Prediction of carbon dioxide concentration in weaned piglet buildings by wavelet neural network models

Because CO2 is a highly autoregressive variable, the best results are obtained with the AM. The AM yield an RMSE of 26.330 ppm and a Pearson's r of 0.995. The EM, with an RMSE of 154.361 ppm and a Pearson's r of 0.895, reveal the importance of indoor and outdoor temperatures and, consequently, of ventilation rate, for CO2 concentration inside the building. In addition, our results show the effects of animal activity on CO2 concentration, which are delayed by 40-50 min. Based on these results, the CO2 concentrations in the animal zone of weaner buildings can be accurately predicted by WNN models. Therefore, WNN modeling could be widely used to predict and understand the dynamics of environmental variables inside livestock buildings.