基于DnCNN声音增强的高坝泄流微弱空化声音信号识别与提取  

Recognition and extraction of weak cavitation sound signals from high dam discharge based on DnCNN sound enhancement

作　　者：刘昉[1] 王润喜 庞博慧 练继建[1] 梁超[1] LIU Fang;WANG Runxi;PANG Bohui;LIAN Jijian;LIANG Chao(State Key Laboratory of Hydraulic Engineering Simulation and Safety,Tianjin University,Tianjin 300072,China;Huaneng Lancang River Hydropower Inc.,Kunming 650214,China)

机构地区：[1]天津大学水利工程仿真与安全国家重点实验室,天津300072 [2]华能澜沧江水电股份有限公司,昆明650214

出　　处：《振动与冲击》2023年第21期52-62,共11页Journal of Vibration and Shock

基　　金：国家自然科学基金(51909185);云南省重点研发计划(202203AA080009-03)。

摘　　要：空化空蚀是水工建筑物泄洪安全监测的重要内容,但是高坝泄流期间产生的强泄流噪声会大幅减弱空化空蚀音频监测方法的效果甚至致其失效。针对该问题提出了基于降噪卷积神经网络(denoising convolutional neural network,DnCNN)声音增强的空化声信号增强方法,该方法依据语音增强思想,通过DnCNN实现带噪音频监测信号中空化声信号的增强。首先对该方法的实现原理和DnCNN网络结构进行了阐述,然后使用采集自空蚀和泄流试验的空化声信号和泄流噪声对该方法的效果进行验证,最后通过支持向量机信号多分类识别试验和单分类支持向量机空化声信号单分类识别试验对该方法的泛化性能和工程实用性进行评价。研究结果表明该方法能够有效提升带噪空化声信号的信噪比,极大地还原空化声信号的频谱结构特征,实现强泄流噪声中微弱空化声信号的识别与提取,同时该方法具有较强的泛化性能和较好的工程实用性。Cavitation and cavitation erosion are important contents of flood discharge safety monitoring for hydraulic structures,but strong discharge noise generated in discharge period of high dam can significantly weaken the effect of cavitation and cavitation error audio-frequency monitoring methods,and even cause their failure.Here,aiming at this problem,a cavitation sound signal enhancement method based on denoising convolutional neural network(DnCNN) sound enhancement was proposed.According to the idea of voice enhancement,this method could realize enhancement of cavitation sound signals in noisy audio-frequency monitoring signals with DnCNN.Firstly,the implementation principle of the proposed method and DnCNN network structure were expounded.Then,the effectiveness of this method was verified using cavitation sound signals and discharge noise collected from self-cavitation and discharge experiments.Finally,generalization performance and engineering practicality of this method were evaluated through support vector machine(SVM) signals multi-classification recognition experiments and single classification SVM cavitation sound signal single classification recognition experiments.The study results showed that the proposed method can effectively improve the signal-to-noise ratio of noisy cavitation sound signals,greatly restore spectral structure characteristics of cavitation sound signals,and realize recognition and extraction of weak cavitation sound signals in strong discharge noise;the proposed method has stronger generalization performance and better engineering practicality.

关 键 词：降噪卷积神经网络(DnCNN) 声音增强 空化噪声 支持向量机 单分类支持向量机 信号识别 

分 类 号：TV65[水利工程—水利水电工程]