Simulating and optimising selection systems

Sugarcane selection, like selection in other breeding programs, involves a multi-stage screening process. The process is time consuming and expensive, aiming to select rare superior individuals from large starting populations. In selection systems, many different parameters can be changed in attempt to make the process more effective and efficient, and it is not feasible to use empirical evaluation to compare all feasible systems. There have been some studies that have examined the effects of a set of parameters by fixing the rest of them, using a mathematical approach (Finney, 1961, 1966; Young 1974, 1976). Martin and McBlair (1991) have doubted the practical applicability of their results, arguing that a complex system has to be observed as a whole rather than analysing its parts separately. Martin and McBlair (1991) have proposed that a simulation of the system is needed to permit a comprehensive examination of multistage selection systems. Moretti and Faveri (1983) used a computer simulation to generate sugar cane clones and concluded that even a simple model, such as the one they used, is able to reproduce similar results to those observed in real popUlations. However, the model was not further employed to simulate a selection system as a whole but rather to analyse a portion of parameters that affect results from selection. Therefore, despite the potential usefulness of simulation of selection systems, there is little evidence of comprehensive investigation, and no evidence of an attempt to optimise such systems using mathematical techniques of optimisation. The study reported here aims to develop a methodology for simulation and optimisation of selection systems using a sugarcane selection system for the Burdekin region in north Queensland as a case study. The computer based sugar cane selection simulation model 'Zucchero' has been written in MS Access and represents the 'reallife' selection system. The model will be used to predict gains of different selection system configurations. The predicted results from each configuration will be used to identify an optimal selection system for the region. An algorithm that is a combination of Operation Research's (OR) optimisation techniques Dynamic Programming (DP) and Branch-and-Bound will be created to preform the optimisation. At the time of writing this paper, the model 'Zucchero' had been written, while the optimisation component of the work had not been completed.