An Improved Univariate Marginal Distribution Algorithm for Dynamic Optimization Problem

In dynamic environments, it is difficult to track a changing optimal solution over time. An improved univariate marginal distribution algorithm (IUMDA) is proposed to deal with dynamic optimization problems. This approach is composed of the diffusion model, which uses the information of current population, and the inertia model, which uses the part history information of the optimal solution. After an environment changed, the strategy is changed by a detecting operator to guide increasing the population diversity. Finally an experimental study on dynamic sphere function was carried out to compare the performance of IUMDA and mutation UMDA. The experimental results show that the IUMDA is effective for the function with moving optimum and can adapt the dynamic environments rapidly.