The Research Of Bad Data Identification In Power System

During past decades, Information & Communication Technology has gotten extensive application in electric power systems. The Information & Communication Technology have brought profounding changes in dispatching and controlling of power system. In a highly automated power system, accurate and precise data acquisition and transmission lay foundation for decision and control of the system. The thesis carries out research on bad data identification in substation automation and dispatching automation system. The outputs of the thesis are listed as follows.A BP Neural Network based approach is proposed to identify bad data in Voltage Quality Controlled (VQC). The context information within a substation is utilized to train a BP neuralnetwork. Thereafter, the real time context data is fed to the trained BP Neural Network to get the induced voltage or reataive power. The acquisied voltage or reataive power deviate far from the induced voltage or reataive power is identified as bad data. First, established a single transformer bad data identification model according to control characteristics of the substation VQC.Second, used the improved BP neural network bad data identification estimator, and corresponded simulation program, using off-line training, real-time online identification of bad data identification. Finally, the actual data of a substation in Guangdong is simulated to verify the validity of the estimator.A Support Vector Machine (SVM) based approach is proposed to identify bad data in Automatic Voltage Control (AVC) System. The context information within a substation is utilized to set up a SVM. Thereafter, the real time context data is fed to the SVM to get the induced voltage or reataive power. The acquisied voltage or reataive power deviate far from the induced voltage or reataive power is identified as bad data. First, established a single generator bad data identification model according to control characteristics of the AVC system.Second, used the support vector machines bad data identification estimator, and corresponded simulation program, using off-line training, real-time online identification of bad data identification. Finally, the actual data of a power plant in Guangdong is simulated to verify the validity of the estimator.A Gentic Algorithm based Radial Basis Function Neural Network (RBFNN) is proposed to identified bad in substation automation system. Simulation result indicates that the proposed approach outperforms Othogonal Least Squared RBFNN. Whereas, the pattern classification capacity of the approach is limited since maximum squared error criterion is adopted. Fix Sized Least Square SVM, which adopt minimize manimum error as critierion, is adopted to identify bad data in real time with higher patter classification capacity.Since there is alays a TA accompanied by a circuit breaker in substation sutomation system, a expert system is proposed to identify fake state of circuit breaker. A simple rule is proposed to verify state variation signal of circuit breaker. Numerical simulation based on data logged in digital fault recorder prove feasibility of the proposed approach.