Prediction Model Of Soil Specific Resistance Based On BP Neural Networks

Soil specific resistance refers to the drag resistance consumed on the unit area of cultivated soil during field work.The traction resistance of agricultural machinery can accurately be estimated by accurate and reliable soil specific resistance data,which is conducive to the efficient development and implementation of matching,type selection and design of tractors and agricultural machinery.The intelligent control and decision-making of agricultural machinery in field operation depend on accurate and reliable soil specific resistance data,which is conducive to improving the level of intelligent and accurate operation of agricultural machinery.The research work of this paper is part of the national key research and development program(2016YFD0700404)and the funded program of Jilin provincial science and technology development program(project number:20170204014NY).On the basis of theoretical analysis and experimental research,taking soil specific resistance as the research object,based on the electro-mechanical and hydraulic integration technology and the introduction of intelligent digital technology and sensor technology,combined with Kalman filter algorithm and BP neural network algorithm,using Matlab and Excel data-processing software,a set of soil specific resistance data acquisition system was designed to obtain the data of soil specific resistance through field experiments and a predictive model of soil specific resistance based on BP neural network was put forward,aiming to provide reference data for the intelligent precise field operation and improve control precision.The main contents are as follows:(1)On the basis of the study of sensor communication interface,communication protocol,and numerical analysis of the return instructions of the sensor,a set of soil specific resistance data acquisition system was designed,mainly including a tractor,a suspended subsoiler,three tension sensors,an inclination sensor and a data acquisition module.The three tension sensors were respectively mounted in the upper link,the left lower link and the right lower link;The inclination sensor was mounted on the left lower link.(2)The relationship between tillage depth and inclination angle of the left lower link,the relationship between soil resistance and pull force of the links,and the relationship between tillage area and number of deep spades,tillage depth and tillage width were established.Then,the ratio of soil resistance to tillage area is soil specific resistance.(3)The Kalman filtering principle and recursive formula were learned and studied,including:the establishment of state equation,prediction equation,Kalman gain equation and iterative correction equation,as well as the determination of initial parameters.The process noise of the sensor was obtained by contrast test,the measurement noise of the sensor through repeated measurements,and the Matlab program of Kalman filter was written.(4)The algorithm principles of artificial neural network and BP neural network were learned and studied,including:determination of the number of input layer,hidden layer and output layer,determination of the number of neurons in each layer,forward transmission of signal and back propagation of error,correction of corresponding weight and threshold,etc.Soil specific resistance prediction models based on BP neural network were proposed by utilizing the first five consecutive soil specific resistance values to predict the sixth soil specific resistance value.(5)With two field trials conducted on June 29th and October 28th in 2018respectively,sensor data was obtained through the soil specific resistance data acquisition system,the Angle sensor data and tension sensor data were preprocessed by Kalman filter to improve the reliability of soil specific resistance data,the soil specific resistance deduced through formula.The soil specific resistance values of the two tests were respectively(30-65)kN/m~2 and(50-120)kN/m~2.The BP neural network was trained and tested off-line by use of Matlab software,that is the trained BP neural network is the soil resistance prediction model.The results show that the model has a good fitting degree for the training samples,with an mean square error(mse)of 0.4203 and 3.6983,and the total goodness coefficient of fit decision R~2 of 0.994 and 0.978,respectively;it has a good prediction effect on the test samples,with an mean relative error of 0.78%and 1.95%,and an mean absolute error of 0.419kN/m~2 and 1.401kN/m~2,respectively.The prediction accuracy of the soil specific resistance prediction model established by BP neural network algorithm is high,which is beneficial to improve the level of intelligent and accurate operation in the field.