基于音频特征的水车室工作状态异常检测
Audio Feature-based Anomaly Detection of Working Status of Turbine Room
曾广栋 1魏学锋 1何林 1孙长江 1张旋2
作者信息
- 1. 溪洛渡水力发电厂,云南 永善 657300
- 2. 北京华控智加科技有限公司,北京 100084
- 折叠
摘要
水电站的水车室包含轴承和顶盖等机械设备,受水力因素影响,水车室的异常工作会带来较大的安全隐患,基于大数据分析的精确维护对于水车室的可靠运行至关重要.针对水车室的异常工作状态,通过模型训练、特征工程和分类模型的开发等过程,采用STFT、Log-Mel、MFCC等方法对音频数据进行了预处理,建立了基于音频数据的异常检测模型,并对溪洛渡水电站水车室工作状态进行了异常检测.结果表明,Log-Mel方法具有有效性.研究结果不仅降低了异常检测的成本,还为水电机组的健康监测提供了参考.
Abstract
The turbine room of hydropower station contains mechanical equipment such as bearings and top cover.Due to the influence of hydraulic factors,the abnormal operation of the turbine room causes a large safety risk.Accurate maintenance based on big data analysis is essential for the reliable operation of turbine room.This paper takes the abnor-mal working condition of the turbine room as the research object and establishes an abnormality detection model based on audio data.The process of building the detection model includes model training,feature engineering and development of classification model.The feature engineering uses methods such as STFT,Log-Mel and MFCC.The experimental results prove the effectiveness of the Log-Mel method through a case study conducted at the Xiluodu hydropower station.This study not only achieves the low-cost purpose of anomaly detection,but also provides a valuable reference for the health monitoring of hydropower units.
关键词
音频数据/水车室/STFT/Log-Mel/梅尔频率倒频谱系数(MFCC)/时域特征/支持向量机Key words
audio data/waterwheel chamber/STFT/Log-Mel/Mel-scale frequency cepstrum coefficients(MFCC)/time-domain features/support vector machine引用本文复制引用
基金项目
三峡金沙江川云水电开发有限公司永善溪洛渡电厂科研项目(4122020006)
出版年
2024
NETL
NSTL
万方数据