提出一种对水轮发电机组水压频率进行综合调节的记忆递归网络灵敏度预测控制器．控制器由预测网络和控制网络组成，预测网络向控制网络提供控制灵敏度信号．两个网络均根据学习和控制的收敛与稳定性定理自适应学习．仿真结果表明控制器能在线实时控制具有非线性、时变、多变量特性的水轮发电机组，有效控制水压． A memory recursive network sensitivity predictive controller is proposed to adjust the hydraulic pressure frequency of hydroelectric generating set. The controller consists of a predictive network and a control network, and the predictive network provides control sensitivity signals to the control network. Both networks learn adaptively according to the convergence and stability theorem of learning and control. The simulation results show that the controller can control the hydro-generator with nonlinear, time-varying and multivariable characteristics on-line in real time, effectively controlling the water pressure.