Optimal Operation Of Water Distribution Systems

Because hourly water demand is affected by weather, weekend and holiday patterns, as well as regular domestic and industrial activities of consumers, it has the characteristics of periodicity and trend. As the ratio of peak samples to total training samples is low, the prediction accuracy of peak load is poor when applying support vector machine (SVM) model to predict a periodical load. A modified SVM model, which can predict peak recognition theory, was proposed in this paper. This model can increase the weight of peak error in the loss function of structural risk minimization, thus improve prediction accuracy of hourly water demand peak. According to the hourly water demand forecasting results of Hangzhou city, the reasonability and effectiveness of this model was proved.Real large water supply system is a complicatedly dynamic nonlinear system, it is influenced by many factors, and these factors are interactional. It is difficult to simulate water distribution networks by using one or several explicit functions. A modified SVM model was proposed and implicit function relationship between nodal pressures and pump station flow, pump station head was established based on modified SVM model. The reasonability of this model was embodied in the practice of networks in Hangzhou and comparison with classical SVM. Furthermore, relationships between pump station flow and pump station head, total system water demand based on SVM were established , which solved the problem of macroscopical modeling for the second part of optimization and operation of water supply system.The direct optimal model of multi-objective mixed discrete variables for lager-scale water supply system was established, by replacing complicated network hydraulic equation with nodal pressures macroscopic model and relationship model between pump stations flow and head In water supply systems without tanks or towers, genetic algorithm was developed to solve the direct optimal model. Methods to manage each constraint condition were put forward, aiming at the premature and slow convergence of genetic algorithm, this algorithm introduced the combination of genetic algorithm and simulated annealing technology, combining with self-adaptive probabilities of crossover and mutation. At the same time, the implemental approaches of the genetic algorithm for water supply systems including both fixed and variable speed pumps were introduced.The direct optimal operation model and two-stage optimal operation model of water supply systems with one or more tanks were proposed, according to the hydraulic characteristic of storage tanks. Appropriate fitness function of genetic algorithm was designed to solve upper level optimization and lower level optimization in two-stage optimal model. The approach of constraint relaxation and adjusting tank level achieving from upper level was proposed to avoid unfeasible solutions of lower-level optimal model. Hierarchic genetic algorithm (HGA) was established to solve direct optimal operation model of system with tanks. HGA is appropriate for optimal model including multi-constraints and multi-variants.