Application of a Novel Grey Model for Forecasting Indoor Air Temperature in Poultry Houses: Model Development

Highlights Novel grey forecasting method was proposed to predict indoor air temperatures. Buffered rolling mechanism, period groups, and variation indices were introduced into the grey model. Grey model could capture variation trends and air temperature fluctuation characteristics in a time series. Proposed novel model exhibits higher prediction accuracy than traditional grey forecasting models. Results can be adopted as a control strategy for thermal environment management for poultry production. Abstract. The indoor climate of a poultry building is essential for the well-being and health of birds and their production performance, but achieving in-time indoor thermal environment control is difficult using the current environmental control systems that are only based on data collected by onsite sensors. Data collected by the onsite sensors can only reflect the current thermal environment conditions. Still, it cannot predict the variation trends in the environment, nor can it reflect whether cold or heat stress will occur in the livestock building. Heat and cold stress have caused significant economic losses to egg production. Predicting the indoor air temperature in poultry houses can aid in forecasting extreme air temperature, formulation of control strategies, saving energy, and reducing losses. This study proposed a novel grey model to forecast indoor air temperature, which captures the essential features of the developing trends and fluctuation characteristics. The novel model was combined with a buffered rolling mechanism, time period groups, and variation indices to enhance the accuracy; the variation indices of each time period group were inserted into each rolling process. The proposed model was employed to forecast the indoor air temperature of a poultry house; the efficacy and reliability of the proposed model were evaluated by conducting field experiments and comparing other similar forecasting models. The results demonstrated that the traditional grey model only showed a growth trend for the measured data but failed to reflect a trend with the fluctuation effect. The absolute percent errors of the conventional and novel grey forecasting models were 16.9%, 13.3%, and 6.3% in the training stage, and 26.3%, 16.4%, and 4.8% in the test stage, respectively. The buffered rolling mechanism and variation indices of each period can reflect the deviation degree in the measured data at each time point from the average trend in real-time, minimize the absolute percentage error, and improve the forecasting performance. The proposed model was superior to the traditional grey forecasting models, exhibiting a more accurate performance based on reduced error (<10%) in both the training and test stage forecasts. The proposed grey model can capture the variation trend and air temperature fluctuation characteristics in a time series. It is a helpful tool for forecasting the indoor air temperature in poultry houses.