Abstract: The optimization problem of the nonlinear large-scale systems is considered. Using linearization and prediction method, the nonlinear system can be converted into a series of linear subsystems, and the nonlinear criteria function can be converted into an equivalent quadratic cost-function. A two-level hierarchical optimization algorithm is proposed by the costate prediction approach, and the solution to the overall problem can be obtained through the iterative computation between the coordinator and the optimization of subsystems. The major advantages are, such as, significantly reducing the computation time.