【摘要】 猪采食量的测定面临设备成本高、日常维护难度大以及测定效率有限等问题。本研究利用机器学习技术构建了四种猪采食量预测模型：多元线性回归（multiple linear regression, MLR）模型、K近邻（K-nearest neighbors, KNN）回归模型、随机森林（random forest, RF）回归模型和支持向量机回归（support vector regression, SVR）模型，并利用320头猪共计390 775次采食数据对各模型的预测准确性进行了评估，旨为开发一种简单高效的猪采食量预测方法。结果表明，SVR模型性能最佳，其均方根误差（root mean square error, RMSE）和平均绝对误差（mean absolute error,MAE）分别为0.056 kg、0.034 kg，决定系数（R~2）达到0.88。MLR模型的预测效果较差，RMSE、MAE和R~2分别为0.088 kg、0.056 kg和0.72。KNN和RF模型的各项评价指标均优于MLR，其中KNN的RMSE、MAE和R2分别为0.078 kg、0.051 kg和0.78,RF模型则分别为0.057 kg、0.035 kg和0.88。综上所述，在这四种典型的机器学习回归模型中，SVR模型在猪采食量预测方面表现最佳。更多还原

【Abstract】 Accurate measurement of pig feed intake has long been challenged by high equipment costs, complicated daily maintenance, and low measurement efficiency. To develop a simple and efficient method for predicting pig feed intake, four machine learning-based prediction models were developed in this study: multiple linear regression(MLR), K-nearest neighbors(KNN), random forest(RF), and support vector regression(SVR). The predictive accuracy of each model was evaluated using feeding data from 320 pigs(390 775 observations in total). The results indicated that the SVR model achieved the optimal prediction performance, with a root mean square error(RMSE) of 0.056 kg, a mean absolute error(MAE) of 0.034 kg, and a coefficient of determination(R~2) of 0.88. The MLR model showed relatively poor accuracy, with corresponding values of 0.088 kg, 0.056 kg, and 0.72, respectively. Both the KNN and RF models outperformed the MLR model; the KNN model yielded an RMSE of 0.078 kg, an MAE of 0.051 kg, and an R~2 of 0.78, while the RF model reached 0.057 kg, 0.035 kg, and 0.88, respectively. In summary, the SVR model exhibited the best performance for pig feed intake among the four typical machine learning regression models.更多还原