Researches and operational application of numerical weather prediction (NWP) are crucial to high quality weather forecasting. The accuracy of NWP lies on the errors of atmosphere models and initial conditions directly. Given evolutionary computation has effective performance for solving numerical optimization problems, the cost function of 3D Var is used as the object function that would be optimized, to balance the background and observation conditions to get optimal analysis conditions. Data Assimilation is the effective method to solve the initial condition problems. Two ideal numerical experiments are implemented with Lorenz 96 and Lorenz 63 The results show that after optimization, errors can be controlled in a minor range.