Practical application of adaptive controllers necessitates the incorporation of substantial monitoring or jacketing software. An expert system is proposed to perform such duties, allowing a distinction to be made between the control and identification algorithms, and the accompanying supervisory functions. The implications of employing expert systems for real-time environments are outlined, along with their consequences for control of a turbogenerator excitation system. An industry standard VME platform linked to a network of Inmos T800 transputers is used to establish the self-tuning expert controller.Illustrations are given of how the performance and robustness of the adaptive controller can be greatly improved through expert system intervention, both in simulation and on a laboratory micromachine system.With the increasing complexity of electric power systems, the need for improved control of power generation equipment has been steadily rising. Large transmission distances and highly complex distribution networks can give rise to dynamic instability, while the consequence of demands for higher efficiency has resulted in generating sets which are less stable than their predecessors, requiring more complex control systems for their operation. Additionally, the recent privatisation of the UK electricity supply industry has necessitated more flexible operation of individual units, with target loads for power stations requiring to be met on a daily basis under threat of financial penalty.Indeed, power stations which compete for the more lucrative peak periods are obligated to operate in 'burst' modes of constant running up and down -a mode of operation for which thermal power stations not intended.Research has that adaptive control can c~~~~~°s ignificant advantages over conventional implementations (Wu and Hogg, 1988a). The highly complex nonlinear nature of power systems, subject to varying loads and generation schedules, can lead to a degradation of performance in fixed parameter controllers, designed at a particular operating point (Finch et al, 1991). Adaptive control allows the parameters of the controller to adjust as the operating conditions change, and has been shown to improve the overall control in turbogenerator systems (Chandra et al, 1991). However, application of self-tuning algorithms reveals that the bare algorithms tend to be somewhat fragile. Particular problems can be associated with the parameter estimator, which is susceptible to numerical difficulties. A common problem is 'bursting' of the parameter covariance matrix due to non-persistently exciting inputs. The self-tuner must also be robust against noise on input measurements, and unmodelled process dynamics and disturbances (Astrom and Wittenmark, 1989), including transducer failures and saturation effects. Given that a power system is frequently affected by small disturbances, such as line-changing, and transformer tap-switching, and occasional major disturbances such as short-circuits or lightning strikes, the estimator may well be asked ...