Research On Scheduling Algorithm Of Melt-cast And Heat Treatment Workshop

In order to realize the leap from a big manufacturing country to a strong manufacturing country,China has proposed the "Made in China 2025" ten-year action plan.Metal manufacturing is an important part of manufacturing.In recent years,the metal manufacturing industry has gradually shifted from the past high-increment and large-scale production methods to flexible production methods,paying more attention to rapid adaptation to the diverse needs of the market.Therefore,how to effectively respond to the diverse needs of the market and how to further improve production efficiency becomes a difficult problem faced by manufacturing companies.This thesis studies the melt-cast and heat treatment workshops scheduling problem which is common in metal manufacturing.Melt-cast and heat treatment are both important processes in the processing of metal profiles.In actual production,they are faced with problems such as excessive order size,a lot of production constraints,and difficulty in formulating production plans.It causes that the scheduling results are difficult to meet the needs of current actual production activities.Therefore,it is necessary to carry out scientific and reasonable scheduling of workshop production,thus shortening the total order completion time,and improving workshop production efficiency.Due to the lack of mature scheduling models and algorithm for solving complex scheduling problems,this thesis seeks new attempts in solving algorithms,and proposes methods based on deep reinforcement learning to solve the scheduling problems of melt-cast and heat treatment workshops.In this study,in order to make the scheduling problem of the melt-cast heat treatment workshop closer to the actual production,machine preparation time and batch processing constraint are considered,and the goal is to minimize the maximum completion time.The main work of this thesis is as follows:Firstly,the scheduling problem of melt-casting and heat treatment workshop is studied,the constraint conditions and objective function of the problem are clarified,the mathematical model of the problem is established,and the coding and decoding method of the problem solution is designed.Then an end-to-end algorithm based and an improved local search algorithm are respectively proposed to solve the melt-casting and heat treatment workshop scheduling problem.The end-to-end algorithm uses a network model based on an encoder-decoder structure and uses REINFORCE with Baseline algorithm to train the model,and the trained end-to-end model can input the problem instance information and directly output the solution of the problem,which has the advantage of fast solving speed.The improved local search algorithm uses a network model based on the self-attention mechanism,and trains the model through the improved Actor-Critic algorithm.The trained iterative search model can guide the selection of solutions during the iterative search process to improve global optimization capabilities of the algorithm.Finally,the effectiveness and feasibility of the designed algorithms are verified through experiments,and they are compared with common shop scheduling algorithms.Experiments show that both algorithms have strong generalization ability.Although the end-to-end algorithm does not exceed the traditional meta-heuristic algorithm,it can give a good result in a very short time.The improved local search algorithm can obtain better results than the traditional meta-heuristic algorithm in a reasonable running time.In addition,the two algorithms can be combined to further improve the quality of solving the problem.