Analysis And Forecasting On Water Consumption Of Areas With Large Water Demand In Urban Water Supply Network

A reliable urban water consumption forecasting, as a key link in urban water supply network, is of great importance in the persistence and effectiveness of urban water system. Among the study on the current management of urban water supply system, most of them only focus on the building and improvement on the prediction model of water consumption. Instead, few researches have centered on the characteristics and prediction of water consumption in small area of regional consumption. Therefore, it is inspiring for reliable management of the entire urban water network and monitoring of leakage on regional water facility by regionalizing and specifying on management of water supply network, in combination with research on characteristics of regional water consumption for realization of the control and forecasting of regional water consumption.This thesis focuses on different community in urban water supply network. First, to receive and save real-time data, it comes up with relative data obtainment and storage and selects a data management process for pre-processing. Second, based on neutral network theory, it uses three widely used neural network prediction methods, including Wavelet neural network prediction method, Elman neural network prediction method and BP neural network prediction based on the genetic algorithm optimization and apply these models into different communities. By analyzing the water prediction results and features of water consumption, the thesis explores the methods for improvement. Third, the thesis puts forward the improved adaptive algorithm of fuzzy c-means cluster, and then makes a case study of a community. Specifically, the thesis makes the clustering analysis on hourly water consumption mode based on the above algorithms to obtain clustering results of different mode curves. Then attribute-reducing algorithm is used to analyze the possible factors in affecting water consumption, such as weather, temperature, rainfall, number of weeks in getting the major factors. Finally, the new prediction model will be built based on the major factors and observance of historical water consumption data. The experiment demonstrates that the new model is good for improving the prediction of water consumption.In the thesis, the water consumption prediction mode based on time series prediction method meets the requirement of reliable prediction in parts of communities. The application of clustering analysis and attribute reduction on hourly water mode curves helps obtain the major factors in influencing the water consumption. By putting these factors into prediction model, it further improves the accuracy to forecast water consumption. The study lays a scientific foundation for predicting the water consumption for those who have large demands in water usage and provides with new methods in automotive monitoring leakage of facility in small areas. The defect of the thesis is that there is still much more to study on the improvement of daily and monthly water consumption. At the same time, since the imperfect data pre-processing, the abnormal data in some of the sample data may disturb the effectiveness of the model, the thesis needs to explore more reliable pre-processing method.