Inteligência artificial na avaliação de respostas produtivas e fisiológicas de frangos de corte submetidos a diferentes intensidades e durações de estresse térmico

Abstract

In aviculture, the thermal environment is responsible for homeothermic process of poultry, and, when subjected to some kind of heat stress their comfort is affected thus compromising the productive performance. Thus, this research aimed to analyze the productive responses, cloacal temperature (t clo, ° C) and surface temperature (t sup , ° C) of broilers submitted to different intensities and durations of air dry-bulb temperature (t db, ° C) throughout their second week of life. An experiment was conducted in the Animal Ambience Laboratory of the Federal University of Lavras, equipped with four air-conditioned wind tunnels that have recirculation and partial air renewal. It was used 240 broilers divided into four stages, where in the second week, for each step there was a different t bs (24, 27, 30 and 33 °C) with a different duration (1, 2, 3 and 4 days). The relative humidity and air velocity were fixed at 60% and 0.2 m s -1 , respectively. In the first and third experimental week, the birds were subjected to thermal comfort conditions, characterized in tdb values of 33 °C and 27 °C respectively. Variance analysis was used to analyze the effect of temperature fluctuations and its duration. Mathematical models have been developed using the fuzzy logic and artificial neural networks (ANN) in which it was possible to predict the t clo , feed conversion (FC, g) and water consumption (C water , ml) depending on the intensities and durations. The results obtained with productive responses showed that in a tdb of 24 ° C (low temperature stress) poultry had more feed intake but obtained a worst feed conversion. Best feed conversion was obtained in poultry submitted to a tdb of 30 °C. It was seen that with tdb of 24 and 27 °C there was a reduction in tclo and tsup , where poultry acclimatization to heat stress occurred from the second day of stress. The tclo values simulated by the fuzzy model had standard deviations and smaller percentage errors of 0.02 and 0.08%, respectively, than those obtained experimentally. For the ANN developed, the coefficients of determination (R 2 ) for tclo, FC and C water were 0.87; 0.79 and 0.97, respectively. These results demonstrated that the templates had high predictive power and could be used to support decision making in control of thermal environment systems.