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

The full power shift gearbox has become the key technology to improve the operation efficiency of heavy tractors. However, the traditional rule-based shift strategy has the inherent defect of insufficient global optimization ability under multi-gear overlap conditions. In heavy tractors, frequent shifting and high fuel consumption caused by the overlap of adjacent gear speed ranges seriously restrict the intelligent development of agricultural equipment. This method relies on expert experience and cannot achieve the optimal working conditions. There are still research gaps in real-time optimization and global optimal collaborative optimization. This study proposed a compound shift strategy integrating dynamic programming (DP) for global optimization and BP neural network (BPNN) for real-time prediction. A working condition classification mechanism was established, dividing shift control into two distinct modes based on operational characteristics: a field mode (utilizing throttle, speed, and slip rate) and a road mode (utilizing throttle and speed). A dynamic programming bi-level optimization model was constructed. Prioritizing fuel economy as the core objective, the model incorporated a shift frequency penalty function. Offline optimization was performed using the real-vehicle plow load spectrum and road transportation conditions modeled on suburban cycle driving profiles. A double-hidden-layer BP neural network gear prediction controller was trained and developed using the DP offline optimization dataset. Validation was conducted via an AMESim-MATLAB/Simulink co-simulation platform. Results demonstrated that the shift strategy met the power demands of the working conditions under the DP offline solution. For plowing conditions, shift frequency decreased by 50.24% and fuel consumption by 5.48%. For road transportation conditions, shift frequency decreased by 13.89% and fuel consumption by 15.80%. In plowing simulations, both DP global optimization and BPNN real-time control effectively tracked vehicle speed and satisfied power demand. However, BPNN maintained a higher minimum gear at low speed and exhibited more frequent shifting. DP achieved significant fuel savings by utilizing higher gears during high-speed segments, while BPNN gained an instantaneous advantage by maintaining higher gears at low speed, albeit resulting in slightly higher total fuel consumption. Under road transportation conditions, both methods also tracked speed accurately. DP utilized the available 24 gears more fully during high-speed sections with fewer shifts. Although the driving resistance was low, placing the engine outside its most economical zone, DP secured more efficient operating points. By optimizing high-gear usage in multiple high-speed segments, DP achieved significantly lower final fuel consumption than BPNN. Collectively, the results indicated that DP attained lower fuel consumption by optimizing gear selection and shift smoothness using global information. While the BPNN real-time strategy showed a gap in high-speed gear utilization and fuel economy compared to DP, the fuel consumption difference between the two strategies across both working conditions was limited, and the increase in shift frequency for BPNN was manageable. This verified that the BPNN strategy met real-time requirements while maintaining acceptable economy. This study successfully integrated global gear optimization with real-time adaptive decision-making. It overcame the limitations of traditional multi-condition strategies reliant on manual calibration rules and the subjective experience dependence of adaptive fuzzy control, providing a valuable technical pathway for intelligent control in agricultural mechanical transmission systems.