Appendix 1 79References 89Index 105
PrefaceSystems research is increasingly being used to investigate and analyse a wide range of real-world problems, including agricultural production systems. Given a valid, verified model of a particular system, optimisation is a logical complement to the modelling exercise. Usually, this takes the form of maximisation of some measure of the system's performance, such as total production or economic gross margin. This book deals with the practical application of optimisation techniques, particularly evolutionary algorithms, to the study and management of these agricultural systems. It should prove useful to practitioners applying these methods to the optimisation of agricultural or natural systems, and would also be suited as a text for systems management, applied modelling, or operations research university subjects. Basic knowledge in systems research, along with some computing and programming skills, are assumed.Models of agricultural systems range widely on both temporal and spatial scales. Farm-level systems have typically been investigated, but models also range out to regional, industry and national scales. Short-term (within-year) profitability and cash-flow issues are common, but the time-frame can be extended to a hundred years or more, to investigate sustainability and long-term effects. In addition to the 'direct' economic maximisation of agricultural systems, optimisation methods have also seen use in the calibration of internal model parameters to observed data, maximising the rate of genetic gain in livestock, in agricultural allocation and scheduling problems, and in the analysis of sustainability issues in natural systems management.Agricultural models present a number of difficulties with regard to optimisation. These problems include complex relationships which are not conducive to the simpler forms of economic modelling (such as linear programming); biological variability, which usually requires a stochastic viii model; the identification of suitable variables to optimise; the high degree of complexity in these systems, which translates to high dimensionality of the search-space; frequent interactions between the effects of the various (assumedly independent) management options; cliffs and discontinuities in the search-space (where the system is over-utilised, and 'crashes' both biologically and economically); and the presence of multiple local optima, caused by very different combinations of management options having similar economic outcomes.Any selected optimisation method is required to deal with all these problems, and reliably return the solution for the global optimum (or a value suitably close to this). Generally, evolutionary algorithms (including genetic algorithms, evolution strategies, and hybrid methods) have proven superior for this task. Depending on the algorithm and the type and usage of the model, some problems do remain, however evolutionary algorithms contain a number of advantageous features which largely circumvent these. For the altern...