Computational & Technology Resources
an online resource for computational,
engineering & technology publications
PROCEEDINGS OF THE SIXTH INTERNATIONAL CONFERENCE ON ENGINEERING COMPUTATIONAL TECHNOLOGY
Edited by: M. Papadrakakis and B.H.V. Topping
Real-Coded Genetic Algorithms Enhanced Using a Niching Strategy for Solving Multi-Modal Problems
A. Kucerová and M. Lepš
Department of Mechanics, Faculty of Civil Engineering, Czech Technical University in Prague, Czech Republic
A. Kucerová, M. LepĀš, "Real-Coded Genetic Algorithms Enhanced Using a Niching Strategy for Solving Multi-Modal Problems", in M. Papadrakakis, B.H.V. Topping, (Editors), "Proceedings of the Sixth International Conference on Engineering Computational Technology", Civil-Comp Press, Stirlingshire, UK, Paper 78, 2008. doi:10.4203/ccp.89.78
Keywords: genetic algorithms, multi-modal problems, niching strategy, differential evolution, reliability, convergence rate.
At present, genetic algorithms belong to the most modern and most popular optimization methods available. They follow an analogy of processes that occur in living nature within the evolution of live organisms during a period of many millions of years. Because in engineering and scientific problems we usually deal with the real-valued parameters, we focus on real-coded algorithms. A very well known real-coded evolutionary algorithm is the so-called differential evolution (DE) . In , Hrstka and Kucerová have proposed an adaptation of the differential evolution called SADE algorithm (Simplified Atavistic Differential Evolution) with the aim of formulating a method which is able to solve optimization problems on real domains with a high number of variables. This algorithm uses the simplified differential operator, but contrary to the differential evolution, the SADE method uses the algorithmic scheme very similar to the standard genetic algorithm. A detailed comparison of these two algorithms with binary genetic algorithms is presented in .
During last few years some modifications and simplifications were proposed to the SADE algorithm with two principal motivations: (i) to increase the convergence rate of the algorithm for smooth objective functions with just one optimum and (ii) to reduce the number of control parameters of the algorithm. A new version called the GRADE algorithm has only three control parameters and results for the set of twenty mathematical functions has shown that for smooth objective functions with one or just several local extremes the GRADE algorithm achieved better convergence rate than the SADE algorithm.
Another enhancement to genetic algorithms was required to increase the reliability of these algorithms for multi-modal problems. A niching strategy  called CERAF (Abbreviation of the French expression CEntre RAdioactiF - the radioactivity center) method was proposed in  to mark previously found local extremes and restart the algorithm. Accordingly, it produces areas of a higher level of "radioactivity" in the neighborhood of all local extremes by increasing the mutation probability in these areas many times. Extensive test computations have shown that this methodology can be considered as a universal technique capable of solving any multi-modal optimization problem.
purchase the full-text of this paper (price £20)