In order to improve the operation stability of mountain tractors in complex farmland environment, based on the Matlab/Simulink simulation platform, a time-domain simulation model of a semi-active suspension tractor with D-level road roughness as the road excitation was established and which had seven degrees of freedom. The simulation time domain model included four-wheel road excitation model, semi-active suspension vibration model, semi-active suspension vibration model, semi-active suspension tractor time domain simulation model, the vertical displacement of body, body inclination angle and body pitch angle were used as the attitude change parameters of the tractor to carry out the simulation test. By constructing the incremental proportion integration differentiation,（PID）controller and the back propagation（BP）neural network PID controller simulation model, the automatic control of the body attitude of the semi-active suspension tractor was realized, and the control performances of the two controllers were respectively evaluated. Using the vertical acceleration of the body and the relative dynamic load of the wheels as the performance evaluation indexes of the semi-active suspension system, the performance of the semi-active suspension under the two control modes was evaluated. The simulation results showed that for the semi-active suspension tractor based on the traditional incremental PID control algorithm, the RMS value of the body vertical displacement was reduced by 42.17%, the root mean square value of the roll angle was reduced by 36.76%, and the root mean square value of the pitch angle was reduced by 57.85%. The vertical acceleration of the vehicle body was 0.0177m/s2, and the RMS values of dynamic load of the four wheels were 0.0284N, 0.0346N, 0.0239N, and 0.0304N, respectively. The semi-active suspension tractor based on the BP neural network PID control algorithm reduced the RMS value of the body vertical displacement by 74.54%, the root means square value of the roll angle by 74.66%, and the root mean square value of the pitch angle by 75.03%. The vertical acceleration of the body was 7.5758×10-5m/s2, and the RMS values of dynamic load of the four wheels were 0.0197N, 0.0235N, 0.0166N and 0.0198N, respectively. Compared with the semi-active suspension tractor controlled by incremental PID, the body stability of the semi-active suspension tractor based on BP neural PID control was better improved, which provided a theoretical basis for the design of the mountain tractor control system.