Performance of Genetic Algorithms and Simulated Annealing in the Economic Optimization of a Herd Dynamics Model

Abstract

This study focuses on replicated exploratory optimizations of a large and difficult beef herd dynamics model, using the net present value over a 10-year planning horizon as the variable of interest. Faced with a practical search-space of the order of 10100 possible management decision combinations, the thorough but slow search pattern of simulated annealing struggled, on average falling 1.2% short of the global optimum of the system. By comparison, the cross-breeding and mutating nature of the genetic algorithm searches usually produced good results, averaging 0.1% from the global optimum. Also, these were achieved with about half the computing time used by the simulated annealing optimizations. Hence, for this problem, genetic algorithms proved the superior method.