| In this thesis, a new identification scheme for nonlinear block structured systems was established. The main idea of this scheme is to represent the memoryless nonlinearities in block structured models with support vector machines (SVMs) and then follow the generic SVM approach to formulate identification algorithms that estimate all of the model's parameters. The convexity of the generic SVM regression problems in addition to the sparseness and robustness of their solutions were the main stimulators of this work. The proposed approach was applied to three different block structured systems. Moreover, large scale SVM regression solutions were investigated and implemented on one of the proposed algorithms. Finally, simulations and data from real systems have been used to test the performance of the developed algorithms under different situations. |
