Urban Water Supply Network Modeling Analysis And Abnormal Events Detection Based On Data

Urban water supply network system is the important infrastructure of city for sustainable development.The model of water supply network is built to better understand its operation rules,and provide scientific guidance for daily scheduling and optimization management.Meanwhile,the abnormal events should be found effectively,then the measures will be taken in time,so as to ensure the urban water supply.This paper studied data driven modeling and abnormal events detection of urban water supply network system,the main researches are as follows:(1)The vector auto-regressive model and wavelet de-noising method are used in urban water supply network for short-term forecast of hydraulic pressure.As a large amount of noises and missing problems are contained in the data which is collected by SCADA system,the wavelet de-noising and interpolation processing are used to reduce the noises in the time series of hydraulic pressure and improve the quality of data.Then VAR model and VARX model are used to forecast the time series of hydraulic pressure,the forecast accuracy of VARX model has a better improvement than VAR model because the external variable(the hydraulic pressure of inlet in water supply network)is added in VARX model.(2)The SPC statistical method and VARX prediction method are attempted to detect the abnormal events in water supply network.The former method obtains the variation rules of hydraulic pressure in normal condition by statistical analysis,an abnormal event will be found when the hydraulic pressure fluctuation does not match these rules.This method has a high sensitivity but easily interfered.The latter method uses VARX model to predict the hydraulic pressure,the differences between predicted values and observed values are regarded as the evidence of abnormal events.This method has a strong disturbance restraint but the accuracy is limited by the size of abnormal events.In the last,the inference method of Bayesian network is used to combine the advantages of these two methods to improve detection accuracy.(3)The influence of noises in the hydraulic/flow of water supply network has been analyzed,then EPANET simulation software and wavelet de-noising method are used to obtain the ideal signal,jump signal and noise from the hydraulic pressure signal of observation point in the water supply network.The effective value of noises in hydraulic pressure is calculated to make sure that the nosies in the water supply network fluctuates within a certain range.Finally,the SNR is used to analyze the result of abnormal events detection in chapter four,the validity of the abnormal events detection is judged by a given threshold of SNR,and the false alarm rate is further reduced.