基于RBF神经网络的种猪体重预测  

Prediction of Pig Weight Based on Radical Basis Function Neural Network

作　　者：刘同海[1,2] 李卓[1] 滕光辉 罗城[1] 

机构地区：[1]中国农业大学水利与土木工程学院,北京100083 [2]天津农学院计算机科学与信息工程系,天津300384

出　　处：《农业机械学报》2013年第8期245-249,共5页Transactions of the Chinese Society for Agricultural Machinery

基　　金：国家自然科学基金资助项目(31072066);国家公益性行业(农业)科研专项资助项目(201003011);中央高校基本科研业务费专项资金资助项目(KYCX2011082)

摘　　要：针对猪体生长参数之间具有一定的自相关性、部分参数与体重间呈非线性关系、通过线性回归模型预测猪体体重存在着自变量间共线性及拟合优度较低等问题,以52头长白母猪的生长参数为基础,通过最近邻聚类算法,构建了基于RBF神经网络的种猪体重预测模型。通过线性回归检验法对种猪体重预测值与实测值进行分析,发现基于RBF神经网络的长白种猪体重预测模型的拟合优度R2为0.998,而线性回归模型的R2为0.891。结果表明:通过RBF神经网络方法建模,消除了线性回归分析中自变量的共线性问题,预测效果优于线性回归模型。There is a certain correlation among the growth parameters of pigs,and some parameters have non-linear relationship with pig weights.When using the simple linear regression model to predict the pig weight,the collinearity among independent variables and low fit goodness were found.For these problems,based on the nearest neighbor clustering algorithm for RBF neural network,a pig weight prediction RBF neural network model was constructed with the growth parameters of 52 Landrace sows.The predicted value and measured value of the pig weight were compared by linear regression test.The regression analysis showed that the goodness of fit（R2） of RBF neural network prediction model for Landrace pig weight was 0.998,while R2 of the linear regression model was only 0.891.The results indicated that the RBF neural network-based modeling method was an effective way to build the prediction model of pig weight.It eliminated the collinearity of the independent variables in linear regression analysis,and forecasted better than linear regression model.

关 键 词：种猪体重 径向基函数 神经网络 回归分析 预测 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] S818[自动化与计算机技术—控制科学与工程]