A control unit (75) for controlling a first state variable of a system, for example an angle of rotation ζ of a throttle valve (7) in a throttle device for use in an internal-combustion engine, the system being influenced by a disturbance variable during operation, for example an oscillating loading torque caused by air-flow fluctuations in the throttle device. The control unit (75) comprises a disturbance observer (125) for calculating a value of the disturbance variable on the basis of a mathematical model of the system, and a disturbance predictor (128) for predicting a value of the disturbance variable for a second point in time which follows a first point in time at which said value in calculated by the disturbance observer (125) by a predetermined time interval. In this manner, a phase shift between the disturbance variable actually influencing the system and a compensation for the disturbance variable realized by the system is prevented at high frequencies of the disturbance variable. In a particular embodiment, the disturbance predictor (128) comprises an artificial neural network (207) having a learning phase and a test phase. In a further embodiment, the disturbance predictor (128) supplies a signal corresponding to the value of the disturbance variable calculated by the disturbance observer (125) when the neural network (207) is in the learning phase, and a signal corresponding to the value of the disturbance variable predicted for the second point in time when the neural network (207) is in the test phase.