基于连续小波分析与深度可分离卷积的水电机组工况识别  

Recognition for Operation States of Hydroelectric Generating Units Based on Continuous Wavelet Transform and Deep Separable Convolution

作　　者：马建军 王彤 王浩宇 唐一中 郭鹏程[2] 李昂 MA Jian-jun;WANG Tong;WANG Hao-yu;TANG Yi-zhong;GUO Peng-cheng;LI Ang(Guoneng Dadu River Maintenance and Installation Limited Company,Chengdu 610041,China;School of Water Resources and Hydroelectric Engineering,Xi’an University of Technology,Xi’an 710048,China)

机构地区：[1]国能大渡河检修安装有限公司,四川成都610041 [2]西安理工大学水利水电学院,陕西西安710048

出　　处：《水电能源科学》2024年第12期166-170,共5页Water Resources and Power

基　　金：国家自然科学基金项目(51839010);国能大渡河检修安装有限公司科技项目(JX-KY-2022-03)。

摘　　要：为快速、准确地判定水电机组运行状态,提出了一种基于连续小波分析与深度可分离卷积相结合的工况识别方法。该方法首先采集水电机组不同运行工况下的振动信号,通过连续小波变换对其进行解析,并获取其多尺度时频联合分布信息。随后,对时频信息进行了数据归一化、几何尺寸变换和格式转换等一系列处理,将其转换为数字图像形式。最后,构建了深度可分离卷积神经网络模型,依据数字图像信息对模型进行参数训练,该模型能够有效区分机组不同出力工况及过渡工况。根据我国西南地区某水电站的一台轴流转桨式水电机组的振动信号,采用所提方法实现了机组多种工况的识别,正确率达到98.06%。To determine the operation states of hydroelectric units quickly and accurately,a recognition method based on Continuous Wavelet Transform(CWT)and Deep Separable Convolution(DSC)was proposed.The acquisition of vibration signals under various operation conditions of units were analyze by CWT,and its multiscale time-frequency distribution information was obtained.In order to convert time-frequency information into digital image format,a series of processes including data normalization,geometric size transformation and format conversion were performed to process the time-frequency information.Finally,a Deep Separable Convolution Neural Network(DSCNN)model was established to train the model according to digital image information.Operation states under different power and transient conditions can be identified effectively.Based on the vibration signals collected from a Kaplan hydroelectric unit of a hydropower station located in Southwest China,the identification of various operating conditions of the unit has been achieved with an accuracy rate of 98.06%.

关 键 词：水电机组 振动信号 连续小波变换 深度可分离卷积 

分 类 号：TV734.1[水利工程—水利水电工程] TK73[交通运输工程—轮机工程] TV738[动力工程及工程热物理—流体机械及工程]