Inteligência artificial aplicada à predição da temperatura superficial de frangos de corte

Abstract

Broiler chickens in the initial growing stage do not have the thermoregulatory system fully developed and, if at this stage they are submitted to thermal discomfort conditions, the performance will be reduced. Surface temperature (Ts) is a physiological, non-invasive response in which the animal can be considered as a biosensor. Thus, the objective of this research was to develop a model based on artificial neural networks to predict the Ts of broiler chickens. To train and validate 100 neural networks, a database with 630 registers was randomly divided for training (70%), test (15%) and validation (15%). Subsequently, the best performing network was chosen. For the development of the RNAs, the input variables were the age of the birds (I) and the air temperature (Tar), and the output variable Ts. The developed RNAs adopted multilayer-perceptron (MLP) architecture with an input layer, a hidden layer and an output layer. The best network is able to predict with high reliability the Ts of young broiler chickens once a coefficient of determination (R2) of 0.9118 was obtained in the validation phase. With the characteristics of the best network, including the neuronal weights, it is possible to develop software that can be shipped in controllers, in order to control the thermal environment inside commercial broiler houses.