基于RFR和VMD-TCN的抽水蓄能机组劣化趋势预测方法  

Deterioration Trend Prediction Method for Pumped Storage Units Based on RFR and VMD-TCN

作　　者：付波[1] 姜奔 赵熙临[1] 李超顺[2] FU Bo;JIANG Ben;ZHAO Xi-lin;LI Chao-shun(College of Electrical and Electronic Engineering,Hubei University of Technology,Wuhan 430068,Hubei Province,China;College of Civil and Hydraulic Engineering,Huazhong University of Science and Technology,Wuhan 430074,Hubei Province,China)

机构地区：[1]湖北工业大学电气与电子工程学院,湖北武汉430068 [2]华中科技大学土木与水利工程学院,湖北武汉430074

出　　处：《中国农村水利水电》2023年第3期208-217,共10页China Rural Water and Hydropower

基　　金：国家自然科学基金项目(51879111)。

摘　　要：抽水蓄能机组具有结构复杂、工况复杂多变、故障复杂多样等特点。利用实时监测数据有效评价抽水蓄能机组的劣化状态并对劣化趋势进行准确地预测仍是一个难题。为此,提出一种基于随机森林回归(RFR)、变分模态分解(VMD)和时间卷积网络(TCN)的抽水蓄能机组劣化趋势预测方法。首先,基于健康状态下的历史监测数据,选择与状态监测数据关联性强的工况参数数据作为健康状态模型的输入,建立基于RFR的健康状态模型;其次,输入实时工况参数数据,根据健康模型输出的标准值与实时状态监测数据计算得到劣化趋势序列;最后,考虑到劣化趋势序列的非线性因数,设计了基于VMD-TCN的时序预测模型,以实现对劣化趋势的精确预测。为验证所提方法的有效性,采集位于中国浙江的抽水蓄能电站真实监测数据进行多组对比实验。结果显示,所提出的方法在建立健康模型时拟合精度达到了0.98,并且在劣化趋势预测任务中,基于VMD-TCN的时序预测模型相比于其他比较模型具有更高的预测精度。Pumped storage units are characterized by complex structure,complex and variable operating conditions,and complex and diverse faults.It is still a challenge to effectively evaluate the deterioration state of pumped storage units and accurately predict the deterioration trend using real-time monitoring data.To this end,this paper proposes a method to predict the deterioration trend of pumped storage units based on Random Forest Regression(RFR),Variational Mode Decomposition(VMD)and Temporal Convolutional Networks(TCN).Firstly,based on the historical monitoring data under the health state,the working condition parameter data with strong correlation with the condition monitoring data are selected as the input of the health state model,and the RFR-based health state model is established.Secondly,the real-time working condition parameter data are input,and the deterioration trend series are calculated based on the standard values output from the health model and the real-time condition monitoring data.Finally,considering the nonlinear factors of the deterioration trend series,a time-series prediction model based on VMD-TCN is designed to achieve an accurate prediction of the deterioration trend.To verify the effectiveness of the proposed method,real monitoring data from a pumped storage plant located in Zhejiang,China,are collected for multiple comparison experiments.The results show that the proposed method achieves a fitting accuracy of 0.98 in building a health model and that the VMD-TCN-based time-series prediction model has higher prediction accuracy compared to other comparative models in the deterioration trend prediction task.

关 键 词：抽水蓄能机组 劣化趋势预测 随机森林回归 变分模态分解 时间卷积网络 

分 类 号：TV734[水利工程—水利水电工程]