Research On Water Distribution System Modeling And Predicting

Water distribution system is a complex, large-scale dynamic network system with strongexternal disturbance. Thus, to model and predict its dynamic behavior possess important practicalsignificance such as enhancing the ability of optimal operation and emergency treatment. And themain contributions and conclusions of this paper are as follows:(1) The present research situation of water distribution system modeling is firstly reviewed.Then the problems and trends are analyzed and presented, and modern control theory anddata-driven method are proposed to apply in this research field.(2) On the basis of chaos theory and wavelet transform method the pressure time series arewell analyzed. Firstly, due to the water pressure data collected from the SCADA (SupervisoryControl and Data Acquisition) contains a lot of noise and some mutation, the wavelet transformmethod is introduced, and it effectively distinguished the pressure mutation parts from noise.Secondly, based on chaotic identification theory, the chaos was verified in the pressure time series.Thirdly, for the complexity of the pressure sequence, the embedded space technology combinedwith neural network modeling method is proposed to predict the pressure time series. Finally, apractical example shows that the prediction method has a good stability and accuracy. Therefore,the verification of chaos in pressure time series provides a way to analyse and predict the behaviorof water distribution system.(3) On the basis of system identification theory, an adaptive robust ARX (Auto-Regressivewith eXogenous Inputs) model for water distribution system is studied, and two robust and stableidentification methods (single forgetting factor method and multiple forgetting factor method) areproposed. In view of that the demand of water distribution system is difficult to be estimated,especially in the context of real time, the ARX model considers the time-variant and uncertain nodaldemand as unmeasured disturbance and updates online to adapt to the system changes by exploitingthe real-time data sets from SCADA based on robust and stable identification methods. A case studydemonstrates the ability of the model to track and predict the system dynamics. Thus it lays a modelfoundation for real-time control of the water distribution system.(4) Based on the data-driven techniques, a NARX (Nonlinear Auto-Regressive witheXogenous Inputs) neural network model is established for water distribution network. At thebeginning, in view of that the demand of water distribution system is difficult to be estimated, amethod of using the system control and state information in history to estimate the variable nodal demand, is proposed. Thus, an offline NARX model is derived. In addition, an improved BPalgorithm is used to help to identify a real system. The results show that the model is feasible andpredicts dynamics of the system well. But it fails to work well under special working conditions.Therefore, an online algorithm with exponent forgetting mechanism for the NARX model isintroduced to adapt to the real-time dynamics. The results illustrate that the improved NARX modelworks better and it tracks and predicts the dynamics satisfactorily under both common and specialworking conditions.