System Fault Diagnosis Considering Uncertainties

Fault diagnosis aims to identify the fault sections and the malfunctioned protective relays and circuit breakers. When a fault occurs in a section or component of a given power system, the dispatchers usually have to handle a flood of alarms from different monitoring systems. It is important for the dispatchers to process the information and estimate the fault section accurately in a short period of time in order to restore the stable power supply as soon as possible. Fault diagnosis for power system has been being an active research area at home and abroad during past two decades, it could be divided into two types according to the object of study. One type is for power transmission system, while the other is for distribution system. The main bottlenecks of existing models and methods for power system fault diagnosis are caused by the uncertainties, i.e., the operating reliabilities of circuit breakers and protection relays, and the correctness and completeness of the received alarm messages. The accuracy of the fault diagnosis result is affected to some extent by the method used to deal with the uncertainties encountered in the diagnostic process.Given this background, based on several types of Artificial Intelligence technology and optimization algorithm, precise and effective models and methods for power system fault diagnosis considering uncertainties are developed. The temporal information of alarm messages as well as the electrical data information (telemetering data) is well utilized in this thesis. Some significant research results are obtained as follows:1) A novel method employing temporal information of alarm messages and cause-effect network (CEN) is developed. CEN is extended to accommodate temporal information and a temporal CEN (TCEN) then proposed. A new method for power system fault diagnosis based on TCEN is next developed. The proposed TCEN based method could not only maintain the advantages, but also improve the fault tolerant capability of the exiting CEN based fault diagnosis method. Moreover, the proposed method could better explain the fault evolution procedure.2) A logic inference approach to fault diagnosis combining temporal and fuzzy inference is proposed. The rationality of using temporal and fuzzy inference in fault diagnosis of power system is analyzed firstly. Then, the thought of alarm data preprocessing is proposed and a temporal and fuzzy Cause-Effect network (TFCEN) is next developed. Finally, a14-bus system is served for demonstrating the essential features of the developed model and method.3) Aiming at the problem of fault section estimation and state identification of unobserved protective relays (FSE-SIUPR) under incomplete information from protective relays, an integrated integer programming model for FSE-SIUPR is developed in the framework of the well-established temporal constraint network. This model could not only diagnose faults and identify the states of unobserved protective relays, but also explain the fault developing process.4) The existing methods for fault location in distribution systems with distributed generators (DGs) cannot well address the situations with incorrect or missing alarms. This thesis presents a two-level model for fault location in distribution systems with DGs. In the upper level, an analytic model is first developed based on telecommunication data, while the model in the lower level is based on telemetering data. The presented two-level model could not only diagnose the faulted section(s), but also identify the incorrect and missing alarms.Finally, several conclusions are obtained based on the research outcomes, and directions for future research indicated.