In this paper the authors propose a new method to avoid premature convergence, a common problem in genetic programming, by increasing the diversity of the population in the process of evolution. This method considered age and adaptability of the solution is the criteria to opimize. Evolution populations based on two-dimensional Pareto includes individuals has the smallest age and highest adaptability. To evaluate the method, the authors conducted experiments on several classes of symbolic regression problems with increasing complexity in terms of structure. Test results show that the solution found by this method is better than the standard genetic programming (SOP) was proposed by Koza, genetic programming method stratified by age ALPS proposed Hornby.