Design of a neuron controller for climatized broiler houses

Abstract

The objective of the present research was to develop a low cost neuron controller for climatized broiler houses. The controller was projected to operate with four control stages, three of ventilation and one of evaporative cooling, as a function of three acquisition points of temperature and relative humidity. The hardware was constructed starting from peripheral interface controllers (PICs), capable of being divided into control module and sensor modules. Artificial neural networks (ANNs) were trained and validated for the prediction of the rectal temperature (Tr) as a function of the thermal conditions (air temperature, Tair; relative humidity, RH; and air velocity, V), thus making possible the decision making in the activation of the control stages. The architecture chosen for that purpose was the multilayer perceptron (MLP), where 139 data, obtained from the literature, were used in the training (94) and validation (45). The selected MLP with three inputs (Tair, RH and V), three neurons in the hidden layer and an output (Tr) presented excellent results, providing estimates with an average percentile error of 0.78% for the training set and 1.02% for the validation set. In the control module, the MLP inputs are obtained starting from a request to the sensor modules. The system developed presented low development cost, stable behavior and absence of random switching under simulated thermal conditions. The neuron controller showed as consistent, seeking comfortable conditions adequate for production and exhibiting the potential of artificial neural network applications to the solution of real problems related to poultry production.