Crossover and mutation operators of real coded genetic algorithms for global optimization problems
This study is primarily aimed at investigating two issues in genetic algorithm (GA) and one issue in conformational search (CS) problems. First and foremost, this study examines the proposed crossover and mutation operators on the problems of slow convergence and premature convergence to suboptim...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2016
|
| Subjects: | |
| Online Access: | http://psasir.upm.edu.my/id/eprint/69322/1/FSKTM%202016%2010%20IR.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | This study is primarily aimed at investigating two issues in genetic algorithm (GA)
and one issue in conformational search (CS) problems. First and foremost, this
study examines the proposed crossover and mutation operators on the problems of
slow convergence and premature convergence to suboptimal solution. Second of
all, this study operates within experimental design with Taguchi method to
discover the optimal design factors for the two proposed genetic operators. On the
other hand, the CS issue focuses on the effects of the combination of the two
proposed genetic operators on two CS problems.
Past studies have revealed that GAs are one of the most prevalently used stochastic
search techniques to date. The strength of the algorithm lies in the fact that it
assists the evolution of a population of individuals who would thrive in the survival
of the fittest towards the next generation. GA has been employed in resolving
many complex combinatorial optimization problems such as CS problems.
However, the lack of diversity in a population and the difficulty to locally exploit
the solutions within a population creates a setback for GA. Apart from that, its
tuning variables are tricky, as it requires intricate setting properties. On another
note, the drawback in CS is in locating the most stable conformation of a molecule
with the minimum potential energy based on a mathematical function. The number
of local minima grows exponentially with molecular size and this makes it that
more difficult to arrive at a solution. As such, this research is aimed at resolving
the issues mentioned.
The rationale behind developing algorithms using real encoding of chromosome
representations is the limitations of binary encoding. In relation to this, Real
Coded GA (RCGA) refers to GAs which incorporate real number vector
representations of chromosomes. Because the representations of the solutions are
similar to the natural formulation, RCGA gets better-customized to the
optimization of problems in a continuous domain. Throughout the years, there has
been a shift in focus on constructing new crossover and mutation operators to
improve the performance of GA in function optimization.
GA operators employ two main strategies; that is, exploration and exploitation to
locate the optimum solutions. This research employed a new generational GA
based on a combination of the proposed Rayleigh Crossover (RX) and proposed
Scale Truncated Pareto Mutation (STPM) called RX-STPM. It is applied in
optimization problems like CS. While RX displays self-adaptive behavior and
possesses exploration capabilities, STPM thrive in its exploitation features. Hence,
RX-STPM becomes an optimal equilibrium between exploration and exploitation
strategies in leading the system towards global optima. The explorative and
exploitative features of the proposed GA are regulated by substantial crossover
probability and mutation rate set up using the Taguchi method. Aside from that,
tournament selections with proper tournament sizes, used in the design of the
proposed operators, also led to strong exploration potentials.
As you will see in this study, the performance of all RCGAs is contrasted to the
standard criteria used in GA literature, which involves accuracy (judged by average
error, mean and standard deviation of the objective function values), efficiency and
reliability (judged by success rate and average number of function evaluation). RX
and STPM operators were separately tested on a dataset of ten benchmark global
optimization problems according to the specified experimental procedure. The
numerical findings gathered from performance evaluations for RX and STPM were
promising and they have shown significantly better results in comparison to the
other crossover and mutation operators found in the literature.
An accurate combination of GA operators is pivotal in securing effective resolution
to the problem. In this study, the GA was analyzed on a few operators. The
numerical results obtained from the performance evaluation indicated that the RX
crossover is the most fitting pair to the STPM mutator in competently solving two
CS problems i.e. minimizing a molecular potential energy function and finding the
most stable conformation of pseudoethane through a molecular model, which
involves a realistic energy function. |
|---|