Metaheuristics for Intelligent Electrical Networks

3Performance Evaluation of Metaheuristics

3.1. Introduction

Metaheuristics have been introduced with the objective of solving difficult optimization problems in a reasonable time. These methods make use of a number of parameters that can be adjusted according to the problem to be solved. Since their introduction, several have been proposed in the literature. Since these methods are of a stochastic nature, the evaluation of their performance must take this aspect into consideration.

The first evaluation can be performed by means of theoretical study through the calculation of complexity [KAR 72]; nevertheless, this study remains insufficient to define the effectiveness of a stochastic algorithm. However, an experimental evaluation must be achieved by means of simulations. Regarding the comparison of different metaheuristics, statistical tests must be conducted. Moreover, an experimental model must be established in order to clarify the objectives as well as the choice of instances implemented, which can be real and/or simulated.

In this respect, several tools are available for the evaluation of metaheuristic performance and the choice of the best tool is not trivial. Accordingly, in this chapter, we are going to present the different performance measures that can be used. Then, we will review tools for statistical analysis. Finally, we will describe a few benchmarks used in the literature which aim to compare the performance of metaheuristics.

3.2. Performance measures

In this ...