Seite - 906 - in Book of Full Papers - Symposium Hydro Engineering

measured displacements and the n extracted principal components   n m are trained in the RBF neural network to make predictions. The main steps are shown in Fig. 1. 2.1. PRINCIPAL COMPONENT ANALYSIS All Generally, there are many related variables involved during the process, and too many input variables will increase the complexity of the calculation. PCA is a kind of analytic method that can transform massive factors into some concentrate ones (principal components). The new principal components are the linear combination of the original variables, which can reflect the information of the original data to the greatest extent. The main steps are as follow: (1) If there are a variables and each one has b groups of data, a two-dimensional matrix( )ij a bX  can be formed. Then the matrix ( )ij a bX  can be standardized to matrix ( )ij a bX   ; (2) Calculate the correlation coefficient matrix ( )ij b bR  of the standardized matrix ( )ij a bX   ; (3) Calculate the eigenvalues i 1,2 . )( ..i p and corresponding eigenvectors il 1,2 . )( ..i p of the correlation coefficient matrix ( )ij b bR  . Arrange p eigenvalues i from the largest one to the smallest one; (4) Calculate the variance contribution rate and the accumulated variance contribution rate of each principal component; (5) Select the principal components according to the accumulated variance contribution rate. 906