基于MLP和注意力机制BiLSTM的水电机组劣化趋势预测  

Prediction of Deterioration Trends in Hydropower Units Based on MLP and Attention Mechanism BiLSTM

作　　者：何一纯 李超顺[2] 杨云鹏 HE Yi-chun;LI Chao-shun;YNAG Yun-peng(State Grid Xinyuan Company Ltd.,Beijing 100052,China;School of Civil and Hydraulic Engineering,Huazhong University of Science and Technology,Wuhan 430074,China)

机构地区：[1]国网新源控股有限公司,北京100052 [2]华中科技大学土木与水利工程学院,湖北武汉430074

出　　处：《水电能源科学》2025年第3期177-181,100,共6页Water Resources and Power

基　　金：湖北省重点研发计划项目(2021BAA193);国网新源集团控股有限公司科技项目(SGXYKJ2023085)。

摘　　要：水电站因工作时间长、内部结构复杂及运行环境等因素导致水电机组部件逐步老化受损,使电站运行存在重大安全隐患。水电机组劣化趋势预测能反映机组的运行安全,为此提出一种基于多层感知机(MLP)和注意力机制的双向长短时记忆(Attention-BiLSTM)相结合的劣化趋势预测模型(MLP-BiLSTM-Attention),首先将机组各工况数据与各个振摆数据进行相关性分析,获取关键部分之间的高度相关性;然后提取较高相关度特征值并输入改进后的MLP模型构建健康模型,利用实际机组运行数据与健康模型数据构建机组劣化度,劣化度信息输入Attention-BiLSTM预测网络实现劣化度预测;最后通过多种模型对比验证了所提模型的可行性和有效性。The long working time,complex internal structure and operating environment of hydropower stationsresult in the gradual deterioration and damage of the unit components,which is a major hidden danger to the operation of the power station.Therefore,it is necessary to predict the deterioration trend of the unit to reflect the operational safety of the unit.This paper proposes a deterioration trend prediction method based on the combination of multilayer perceptron(MLP)and Attention-BiLSTM network model.Firstly,the data of each working condition of the unit is correlated with the data of each oscillation to obtain the high correlation between the key parts.The higher correlation eigenvalues are extracted and input into the improved MLP model forestablising the health model.The deterioration degree is generated by combining the actual unit operating data with the health model data.The deterioration degree information is input into Attention-BiLSTM network to predict the deterioration degree.Finally,the feasibility and effectiveness of the proposed model are verified by comparing various models.

关 键 词：水轮机组 劣化预测 健康模型 多层感知机 双向长短时记忆网络 

分 类 号：TV734.21[水利工程—水利水电工程]