Abstract
Environmental quality in nursing sow buildings has been a crucial determinant of the health and growth performance of piglets in large-scale pig production systems. Given that the environment within sow buildings comprises numerous interrelated factors, predicting and evaluating environmental comfort presents significant challenges. Consequently, accurate assessment of environmental quality and timely regulation of environmental conditions are essential, particularly for optimising breeding efficiency under minimal environmental stress. An analytical method was proposed using multivariate data fusion. The data repair, Grubbs criterion and a batch estimation adaptive weighted calculation method were employed to fuse the multiple sensor data, so as to eliminate abnormal values and redundant data. The Random Forest (RF) model was selected for the feature selection. There were six feature factors that were closely related to environmental quality, including temperature, relative humidity, concentrations of NH3, CO2,H(2)S and air speed. An adaptive MSCCS-RF-MK-LSSVR model combining Mutative-Scale Chaotic Cuckoo Search, Random Forest, and Multiple Kernel Least Squares Support Vector Regression was proposed for predicting and evaluating environmental quality of nursing sow buildings. The validation test results indicate that this model outperformed four other models, achieving a coefficient of determination (R(2)) of 0.9086, a Mean Absolute Error (MAE) of 0.0639, a Root Mean Squared Error (RMSE) of 0.1787, and a computational time of 7.5862 s. Compared to the GS-RF-LSSVR model combining Grid Search, Random Forest, and Multiple Kernel Least Squares Support Vector Regression , the MAE, RMSE, and computational time were reduced by 62.89%, 51.81%, and 24.98%, respectively, while R(2) was improved by 36.80%. Most importantly, the adaptive MSCCS significantly improves computational efficiency and accelerates LSSVR convergence. Thus, the MSCCS-RF-MK-LSSVR model more effectively captures nonlinear relationships between interrelated environmental parameters and environmental quality . This method can also function as an intelligent decision support tool for real-world applications, such as adaptive ventilation control, environmental stress mitigation, and early warning systems.