Generalized benchmark generation for dynamic combinatorial problems

Abstract

Several general purpose benchmark generators are now available in the literature. They are convenient tools in dynamic continuous optimization as they can produce test instances with controllable features. Yet, a parallel work in dynamic discrete optimization still lacks.In constructing benchmarks for dynamic combinatorial problems, two issues should be addressed: first, test cases that can effectively test an algorithm ability to adapt can be difficult to create; second, it might be necessary to optimize several instances of an NP-hard problem. Hence, this paper proposes a method for generating benchmarks with known solutions without the need to re-optimize. Consequently, the method does not suffer the usual limitations on the problem size or the sequence length.The paper also proposes a general framework for the generation of test problems. It aims to unify existing approaches and to form a basis for designing newer benchmarks. Such a framework can be more appreciated knowing that combinatorial problems tend to assume very distinct structures, and hence, relevant benchmarks are basically too specific to be of interest to the general reader.

A survey of swarm intelligence for dynamic optimization: Algorithms and applications

“…The benchmark problems that generate dynamic environments following this methodology are specified for a single problem. In some other cases, researchers prefer to create their own customized benchmark problems that aim to model some realworld scenarios [65,66,37,67,68,69] which again are developed for a specific problem, or even a specific instance of the problem.…”

Adjustability of a discrete particle swarm optimization for the dynamic TSP

“…Each change can modify the optimal route, and the knowledge of the optimum is useful to be able to monitor the quality of the results that are being produced by an algorithm. Younes et al (2003) proposed a procedure in which the optimal route does not have to be known for each subproblem. A modification is carried out for half the number of subproblems (adding, removing or changing a given distance), and each of these changes is undone in the second half of the procedure.…”

“…As the method affects the optimum, evaluated their algorithms based on the relative differences in the length of the successive solutions. A more general method for generating dynamic versions of combinatorial optimization problems (COPs), including the TSP, was proposed by Younes et al (2005). The basic idea, in the context of genetic algorithms, exploits the fact that most optimization methods involve some form of mapping from the problem solution space to the individuals used in the algorithm, e.g., a permutation of nodes.…”

Pre-scheduled Colony Size Variation in Dynamic Environments

“…In [6,21], only the weights of arcs that belong to the best tour increase or decrease accordingly. Younes et al [26] introduced a benchmark generator for the DTSP with different modes: 1) topology changes as in [10], 2) weight changes in [6], and 3) swap cities. Based on the last mode (i.e., swap cities) of the aforementioned benchmark generator, a general dynamic benchmark generator for permutation-encoded problems (DBGP) was proposed that can generate test cases with known optima [15].…”