论文部分内容阅读
Orthogonal Least Squares (OLS) is a general and powerful algorithm for solving the output layer weights of a wavelet network. In this paper, the Recursive Orthogonal Least Squares (ROLS) method is used to orthogonalize the wavelet regressors. With the result of ROLS method, it is possible to compute which wavelets are important, and which are redundant and can be eliminated from the wavelet network. A structure identification algorithm is carried out based on OLS for the reduction of network. Akaike ’s Information Criterion (AIC) is introduced in the process of structure identification to seek a compromise between network complexity and accuracy. The final network models obtain acceptable accuracy with a relatively small number of significant wavelets. Numerical example is given to illustrate the effectiveness of the method mentioned above.