The major dynamics of the heat recover steam generator in combined cycle power plant include nonlinear behavior, time delays, and uncertainties. Traditional control strategy could not offer satisfactory result. The solution in this paper, called the supervisory predictive control, is based on the set-point optimization method. It consists of a two-level control diagram. The supervisory level dynamically optimizes a general objective function to determine the optimal set-point for a given regulatory level. The algorithm of the supervisory predictive control is deduced in detail, and used to control the steam boiler system under different load conditions. Comparisons with generalized predictive control and conventional PID approaches were made. Simulation result shows excellent tracking results.