Optimisation of a Large, Constrained Simulation Model using Compressed Annealing

Simulation models are valuable tools in the analysis of complex, highly constrained economic systems unsuitable for solution by mathematical programming. However, model size may hamper the efforts of practitioners to identify efficiently the most valuable management strategy. This paper investigates the efficacy of a new stochastic search procedure, compressed annealing, for the identification of profitable solutions in large, constrained systems. The algorithm is used to examine the value of incorporating a sown annual pasture, French serradella (Ornithopus sativus Brot. cv. Cadiz), between extended cropping sequences in the central wheatbelt of Western Australia. Compressed annealing is shown to be a reliable means of considering constraints in complex optimisation problems relative to the incorporation of fixed penalty factors in standard simulated annealing and genetic algorithms. French serradella is found to be an economic break pasture in the study region when weed populations are high or sheep production is lucrative.