Effect Of Soil Compaction On Soybean Yield Based On Stacking Ensemble Learning

Soil is the material basis for human survival and is vital for the sustainable development of agriculture.In recent years,with the popularization of agricultural modernization and the promulgation and implementation of a series of agricultural machinery purchase subsidy policies,China’s agricultural mechanization has achieved unprecedented development.While large-scale agricultural machinery operation improves production efficiency,the risk of soil compaction is also rising.Over mechanical compaction will lead to changes in soil physical properties,increase soil penetration resistance,reduce porosity,weaken water permeability,hinder the growth of crop roots,and therefore reduce crop yield.In order to further explore the relationship between soil compaction caused by mechanical operation and crop yield,based on the reality of soybean mechanized production in Heilongjiang Province,this research designed compaction tests using different types of tractors and at different times.On this basis,the impact of soil compaction on soybean yield was predicted based on the improved Stacking ensemble learning method.The research included the following three specific aspects:(1)Gradient compaction test design.Combined with the reality of soybean production mechanization in Heilongjiang,soil compaction experiments tests were designed with different gradients.Large,medium and small tractors were used for 2,4,6,8,10 and 12 times of compaction operations respectively,so as to simulate the impact of mechanical operation on soil and crops in the soybean production mechanization,and the data of soil penetration resistance data at different depths and the data of soybean yield were collected.(2)Construction and performance evaluation of prediction model for the impact of soil compaction on soybean yield.A prediction method of soil penetration resistance to soybean yield based on improved stacking ensemble learning is proposed.Taking soil penetration resistance at different depths as the input variable and the change rate of soybean yield in the compaction test group and the control group as the output,the impact prediction of soil compaction on soybean yield was realized.In the design of Stacking structure,Lasso regression,random forest,Xgboost,support vector regression and k-nearest neighbor algorithm were used as the candidate set of base learners.Bayesian optimization method was introduced to optimize the model super parameters.The performance correlation of base learners was calculated through Spearman correlation coefficient,and the models with large differences were selected as base learners,so as to make full use of the advantages of different models and improve the prediction accuracy.The performance of the improved Stacking ensemble learning model was compared with that of a single learner.It was concluded that the improved Stacking ensemble learning model has better robustness in predicting the impact of soil compaction on soybean yield,and could effectively solve the problems of low prediction accuracy and weak generalization caused by plot difference and variable coupling.(3)Feature importance evaluation scheme based on multi model fusion.A multi model fusion feature importance evaluation method based on Lasso regression,random forest and XGBoost is proposed to evaluate the importance of input features.The results showed that in the process of soybean growth,the soil penetration resistance of the surface layer(0～30cm)had the greatest impact on soybean yield.Using two-way ANOVA and Spearman correlation coefficient analysis the factors affecting the firmness of topsoil were further analyzed.It was known that agricultural machinery was the main factor causing the change of penetration resistance of topsoil.In the actual production operation,we should make rational use of machinery and equipment,reduce the unnecessary usage of agricultural machinery,do a good job in the loosening of topsoil,improve the topsoil structure in order to reduce the impact of soil compaction on crop growth,and therefore improve economic benefits and the quality of agricultural production.In conclusion,the research results may provide theoretical and technical support for the field operation of agricultural machinery,which is of great significance to improve the quality of soybean,protect soil resources and promote the development of quality-benefit agriculture.