A broad range of topics concerning fuzzy logic systems, nonlinear system identification and adaptive control are addressed in this thesis to achieve our main objective, which is to develop effective and reliable fuzzy logic approaches for identification and control of ill-defined, nonlinear dynamic systems.; First, improvements are made of existing fuzzy logic systems, which include defining two on-line quantitative measures for IF-THEN rule performance, and introducing a statistical confidence measure for approximation accuracy of FLS estimators. To facilitate on-line applications, a simplification is proposed of fuzzy inference computation, and the bounds of approximation errors are derived. Next, a complete procedure is presented for formulating expert knowledge based fuzzy logic controllers, and experimental results are demonstrated on a real mechanical system. The empirical FLC is also compared with PD controllers, and their respective properties discussed. Following is an optimal training scheme for fuzzy logic systems which combines a backpropagation algorithm with a least square estimation technique, synergistically combining them.; Observing the fact that the fuzzy logic systems being used to date are static in nature while the physical systems of interest are generally dynamic, a novel fuzzy logic system structure, the DFLS, which is characterized by inclusion of dynamics, is proposed, and its universal approximation property proved. Based on the DFLS, an identification algorithm is further developed, and its stability properties analysed theoretically. Its application to nonlinear, ill-defined dynamic systems is illustrated via a variety of examples, where the significance of human expert knowledge in improving system performance is demonstrated and a comparison of performance between DFLS and FLS identifiers is presented. In addition, a novel DFLS based indirect adaptive control scheme is developed, and its closed loop system performance and stability properties theoretically analysed. Two approaches are presented to estimate an unknown control gain function, g. One is based on a self-tuning scheme, the other is a FLS approach, and their respective properties are discussed. The DFLS adaptive control algorithm is applied to a variety of nonlinear systems, including a real mechanical system, and satisfactory results are observed in all situations. which demonstrates the effectiveness of the proposed control approach in dealing with nonlinear, ill-defined systems. Finally, a recurrent DFLS, the RDFLS, is introduced, its universal approximation property proved, and a RDFLS based stable identification algorithm developed. The stability properties of the RD-FLS identifier are theoretically analysed, and its application to nonlinear systems is demonstrated via simulation examples.
展开▼
