Evolutionary Computation Targeting Market Fat Specifications In Beef Steers

Increasing the efficiency and profitability of beef production through targeting market specifications for fat and weight are important industry issues. Typically, young animals are drafted into pens within a feedlot situation, and then grown out until they reach a marketable size. There are considerable price penalties for animals which do not achieve the set weight and fat depth specifications. The Cooperative Research Centre (CRC) for Beef Genetic Technologies in Armidale is promoting the use of systems models to better manage the overall production of marketable animals. In this study, the Davis Growth Model was adopted for the simulation of a feedlot scenario. This model simulates the growth, composition and fat distribution of beef cattle over time. A typical mixed-breed cohort of 306 animals was taken as the intake to the feedlot, and growth was then simulated for 60 to 140 days. The overall value of the carcasses and the variable feedlot operational costs were estimated, and the gross margin calculated as the difference between these. The optimal gross margin for the single (non-segregated) herd was $237.19 per animal, which occurred at 84 days on feed. Whilst this value initially seems attractive, note that all of the fixed costs of feedlot operations have not yet been included. Segregating this herd by breed type into separate feedlot pens, and growing each pen to its respective optimal days on feed, resulted in relatively little improvement (0.5%) in the average gross margin. However, segregation by initial liveweight allowed the pens to be marketed at a wider range of times, and this increased the overall gross margin by 2.2%. These results indicate that, across the breed types used in this study, initial size has more of an influence on ultimate profitability than actual breed. Evolutionary computation is a biologically-inspired optimisation technique, based on simulated natural selection. Given our animal intake data and feedlot model, differential evolution (a particularly efficient and robust evolutionary algorithm) was used to seek the optimal allocation of animals to pens, as well as days on feed for each pen. Despite the size and complexity of the potential search-space, this efficient algorithm repeatedly converged to the global optimum for this system, in less than two hours of computation time. The best allocation was somewhat related to initial weight, but also incorporated the interacting aspects of breed, frame size, and initial body condition and fat depth. This optimal allocation resulted in a gross margin of $244.27 per animal, which is a 3.0% improvement over the base management scenario, and worth about $2,200 for this particular intake of animals. This study demonstrates that feedlot scenario modelling can profitably be used to investigate the likely outcomes of alternate management strategies. Also, it is shown that formal economic optimisation is a useful and logical extension, and for this we recommend differential evolution as a proven robust optimisation algorithm. Future CRC research will include simulating more of the 'real-world' feedlot management options, after consultation with feedlot operators.