Evolutionary Constrained Multiobjective Optimization: Test Suite Construction and Performance Comparisons

By Zhongwei Ma, Yong Wang. IEEE Transactions on Evolutionary Computation, 2019

377 citations recorded by the source corpus.

For solving constrained multiobjective optimization problems (CMOPs), many algorithms have been proposed in the evolutionary computation research community for the past two decades. Generally, the effectiveness of an algorithm for CMOPs is evaluated by artificial test problems. However, after a brief review of current artificial test problems, we have found that they are not well-designed and fail to reflect the characteristics of real-world applications (e.g., small feasibility ratio). Thus, in this paper, we first propose a new constraint construction method to facilitate the systematic design of test problems. Then, on the basis of this method, we design a new test suite consisting of 14 instances, which covers diverse characteristics extracted from real-world CMOPs and can be divided into four types. Considering that the comprehensive performance comparisons among the constraint-handling techniques (CHTs) remain scarce, we choose several representative CHTs and compare their performance on our test suite. The performance comparisons identify the strengths and weaknesses of different CHTs on different types of CMOPs and provide guidelines on how to select/design a CHT in a specific scenario.

Subfields: Artificial Intelligence, Computational Theory and Mathematics

View Paper: https://www.semanticscholar.org/paper/a3865b30adf37516ac86cf4ea7bb7e06927cc769

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Yaochu Jin
2. Xing-yi Zhang
3. Ye Tian
4. Ran Cheng
5. Kay Chen Tan
6. Yansen Su
7. B. Qu
8. Jing J. Liang

In Computational Theory and Mathematics

1. Yaochu Jin
2. Xing-yi Zhang
3. Ye Tian
4. Kay Chen Tan
5. Ran Cheng
6. Yansen Su
7. B. Qu
8. Jing J. Liang

Citations in the corpus

• Balancing Objective Optimization and Constraint Satisfaction in Constrained Evolutionary Multiobjective Optimization (2021, background): “…algorithms (CMOEAs) on simple CMOPs, they are incapable of balancing objec-tive optimization and constraint satisfaction on complex CMOPs [12]–[14], whose feasible regions may be disconnected (e.g., LIR-CMOP9 [12]) or disjoint from the unconstrained Pareto front (e.g., MW13 [13]) as shown in Fig.”
• A Coevolutionary Framework for Constrained Multiobjective Optimization Problems (2021, methodology): “More specifically, existing MOEAs may be incapable of balancing constraints and objectives on CMOPs with small feasible regions [e.g., the feasible region is discrete or far from the unconstrained Pareto front (PF)], which leads to a bad convergence or diversity of the population [6].”
• A Survey on Evolutionary Constrained Multiobjective Optimization (2023, methodology): “…[135], CTP test set [136], CF test set [137], NCTP test set [138], C-DTLZ test set [139], CMOP test set[140], LIR-CMOP test set [22], MW test set [18], DOC test set [96], DAS-CMOPs test set [23], CF test set [17], some test problems derived from actual problems [141], and equality constrained…”