为有效调度进出港船舶以解决港口拥堵问题,提出一种基于强化学习的自学习遗传算法( GA-RL),在 GA- RL 中,以遗传算法为基本优化模型,利用 Q-learning 算法自适应调整交叉和变异参数来提高算法的搜索能力;同时,构建可动态调参的马尔科夫决策过程(MDP)模型,在MDP 模型中,为全面评估种群性能,提出基于种群适应度函数的状态集,并设计了有效减少目标值的奖励机制;最后,以黄骅港综合港区为例,选取不同组算例进行仿真实验,结果验证了模型和算法的有效性,该方法可显著减少船舶在港等待时间并提升港口通航效率。

s to solve the p roblem of port congestion , a self-learning genetic algorithm based on reinforce ment learning (GA- RL ) was proposed. In GA-RL, taking the genetic algorithm as the b asi c optimizat ion m odel , and Q-learn ing algo-rithm was use d to ada ptive ly adju st the c ros sove r and mutat ion par ameter s to improve the search ability of the algo rithm. At the same time, a dynamically adjustable Markov decision pr ocess  (M DP) model was  construct ed.  In the MD P model, in ord er to c omp rehens ivel y e valuate the popula tion perf orm- ance, a st ate se t based on the pop ulat ion fitness function was propo sed, a nd a reward me chanism for effe ctiv ely reducing the targ et va lue was desig ned. Fi nal ly, taking the comp rehens ive port area of H uanghua p ort as an exam ple, different group s of exam ples w ere selected for s imu latio n exper imen ts. The re- sults verif y the effect iveness of the m odel an d alg orit hm,  whi ch can si gnif icantly red uc e the waiting time of ships in port and improve the efficiency of port navigatio n.