Support Vector Machines for Identification and Classification Problems in Control Engineering
This dissertation explores the use of support vector machines (SVMs) in control engineering - mainly as regression methods for system identification, but also for classification tasks. In general, SVMs serve as data-driven process models and have become an important tool of statistical data analysis and machine learning. Besides the comparison with existing approaches, the thesis is focused on the development of optimization algorithms which are crucial for the success of the method. As a substantial innovation, this is done with active-set strategies and gradient projection. Another novelty is the use of SVMs for the robust estimation of output error models for linear dynamic processes. Finally, the new algorithms are tested and compared on three experimental plants: A dialysis machine, a hydraulic servo axis, and a condensing boiler.