To simulate the broiler growth the input variables were: day of year, vents opening, wind velocity, external temperature and absolute humidity, the maximum, average and minimum of the internal temperature and absolute humidity. For that purpose, two techniques were applied, a multi-layer perceptron (MLP) static Neural Network (NN) and the Layered Digital Dynamic Network (LDDN) which were applied to a set of experimental data obtained from a broiler cycle of production. The performance for both techniques was compared using: mean squared error (MSE), mean absolute error (MAE) and model efficiency (EF). The model evaluation measurements showed the superiority of the LDDN compared with MLP. The results from the sensitivity analysis found that the variable day of year was the most important variable to predict the broiler growth rate, so using this variable as the only input variable in the model, efficiency of 0.995 was reached for simulation.
Read full abstract