Research On Fault Diagnosis Of Transmission Lines Based On Machine Learning

Accurate diagnosis and discrimination of high-voltage transmission line fault types are the prerequisites for quickly locating the fault location,removing the fault segment,and restoring power supply.It is also the key to effectively reduce the economic loss of users,ensure stable operation of the power system,and provide safe and reliable power supply.High-voltage transmission lines serve as backbone grids connecting regional power grids,spreading all over every corner of the power grid.Most of them are in outdoor environments or even remote mountainous areas with harsh and complex environments.Usually,manual maintenance is difficult,resulting in transmission lines having the highest probability of failure in the power grid.With the vigorous development of computer technology and artificial intelligence algorithms,machine learning methods are used in all aspects of human production and life,and have achieved fruitful results.Therefore,research on fault diagnosis of transmission lines based on machine learning is of great significance to the stable operation of power systems.The specific work is as follows:Based on the analysis and research on the types of transmission line faults,short-circuit causes,operation modes and mathematical models,this paper builds a Π-type equivalent line digital display simulation model of 220 k V high-voltage double-ended power supply on the Matlab / Simulink simulation platform.Simulate common faults and analyze their fault transient signal characteristics.In a small time window,collect three-phase transient voltage and current data of transmission lines under different working conditions and different operating states.In order to effectively distinguish between two-phase ground faults and twophase short-circuit faults,the introduction of zero-sequence component features for auxiliary judgment.According to the characteristics of different machine learning algorithms,different types of transmission line fault data sets are established through programming.Using the characteristics of multi-scale array entropy(MPE)sensitivity to weak signal detection,to overcome the problem of weak voltage signal on the strong power supply side during the diagnosis of fault type.This paper proposes to construct a multi-scale permutation entropy value of the three-phase voltage and zero-sequence component signals of the transmission line fault as the fault feature,and combined with the traditional particle swarm algorithm to optimize the support vector machine(PSO-SVM)transmission line fault diagnosis Program.The collected sample data and a small amount of actual data were used to verify the accuracy and engineering applicability of this method.And compared with other machine learning algorithms,analyze the excellent performance of the fault diagnosis scheme constructed in this paper when a small amount of sample data.For shallow machine learning models such as the SVM algorithm,there are limitations that require artificial feature extraction,and have good performance when targeting small amounts of data.However,for large data sets,the CNN model in deep learning is used to extract and identify powerful features in data processing.In this paper,a CNN diagnostic model for identifying transient fault current signals of transmission lines is proposed.The specific structure and optimal parameters of the network model are determined by setting comparison experiments.The original three-phase transient current and zero-sequence component fault data are used for learning and training,and compared with other traditional machine learning models,confirming the CNN constructed in this paper.The model can effectively complete the fault type identification and fault phase selection,and is effective,practical and advanced.