TLBO with variable weights applied to shop scheduling problems

CAAI Transactions on Intelligence Technology

  ^{1    2} 

[1] Computer Science Department, Federal University of Ceará, Rua Campus do Pici, Sn, Fortaleza 60440-554, Brazil;Electronics Division, Institute of Aeronautics and Space, Praça Marechal Eduardo Gomes, 50, São José dos Campos 12228-904, Brazil;

关键词: learning (artificial intelligence);    teaching;    flow shop scheduling;    search problems;    statistical testing;    simulated annealing;    job shop scheduling;    scheduling;    optimisation;    assigning zero;    variable weights;    teaching–learning-based optimisation algorithm;    population-based;    teaching–learning process;    variant version;    different weights;    student phase;    higher weights;    assign weights;    flow-shop;    job-shop scheduling problems;    original TLBO algorithm;    solution quality;    original version;    fixed weight;   

DOI  :  10.1049/trit.2018.1089

来源: publisher

【 摘 要 】

The teaching–learning-based optimisation (TLBO) algorithm is a population-based metaheuristic inspired on the teaching–learning process observed in a classroom. It has been successfully used in a wide range of applications. In this study, the authors present a variant version of TLBO. In the proposed version, different weights are assigned to students during the student phase, with higher weights being assigned to students with better solutions. Three different approaches to assign weights are investigated. Numerical experiments with benchmark instances of the flow-shop and the job-shop scheduling problems are carried out to investigate the performance of the proposed approaches. They compare the proposed approaches with the original TLBO algorithm and with two variants of TLBOs proposed in the literature in terms of solution quality, convergence speed and simulation time. The results obtained by the application of a Friedman statistical test showed that the proposed approaches outperformed the original version of TLBO in terms of convergence, with no significant losses in the average makespan. The additional simulation time required by the proposed approaches is small. The best performance was achieved with the approach of assigning a fixed weight to half the students with the best solutions and assigning zero to other students.

【 授权许可】

CC BY