INGAR   05399
INSTITUTO DE DESARROLLO Y DISEÑO
Unidad Ejecutora - UE
congresos y reuniones científicas
Título:
CONTROL SYSTEM PERFORMANCE MONITORING USING BAYESIAN SURPRISE
Autor/es:
LUIS AVILA; ERNESTO CARLOS MARTINEZ
Lugar:
Buenos Aires
Reunión:
Congreso; AADECA 2012 Semana del Control Automático –23º Congreso Argentino de Control Automático; 2012
Institución organizadora:
Asociación Argentina para el Control Automático
Resumen:
Optimal action selection chooses the controller design that makes the closed-loop behavior as close as possible to the desired one. A fully probabilistic design framework is proposed to describe the closed loop behavior in probabilistic terms using Bayesian surprise to fast detect performance degradation. Bayesian surprise is significantly affected by any deviation from optimal operation in a controlled system or agent behavior, which allows its use not only for monitoring regulation/tracking control systems but also in supervising Real-time Optimization (RTO) and dynamic optimization schemes. Blood glucose control in the human body is used to exemplify the proposed approach for control loop monitoring.