D. Karimanzira, P. Otto, and J. Wernstedt (Germany)
Neural networks, predictivecontrol, self-tuning control and instantaneouslinearization, cascade fuzzy adaptive control, real time.
The combination of neural network models, predictive and self-tuning control has frequently been discussed in the neural community e.g., [1], [3], [4]. Self-tuning algorithms, based on linearized models, can be used to automatically adapt the controller parameters to overcome process variations, both non-linear and time variant [7]. However, the practical implementation of self-tuning algorithms requires careful jacketing to avoid possible stability problems and, when operating conditions are altered and non-linearities are a factor, the problem of deteriorating control performance still remains. This work proposes a predictive self-tuning control scheme, which facilitates the implementation and gives a substantial reduction in the required amount of computations. The idea is to use an instantaneous linearization technique to extract a linear model from an a priori trained non-linear neural network model at each sample [6], then use it in a gain scheduling type controller [8]. Implementation of the self tuning predictive control strategy (StGPC) incorporating the identified neural network model and the instantaneous linearization technique is described, and real time control results illustrate the improvements in control performance that can be achieved when compared to a cascade fuzzy adaptive PID Control (CFuPID) and a conventional neural model predictive control (MPC).
Important Links:
Go Back
