Cable joint partial discharge ultrasonic sequence prediction method

Abstract

The invention discloses a cable joint partial discharge ultrasonic sequence prediction method and belongs to the technical field of terminal joint fault early warning of a cable system. According to the method, ultrasonic on-line monitoring is performed on a cable joint through an ultrasonic sensor, and an ultrasonic sequence is formed by ultrasonic signal data detected by the ultrasonic sensor and collected according to a certain period. The method is used for predicting an ultrasonic sequence caused by partial discharge of a cable joint based on an LSTM neural network algorithm. The method comprises the following steps of: a, establishing an LSTM neural network algorithm framework; b, optimizing hyper-parameters in an LSTM neural network algorithm framework; c, determining an objective function in the LSTM neural network algorithm; d, predicting a model and hyper-parameter optimized LSTM neural network architecture; e, carrying out particle swarm division on a hyper-parameter solution space to be optimized; and f, predicting the partial discharge ultrasonic sequence of the cable joint. The prediction precision of the neural network can be improved.

Description

technical field [0001] The invention relates to the technical field of early warning of terminal joint faults of cable systems. Background technique [0002] In recent years, with the rapid development of computer technology and power electronics technology, the on-line monitoring technology of power equipment relying on various sensors has been widely used to monitor the operating status and health status of power equipment, such as switchgear cable joints and parts of power equipment. Discharge monitoring sensor. Based on the characteristic parameters monitored by various sensors, the current operating state and health level of power equipment can be analyzed; in addition, using online monitoring data, the future health state can also be judged by predicting the trend of the monitored characteristic quantities. [0003] Long Short-Term Memory (LSTM) neural network has the ability of deep learning. As a variant model of recurrent neural network (RNN), LSTM neural network e...

AI Technical Summary

Problems solved by technology

The current neural network hyperparameter optimization method has shortcomings. For deep learning models, repeated experiments are not only inefficient but also time-consuming.

Images