Nonlinear system identification by Haar wavelets. (Q434279): Difference between revisions

The block-oriented approach to system modelling, which takes into account both linear and nonlinear properties, is considered. The Hammerstein system is described by the input-output equation \[ y_k=\sum_{i=0}^\infty \lambda_i m(u_{k-i}) + z_k, \] where \(\{(u_k,y_k),z_k\}\), \(k=1,2,\dots\) are pairs of the system input--output measurements and an additive noise, \(m(u)\) is a cascaded nonlinear block followed by linear dynamics \(\{\lambda_i\}\), \(i=0,1,\dots\). Under assumptions that \(u_k\) is a stationary white noise signal and \(z_k\) is a zero-mean stationary signal the nonlinear block \(m(u)\) may be extracted as nonparametric regression function \[ E(y_k|u_k=u)=\lambda_0 m(u)+ E\{m(u_1)\}\sum_{i=1}^\infty \lambda_i, \] where \(E\) stands for mathematical expectation. In the book, four Haar wavelet identification algorithms are proposed to the system nonlinearity \(m(u)\) identification: quotient orthogonal series algorithm, order statistics algorithm, empirical distribution algorithm and empirical orthogonal series algorithm. Computational properties, such as convergence rates and complexity are given, and a comparative analysis of algorithms is provided.