Method for matching double-end fault recording data of power transmission line based on convolutional neural network

Abstract

The invention relates to a method for matching double-end fault recording data of a power transmission line based on a convolutional neural network. The method comprises steps of: firstly creating a line fault simulation model, traversing parameters, forming a parameter matrix, using current data at both ends during failure as the input of a sample and using a matching result as the output of the sample, and generating a test sample in a similar way; secondly, listing network structures, training and testing a network to obtain the error rate of the test sample, and using the network structure with the lowest error rate as the optimal network structure; then, changing the batching processing number under the optimal network structure, increasing the number of training times, saving a trained network structure parameter with a low error rate and a weight bias matrix; finally, inputting the data required to be matched into the trained network after each failure so that a matching result can be output. The method requires less electrical quantity, and is not affected by system frequency, transition resistance, fault positions and fault types. The data matching accuracy is higher than that of a data matching triple criterion result.

Description

technical field [0001] The invention relates to a fault matching method of a transmission line, in particular to a method for matching wave recording data of a double-terminal fault of a transmission line based on a convolutional neural network. Background technique [0002] The fault information system contains a large amount of fault recording data, which often has the characteristics of timing, clock asynchrony, inconsistency, incompleteness, redundancy, etc. The recording data at both ends of the transmission line under the same time scale are not must be a match. Match the double-terminal fault recording data at both ends of the transmission line, and apply it to double-terminal fault distance measurement, protection behavior analysis, fault playback, and equivalent value verification under accident status, etc., will be able to better play the value of the data, Fault analysis and fault recovery are of great significance. [0003] The current fault data matching meth...

AI Technical Summary

Problems solved by technology

As one of the typical models of deep learning, the product neural network has a strong learning generalization ability and has been applied in the power industry. Among them, the application research of wind farm power prediction, power

Images