The scope of this research is to decrease instantaneous output torque (dynamic inertia moment) oscillations influencing a rolling mill drive. Such oscillations in the drive mechanics lead to fast equipment wearing and even might result in its failure. They can be reduced with the help of the drive adaptive control system. In particular, speed PI-controller parameters can be adjusted to compensate dynamic load impact using a neural tuner. It calculates Kp and KIusing a neural network which is trained online with a learning rate value computed by a rule base containing knowledge of an expert who is responsible for controller adjustment. Such rule base to minimize instantaneous output torque oscillations is developed. It is also useful to reject disturbances. A model of second rolling mill drive of the rolling stand 350 of the Oskol electro metallurgical plant is used as a plant. Modelling is conducted in Matlab Simulink using a two-mass mechanical part model of this drive. A test bench with the drive of the same kind is also used to prove the tuner effectiveness. The results of the experiments show that the neural tuner application allows one to reduce the instantaneous output torque fluctuations averagely by 13.7% for their first peak and 52.7% - for the second one comparing to the system without tuning, and also increase the energy efficiency of the perturbation rejection by 3.8%.