Abstract: Rheological behavior can be used to characterize the multilevel composite agglomerates of cultivated soil for perennial sugarcane in South China. The intrinsic parameters, particle size distribution, and cohesive properties of soil were obtained using various physical and mechanical experiments, such as soil density, moisture content, clay particle size gradation screening, angle of repose and triaxial compression. According to the distribution of particle size within the ranges of different tillage depth and the Hertz Mindlin with JKR particle contact model, a systematic simulation was conducted on the angle of repose under clay flow. The contact factors were also screened and calibrated by the Plackett-Burman, the steepest climb and the Box-Behnken test. A multilayer model with nonhomogeneous discrete element was constructed using agglomerate geometric and mechanical properties. The discrete element parameters of heavy clay were used to explore the quantitative agglomerate generation, settling, platen undermining and layered stacking. According to the geometric composition of unsaturated soil particles and parameter assignments, such as JKR, the rheological constitutive behavior of soil was verified using the coupled triaxial compression on RecurDyn-EDEM. The rheological model of cane field tillage was verified to predict the tillage performance, and then to analyze the propulsion resistance of bionic wall plate. The angle of repose and soil triaxial compression showed that the JKR surface energy, coefficient of recovery and kinetic friction factor of the soil were the significant influencing factors on the rheological response of the heavy clay soil, with the contribution rates of 48.24%, 30.09% and 7.80%, respectively. The measured angles of repose of the soil materials in different layers were used as the objective for optimization. The optimal solution of contact parameter was calibrated to evaluate the differences of the parameters among the different layers of the cultivated field. There was the similar between the simulated and experimental stacking shape for the angle of repose using the optimal calibration parameters, where the prediction error of angle was less than 1.67%. The discrete element model was contained the agglomerate gradation distribution, intrinsic contact and cohesive properties. The composite rheological properties of multi-till layers were characterized to obtain the intrinsic response deviation. The triaxial behavioral error was less than 9.27%. The relative error was only 8.3% between the calculated using this model and the measured of propulsion resistance of the bionic wall plate. The error was within the acceptable range, indicating the effective and accurate prediction of the interaction relationship between the machine tool and the soil material. The finding can also provide the data support and model reference for the adaptive design and dynamic optimization of contact soil components.