Genetic Algorithms
Most problems in physics involve at some point the optimization of parameters. The idea of genetic algorithms consists in working with
a population of individuals. Each individual is representative of a given set of physical parameters. The best individuals are selected. They generate new individuals for the next generation. Random
mutations in the coding of parameters are finally introduced. This evolutionary strategy is applied from generation to generation in order to determine the global optimum of a problem.
Language : Fortran 90 (OpenMP & Multi‐Agent) & MATLAB
A multi‐objective genetic algorithm was implemented.
High performance on the CECI and Tier‐1 supercalculator was achieved
(massively parallel computing).
Applications :
• optimization of photovoltaic pannels (A. Herman, J. Muller, O. Deparis)
• optimization of solar thermal collectors (L. Gaouyat, O. Deparis)
• optimization of light‐emitting diodes (A. Bay)
Email : [email protected]
population
selection
crossover
mutation
Benchmarking in 5, 10 and 20 dimensions
We reach a target accuracy of 10-4 on
the global optimum of these analytical
functions with a probability of success
in one run of the order of 97-99%,
thanks to different mathematical tricks.
Sphere
Rotated Hyper-Ellipsoid
Rosenbrock
Dixon-Price (modified)
Square Cosines
in a Gaussian Envelope
Schwefel
Levy
Rastrigin
Ackley
Griewank
Worker%1
Worker%2
Worker%3
Worker%4
…
Worker%100
The GA can be coupled easily with external softwares.
It will run in parallel on the CECI/Tier-1 super-calculators
(10.000 cores available).
We can therefore address complex optimization problems.
Optical Engineering with Genetic Algorithms
Bio-inspired surface texturation of Light Emitting Diodes
(LED) for the optimization of light extraction
Waffle-shaped structures with conformal coatings
of cermets and SnO2 for the development of highperformance solar thermal collectors
Optimized structures based on thin
films of c-Si for the development of
high-performance photovoltaics
A. Mayer and A. Bay, Journal of Optics 17, 025002 (2015)