Wavelet networks based general multi-variable non-linear system identification arithmetic and its application (Q2785122)

The authors present an algorithm for identifying general multi-variable nonlinear systems based on the approximation property of wavelet functions in \(L_2 (\mathbb{R})\). They only use the scaling function \(\phi (x)\), which generates a multiresolution analysis of \(L_2 (\mathbb{R})\), and its dyadic transforms \(\phi_{m,n}(x)\), so the number of base functions for the network is decreased significantly and the training method is just a linear least squares problem. Therefore, it can make the nonlinear process modeling method general and easy to use. This guarantees that the neural network has a good generalization ability in the condition of training data seriously being contaminated by noise. Both the numerical simulation and the practical application show that the proposed identification method has good accuracy and generalization ability.