11 Parameter Control in Evolutionary Optimisation
363
– The number of iterations for reproduction and recombination of solutions
during the optimisation process.
– It is very often connected to the population size, especially when the total
number of evaluations is limited.
– In general, when using a higher number of generations, it is more probable
that the optimisation process will converge towards a better/optimal solution,
but also the computational time is increased.
Whenever some control parameter or other component of the optimisation
algorithm is changed, then, depending on the implementation, it can influence either
an individual solution, or it can influence the whole population of solutions.
11.2.2 What Else can be Adapted
Further on, to respond to changes and to improve the optimisation convergence, one
can also adapt:
• Fitness function
– Sometimes it is needed either to adapt to changing environment with modified
fitness function, or it is needed to add some constraints into the fitness function
(e.g. penalty terms for constraints) [15].
• The representation of the problem
– The representation of the solution can be changed to reflect either the
simplifications due to already optimised parts of the solution or to increase
the force of promising parts for the next generations.
– The proper encoding (representation of the problem) ensures that the optimised variables are modified efficiently and that the heuristics can be easily
applied during the optimisation. Different representations of the same problem
can give a different insight into the properties, including redundant encodings,
synonymity, locality and connectivity, as well as their interrelationships [16].
• The components of the algorithm
– An evolutionary algorithm can use different combinations of parent selection,
crossover operators, mutation operators, survival selection and termination
condition [17]. For example, the implemented versions of crossover and mutation operators influence the way how the encoded information is modified.
363
– The number of iterations for reproduction and recombination of solutions
during the optimisation process.
– It is very often connected to the population size, especially when the total
number of evaluations is limited.
– In general, when using a higher number of generations, it is more probable
that the optimisation process will converge towards a better/optimal solution,
but also the computational time is increased.
Whenever some control parameter or other component of the optimisation
algorithm is changed, then, depending on the implementation, it can influence either
an individual solution, or it can influence the whole population of solutions.
11.2.2 What Else can be Adapted
Further on, to respond to changes and to improve the optimisation convergence, one
can also adapt:
• Fitness function
– Sometimes it is needed either to adapt to changing environment with modified
fitness function, or it is needed to add some constraints into the fitness function
(e.g. penalty terms for constraints) [15].
• The representation of the problem
– The representation of the solution can be changed to reflect either the
simplifications due to already optimised parts of the solution or to increase
the force of promising parts for the next generations.
– The proper encoding (representation of the problem) ensures that the optimised variables are modified efficiently and that the heuristics can be easily
applied during the optimisation. Different representations of the same problem
can give a different insight into the properties, including redundant encodings,
synonymity, locality and connectivity, as well as their interrelationships [16].
• The components of the algorithm
– An evolutionary algorithm can use different combinations of parent selection,
crossover operators, mutation operators, survival selection and termination
condition [17]. For example, the implemented versions of crossover and mutation operators influence the way how the encoded information is modified.