Research On Fault Diagnosis Method For Power Distribution Network Based On Multiple Data Sources Information Fusion

At present, power grid automation is developing at a high speed. So fault information is becoming more and more convenient to be obtained. Once complex fault occurs, a lot of alarm and fault information swim into dispatching center in a short time, in which circumstance dispatching personals need to recognize the core alarm information. Apparently this is hard to identify all kinds of faults quickly and correctly, leading to inaccurate diagnosis inevitably. Dispatching workers need to rely on efficient fault diagnosis system to offer decision reference. This will be regarded as assistant judgment to ensure the safety of power system.Traditional researches and methods of fault diagnosis are mostly based on the information of protection and circuit breaker action to recognize the fault elements, using kinds of intelligent algorithms. It is highly demanding for the completeness of switching-status information. The degree of completeness and accuracy of alarm information will exert a great influence on fault diagnosis. Based on the full consideration of the characteristics of electrical measurement before and after fault, this thesis adds continuous-time data to fault diagnosis of power grid, and designs a multi-source information infusion method to realize the diagnosis of faults comprehensively.In consideration of the development of power system automation and communication and wide area measurement technology, this thesis presents a fault diagnosis method for power grid with information fusion based on multi-data resources, which contains the switching-status data and continuous-time data analysis derived from fault recorder. Based on electrical value analysis with fault information and switch value diagnosis, this method employs the wavelet energy analysis to extract features from the continuous-time data and uses Bayesian network to perform fault reasoning on switching-time data of protection relays and breakers. During this process, wavelet fault degree(WFD), wavelet singularity degree(WSD), wavelet energy degree(WED), and fault degree based on Bayesian networks are described to indicate the situation of the faults on the lines, which are taken as evidences for information fusion by using improved D-S theory. Then the fuzzy C-means clustering method is involved to handle the fusion result and gives decision-making. PSCAD simulation and calculations based on MATLAB programming show that this new approach for fault diagnosis significantly improves the diagnostic accuracy compared with conventional ways based merely on switching-time data, and has practical value and good application prospects.