EPSILON-AND NU-SUPPORT VECTOR REGRESSION ALGORITHMS FOR MOLDBOARD PLOW DRAFT-FORCE PREDICTION.

The draft force acting on a moldboard plow plays an important role in the design of more efficient plows to facilitate attainment of optimum results when implementing size matching to estimate the required tractor power. The purpose of this study, therefore, was to investigate e-SVR and u-SVR algorithms, in predicting the draft force acting on the moldboard plow considering plowing speed, plowing depth, and soil moisture content as governing parameters. The experimental design was randomized block with two replications and the treatments were: 1) three plowing speeds, 2.71, 3.21, and 4.32 km h-1; 2) three tillage depths, 9, 14, and 16 cm; and 3) three levels of soil-moisture content, 9.9, 10.9, and 13.2% db. The data were analyzed with the SPSS Software and treatment means were compared with the Tukey test (p£0.05). The maximum draft force, 8.57 kN, was observed under conditions corresponding to a plowing speed of 4.32 km h-1, 16-cm plowing depth, and soilmoisture content of 9.9% db. A prediction model was developed using two learning algorithms, e-SVR and u-SVR, to facilitate accurate prediction of the moldboard-plow draft force through use of different kernel functions: linear, polynomial, sigmoid, and radial-basis-function kernels. The prediction model based on e-SVR and the use of the radial basis function demonstrated optimum performance with regards to draft-force prediction, and key parameters corresponding to the said optimum model assume following values: C = 1604.774, e= 0.001, l= 0.123, and p = 0.012. The total analysis run time was 3.93 min. The root mean squared error in predictions made by the proposed model was 0.288 kN, and the correlation coefficient between actual and predicted draft forces was 0.969, thereby confirming a satisfactory performance of the proposed model. [ABSTRACT FROM AUTHOR]