Abstract: In PWR nuclear power plant, whether the liquid level of the steam generator can be stably controlled within the set range is directly related to the safe and economic operation of nuclear power plant. Aiming at the complex problems of “false liquid level”, time-varying and input constraints in steam generator liquid level control, an adaptive generalized predictive control algorithm with constraints is designed. In this algorithm, a two input single output mathematical model is constructed to describe the influence of steam flow and the feedwater flow on the liquid level of steam generator; The minimum recursive multiplication with variable forgetting factor is introduced to identify and correct the parameters of the liquid level object model of the steam generator on line; The mutation strategy of inverse learning of the center of gravity per dimension is introduced to improve the convergence accuracy of the standard particle swarm optimization algorithm, and then the improved particle swarm optimization algorithm is used to calculate the optimal value of the feedwater flow increment within the constraint range. The simulation results show that the improved generalized predictive controller has better control effect than the three impulse PID controller for the complex control problem of steam generator liquid level.