基于深度门控长短时记忆网络的继电保护装置寿命预测  

Life Prediction of Relay Protection Device Based on Depth Gated Long Short-term Memory Network

作　　者：王昭雷 张惠山 赵智龙 付炜平 孟荣 WANG Zhaolei;ZHANG Huishan;ZHAO Zhilong;FU Weiping;MENG Rong(State Grid Hebei Electric High Voltage Company,Shijiazhuang 050071,China)

机构地区：[1]国网河北省电力有限公司超高压分公司,河北石家庄050071

出　　处：《河北电力技术》2023年第6期28-35,共8页Hebei Electric Power

基　　金：国家电网有限公司科技基金项目(kj2021-055);国家重点研发计划(2019YFB1505404)。

摘　　要：为了保证继电保护设备可靠性以及电力系统安全、稳定运行,建立了基于深度门控长短时记忆网络的继电保护设备寿命预测模型,针对继电保护设备特征量的时间序列特性,采用相对熵对特征量权重进行分析,得到了各个特征的赋权结果,提出了各个特征量的预测结果加权求和计算综合特征的方法,构建了设备寿命预测和评估框架。结果表明:剔除专家E_(4)的评分后,专家E_(3)的评分熵值最大,其值为1.42,对应的评分合理性为76%,该模型具有较好的泛化能力和逼近能力,利用模型预测得到的综合特征量能够准确评估继电保护设备未来发展状态。To ensure the reliability of relay protection equipment and stable operation of the power system,a life prediction model based on a depth gated long short-term memory(LSTM)network is established.Consideriing the time series characteristics of relay protection devices,relative entropy is utilized to analyze the weight of feature quantity,and the empowerment for each feature was obtained.This paper proposes a method for calculating comprehensive features,and constructs a framework for equipment life prediction and evaluation.The results reveal that the score entropy value of expert E_(3) is the largest,reaching 1.42,after excluding the score of expert E_(4),the corresponding score rationality is 76%.The life prediction model,based on the depth gated LSTM network,demonstrates excellent generalization and approximation capabilities.The comprehensive feature quantities obtained through the model's prediction accurately evaluate the future development state of relay protection equipment.This approach provides valuable support for predicting and evaluating the future life of the relay protection equipment.

关 键 词：继电保护设备 寿命预测 状态评估 长短时记忆网络 相对熵 

分 类 号：TM774[电气工程—电力系统及自动化]