eduzhai > Applied Sciences > Engineering >

Online Reinforcement Learning Control by Direct Heuristic Dynamic Programming from Time-Driven to Event-Driven

  • Save

... pages left unread,continue reading

Document pages: 7 pages

Abstract: In this paper time-driven learning refers to the machine learning method thatupdates parameters in a prediction model continuously as new data arrives.Among existing approximate dynamic programming (ADP) and reinforcement learning(RL) algorithms, the direct heuristic dynamic programming (dHDP) has been shownan effective tool as demonstrated in solving several complex learning controlproblems. It continuously updates the control policy and the critic as systemstates continuously evolve. It is therefore desirable to prevent thetime-driven dHDP from updating due to insignificant system event such as noise.Toward this goal, we propose a new event-driven dHDP. By constructing aLyapunov function candidate, we prove the uniformly ultimately boundedness(UUB) of the system states and the weights in the critic and the control policynetworks. Consequently we show the approximate control and cost-to-go functionapproaching Bellman optimality within a finite bound. We also illustrate howthe event-driven dHDP algorithm works in comparison to the original time-drivendHDP.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...