|M.Sc Student||Skulovich Olya|
|Subject||Optimal Layout of Surge Contol Devices in Water|
|Department||Department of Civil and Environmental Engineering||Supervisors||Professor Avi Ostfeld|
|Dr. Polina Sela|
|Full Thesis text|
When designing and operating a water distribution system, dynamic transient analysis is often threated extemporary, despite its potential catastrophic impact on the system. In this work, optimization techniques are introduced for optimal valve closure operation and closed surge tanks’ (CST) placement and sizing for surge control.
The effect of downstream valve closure scheduling was analyzed to find optimal closure parameters that lead to the minimal maximum pressure head. Several valve closure strategies were explored, combining known valve performance curve (change in flow as a function of change in valve’s opened area) with unknown valve closure curve (change in valve’s opened area as a function of time). Second order polynomial, power function, and piece-wise linear curves were implemented and compared. Genetic algorithm (GA) and Quasi-Newton optimization methods are applied. The methodology was tested for three networks, including looped gravitational and pressurized networks. The results demonstrate that flexible multi-parametric valve closure curve and Quasi-Newton optimization method are more effective in minimizing the maximum pressure head in the system.
Proper selection of CST is carried out with GA, bi-level optimization and repeated mixed-integer linear programming (MILP) approaches. Last two techniques were developed specifically for this work. Bi-level optimization method exploits natural hierarchic structure of the decision variables in devices placement and design. In repeated MILP approach, a piece-wise function-fitting model is integrated with mixed integer programming. The key feature of the algorithm is model-driven discretization of the search space with subsequent linear approximation of the nonlinear non-smooth transient hydraulic model and mixed integer linear programming optimization. The proposed approaches are implemented using two case studies. Although the modeling assumptions for surge control devices in this work include only closed surge tanks, the methodology is fairly general and could in theory address any surge control device if smooth relationship between the maximum pressure head and device’s design parameter is detected. The repeated MILP algorithm appears to provide computational advantage and more problem-specified concept over general GA approach. For bi-level optimization approach, number of function evaluations depend on algorithm starting point. Since the algorithm utilizes GA at the inner level (devices’ sizing), the overall number of transient simulation runs may exceed the number of transient simulation runs of GA used for complete optimization (location and sizing).
Overall, results indicate the potential of the developed approaches for optimal surge control design problem in water systems in terms of computational efficiency and model reliability.