Numerical optimization using differential evolution

• Main Author: Awad, Noor Hussien Ali
• Other Authors: Ponnuthurai N. Suganthan
• Format: Theses and Dissertations
• Language: English
• Published: 2019
• Subjects: DRNTU::Engineering::Computer science and engineering::Theory of computation::Analysis of algorithms and problem complexity; DRNTU::Engineering::Computer science and engineering
• Online Access: https://hdl.handle.net/10356/83265; http://hdl.handle.net/10220/48011
• Institution: Nanyang Technological University

Engineers and scientists from all disciplines often have to tackle numerous real- world applications. Developing efficient evolutionary algorithms for this target has attracted many researchers due to the fact that many real-world applications can be stated as optimization problems. Differential evolution (DE) has become one of the most effective metaheuristics during the last decade, due to its ability to solve complex optimization problems with diverse characteristics. In this thesis, novel efficient differential evolution variants that can be successfully applied to solve numerical optimization problems are studied. The aim is to develop new improved differential evolution algorithms through mitigating well- known problems that DE suffers from, such as easily getting stuck in local optima, and being easily influenced by the choice of its control parameters. Such improvements should empower these new variants to solve challenging optimization problems efficiently when compared to other existing state-of-the- art algorithms. Different ideas were employed in building such new variants such as: hybridizations that combine the strengths of different canonical algorithms, new ensemble control parameter settings, an improved crossover strategy that is used to build a suitable coordinate system during the search and an assistant surrogate model to mimic the response of the objective function. To validate the performance of the developed algorithms, different challenging test suites from recently developed IEEE-CEC benchmarks were used. Those benchmarks are among the widely used benchmarks by many researchers to test their developed algorithms. Each of them constitutes problems that are tested on different dimensionalities, with a various set of problem features and characteristics, including ruggedness, noise in fitness, multimodality, ill-conditioning, interdependence and non-separability. Moreover, a variety of real-world optimization problems taken from diverse fields are also used. The results of the comparative study statistically affirm the efficiency of the proposed approaches to obtain better results compared to other state-of-the-art algorithms from the literature.