Optimizing operational set points for complex distribution systems using genetic algorithms

Abstract

Considerable research has been carried out into the optimization of water distribution systems. Early research in this area used traditional optimization techniques such as linear and non-linear programming. More recently, genetic algorithms (GAs) have been applied successfully to real-world network planning and design problems. Genetic algorithms belong to a class of guided search techniques that utilize the full power of distribution network simulation technology. This enables all practical design considerations such as multiple loading patterns, extended period analysis, pump and valve operations, tank locations, and water quality criteria to be included in the optimization study. In fact, any distribution system features or decisions that can be simulated on the computer can be optimized using GAs. Since 1994, a number of practical applications of GA optimization have been reported in the literature. The focus of these papers has been on the use of the GA optimization technique to identify optimized capital improvement alternatives as part of a system expansion study. The reported results have been very promising with projected capital cost savings of up to 50% when compared to traditional design procedures. This paper describes another important application of GA optimization distinct from the identification of optimized capital improvement plans. This new GA application involves the investigation of operational set points and operating procedures to achieve the best overall solution for a distribution system problem. Improvements to system operations may be based on objectives such as reducing operating and energy costs, improving the level of service, lowering risks, providing for maintenance scheduling, or optimizing the utilization of existing assets. Three case studies are presented to illustrate how the GA can be formulated to optimize different operational decisions with the objective of minimizing the lifetime cost of the project. The first case study involves a fairly simple distribution system that includes a pump station, an elevated tank and a looped pipe network. The GA was used to optimize the normal operating level of the tank taking into consideration the operational cost of pumping to the tank and the capital cost of the tank and planned new pipes for rehabilitation of the existing network. A lower tank water level would result in lower pumping costs, but higher overall capital costs as the pipes would need to be larger to supply required demands. Various alternative layouts of the pipe network were later investigated. The second case study involves a complex municipal distribution system comprising two pressure zones, 300+ pipes, three ground storage reservoirs, two pressure reducing valves (PRV), and one combination pressure reducing/flow control valve (PRV/FCV). The GA optimization was applied to an extended period simulation of the maximum day event to develop alternative improvement options for the system. The constraints included not only system pressures and demands, but also a requirement that the reservoirs be drawn down at least 5.0 feet during the 24-hour event to ensure adequate turnover of storage. Various operational settings and procedures were optimized directly by the GA, including the hours for opening/closing the critical PRV and the high and low flow settings and corresponding hours for operating the FCV. In addition, alternative PRV capacities were investigated to see if further operational improvements could be made. The third case study involves an expansive water delivery system that will supply water to farms and country towns through a looped network totaling 1,500 km (930 mi) of new pipelines. The GA first analyzed five different concept design options based on assumed layouts of re servoirs, elevated tanks, pump stations and pipeline routes. The decision variables considered by the GA included the pump station heads, the normal operating levels in the elevated storage tanks, an...