Experimental Analysis of Hydraulic Solver Convergence with Genetic Algorithms

Abstract

A procedure for the experimental convergence evaluation of a hydraulic-network solver is proposed, based on using genetic algorithms to search for network parameter values that maximize the number of iterations of the hydraulic-network solver under test. The efficiency of the method is demonstrated by the example of convergence evaluation for the EPANET hydraulic simulator. Examples of a pipe network and of combinations of parameter values for which the static solver of the simulator fails to converge in a reasonable number of iterations are given. The features of the EPANET 2.00.12 solver responsible for loss of convergence are discussed. New criteria for the automatic start of solution damping aimed at improving the convergence of the solver are proposed. The better convergence of the EPANET solver modified in accordance with these criteria is confirmed by the random and the proposed search-based testing method.