An accelerated asynchronous advantage actor-critic algorithm applied in papermaking

The basis weight and moisture system with multivariate, strong coupling, large time lag and large inertia has always been tricky in papermaking. In this paper, a framework which applies the reinforcement learning methods is introduced to solve. Inspired by the previous work, we improve the Markov Decision Process (MDP) in our model, moreover we propose a new form of eligibility traces which can realize data multiplexing to speed up learning, aiming at remitting the costly expense of industrial process online training, where we use it in asynchronous advantage actor-critic (A3C) for illustration. The effectiveness of accelerated A3C in aforementioned system and the superiority of fast learning have been proved in the simulation.