Research On Dynamic Adaptive Scheduling Optimization Algorithm Of Hybrid Flow Workshop Based On DDQN

Hybrid Flow Shop（HFS: Hybrid Flow Shop）is a common and mainstream mass customization production（MCP: Mass Customization Production）organizational model.Solving the scheduling problem of Hybrid Flow Shop has always been a research hotspot in the field of intelligent manufacturing.At present,most of the problems are solved by traditional mathematical programming and heuristic algorithms.However,because HFS has the characteristics of large scale,many resource constraints,and strong dynamics,on the one hand,it will cause difficulty in solving due to the disaster of dimensionality.This results in insufficient timeliness of system response scheduling.This thesis intends to use the deep reinforcement learning method to solve such problems,first learn by mining the value of the historical production data of the workshop,and then apply it to the matching scheduling strategy,which can not only achieve the effect of solving the dimension disaster,but also in the process of matching scheduling,the response time is extremely short,which can effectively solve the problem of insufficient timeliness.The overall research idea of this thesis is to aim at improving processing efficiency and enhancing adaptability.On the basis of using MPN（Manufacturing Petri Net）modeling and data state modeling for the workshop system,a new method based on DDQN（Double Deep Q-Net）is proposed.Learning dynamic adaptive scheduling optimization method of hybrid flow workshop is studied.The specific research work of this thesis is as follows:（1）First,the MPN model is used for modeling according to the hybrid flow workshop actually produced by the enterprise.It abstracts the key points in the actual hybrid flow workshop,restores the workshop production logic and mechanism,compresses the solution space,and summarizes the processing constraints of the workpiece.Make system modeling both realistic and accurate in description and intuitive and convenient when conducting research.（2）Secondly,according to the historical data of workshop operations,accurately construct the data state based on the Markov state process.Aiming at the problem of matching data state modeling,the multi-source heterogeneous state information and behavior event information such as people,machines,materials,processes,and environments obtained by the workshop manufacturing node acquisition system are linked to form a multi-dimensional inline The state of the data to reflect the production logic in the manufacturing process is used as the model training data set.（3）Then,based on the DDQN algorithm and combined with the workshop scheduling scenario,the training steps are designed,and the deep knowledge network DQN for guiding scheduling is generated through model training.Aiming at the problem that the traditional Q-learning algorithm is prone to overestimation,this thesis uses a CNN-based two-layer deep Q network algorithm to train the model,making full use of the value contained in the data generated in workshop production,and the deep knowledge DQN obtained from training can guide Workshop intelligent production.（4）Finally,the experiment is carried out in combination with the production case of CCL,and the conclusion is drawn by analyzing the experimental results.By carrying out experiments with examples,on the one hand,it is to verify the guidance effect of the deep knowledge network DQN,and to analyze the effectiveness of the deep knowledge network DQN;on the other hand,it analyzes the advantages of the dynamic adaptive scheduling system based on DDQN through comparative experiments with other traditional algorithms and heuristic algorithms.The experimental data show that the DDQN algorithm can effectively solve the HFS scheduling problem.Compared with the ant colony algorithm and the genetic algorithm,the optimization time is shorter and the optimization effect is more stable.In terms of improving the timeliness of the system,compared with the reinforcement learning algorithm based on heuristic rules,it has a faster response speed and can improve the adaptability to the dynamic environment.