Derivation of near-optimal pump schedules for water distribution by simulated annealing

Abstract

The scheduling of pumps for clean water distribution is a partially discrete non-linear problem with many variables. The scheduling method described in this paper typically produces costs within 1% of a linear program-based solution, and can incorporate realistic non-linear costs that may be hard to incorporate in linear programming formulations. These costs include pump switching and maximum demand charges. A simplified model is derived from a standard hydraulic simulator. An initial schedule is produced by a descent method. Two-stage simulated annealing then produces solutions in a few minutes. Iterative recalibration ensures that the solution agrees closely with the results from a full hydraulic simulation.