Recursive neuro fuzzy techniques for online identification and control

Dissertação para obtenção do Grau de Mestre em Engenharia Electrotécnica e de Computadores

The main goal of this thesis will be focused on developing an adaptative closed loop control solution, using fuzzy methodologies. A positive theoretical and experimental contribution, regarding modelling and control of fuzzy and neuro fuzzy systems, is expected to be achieved. Proposed non-linear identification solution will use for modelling and control, a recurrent neuro fuzzy architecture. Regarding model solution, a state space approach will be considered during fuzzy consequent local models design. Developed controller will be based on model parameters, being expected not only a stable closed loop solution, but also a static error with convergence towards zero. Model and controller fuzzy subspaces, will be partitioned throughout process dynamical universe, allowing fuzzy local models and controllers commutation and aggregation. With the aim of capturing process under control dynamics using a real time approach, the use of recursive optimization techniques are to be adopted. Such methods will be applied during parameter and state estimation, using a dual decoupled Kalman filter extended with unscented transformation. Two distinct processes one single-input (SISO) other multi-input (MIMO), will be used during experimentation. It is expected from experiments, a practical validation of proposed solution capabilities for control and identification. Presented work will not be completed, without first presenting a global analysis of adopted concepts and methods, describing new perspectives for future investigations.