菠萝果园多田块全覆盖作业路径规划方法研究

Multi-field Complete Coverage Operation Path Planning Method of Pineapple Orchard

摘要:

为了优化菠萝老苗清理农机在菠萝果园中的作业路线，减少非工作区域消耗，本研究设计了混合改进的遗传粒子群优化(Genetic particle swarm optimization, GPSO)算法和Q-Learning(QL)算法全覆盖作业路径规划方法。利用无人机航拍获取目标菠萝果园田块图并转化为轮廓坐标地图;对各田块从0°到355°每间隔5°遍历覆盖路径寻找最优行进角度;针对传统GPSO算法收敛速度慢、易陷入局部最优等问题，提出改进遗传粒子群优化(Improved hybrid genetic particle swarm optimization, IHGPSO)算法获得菠萝果园多田块作业最佳遍历顺序;针对传统QL算法探索效率低、收敛慢的问题，提出改进的QL算法(Order initialization Q-Learning, OI-QL)完成田块之间的路径连接。仿真结果表明，在子区点数为30求解最佳遍历顺序时，IHGPSO算法适应度小于GPSO算法。OI-QL算法相比QL算法路径长度缩短约28.1%，平均收敛迭代次数减少约37.3%，平均收敛时间减少约25.7%，表明本文方法能有效地实现菠萝果园多田块全覆盖作业路径规划。

Abstract:

In order to optimize the operation routes of agricultural machinery for cleaning old pineapple seedlings in pineapple orchards and reduce consumption in non-working areas, a path planning method integrating improved genetic particle swarm optimization (GPSO) and Q-Learning(QL) algorithms was proposed to optimize the complete coverage operation path of unmanned machines in multi-field pineapple orchards. Images of the target farmland were obtained by unmanned aerial vehicle and converted into a coordinate map. Each field was traversed at 5° intervals from 0° to 355° to find the optimal travel angle. To address the inherent problem such as slow convergence and getting stuck into a local optimum in traditional GPSO algorithms, an improved GPSO algorithm, namely IHGPSO, was proposed to improve the generation and selection method of initial populations, a multi-objective weighted fitness function was designed, and a dynamic adjustment probability was added to particle exchange mechanism, thereby obtaining the optimal traversal order for multi-field complete coverage operation of pineapple orchard. To address the inherent problem in original QL algorithm, such as low exploration efficiency and slow convergence, an improved QL algorithm, namely order initialization Q-Learning(OI-QL) was proposed, which improved initialization of Q-table, proposed learning rate, and designed a reward function, thereby connecting multi-field complete coverage operation path of pineapple orchard. Simulation results showed that when the sub-area point groups contained 30 points, the IHGPSO algorithm outperformed the GPSO algorithm in solving the problem of the optimal traversal order. The average path length planned by OI-QL algorithm was 28.1% shorter than that planned by QL algorithm, and the average number of convergence iterations and the average convergence time by using OI-QL algorithm was 37.3% and 25.7% less than those by using QL algorithm, respectively. These results indicated that the method can effectively complete the complete coverage operation path planning of pineapple orchard.