An improved crowding-based differential evolution for multimodal optimization

Traditional optimization technologies usually try to find a global optimum, however, many optimization problems are multimodal with many global or local optima. In real world, multiple optima are usually interested in and can give people multi-choices. Crowding-based differential evolution (CRDE) algorithm is a simple but very powerful for multimodal optimization. CRDE has good explorative ability to find the optima in search space. The main shortcoming of CRDE is the convergence speed is low. To welcome this, an improved CRDE with local search on the individuals nearest optima in the population is introduced. Local search uses Gaussian mutation whose mutation range decreases linearly with iteration. It makes refined search in the area around the optima and improves the exploitable ability. To identify the best individuals around the optima in the current population, the idea of specifying the seeds of species (i.e. the best individuals in niches) in species-based particle swam optimization (SPSO) is adapted. The introduced algorithm is tested on multimodal benchmark problems CRDE used and the test shows it outperforms CRDE in convergence speed greatly.