Adaptive control for time-varying systems

This dissertation proposes three innovative methods to improve the performance of adaptive control under time-varying environment. The main contributions in the dissertation are: (1) The bootstrap adaptive control algorithms for time-varying systems. The adaptive control algorithms employ two parameter estimators, the first one generates the estimates of process parameters and the second one generates the estimates of closed-loop system parameters. The estimates of closed loop system parameters first are converted to a second set of estimates of process parameters. Then, with a carefully devised combination procedure, the two sets of estimates are combined to obtain the better estimates. The combined estimates are then used in designing the adaptive controller. The bootstrap algorithm outperforms the existing adaptive algorithms in the moderate time-varying environment. It also has the capability of controlling certain fast time-varying systems with excellent control performance. (2) Signal compensation methods for time-varying adaptive control systems. These adaptive control algorithms also employ two parameter estimators. But they differ from the previous methods in the way in which the estimates are utilized. The estimates of the closed-loop system parameters are used to design a second adaptive controller called compensator. The compensator takes the command signal and generates the compensated command signals which in turn are fed to the inner adaptive system. The compensated command signal has the effect of offsetting the error caused by using inaccurate estimates in the inner controller design. Consequently, the performance of overall adaptive control system is improved. (3) Adaptive control for periodical time-varying systems with unknown frequencies. This algorithm is a self-tuning technique designed to handle the periodical fast time-varying systems. The frequencies of time-varying parameters are estimated by a frequency detector which consists of a crude estimator, a filter, a peak detector and a monitoring program. The estimated frequencies in turns are used in an estimator with time-varying model to obtain the finer estimates of system parameters. These finer estimates are then used to compute the parameters of the adaptive controller. The result of this combination produces a satisfactory control performance unattainable otherwise.; The algorithms proposed are backed by the analytical results and the computer simulations.