Sparse automatic encoder optimized neural network-based power transmission line fault identification method

Abstract

The embodiments of the invention disclose a sparse automatic encoder optimized neural network-based power transmission line fault identification method. The method includes the following steps that: S1, data acquisition is performed: fault current traveling waves generated on a power transmission line after the failure of the power transmission line are acquired; S2, a fault waveform feature quantity is extracted: the feature quantity of the fault current traveling wave data of the power transmission line which are acquired in the step 1 is extracted; S3, a sparse automatic encoder is trainedwith a random gradient descent algorithm, so that the sparseness expression of intrinsic feature information in the feature quantity in the step S2, the weight W and offset b of the sparse automatic encoder which are obtained after the training is finished are saved; S4, the input layer and hidden layer of a neural network are initialized with the weight W and the offset b of the sparse automaticencoder in the step S3, a forward iterative algorithm and a reverse iterative algorithm are adopted to train the neural network; and S5, the optimized neural network and a BP neural network are adopted to identify the fault type of the transmission line under the condition of different quantities of training samples.

Description

technical field [0001] The invention relates to the field of power transmission and distribution, in particular to a transmission line fault identification method based on a neural network optimized by a sparse automatic encoder. Background technique [0002] Power system transmission lines have long distances, wide geographical distribution, and complex environments. They are often affected by factors such as lightning, animals and plants, ice and snow weather, etc., causing transmission line faults to trip. The types of transmission line faults can be divided into lightning strike faults and non-lightning strike faults, and the protective measures for different faults are different. Therefore, it is particularly important to accurately identify the fault types of transmission lines. [0003] Transmission line fault type identification is mainly divided into two aspects: fault feature extraction and fault type identification. The characteristic quantities of transmission l...

AI Technical Summary

Problems solved by technology

Neural network has strong learning and generalization capabilities, so it is widely used in various classifiers. However, the premise of supporting this method is to extract appropriate feature parameters and require a large number of data samples to realize network training. However, in practice, transmission lines Fault current traveling wave label data is less and the acquisition process is cumbersome, resulting in limited data samples, and the application of neural networks tends to lead to low recognition ability

Images