Desenvolvimento de modelos de suporte à decisão para predição de respostas fisiológicas em bovinos leiteiros

Abstract

The objectives of this study were to develop and validate decision support systems using systems based on artificial intelligence: fuzzy logic, artificial neural networks, neuro-fuzzy networks, and regression models for the prediction of rectal temperature and respiratory frequency of dairy cows in confinement. All systems were developed based on two input variables: dry bulb air temperature (tdb) and relative humidity (RH), with rectal temperature (trectal) and respiratory rate (RR) as output variables. The fuzzy inference system was carried out using the Mamdani method, which consisted of elaborating 192 rules and defuzzification through the center of gravity. Data obtained from the literature and data observed in the field were used to manufacture the artificial neural network and the neuro-fuzzy network, where membership functions of the neuro-fuzzy system were of the triangular type. The regression models were developed in computing environment R. Experimental results were used to validate the models, and showed that the average standard deviations between the simulated and measured values of trectal for the regression model, the fuzzy system, the artificial neural network and the neuro-fuzzy network were 0.2 °C, 0.1 °C, 0.1 °C and 0.2 °C, respectively. For the values of RR, the average standard deviations were 5.0, 4.3, 3.2, and 3.5 breaths min-1 for the regression model, fuzzy model, artificial neural network and neuro-fuzzy network, respectively. Of the models developed, the artificial neural network and the neuro-fuzzy network showed the fewest prediction errors; therefore, these models are the most suitable for the prediction of rectal temperature and respiratory rate on the basis of the two climatic variables (tdb and RH), and can be used in decision support.