华中农业大学学报2026，Vol.45Issue(3)：56-67,12.DOI:10.13300/j.cnki.hnlkxb.2026.03.005
融合随机森林-递归特征消除的冬黑麦产量预测模型的建立
Establishment of a model for predicting yield of winter rye with a hybrid random forest-recursive feature elimination
摘要
Abstract
A two-factor field gradient experiment was conducted to solve the problems of traditional models being difficult to analyze the nonlinear effects of interaction between row spacing and sowing rate.The biological traits and yield components at the stages of key growth in winter rye were systematically measured.The patterns of interaction between population dynamics and yield components during these stag-es were analyzed to construct a recursive gradient fusion-based random forest model for predicting the yield of winter rye.The results of field experiment showed that the combination of a row spacing of 40 cm and a sowing rate of 535-680 g/40 m2 can optimize the light transmittance of canopy,balance the"source sink"relationship,and improve the resistance to stress,achieving coordinated regulation of high and stable yield of winter rye.Samples of plant were systematically collected from different treatment groups at the stage of full maturity in winter rye.A standardized experimental procedure was used to quantitatively characterize key agronomic traits including spike length,spike mass,number of spike nodes,plant height,root length,effective tiller count,leaf count,and seeds per plant.540 sets of valid data were obtained.A multidimen-sional feature set encompassing sowing row spacing,seeding rate,spike length,spike mass,number of spike nodes,plant height,root length,effective tiller count,leaf count,and six derived features was con-structed using seeds per plant as the prediction target.A hybrid random forest-recursive feature elimination(RF-RFE)with excellent predictive performance(R²=0.951,RMSE=2.28)and generalization ability(OOB-RMSE=4.33)was constructed.The model for predicting yield was used to invert and obtain a three-dimensional response surface for row spacing and seeding rate to identify the high-yield parameters as row spacing ranging from 35 to 49 cm and seeding rate between 535 and 680 g/40 m2.The results of mar-ginal effect showed that a 1 cm increase in row spacing beyond 45 cm led to a yield decrease of 1.23 kg,and the marginal benefit decreased as the seeding rate exceeded 650 g/40 m2.Therefore,it is recommend-ed to control the row spacing within 40-45 cm and the seeding rate between 600-650 g/40 m2 in the prac-tice of production,which can ensure high yield and leave room for the tolerance of operational error.关键词
冬黑麦/随机森林/递归梯度融合/产量预测/边际效应/三维响应曲面Key words
winter rye/random forest/recursive gradient boosting/yield prediction/marginal ef-fects/three-dimensional response surface分类
农业科技引用本文复制引用
吕林有,李佳慧,赵艳,马艳,何梓源..融合随机森林-递归特征消除的冬黑麦产量预测模型的建立[J].华中农业大学学报,2026,45(3):56-67,12.基金项目
国家重点研发计划项目(2024YFD1501405) （2024YFD1501405）
辽宁省科技攻关专项(2023JH1/10400001) （2023JH1/10400001）
