11 Parameter Control in Evolutionary Optimisation
359
Fig. 11.1 Simplified flowchart of an evolutionary algorithm: (a) Parents are selected from the
population and produce offspring, through mutation and recombination. The most suitable are
selected to form the next population (inspired by [4]). (b) The chromosomes of two parents, the
yellow and red dots, are recombined and mutated to produce two unique offspring
over time the most successful of these individuals will evolve to discover the
optimal solution. Each iteration of an EA involves a competitive selection that
filters out weak solutions. The fittest individuals are more likely to be chosen for
reproduction and being modified by crossover and mutation, eventually leading
potentially to superior ones. The selected solutions are recombined with other
solutions by swapping parts of a solution with another (crossover). The solutions
can further be mutated by making a small change to a single element of the solution
(mutation).
A simplified flowchart of one of the most popular EA, i.e. genetic algorithm
(GA), is shown in Fig. 11.1, along with the concept of chromosome representation
of a candidate solution, crossover, and mutation in an illustrative way.
11.1.2 Exploration and Exploitation
Exploration and exploitation are the two main aspects of evolutionary problemsolving [6]. The exploration and exploitation of a search space must be addressed
by every search algorithm. Exploration is the process of visiting completely new
regions of a search space, while exploitation is the process of visiting those regions
of a search space within the neighbourhood of previously visited points (see
Fig. 11.2). A search algorithm needs to establish a good ratio between exploration
and exploitation in order to be successful.
Different metaheuristics present unique exploration and exploitation capabilities,
i.e. they have various forms of exploring and exploiting the search space. What
works to solve a particular problem may be not good for tackling another one.
Moreover, each problem can demand a particular set of control parameters for each
algorithm. In this context, adaptive algorithms have appeared trying to solve as
359
Fig. 11.1 Simplified flowchart of an evolutionary algorithm: (a) Parents are selected from the
population and produce offspring, through mutation and recombination. The most suitable are
selected to form the next population (inspired by [4]). (b) The chromosomes of two parents, the
yellow and red dots, are recombined and mutated to produce two unique offspring
over time the most successful of these individuals will evolve to discover the
optimal solution. Each iteration of an EA involves a competitive selection that
filters out weak solutions. The fittest individuals are more likely to be chosen for
reproduction and being modified by crossover and mutation, eventually leading
potentially to superior ones. The selected solutions are recombined with other
solutions by swapping parts of a solution with another (crossover). The solutions
can further be mutated by making a small change to a single element of the solution
(mutation).
A simplified flowchart of one of the most popular EA, i.e. genetic algorithm
(GA), is shown in Fig. 11.1, along with the concept of chromosome representation
of a candidate solution, crossover, and mutation in an illustrative way.
11.1.2 Exploration and Exploitation
Exploration and exploitation are the two main aspects of evolutionary problemsolving [6]. The exploration and exploitation of a search space must be addressed
by every search algorithm. Exploration is the process of visiting completely new
regions of a search space, while exploitation is the process of visiting those regions
of a search space within the neighbourhood of previously visited points (see
Fig. 11.2). A search algorithm needs to establish a good ratio between exploration
and exploitation in order to be successful.
Different metaheuristics present unique exploration and exploitation capabilities,
i.e. they have various forms of exploring and exploiting the search space. What
works to solve a particular problem may be not good for tackling another one.
Moreover, each problem can demand a particular set of control parameters for each
algorithm. In this context, adaptive algorithms have appeared trying to solve as