Research And Implementation Of Power Grid Fault Diagnosis Based On PMU Data Graph Semantic Extraction

With the rapid development of UHV/UHV DC transmission,flexible DC transmission and DC grid construction,my country has built the world’s largest and highest voltage AC/DC hybrid power grid.With the rapid development of the power grid,the structure of the power grid is becoming increasingly complex.For the safe and stable operation of the power grid,when the power grid fails,it is necessary to accurately and quickly realize the fault diagnosis of the power grid.For this reason,this article aims to realize intelligent power grid fault diagnosis.In view of the shortcomings of traditional power grid fault diagnosis methods,a deep learning model is introduced,and the discrete data collected by the Phasor Measurement Unit(PMU)is drawn into a radar chart to generate continuous image data,and the convolutional autoencoder is used Carry out unsupervised pre-training and realize fault device location through supervised fine-tuning,and use the improved Inception-Resnet model to judge the fault type.The main work of the thesis is as follows:The inherent characteristics of PMU data are studied,and the PMU data is converted into a radar chart,so that the discrete data features are converted into trend features to form a more clear expression of semantic features,and the PMU data is changed from discrete features to continuous features,from quantitative analysis to quantitative analysis.Conversion of qualitative analysis.Experiments show that the use of image data not only effectively enhances the accuracy of faulty equipment location and fault type determination,but also greatly accelerates the speed of model convergence.According to the needs and characteristics of power grid fault diagnosis,the theoretical principle and algorithm process of the convolutional autoencoder are studied,and a power grid fault location method based on the pre-training of the convolutional autoencoder is designed.This method can be based on the extremely unbalanced data.Effectively extract data features to achieve precise positioning of faulty equipment.A power grid fault type judgment model based on improved Inception-Resnet to extract the semantics of PMU data is designed.This model trains fault cases,learns the fault characteristics in the samples autonomously,and realizes the judgment of the fault type based on the learned characteristics of the fault samples.And use actual failure cases to retrain the improved Inception-Resnet model to supplement the characteristics of the measured failure samples,so as to improve the judgment model.Based on the data collected by the IEEE 39-node model,the paper realizes the end-to-end power grid fault equipment location and fault type judgment.The improved deep learning network is used to improve the diagnosis efficiency of the power grid fault diagnosis system.Finally,a field fault case in Sichuan is tested to verify the practicability of the module.