Predictive-feedforward energy management for a hybrid agricultural machine operating in mountainous orchards

Hybrid agricultural machines in mountainous orchards operate under low-speed, high-load conditions with coupled gradient and time-varying rolling-resistance disturbances. Their power demand is therefore nonstationary and may change abruptly. Rule-based and instantaneous strategies provide limited preview, whereas model predictive control (MPC) can transmit prediction errors directly to actuator commands and imposes substantial online computation. To address these limitations, a forward model was developed for a power-split hybrid agricultural machine comprising an engine, two electric machines, a traction battery, and a planetary gear set. A predictive-feedforward, multi-objective energy-management strategy was proposed. Short-horizon speed and gradient predictions are normalized, weighted, and saturated to form a bounded load-risk indicator. Rather than entering low-level torque calculation directly, this indicator modifies the dynamic state-of-charge (SOC) safety boundary, power-source mode, and transmission-ratio switching threshold, thereby softly decoupling prediction and execution. Engine high-efficiency operation, two transmission ratios, and actuator constraints are then coordinated for computationally efficient online power allocation. With offline dynamic programming (DP) as a common benchmark, the strategy achieved 93.29-95.59% of the DP performance over cycles C1-C3, exceeding MPC by 1.41-3.19 percentage points and adaptive equivalent-consumption minimization by 3.62-4.99 percentage points. Mean execution time was 0.128-0.191 ms per step, 93.54% lower than that of MPC, while terminal SOC remained near 0.600. No SOC, battery-current, or power-source torque constraint was violated under prediction noise or abrupt equivalent rolling-resistance disturbances. The conclusions apply to maintained, firm orchard lanes with low track slip; soft-soil operation involving high traction, severe sinkage, or substantial slip requires further validation. Bounded load-risk feedforward thus retained short-horizon preview while limiting direct error propagation, providing a practical real-time approach for hybrid agricultural machines with resource-constrained embedded controllers.