Otimização do clima avícola para maior produtividade

Abstract

Machine learning, search and mutation methods were applied to find ambience parameters (also called action plans) for broiler house controllers, capable of providing greater weight gains by feed consumption, with the potential to increase productivity. Setting temperature and humidity parameters that are capable of providing the necessary thermal comfort for the good development of broilers traditionally depends on the empirical knowledge of a poultry specialist. The study carried out here is based on previous literature on the application of machine learning, search and optimization methods using genetic algorithms, but applied to a real database of an automation company in the poultry sector, collected automatically, covering the stages of data engineering, selection of relevant learning characteristics, learning model training, productivity prediction based on ambience parameters, search and mutation using genetic algorithms to obtain candidates for action plans for higher productivity. At the end, the expected productivity for the proposed action plan is compared with a sample from the database and with the literature. Aspects that could improve forecasts for future work in the area are also presented.