Flow shop scheduling method based on deep reinforcement learning

Abstract

The invention discloses a flow shop scheduling method based on deep reinforcement learning. Each workpiece is used as a node, and processing time of a working procedure is used as node information to be aggregated to obtain embedded representation. A strategy network is fit by using a pointer network, the maximum completion time is used as an award, the strategy network is trained and parameters are stored. In the actual problem, the embedded representation of each workpiece is used as the input of the strategy network, and the workpiece with the highest probability is selected in sequence until all the workpieces are selected to obtain a complete scheduling sequence. According to the flow shop scheduling method, a near-optimal solution can be obtained on a small-scale problem, a better solution superior to a heuristic algorithm and a genetic algorithm can be obtained on a large-scale problem, and the flow shop scheduling method is expanded to the problems of different machine numbers and job numbers; the limitation that the model needs to be trained again on different problem scales and problem data is broken through, and the universality is wider.

Description

technical field [0001] The invention relates to the technical field of flow shop scheduling, in particular to a flow shop scheduling method based on deep reinforcement learning. Background technique [0002] The flow shop scheduling problem is one of the classic combinatorial optimization problems, which is difficult to find the optimal solution and has high computational complexity. In order to increase the productivity of the system, effective and efficient manufacturing and scheduling technology solutions must be used. Many researchers have proposed various algorithms for this problem. The exact algorithm can get the optimal solution, but the solution time is often too long, so it is not suitable for large-scale problems; the NEH heuristic algorithm is a kind of flow shop scheduling algorithm that is widely used at present, and can be used to solve the problem with the minimum processing time (makespan) optimization problem. The heuristic algorithm is difficult to guar...

AI Technical Summary

Problems solved by technology

Using a multi-layer perceptron to fit the policy network in reinforcement learning requires manual design of state features and reward functions, and directly using indicators such as the number of machines and jobs as state features will cause large changes in the input of the network, and it is difficult to expand to different numbers of machines. In terms of the number of tasks and the number of jobs, the model may need to be retrained on different problem scales and problem data, and there are still limitations in versatility

Images