Dynamic evolutionary optimization with particle swarm optimizatio

Many real world optimization problems have to be solved in the presence of uncertainties. An optimization algorithm has to perform satisfactorily under the presence of such dynamic changes in the environment. In addition to it, the algorithm also has to justify for the additional computational cost incurred. Multi population approaches are found very effective in tracking and locating dynamic optima. In addition, it is necessary to reuse the information from the past evolutions as it facilitates a faster and effective convergence after the occurrence of the change. This thesis proposes a new dynamic particle swarm optimization technique that uses multiple swarms to locate a set of optimal solutions and effectively exploits the past information and adapts the population to the corresponding new locations using the concept of relocation radius. The proposed algorithm uses an adaptive hierarchical clustering mechanism to form multiple swarms. The relocation radius is determined based on the change in the functional values of the particles due to change in the environment and the average sensitivities of the decision variables to the corresponding change in the objective space. The newly adapted population is fitter compared to the original population or a randomly initialized population. The algorithm is tested on dynamic benchmark functions and compared to some of the state-of-the-art dynamic evolutionary algorithms and the results are found to be promising. The algorithm performs better than most of the existing algorithms proposed in literature.