Optimization of the automatic shifting strategy for heavy tractors based on dynamic programming and BP neural network

Full-power shift gearbox has been one of the key configurations to improve the operation efficiency of heavy tractors. However, the conventional shift strategy cannot fully meet the requirements of the global optimization under multiple-gear overlap. In heavy tractors, the frequent shifting and high fuel consumption can be caused by the overlap of adjacent gear speeds, seriously restricting the intelligent equipment in modern agriculture. It is often required for the global collaborative optimization of optimal working conditions in real time. In this study, a compound shift strategy was proposed to integrate the dynamic programming (DP) with the global optimization and back propagation neural network (BPNN) for real-time prediction. The shift control was divided into a field mode (throttle, speed, and slip rate) and a road mode (throttle and speed), according to the operational classification. A DP bi-level optimization was constructed to incorporate a penalty function of the shift frequency. Fuel economy was prioritized as the core target. Offline optimization was also performed on the real-vehicle plow load spectrum and road transportation from the suburban cycle driving profiles. A double-hidden-layer BPNN controller for the gear prediction was trained using the DP offline optimization dataset. A series of tests was conducted to validate the optimization via an AMESim-MATLAB/Simulink co-simulation platform. Results demonstrated that the shift strategy fully met the power demands of the working conditions under the DP offline solution. In plowing, the shift frequency and fuel consumption decreased by 50.24% and 5.48%, respectively. In road transportation, the shift frequency and fuel consumption decreased by 13.89% and 15.80%, respectively. Both DP global optimization and BPNN real-time control effectively tracked the vehicle speed to meet the demand of the power after plowing simulation. Furthermore, the BPNN was maintained on the minimum gear at the low speed, indicating the more frequent shifting. The higher gears of the DP were utilized for the significant fuel savings during high-speed segments. While the BPNN instantaneously maintained the higher gears at the low speed, this resulted in slightly higher fuel consumption. Both DP and BPNN were also accurately track the speed under road transportation. The available 24 gears were more fully utilized in the DP with fewer shifts during high-speed sections. Although the driving resistance was relatively low, the engine was placed outside the most economical zone, in order to secure the more efficient points of the DP. The high-gear configuration was optimized in the multiple high-speed segments. The DP achieved significantly lower fuel consumption than the BPNN. Collectively, the gear selection and shift smoothness of the DP were optimized to attain the lower fuel consumption using global information. While the DP real-time strategy shared much higher-speed gear utilization and fuel economy, compared with the BPNN. There were only a few differences in the fuel consumption under working conditions. There was some increase in the shift frequency for the BPNN. The BPNN strategy was verified to meet the real-time requirements with acceptable fuel economy. The global gear optimization was also integrated with the real-time adaptive decision-making. Conventional multi-condition strategies on manual calibration were reduced by the adaptive fuzzy control. The finding can also provide a valuable technical pathway for the intelligent control of the mechanical transmission in modern agriculture.