【摘要】 支持向量机回归算法是一种基于统计学习理论,具有较强预测性能的数据建模方法。采用该方法构建利用亲本主要经济性状的中亲值预测家蚕杂交一代经济性状成绩的模型,简化传统家蚕杂交育种需对所有可能的杂交组合进行测交测试的过程,提高育种效率。收集7次家蚕新品种选育的杂交组合选配试验中,共325个杂交组合的12项经济性状成绩及亲本品种15项经济性状的中亲值,利用支持向量机回归算法建立了家蚕杂交一代12项经济性状成绩的预测模型。模型经5折交叉验证法证明:对于训练集,其平均标准化均方误差在0.046~0.574之间,表明模型具有较好的拟合性能,暗示家蚕杂交一代的经济性状成绩与其亲本多项性状的中亲值之间存在着某种非线性关系;对于测试集,模型预测值与实际观察值的Pearson相关系数,除虫蛹率性状预测模型略低(R=0.534)外,其余11项性状预测模型的相关系数在0.753~0.955之间,显示了模型优良的预测效果和应用价值。更多还原

【Abstract】 The support vector machine regression algorithm is a kind of data modeling method based on statistical learning theory and has strong prediction performance. Using this method,the model for prediction of economic trait performance of silkworm F1 hybrid was build based on the mid-parent value of main economic trait performance of parentral varieties to simplify the traditional cross breeding of silkworm for improving breeding efficiency. During the traditional cross breeding of silkworm,all possible hybrid combinations should be tested. In the present study,data sets of 12 economic traits of silkworm hybrid combinations and mid-parent value of 15 economic traits of parental varieties were collected from325 hybrid combinations in 7 crossing tests of new silkworm variety breeding. The model for prediction of 12 economic trait data of silkworm F1 hybrid was constructed by support vector machine regression algorithm based on the collected data sets. Five fold cross-validation method proved that the average normalized mean square error( NMSE) ranged from0. 046 to 0. 574 in the training set. It was shown that this model had high fitting degree,and there existed a nonlinear relationship between economic trait data of silkworm F1 hybrid and mid-parent values of multiple economic traits of parentralvarieties. In the test set,Pearson correlation coefficient between the predicted value and actual observed values showed that,except the larva-pupa rate model( R =0. 534),the correlation coefficient of 11 other models ranged from 0. 753 to 0. 955,showing good prediction effect and application value.更多还原