Neurofuzzy Reinforcement Learning Control Schemes for Optimized Dynamical Performance
Mohammed Abouheaf, Wail Gueaieb · 2019
Tracking control mechanisms employ optimization approaches that rely on tracking error signals to advise appropriate control decisions. However, these processes often neglect other important criteria, such as the control effort required to optimize the overall dynamic performance and the response's transient characteristics. A fuzzy control mechanism is developed to fulfill the aimed tracking objectives. Then, it is integrated with two other supporting artificial intelligence schemes to optimize the overall performance during the tracking process. One is based on a Q-learning approach, while the other uses a neural network architecture. Both methods are tested to adjust the main fuzzy tracking control signal to minimize energy dissipation within the dynamical system. These supporting mechanisms revealed improvements over the standalone tracking fuzzy system. The fuzzy-neural network approach, which is based on an optimized future dynamical cost function, exhibited superior results compared to the fuzzy-Q-learning technique.