Multiple model adaptive control with mixing.

摘要

Despite the remarkable theoretical accomplishments and successful applications of adaptive control, the field is not sufficiently mature to solve challenging control problems requiring strict performance and safety guarantees. Towards addressing these issues, a novel deterministic multiple-model adaptive control approach called adaptive mixing control is proposed.;In this approach, adaptation comes from a high-level system called the supervisor that mixes into feedback a number of candidate controllers, each finely-tuned to a subset of the parameter space. The mixing signal, the supervisor's output, is generated by estimating the unknown parameters and, at every instant of time, calculating the contribution level of each candidate controller based on certainty equivalence.;The proposed architecture provides two characteristics relevant to solving stringent, performance-driven applications. First, the full-suite of linear time invariant control tools is available. A disadvantage of conventional adaptive control is its restriction to utilizing only those control laws whose solutions can be feasibly computed in real-time, such as model reference and pole-placement type controllers. Because its candidate controllers are computed off line, the proposed approach suffers no such restriction. Second, the supervisor's output is smooth and does not necessarily depend on explicit a priori knowledge of the disturbance model. These characteristics can lead to improved performance by avoiding the unnecessary switching and chattering behaviors associated with some other multiple adaptive control approaches.;The stability and robustness properties of the adaptive scheme are analyzed. It is shown that the mean-square regulation error is of the order of the modeling error. And when the parameter estimate converges to its true value, which is guaranteed if a persistence of excitation condition is satisfied, the adaptive closed-loop system converges exponentially fast to a closed-loop system comprised of the plant and some linear time invariant controller that satisfies the control objective.;Three examples are presented. First, a pedagogical example introduces the proposed approach in a tutorial manner. Second, to demonstrate the performance capabilities of an proposed scheme, a benchmark control example is considered. The third example is a robust adaptive velocity and altitude tracking controller for a multiple-input/multiple-output airbreathing hypersonic flight vehicle model.