Research On Manufacturing Process Modeling Method Based On Multi-Agent Reinforcement Learning

With the development and application of a new generation of information technology(artificial intelligence,big data,etc.),a new round of industrial revolution is emerging.Governments around the world have introduced policies to promote the transformation and upgrading of domestic manufacturing enterprises,such as Germany's "Industry 4.0." In the context of “Industry 4.0”,China has introduced the “Made in China 2025” plan to promote the transformation of Chinese manufacturing companies into smart manufacturing.Aiming at the problem that manufacturing resources are difficult to share and the manufacturing process is difficult to coordinate,this paper takes process industry manufacturing process as an example,introduces multi-agent technology and reinforcement learning algorithm into process industry manufacturing process,and builds on process industry manufacturing process intelligent control model.Focus on solving key issues such as dynamic task assignment in multi-agent collaboration and collaboration in actual production,and promote the intelligent development of manufacturing enterprises.A multi-agent distributed hierarchical intelligent control model for manufacturing system integrating multiple production units is constructed based on MAS technology for the problem that the process industry production process is difficult to coordinate.The model organically combines multiple intelligent agent modules and physical entities to form an intelligent control system with certain functions.The model consists of system agent,workshop control agent and on-site agent.This paper focuses on multiagent collaboration and dynamic task assignment.Aiming at the multi-agent collaboration problem in process industry production process,based on the hierarchical distributed collaborative control model constructed in this paper,the multi-agent depth deterministic strategy gradient(MADDPG)algorithm is introduced.In MADDPG,we introduce distributed asynchronous priority batch processing.The idea is to build a multi-agent algorithm based on improved MADDPG.In order to verify the effectiveness of the algorithm,we use OpenAI's multiagent environment to simulate the MADDPG algorithm based on the pursuit-escape in the experimental environment.The simulation results show that the Agent trains through the improved MADDPG algorithm.In order to obtain more rewards,multiple agents can cooperate with each other to achieve the round-up of the target agent,and the improved algorithm's average round reward is higher than the previous algorithm.Aiming at the problem of production task assignment of multi-agent system in industry production process,based on the analysis of process industry production task scheduling problem,a genetic algorithm based on reinforcement learning mechanism(QGA)is proposed for multi-agent production task scheduling.Taking a production line to process a production order as an example,the effectiveness of the algorithm is verified by comparing the genetic algorithm task assignment strategy with the QGA algorithm based task assignment strategy.