Adaptive Model-Free Control of Information-Poor Systems
12.1 Introduction to Model-Free Adaptive Control of Non-Linear Systems
The model-based (or indirect) approach to adaptive control described in the previous chapter has two main weaknesses. The first is that the goal of the online adaption is to reduce the modelling errors (the differences between the actual and predicted plant output) and not the tracking errors (the differences between the plant output and the setpoint) (Wu and Dexter, 2008). The second is the difficulty of identifying a plant model online from normal operating data, which are usually incomplete and may be corrupted by unmeasured disturbances and measurement noise (see Section 11.4.3). Model-free (or direct) adaptive control algorithms have just one goal: to minimize the tracking errors.
Several fuzzy model-free control schemes (Abonyi et al., 1999; Ordonez and Passino, 1999; Tan and Dexter, 2000; Angelov, 2004; Pomares et al., 2004; Wai and Chen, 2006; Wai and Lee, 2008) have been proposed but most are concerned with the problem of controlling systems whose behaviour is non-linear; relatively few take explicit account of the uncertainties (Yang, 2005).
Schemes that are based on feedback error learning use a fixed feedback controller and an adaptive feedforward controller, which is trained using the control signal generated by a feedback controller as its target (Kawato and Gomi, 1992). The aim is to train the feedforward controller to act as an inverse ...