Research On Data Preprocessing Technology For Power Grid Fault Diagnosis System

Under the background of the development of smart grid,on the one hand,the large number of intermittent energy access and the application of information sensing technology make the scale and structure of power grid more complex.The establishment of fault recorder information network and the popularization of intelligent electronic measurement devices make the fault data diversified;On the other hand,channel interruption or abnormality,inadequate measurement points,misoperation,and correlation failures occur during the uploading process of the control platform,which will result in incomplete and inaccurate data source cross-section received by the system.Therefore,in order to improve the adaptability and fault tolerance of power grid diagnosis system in complex data environment,the analysis and processing of fault information are deeply studied in this paper.Based on the actual requirement of the power grid fault diagnosis and recovery system developed by our research group,a fault information preprocessing module is designed according to the complex data environment.By combining the basic principles of fault diagnosis with the actual fault cases,a large number of fault samples and related information are extracted by the integrated simulation system of regulation and control,thus providing complete data source for the analysis of fault information diagnosis mechanism.Aiming at the low-value-density fault information received by the dispatching platform,a denoising analysis method is proposed to deal with the alarm information.The text similarity algorithm is used to correlate and match the non-diagnostic keywords in the information sequence.The method reduces the workload of fault information modeling and improves the robustness of diagnosis calculation.At the same time,according to the network topology,the sequence of protection-circuit breaker related events is established to provide compiled data link for the next step of anomaly event identification.In order to identify and process anomalous information,the maximum entropy hidden Markov model(ME-HMM)is constructed.Based on the relevancy relation between protection and circuit breaker,the type of information to be diagnosed and the pattern of abnormal information are defined.Then the eigenfunction vector is constructed by combining the electrical quantity information and context information,and the sub-model parameters are trained by Baum-Welch algorithm to estimate the probability of the abnormal patterns hidden in the fault data.The example analysis proves that this method can simplify the original fault data,effectively identify the abnormal information including information distortion and incorrect action of protection circuit breaker,and provide a new idea for the development and design of adaptive fault diagnosis system driven by data.