基于滚动时域优化的果园多机协同动态调度方法

Dynamic Scheduling Method for Orchard Multi-robot Coordination Based on Receding Horizon Optimization

摘要:

针对无存储式采摘-运输果园机器人协同作业中时效与转运成本冲突问题，本文提出一种基于阈值-运力状态解析耦合理论的事件驱动滚动时域优化(RHO)协同调度方法。建立以接替触发阈值为动态决策变量的多目标混合整数非线性规划模型，推导给定候选运输机器人状态下单次接替事件阈值的闭式解析解;通过将传统的阈值网格枚举求解替换为解析计算，显著降低在线决策的评估规模;构建滚动预测-解析计算-反馈校正的事件驱动在线调度框架，结合综合代价评估与贪心指派实现快速决策。基于真实果园参数的仿真结果表明，固定阈值策略对参数取值高度敏感，81组配置中非支配解仅占24.7%;所提方法与网格搜索RHO相比，停机时间偏差不大于1.60%、行驶距离偏差不大于0.24%;与遗传算法基线相比，解质量偏差小于1.3%，单次决策耗时由秒级降至毫秒级，计算效率提升约3个数量级。多机协同实物试验验证表明，单采摘机器人场景下，动态阈值策略较固定阈值策略将累计采摘停机时间降低82.4%;完整多机协同场景的20次接替决策中，实物与仿真指派一致率达85%，解析阈值实测均值与仿真值偏差不超过0.02，表明本文方法在物理平台上具有较好的工程实现可行性。通过调节单一权重系数，可实现时效优先与成本优先策略之间的灵活切换;结合在线装载速率估计与闭环反馈机制，在本文试验条件下本文方法表现出对作业扰动的适应能力，为结构化果园采摘运输协同调度提供了一种可行思路。

Abstract:

Aiming to address the timeliness-cost conflict in non-storage collaborative harvesting-transport operations in orchards, an event-driven receding horizon optimization (RHO) scheduling method based on analytical threshold-fleet-state coupling was proposed. A multi-objective mixed-integer nonlinear programming model was established by using the replacement trigger threshold as a dynamic decision variable. A closed-form analytical expression for the threshold of single replacement events under given candidate transport robot states was derived. The online evaluation scale was significantly reduced by transforming the original threshold grid enumeration into analytical calculations. Thus, an event-driven framework integrating rolling prediction, analytical calculation, and feedback correction was constructed, achieving rapid decision-making via comprehensive cost evaluation and greedy assignment. Simulations using real orchard parameters showed that the fixed-threshold strategy was highly sensitive, with non-dominated solutions yielded in only 24.7% of 81 configurations. Compared with grid-search RHO, deviations in idle time and travel distance were no more than 1.60% and 0.24%, respectively. Against a genetic algorithm baseline, solution-quality deviation remained within 1.3%, while single-decision computation time was compressed from seconds to milliseconds, improving efficiency by approximately three orders of magnitude. Physical multi-robot experiments verified that cumulative harvesting idle time was reduced by 82.4% by using the dynamic-threshold strategy. Across 20 decisions, a physical-simulation assignment consistency of 85% was reached, and measured analytical threshold deviations were under 0.02, demonstrating excellent engineering feasibility. Flexible switching between timeliness- and cost-priority strategies was enabled by adjusting a single weight. Integrated with online loading-rate estimation and closed-loop feedback, the proposed method exhibited good adaptability to operational disturbances, providing a feasible cooperative scheduling approach for structured orchards.