文章摘要
改进差分进化算法及核动力船舶管路优化布置研究
Research on improved differential evolution algorithm and optimization layout of nuclear power ship pipeline
投稿时间：2024-03-06  修订日期：2024-04-03
DOI：
中文关键词: 核动力船舶  管路路径  布局优化  深度强化学习  差分进化  核动力船舶管路Agent模型  
英文关键词: Nuclear-powered ship  pipeline path  layout optimization  deep reinforcement learning  differential evolution  Agent model of nuclear-powered ship pipeline  
基金项目:国家重点实验室专项基金自主项目(S18315);中核绿色建造技术与装备重点实验室开发基金 （CNNC-STGCL-KFKT-2022-001）
作者单位部门
杨明君 大连理工大学 船舶工程学院
林焰* 大连理工大学船舶工程学院 
摘要点击次数: 20
全文下载次数: 0
中文摘要:
      摘要：【目的】为了将进化算法过程中种群积累的经验充分利用，且能在短时间内提供更多的核动力船舶管路布局方案，本文提出了一种将深度强化学习与差分进化相结合的算法（DRL-DE）。【过程和方法】该方法基于无网格空间，提出核动力船舶管路Agent模型，将管路Agent中的code属性作为连续状态空间，差分进化算法中的变异策略为离散动作空间，构造了奖励和差异率函数的趋势曲线，利用深度强化学习中的神经网络自主选择最佳策略，通过奖励函数充分利用种群的历史经验信息来提高搜索效率，并引入了外部档案优化机制来确定分支管路起点的优化策略。【结果】通过实际的核动力船舶工程案例验证，DRL-DE(DQN)可将算例1、算例2、算例3的管路实际长度分别减少15.1%、33.8%、7.3%， 【结论】针对分支管路提出的外部档案制有效减少分支管路的弯头点，DRL-DE(DQN)相比于DRL-DE(PPO)在核动力船舶管路布局中展现出了更好的性能和效果，并在核动力船舶管路布局方案生成中提供了丰富的选择与机会。
英文摘要:
      [AIMS] In order to make full use of the accumulated experience of the population in the process of evolutionary algorithm and provide more nuclear-powered ship pipeline layout scheme in a short time, this paper proposes an algorithm combining deep reinforcement learning with differential evolution (DRL-DE). [Process and method] This method is based on no grid space, proposed the Agent model of the nuclear-powered ship pipeline , used the code attribute in the line Agent as a continuous state space ,the variation strategy in the difference algorithm is the discrete action space, using the neural network in the deep reinforcement learning to improve the search efficiency through the reward function, and introduces the external file optimization mechanism to determine the starting point of branching pipeline. [Results] Through the actual nuclear power ship engineering case verification, DRL-DE (DQN) can reduce the actual length of examples 1,2 pipeline by 15.1%, 33.8% respectively,[Conclusion]For the external archives proposed for branch pipeline effectively reduces the elbow points of branch pipeline, DRL-DE (DQN) shows better performance and effect than DRL-DE (PPO) in nuclear power ship pipeline layout, and provides a rich choice and opportunities in the generation of nuclear power ship pipeline layout scheme.
View Fulltext   查看/发表评论  下载PDF阅读器
关闭