A machine learning approach for flow shop scheduling problems with alternative resources, sequence-dependent setup times, and blocking
In proposing a machine learning approach for a flow shop scheduling problem with alternative resources, sequence-dependent setup times, and blocking, this paper seeks to generate a tree-based priority rule in terms of a well-performing decision tree (DT) for dispatching jobs. Furthermore, generating...
Lưu vào:
| Tác giả chính: | |
|---|---|
| Đồng tác giả: | |
| Định dạng: | BB |
| Ngôn ngữ: | en_US |
| Thông tin xuất bản: |
Springer Nature
2020
|
| Chủ đề: | |
| Truy cập trực tuyến: | https://doi.org/10.1007/s00291-019-00567-8 http://tailieuso.tlu.edu.vn/handle/DHTL/9422 |
| Từ khóa: |
Thêm từ khóa bạn đọc
Không có từ khóa, Hãy là người đầu tiên gắn từ khóa cho biểu ghi này!
|
| Tóm tắt: | In proposing a machine learning approach for a flow shop scheduling problem with alternative resources, sequence-dependent setup times, and blocking, this paper seeks to generate a tree-based priority rule in terms of a well-performing decision tree (DT) for dispatching jobs. Furthermore, generating a generic DT and RF that yields competitive results for instance scenarios that structurally differ from the training instances was another goal of our research. The proposed DT relies on high quality solutions, obtained using a constraint programming (CP) formulation. Novel aspects include a unified representation of job sequencing and machine assignment decisions, as well as the generation of random forests (RF) to counteract overfitting behaviour. To show the performance of the proposed approaches, different instance scenarios for two objectives (makespan and total tardiness minimisation) were implemented, based on randomised problem data. The background of this approach is a real-world physical system of an industrial partner that represents a typical shop floor for many production
processes, such as furniture and window construction. The results of a comparison of the DT and RF approach with two priority dispatching rules, the original CP solutions and tight lower bounds retrieved from a strengthened mixed-integer programming (MIP) formulation show that the proposed machine learning approach performs well
in most instance sets for the makespan objective and in all sets for the total tardiness objective. |
|---|