Premium
Continuous‐Action XCSR with Dynamic Reward Assignment Dedicated to Control of Black‐Box Mechanical Systems
Author(s) -
Hashemnia Saeed,
Shariat Panahi Masoud,
Mahjoob Mohammad
Publication year - 2018
Publication title -
asian journal of control
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.769
H-Index - 53
eISSN - 1934-6093
pISSN - 1561-8625
DOI - 10.1002/asjc.1659
Subject(s) - settling time , overshoot (microwave communication) , inertia , control engineering , control theory (sociology) , computer science , fuzzy logic , classifier (uml) , fuzzy control system , engineering , artificial intelligence , control (management) , step response , physics , telecommunications , classical mechanics
A real‐value classifier system (CSR) is improved by the introduction of a continuous domain of actions to be employed for control of mechanical systems where there is no information concerning the system's mathematical model. To enable the classifier system to handle real‐world control problems where continuous (non‐discrete) actions are required, the exploitation of fuzzy membership functions is proposed. To cope with the dynamic system's delayed response due to its mass inertia, a dynamic reward assignment mechanism is incorporated into the proposed CSR. This allows the rapid calculation of the reward and hence enables the controller to be used in such real time applications. To demonstrate the efficiency of the developed enhanced CSR, it is employed as the controller to balance an unmanned bicycle, without using bicycle properties for the design process of the enhanced CSR. Simulation results show that in terms of overshoot and settling time, the proposed classifier system outperforms traditional XCSR as well as some of the more common balance‐control strategies reported in the literature, as verified using ADAMS software.