A New On-Line Subspace-Based Identification Algorithm for Multivariable Hammerstein Models

This paper presents a new on-line subspace-based identification algorithm for multivariable nonlinear systems modeled with Hammerstein model structure. The developed algorithm makes use of the subspace identification method via principal component analysis (SIMPCA) to estimate the Hammerstein linear dynamic block. Then, A genetic algorithm (GA) is utilized to optimally estimate the nonlinear mapping due to the Hammerstein nonlinear block by a set of radial basis functions (RBFs). A new approach based on a sliding data block window is developed to facilitate the on-line implementation of the proposed identification algorithm. The performance of the proposed algorithm is evaluated on a simulated binary distillation column benchmark problem with complicated multivariable nonlinear dynamics. The obtained results demonstrate the effectiveness of the proposed algorithm.