Key Words: Adaptation; integration; Machine Learning; modeling; optimization; Reinforcement Learning; tuning.
Abstract. The article examines the main directions for integrating the Reinforcement Machine Learning (RL) as an extension of the already existing advanced industrial automation systems based on the classical PID and MPC regulators. The appropriate areas of application, functionality, advantages and disadvantages of the application of RL in direct process control, hybrid regulation, as well as in local coordination of complex multidimensional, nonlinear, with parametric variability and large number of constraints objects with frequent changes of the control objectives are analyzed. The goal of integrating RL is to achieve optimality, adaptability and security in all possible operational situations from technological and business aspects that are unattainable for traditional systems with PID and MPC regulators. An analysis was made and the prospects for the gradual transformation of RL into the third main pillar of industrial automation along with PID and MPC were assessed.
