基于自适应门控循环神经网络的继电保护在线监测技术研究

Research on Online Monitoring of Relay Protection Based on Self-adaptive Gated Recurrent Neural Network

针对继电保护装置可靠性在线监测难题,研究了继电保护评价指标体系和智能在线监测技术,提出了继电保护装置可用时间率评价要素和基于自适应门控循环网络的在线监测方法,构建了基于自适应门控循环网络的继电保护可用时间率在线监测模型框架。通过实例验证,结果表明:自适应差分进化算法能够准确得到最佳学习率和隐患层节点数,采用自适应差分进化算法得到的权值和阈值使得迭代步数减少15步、Loss函数值降低0.1%;可用时间率在35%以上时预测相对误差控制在±5%,且预测值对实际值具有较强的逼近能力,实现了继电保护装置可靠性准确预测和在线监测。

The study addresses the challenge of ensuring reliable online monitoring of relay protection devices through the development of a comprehensive relay protection evaluation index system and intelligent online monitoring technology.The study introduces assessment elements for evaluating the available time rate of relay protection devices and proposes an innovative online monitoring method based on a self-adaptive gated recurrent unit(GRU)network.This method establishes a novel framework for an online monitoring model centered around the available time rate of relay protection,leveraging the capabilities of the self-adaptive GRU network.The results demonstrate the effectiveness of obtaining the optimal learning rate and the number of hidden danger layer nodes through the application of a self-adaptive differential evolution algorithm.By utilizing the weights and thresholds derived from this algorithm,the number of iterative steps is reduced by 15,and the loss function value decreases by 0.1%.Moreover,the proposed method ensures a prediction relative error within ±5% for available time rates greater than 35%,resulting in prediction values that closely approximate the actual values.These findings signify a significant advancement in accurately predicting and online monitoring the reliability of relay protection devices.