基于DDPG算法的桥式吊车自适应防摆控制  

Adaptive Anti-sway Control of Bridge Cranes Based on DDPG Algorithm

作　　者：辛增淼 万思成 高永锹 王天雷[2] 郝晓曦[1] 邱光繁 XIN Zengmiao;WAN Sicheng;GAO Yongqiao;WANG Tianlei;HAO Xiaoxi;QIU Guangfan(School of Mechanical and Automation Engineering,Wuyi University,Jiangmen 529020,China;School of Electronic and Information Engineering,Wuyi University,Jiangmen 529020,China;Jiangmen Mengde Electric Co Ltd,Jiangmen 529040,China)

机构地区：[1]五邑大学机械与自动化工程学院,广东江门529020 [2]五邑大学电子与信息工程学院,广东江门529020 [3]江门市蒙德电气股份有限公司,广东江门529040

出　　处：《五邑大学学报(自然科学版)》2025年第1期24-30,共7页Journal of Wuyi University(Natural Science Edition)

基　　金：江门市科技计划项目(2022JC01021);江门市科技特派员科研合作项目(2023760300180008278);江门市扶持科技发展资金项目(江科(2023)117号2320002000601)。

摘　　要：为解决桥式吊车控制中存在的台车定位速度慢和负载摆动大等问题,提出一种基于深度确定性策略梯度(DDPG)的控制策略.将桥式吊车设为强化学习的智能体;通过设定状态量及其误差作为智能体观测目标,设计合适的惩罚与奖励函数;智能体通过对控制系统的实时运行情况进行响应来生成控制动作.仿真结果表明,与传统控制算法相比,基于DDPG强化学习的算法在定位速度和摆角抑制方面的表现都优于传统方法,展现出较传统控制算法更高的性能和应用潜力.To overcome the challenges of slow trolley positioning and pronounced load swinging in overhead crane control,a control strategy based on Deep Deterministic Policy Gradient(DDPG)is proposed.Initially,the overhead crane is conceptualized as an agent within the framework of reinforcement learning.Subsequently,by establishing state variables and their associated errors as the observation targets of the agent,appropriate penalty and reward functions are meticulously designed.Furthermore,the agent generates control actions by responding dynamically to the real-time operational conditions of the control system.Simulation results reveal that,in comparison with traditional control algorithms,the algorithm based on DDPG reinforcement learning exhibits superior performance in terms of positioning speed and swing angle suppression.

关 键 词：桥式吊车 强化学习 深度确定性策略梯度 自适应控制 防摆控制 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]