Abstract:As the sow building environment is a complex, nonlinear and timevarying system, consisting of multiple coupling factors, it is difficult to predict the environment comfort reasonably. Therefore, a prediction model was built to determine the variation trend of environment comfortable degree. The assessment index system was constructed, and the parameter optimization of the least squares support vector regression (LSSVR) with mixed kernels was presented based on mutative scale chaos cuckoo search (MSCCS) algorithm to find optimal parameters γ and σ. The model was exploited to predict the sow house environmental comfort. Three models of particle swarm optimization (PSO-LSSVR), genetic algorithm (GA-LSSVR) and traditional LSSVR were compared with the proposed prediction model. The experimental results showed that MSCCS-LSSVR had a higher accuracy and more reliable performance than the other three models, the mean absolute error (MAE) were 0.0611, 0.0972, 0.1306 and 0.1681, respectively. To facilitate the use of prediction model for farmers, a comfort assessment and prediction system graphical user interface (GUI) based on Matlab was developed. Farmers could download the historical data from a webserver and then exploit them as training and testing data, the assessment and prediction results at different time calculated and displayed on the GUI. A prediction model was exploited in Zhenjiang, Jiangsu Province, China, and it performed well. It can reflect the air quality reasonably and also provide decision support for precise regulation of a swine house environment. It can help farmers decrease the risk of livestock breeding.
