基于时频分析的水质多尺度特征提取和异常检测方法研究

发布时间：2018-03-12 17:14

本文选题：供水管网　切入点：水质时间序列　出处：《浙江大学》2015年硕士论文　论文类型：学位论文

【摘要】：水质安全问题事关民生,在全球范围内都受到了高度的重视。水质污染异常事件往往发生突然,而且易于在短时间内带给人们巨大危害。因此,准确、快速地检测出水体中潜在的水质异常,实现及早预警是保障民生的前提,也是当前大家共同关注的重大课题。由于供水管网中的水质异常发生时会在水质指标时间序列上有所反映,为此本文主要基于小波分解和经验模态分解方法,在时域和频域上对水质时间序列进行分析,提取出水质异常发生时水质指标时间序列在时频域上的特征,采用阈值法、能量谱分析法等进行特征判别,从而实现更为有效的水质异常检测。本文的主要研究工作和成果如下：(1)针对水质时间序列中所隐含的不同时频特性,引入并研究了基于小波分析的水质监测数据预处理和阈值超标异常检测方法,进行了水质单指标和多常规指标融合的异常事件判断,并用受试者工作特征曲线(ROC)对其检测效果进行评估。论文首先通过小波变换对水质信号中存在的离群点与基线漂移等进行预处理,然后利用小波包分析方法分解水质单指标和多指标时间序列,再根据各频段上水质信号的强度分布进行信号异常与否的判断。此后,论文利用美国国家环保署Canary软件模拟的管网水质数据,进行了单指标和多指标异常检测仿真分析,以验证该方法对水质异常检测的有效性。(2)为了充分利用水质信号在不同频段上体现出的能量特征,在进行水质数据小波分析预处理的基础上,研究了基于小波包能量谱(能量特征向量)的水质时间序列分析方法。论文将水质监测数据和背景数据(正常数据)在不同时频段内的能量谱进行比较,以统计分布标准差大小为判断准则来综合判断水质波动是否隐含异常事件。论文利用多指标水质数据对算法准确率等性能进行了讨论。(3)针对水质信号常常表现出的周期性特征,研究了面向周期性水质波动和异常的分析和检测方法。论文首先研究了基于傅里叶频谱分析的水质监测数据周期性波动判定方法,利用基于经验模态分解的水质信号周期性分量提取和周期模式判断技术,以统计分布标准差大小为判断准则,实现了周期模式的匹配与判断。最后采用水质时间序列数据,展开了水质周期性波动异常事件检测实验,并与时间序列递增、线性滤波等算法的检测结果进行了对比分析。论文工作分别从水质时间序列不同时频段特征分析、不同时频段能量谱分析以及不同时域周期性特征分析的视角,研究了相应的水质异常检测方法,并进行了验证实验,为各型水质异常波动的有效检出提供了技术积累。
[Abstract]:Water quality and safety issues related to people's livelihood, in the global scope is highly emphasized. Water pollution incident often occurs suddenly, in a short time and easy to bring people great harm. Therefore, accurate and rapid detection of water quality in water potential abnormal, and early warning is a prerequisite for the protection of people's livelihood, but also a major issue the current common concern. Due to the water quality of the water supply network system exception occurs will be reflected in the water quality index time series, this paper mainly based on wavelet decomposition and empirical mode decomposition method, time domain and frequency domain on time series of water quality analysis, water extract abnormal frequency characteristics on the time when water quality index sequence, using the threshold method for the discriminant feature energy spectrum analysis, in order to achieve a more effective water anomaly detection. The main research work and The results are as follows: (1) according to the frequency characteristics of the implied time series of water quality in different time, we introduce and study the wavelet analysis of water quality monitoring data preprocessing and threshold exceed the standard anomaly detection method based on the integration of water quality incident single index and multi routine index of judgment, and the receiver operating characteristic curve (ROC) to evaluate its detection effect. Firstly, using wavelet transform to preprocess the existing water quality signal outliers and baseline drift, and then use the wavelet packet decomposition quality of single index and multi index time series analysis method, according to the water quality signals of the frequency band of the intensity distribution of abnormal signal judgment. Then the water quality simulation data, the national environmental protection agency of Canary software, the single and multi indicator anomaly detection simulation analysis to verify the method of water quality anomaly detection The validity of (2). In order to fully reflect the characteristics of the energy in different frequency band signals in water use, water quality data preprocessing based on wavelet analysis, the wavelet packet energy spectrum based on (energy eigenvector) water quality time series analysis method. The water quality monitoring data and background data (normal the data were compared in different time) frequency band energy spectrum, the statistical distribution of standard deviation as the criterion to judge whether the implied fluctuation of water quality anomalies. Multi index data of water quality of accuracy using the algorithm performance are discussed. (3) the periodic characteristics for water quality signals often exhibit, study the periodic fluctuation of water quality and anomaly analysis and detection methods. The thesis firstly studies the method to determine the Fourier spectrum analysis based on the monitoring data of water quality fluctuation, based on empirical mode The water quality state decomposition periodic signal component extraction and periodic pattern prediction technology to the statistical distribution of standard deviation as the criterion, to achieve matching cycle model and judgment. Finally, the water quality time series data, the quality of the periodic fluctuation of abnormal event detection experiments, and increasing with time series, linear filtering detection results methods are compared and analyzed. This paper respectively from the water quality time series and frequency characteristic analysis, time frequency energy spectrum analysis and time-domain periodic characteristics of the perspective of water by the method of anomaly detection, and the results were validated, provides technical accumulation for effective detection of abnormal fluctuations of various types of water quality.

【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：X832

【相似文献】