Is there any new research talk about the relationship between GA parameters, such as: population size with mutation ratio, or crossover ratio or Number of generations ?
In problems such as the TSP, do you need to put a different size of the population for each city? What if we set Fixed size of the population of all cities? How can extract the relationship between those parameter?
Although there are some advises on determining GA parameters, there are no generally accepted relationships between them. For example one may suggest to take the population size equal to at least k-1 or set the mutation rate to 1/population size (k is the problem dimension). But these don't work in all problems and the only way to obtain a set of good parameter values is tuning the parameters.
About TSP, I guess you made a mistake in defining chromosome representation. For solving TSP using GA, each chromosome is considered as a tour and the cities are genes. So, population size is the number of initial tours and don't depend on the number of cities.
if you want to understand GA and the evolutionary computation in general I suggest reading the following book:
Evolutionary Computation 1st Edition
by Kenneth A. De Jong (Author)
The book shows detailed explanation of the issues that you asked about them before such as what will happen if the population size is increased, number of iterations, change to mutation and crossover rates, changing type of cross over etc.
In general
the author's idea to model the evolutionary process with a dynamic (nonlinear!)system so that "...given particular initial conditions the system follows a trajectory over time through a complex evolutionary state space. One can than study various aspects of these processes such as their convergence properties, their sensitivity to initial conditions, their transient behavior and so on.
Hello everyone, thanks very much for all.
But, I am not asked what the definition of the GA, but I want to understand the relationship between these parameter by each other, such as: population size with mutation ratio. Trail and error are very complex task, and we cannot draw fast conclusion or true conclusion about those parameter. I asked you about new research which help us to draw simple result.
Genetic Algorithm control parameters can be set using Taguchi Design of Experiment methods (DOE) . examples are Taguchi methods and orthogonal arrays.the following article solve TSP
Doughabadi, M. H., Bahrami, H., & Kolahan, F. (2011). Evaluating the Effects of Parameters Setting on the Performance of Genetic Algorithm Using Regression Modeling and Statistical Analysis. Industrial Engineering, 61-68.
I have attached a few references. You can find many more using google scholar advance search.
Best Regards,
Behrouz
http://profdoc.um.ac.ir/articles/a/1025158.pdf
http://link.springer.com/chapter/10.1007%2F978-3-642-25566-3_21#page-1
http://www.ijcaonline.org/research/volume127/number5/majumdar-2015-ijca-906383.pdf
Article Calibration of GA Parameters: The Design of Experiments Approach
Article Design of experiments to identify optimal parameters of gene...
- Badges
- Science topic
- Similar topics
- Biological Science
- Genetics
- Genetic Analysis