Research On Prediction Model Of Tillage Depth Based On An Improved Random Forest

Artificial intelligence agriculture(AI agriculture)is a production method for the future of intelligent agriculture.It uses research on intelligent search,natural reasoning and automatic programming to produce precision agriculture.Intelligent systems for diagnosis and management in agriculture can effectively save production costs.Sensor technology is gradually enhancing the technology of agricultural equipment.For example,intelligent tractors are equipped with intelligent systems to complete intelligent operations,such as deep ploughing and ploughing of large plots of land and large farms.Because of the differences in topography and soil moisture in all provinces of the country,this leads to poor levelling of the ground after the implements have been rolled over.There is a problem of unstable ploughing depth,which affects the loosening of the soil,the ability to store water and the stability of the implements.To solve this problem,a predictive analysis of tillage depth is required.The aim is to construct a prediction model for ploughing depth that has a good predictive effect and provides indicators for intelligent regulation systems.First of all this paper is based on the intelligent tractor collecting system and ploughing depth data in a test field in Nong’an County,Changchun City,in order to verify the validity of the data and test the stability of the system.Smoothness test,missing value treatment,outlier treatment,useless factor deletion and standardized treatment were used for pre-processing analysis.To ensure the excellence of the subsequent model training effect,the thesis used Pearson coefficient method and Spearman coefficient method to investigate the correlation of the research subjects.The features with weak correlations were removed.In using the importance ranking of random forest,features with importance scores below 0.01 were removed.To avoid the appearance of invalid features,thus preventing overfitting from occurring and improving the model training effect.Secondly,machine learning algorithms were applied to the field of tillage depth,and five new models of tillage depth prediction were constructed by Decision Tree(DP-DT),Random Forest(DP-RF),Bagging Algorithm(DP-Bagging),DP-Boosting,Support Vector Machine(DP-SVM).Evaluation indexes(RMSE,NMSE,R~2 and MAE)were selected as evaluation criteria for the construction and improvement of subsequent prediction models.After the initial prediction of the five constructed tillage depth prediction models,the comparison with the traditional tillage depth prediction models(multiple linear regression models)was evaluated and the six models were found to have advantages and disadvantages.Overall,the random forest tillage depth prediction model was the most effective.Finally,in order to improve the prediction accuracy and effectiveness of the tillage depth model,the random forest tillage depth prediction model with the best prediction effect was selected from the six models for improvement.To avoid the problems of data noise and high running costs of the model,it was chosen to be initially improved by external weighting of the principal component weights(PCA-FRF).In view of the slow running speed of the random forest and the existence of black box problems,the Bayesian optimization fusion grid search method was chosen to optimize its parameters.The final model(PCA-BGS-FRF)is a Bayesian fused grid search optimized random forest tillage depth prediction model with principal component weighting improvement.The final result of the improved model prediction has a good fit of 92%,a matching degree of around 1,and a minimum error between the predicted and actual values within 0.1%.This not only avoids the occurrence of over-fitting,but also improves the accuracy of the prediction.It can also determine anomalies,provide adjustable system indicators and solve the problem of unstable tillage depth.In summary,the PCA-BGS-FRF model proposed in this paper can improve the accuracy of tillage depth prediction.Effective prediction of tillage depth.It extends the field of application of statistical methods and machine learning algorithms.Therefore,this model proposed in this paper can be a reference for the data prediction processing of the data piled up in the sensors of sprinklers,seeders,etc.