Data-based control design for linear and recurrent neural network models with robust stability guarantees
William D’Amico
PHD Student
DEIB - Conference Room "E. Gatti" (Building 20)
March 17th, 2023
11.50 am
Contacts:
Simone Formentin
Research Line:
Control systems
PHD Student
DEIB - Conference Room "E. Gatti" (Building 20)
March 17th, 2023
11.50 am
Contacts:
Simone Formentin
Research Line:
Control systems
Sommario
On March 17th, 2023 at 11.50 am William D’Amico, PHD Student in Information Technology, will give a seminar on "Data-based control design for linear and recurrent neural network models with robust stability guarantees" in DEIB Conference Room.
In automation and control, data-based techniques are becoming increasingly popular since they allow one to design controllers providing satisfactory results with moderate time and computational effort. In this presentation we first show the application of the design of recurrent neural network (RNN) controllers using the virtual reference feedback tuning (VRFT) method in an experimental case study, i.e., the electronic throttle body. Particular focus is given to the capability of this class of controllers to handle the actuator limitations by design. Secondly, we propose a methodology based on set membership identification which allows one to apply VRFT to linear discrete-time systems while guaranteeing the robust stability of the closed-loop system by design. Finally, we provide some hints to a more recent work dedicated to provide stability guarantees in case of systems described by RNNs.
In automation and control, data-based techniques are becoming increasingly popular since they allow one to design controllers providing satisfactory results with moderate time and computational effort. In this presentation we first show the application of the design of recurrent neural network (RNN) controllers using the virtual reference feedback tuning (VRFT) method in an experimental case study, i.e., the electronic throttle body. Particular focus is given to the capability of this class of controllers to handle the actuator limitations by design. Secondly, we propose a methodology based on set membership identification which allows one to apply VRFT to linear discrete-time systems while guaranteeing the robust stability of the closed-loop system by design. Finally, we provide some hints to a more recent work dedicated to provide stability guarantees in case of systems described by RNNs.
