|
Copyright © 2023 Insight Numerics, LLC | All Rights Reserved |
Description of the properties of the evolution operators section of the Genetic Algorithm Tab in the Optimization Controller:
The number of crossover points used when combining two parents to form offspring. Any value is sensible, as long as it is greater than, or equal to, 1.
By default, the value is 2, representing two-point crossover.
Elite individuals are chosen from the fittest individuals in a population. If an individual is marked as elite, it is not subject to mutation. This means that the individual is carried through directly to the next generation.
Sensible values range between 1 and 10% of the number of individuals in the generation (i.e. 3 elite individuals for a population of 30). However, increasing the number of elite individuals to be greater than 1 may have the effect of reducing diversity, as the elite individuals may be very similar. Any reduction in diversity is generally something to be avoided.
The default value is 1.
The likelihood that a random change will occur in the individual in the current generation. Higher mutation rates lead to greater diversity in the population, and tend to reduce the mean fitness of the population as most random changes result in worse layouts. However, some random changes will lead to improvements, and it is vital that this component is non-zero so that the Genetic Algorithm does not get stuck too early in a local maximum.
Sensible values commonly used in other Genetic Algorithms vary between 0.001 and 0.1.
The default value is 0.01.
The method used for selecting parents for the next generation. Available options are ‘Roulette Wheel’, ‘Tournament’ and ‘Rank’.
Roulette Wheel selection tends to strongly favor fitter individuals, which can reduce diversity too quickly.
Rank selection is at the other end of the scale, and provides a more even selection process, which can result in fitter individuals not being chosen enough, which slows convergence.
Tournament selection is a balance between these two approaches.
The default value is Tournament.
The proportion of the generation that is directly carried through to the next generation. For example, if this value is 0.1 then new generations are comprised of 10% of the fittest individuals from the previous generation, and 90% offspring generated by parents in the previous generation. It is generally useful to have some ‘carry over’ from one generation to the next. Too much will lead to slow convergence.
The default value is 0.1.