Differential evolution (DE) was introduced by Stone and Price in 1995 as a population-based stochastic search technique for solving optimization problems over continuous space. DE is an effective solving method for many problems even though it has few control parameters to be set. DE belongs to a group of evolutionary algorithms and is similar to Genetic Algorithm (GA). The generation alternation model of DE is a discrete generation model. In a differential operation, a target vector, a base vector and differential vectors are chosen as parents and generate a child called trial vector. The children are evaluated and replaces its parent if its fitness is better than that of its parent. In this paper, we propose a new DE generation alternation model called Roulette Selection Based on Evolutionary Advance Level (REAL) for reducing the number of function evaluations. Firstly, we design the DE/MGG model which based on a policy of Minimal Generation Gap (MGG). In the differential operation, a number of children are generated and the target vector is alternated to the best individual from the family. Secondly, we define an evolutionary advance level as a degree of progress of the alternation of generations and modify DE/MGG as a propose model which make a difference in the evolutionary advance level of the individuals. In the proposed model, parents are selected based on its evolutionary advance level and the number of childlen is also decided depending on the evolutionary advance level of the target vector. Finally, we compare the proposed model to conventional DE and DE/MGG through experiments on several test functions and show a searching performance of proposed model.