Application Of Reinforcement Learning In Rebar Arrangement Of Complex Nodes

In recent years,reinforcement learning has received many attentions in both academia and industry.It has strong self-learning and adaptive capability in a complex environment,which makes reinforcement learning successfully been applied in many real-world applications.In the construction industry,the problem of arrangement of rebar with complex nodes is one of the pain and challenging issues of actual production.Practitioners in the construction industry struggle to manually adjust the shape of the rebar position to ensure that the rebar does not collide with each other or obstacles.This traditional approach is labor-intensive,inefficient,and error-prone.It is thus inevitable to apply more intelligent artificial intelligence technology as the solution of this problem,so as to realize the function of automatic and intelligent rebar arrangement.In order to achieve the above goals,this thesis combines the Q-learning method and the actual production situation,and develops a reinforcement learning method for rebar arrangement in the complex node reinforcement arrangement.The achieved results are as follows:(1)A new modeling method is proposed.The model considers the similarity of the shape of the generated steel bar and the trajectory of the agent,and transforms the problem of complex node reinforcement arrangement into a multi-agent path planning problem.(2)A multi-agent path planning algorithm based on reinforcement learning is proposed.The algorithm uses a classic Q-learning theory framework.Through careful design of the state space,action space and reward mechanism of the agent,the algorithm achieves the optimal obstacle avoidance path of each agent in a reasonable time.The analysis of the simulation experiment results shows that the algorithm is feasible and efficient.(3)A method of knowledge transfer is proposed,which improves the efficiency off the proposed reinforcement learning to solve the problem of complex node reinforcement arrangement.This method considers the reinforcement learning process of a single agent as a task.By defining the characteristics of tasks and measuring the similarity between tasks,this method conducts knowledge transfer between agents.Simulation experiments based on real engineering data verify the effectiveness of this method in improving the efficiency of problem solving.In summary,this thesis uses reinforcement learning to explore the complex node reinforcement arrangement problem in the construction industry,and accelerates the problem solving efficiency through knowledge transfer.This thesis attempts to provide some ideas for the landing of artificial intelligence technology in the construction industry.