Q-learning is a Reinforcement Learning method where the learner builds incrementally the Q-function which estimates the future rewards for taking actions from a given state. This paper studies the application of Q-lea...
详细信息
Many processes, even being of a continuous nature, involve in its operation signals or rules different from the classical continuous variables represented by real variables and modelled by DAE. In practice they includ...
详细信息
Many processes, even being of a continuous nature, involve in its operation signals or rules different from the classical continuous variables represented by real variables and modelled by DAE. In practice they include on/off valves or other binary actuators, are subjected to logical operational rules, or are mixed with sequential operations. As a result, classical control does not fit very well with the overall operation of the plant. In this paper we consider the problem of hybrid control from a predictive control perspective, showing in a practical non trivial example with changing process structure, how the problem can be stated and solved.
Direct adaptive control algorithms using state variables in control structure are known for their good adaptation capability and tracking performances that result from the fact that they use plant state variables in c...
详细信息
Direct adaptive control algorithms using state variables in control structure are known for their good adaptation capability and tracking performances that result from the fact that they use plant state variables in control law as well as in adaptation law. The implementation of standard model reference adaptive control (MRAC) with state variable control structure requires some a priori knowledge about the plant to be controlled including the plant order. This information is crucial for the proper choice of reference model describing the desired closed loop dynamical behavior and consequently for the adaptive system performances. The aim of our paper is to propose the fuzzy adaptation law for MRAC with state variable structure of control law that is able to ensure the adaptation process convergence and tracking capability even in the presence of unmodelled dynamics.
Q-learning is a Reinforcement Learning method where the learner builds incrementally the Q -function which estimates the future rewards for taking actions from a given state. This paper studies the application of Q -l...
详细信息
Q-learning is a Reinforcement Learning method where the learner builds incrementally the Q -function which estimates the future rewards for taking actions from a given state. This paper studies the application of Q -learning on Process control problems, more precisely on Neutralization Processes. As the process to be studied is non-linear, 9 states are selected. Each state has 5 possible actions (unless in goal state, which has 3 actions), that corresponds to variations of the control signal. Softmax and ε-greedy policies are applied to select the actions on a laboratory pH plant. On-line results show that the controllers are able to learn how to control adequately the process.
Traditionally, when approaching controller design with the Youla-Kucera parametrization of all stabilizing controllers, the denominator of the rational parameter is fixed to a given stable polynomial, and optimization...
详细信息
Traditionally, when approaching controller design with the Youla-Kucera parametrization of all stabilizing controllers, the denominator of the rational parameter is fixed to a given stable polynomial, and optimization is carried out over the numerator polynomial. In this work, we revisit this design technique, allowing to optimize simultaneously over the numerator and denominator polynomials. Stability of the denominator polynomial, as well as fixed-order controller design with H/sub /spl infin// performance are ensured via the notion of a central polynomial and LMI conditions for polynomial positivity.
This paper focuses upon the diagnosis of an assembling/dissembling robot, using analytical methods based on observers. The paper presents a modality of detecting incidents, using the mathematical model of the robot, w...
详细信息
Due to the complexity of neural networks the problem of large computation requirements on computational devices where the network is trained or evaluated and slow network evaluation/training response can appear. The p...
详细信息
Two novel linear matrix inequality (LMI)-based procedures to receive stabilizing robust output feedback gain are presented. one of them being modification of previous results of (Oliveira et al..1999). The proposed ro...
详细信息
The paper addresses the problem of robust output feedback controller design with a guaranteed cost and parameter dependent Lyapunov function quadratic stability for linear continuous time polytopic systems. The propos...
详细信息
New developments in computer networks and communications provide new possibilities also for control purposes. controlsystems for highly complex plants are themselves very complex and heterogeneous. A new software and...
详细信息
暂无评论