A Hybrid GA-TS Technique with Dynamic Operators and its Application to Channel Equalization and Fiber Tracking

Abstract

1. Chapter description This chapter is based on the author’s research activities during the last decade on the fields of Evolutionary Computation (EC)-based techniques applied to digital communications and medical image processing. Specifically, the chapter is organized in three main sections: • The first section (section 2) presents a conc ise introduction to metaheuristic EC-based strategies, mainly genetic algorithms (GA) and tabu search (TS) - for the sake of comparison, brief comments to simulated annealing (SA) are included, as well. Section 2.1 shows a general description of the standard GA while section 2.2 focuses on the basic TS algorithm. • Next, section 3 develops the proposed hybrid GA-TS method. It begins with the description of a genetic algorithm with notable reduced complexity (known as a micro genetic algorithm, μGA) that uses a modification of the standard genetic operators in order to improve its convergence rate and computational load. Such features are achieved by on-line tuning up the probabilities of mutation and crossover by means of analysing the population individuals’ fitness entropy. This way, a new method to control and adjust the diversity of the population is obtained. The μGA here described was partially developed in (San Jose, 2005). Once the GA is obtained, it is then modified and improved using the main distinctive concepts of TS. Specifically, we introduce a systematic use of memory in order to keep information on the last visited solutions as well as on the concrete parts of the chromosomes, or population’s individuals, that have experimented alterations that have positively or negatively affected the fitness function. Besides, memory keeps track of the genes affected by the genetic operators and the tabu tenure depends on the explorative or expl oitative sense of the search, which is estimated from the mean population fitness entropy previously described. This way, the TS main ideas will help to avoid both cycling and processing of non-interesti ng regions of the solutions’ space. The hybrid algorithm thus developed will be denoted as “GA-TS”.