Improvement Of Differential Evolution And Its Application On Dynamic Optimization Problems

To tackle complex computational problems, researchers have been looking into nature for many years to find out inspiration. Optimization is the heart of many natural evolution process like Darwinian evolution itself. Through millions of years, every species had to adapt their physical structures to fit to the environments they were in. Researchers developed the evolutionary computing techniques through a keen observation of the underlying relation between optimization and biological evolution, such as evolutionary programming, evolution strategies and genetic algorithm which are called evolutionary optimization algorithms.Differential evolution (DE) is one of the most powerful and effective optimization algorithms. Compared to other optimization algorithms, it is simpler to implement and the number of its control parameters is very few, which triggered the popularity of DE among researchers all around the global within a short span of time. However, it also has some drawbacks, for example the performance of DE highly depends on control parameters and it is easy to get stuck in local optimal for some problems. Hence it has significant impact on real world application to conduct research on how to remove these drawbacks and improve the performance of DE.The main work of this thesis are summarized as follows:we improved the scheme of control parameters in DE and proposed a variant, namely, dissipative differential evolution with self-adaptive control parameters in which the values of control parameters are adjusted by the fitness information between the target individual and the trial individual. In addition, dissipative scheme is employed to make the evolution process escape from the equilibrium state. We investigated impact of one-to-one survival selection on DE and proposed the Subset-to-Subset selection operator which randomly divides target and trial populations into several subsets and employs the ranking-based selection operator among corresponding subsets. The proposed framework gives more survival opportunities to trial vectors with better objective function values. We proposed a new DE algorithm (L-STS-DSDE) by integrating the proposed method of control parameters with the Subset-to-Subset selection operator and introducing population size liner reduction scheme and elite individuals saving method. Experimental results on CEC 2009 benchmark functions show the efficiency of the proposed method when dealing with dynamic optimization problems.