Abstract: Taking a 600 MW supercritical unit as an object of study, a mathematical model was set up for the unit load and steam pressure characteristics based on neural network, which was subsequently trained with large amount of data obtained under wide-range varying load conditions. Simulation results show that the model can well fit the complex non-linear dynamic characteristics between unit load, main steam pressure and the fuel supply, feedwater flow and turbine governing valve opening with high precision and strong generalization ability, which therefore may serve as a prediction model for design of intelligent controller in supercritical units.