Abstract: To address the challenge of rising scheduling costs for agricultural machinery caused by dynamic operational time windows (e.g., due to weather and crop maturity), an improved particle swarm optimization algorithm for dynamic time windows in agricultural machinery (Set-Based Particle Swarm Optimization with Neighboring Nodes, S-PSO-N) is developed. The objective is to dynamically adjust scheduling schemes to minimize total costs within a given time window. To rapidly generate high-quality solutions in dynamic environments, a historical learning strategy is introduced to guide particle swarm initialization by analyzing historical optimal solutions for accelerating convergence. Furthermore, a velocity updating mechanism based on the position information of neighboring nodes is designed to improve the particles' search behavior, effectively balancing global exploration and local exploitation. Moreover, an adaptive parameter adjustment strategy iss constructed by integrating Q-learning module to enhance the algorithm's environmental adaptability. Simulation experiments were conducted under two time windows change distribution scenarios: random and mixed distributions, and compared with an ant colony algorithm and a genetic algorithm. The S-PSO-N algorithm achieved a 2.9% and 5.2% reduction in average total cost compared to the best-performing baseline algorithm (DVRPTW-ACS) for time windows change instances with random distributions (R) and mixed distributions (RC), respectively.

摘要： 为解决农业生产中因异常天气、作物成熟度等因素所导致的农田作业时间窗的动态变化问题，提出基于改进集合粒子群算法的可变时间窗农机调度算法(Set-Based Particle Swarm Optimization with Neighboring Nodes,S-PSO-N)。该算法旨在动态调整调度方案，以最小化农机调度数量与农机运行路径成本，同时在时间窗变化环境中快速生成高质量解。首先，算法引入历史学习策略，通过分析历史最优解以指导新环境下的粒子群初始化以加速收敛。其次，设计了基于节点邻居位置信息的速度更新机制以改善粒子的搜索行为，有效平衡算法的全局探索与局部搜索能力。此外，为提升算法的环境适应性，融合Q-learning模块构建了一种自适应参数调整策略。仿真实验在随机与混合两种时间窗变更分布场景下进行，并与最新蚁群算法、遗传算法对比。实验结果表明，在处理随机分布(R类)和混合分布(RC类)的时间窗变更实例时，S-PSO-N算法的平均总成本相较于表现最优的动态蚁群（DVRPTW-ACS）算法分别降低了2.9%和5.2%。