Managing Two-Level Manufacturing in Transportation with Reinforcement Learning

We consider a model of hierarchical transport corporation, in which the Executive and his subordinate factory manager (the Manager) organize two-level manufacturing. They are controlled by corporation's governance (in short, the Chief) with the Consultant. The Executive knows the capabilities of increasing manufacturing output only at own level. The Manager knows the manufacturing potential of the factory. Neither the Chief nor his Consultant knows the Executive and Manager capabilities. So the Executive can manipulate own output in order to influence the decisions of the Consultant and the Chief to growth own reward. But the Executive itself does not know the potential of the factory. This also can be used by its Manager in own favor. So, the Executive needs to learn to control the Manager. The interests of the Executive and the Manager are reflected in the model by introducing their reward functions. On the basis of this model, sufficient conditions have been found for the synthesis of a hierarchy of controllers for managing output, ensuring the use of random possibilities of its increasing. In these conditions, the Chief needs consultations with the help of a self-learning Consultant. Through these consultations, the Chief can norm the Executive's output. The Learning controller encourages the Executive, firstly, to maximize auxiliary output and, secondly, to implement an adaptive controller for the Manager. Such an adaptive controller includes planning and incentive procedures that encourage the Manager to maximize the manufacturing in the factory. The application of this approach is illustrated by the example of managing output in a car repair corporation.