Orthogonal polynomials neural network for function approximation and system modeling
Chu Kwong Chak, Gang Feng, Chi Ming Cheng · 1995
By using a series of orthogonal polynomials, the architecture of a neural network can be developed for function approximation and system modeling. Due to the orthogonality properties, the regression matrix for parameter estimation is not of column degeneracy and the magnitude of the estimated parameters is small. This makes the proposed neural network useful in practical applications. The orthogonal least squares technique is applied for parameter estimation and model structuring. The neural network can be constructed to meet some pre-specified root mean square errors in one pass. Some simulations are done to support and illustrate our approach.