Aiming at the problem of low accuracy in shock environment prediction using neural networks because of the strong nonlinear characteristics of ship underwater explosion shock,a method based on principal component a-nalysis is used to improve accuracy by downscaling the input parameters of the network model.Using mathematical matrix eigenvalue extraction and matrix transformation,original data samples are subjected to dimensionality reduc-tion by principal component analysis and factor analysis.Then,the adapted network is selected for the fast forecas-ting of shock spectral values.The experimental results show that the selection of principal components mainly con-siders the size and decreasing trend of the eigenvalues,retains the eigenvalues of the steeply decreasing section,and analyzes the trade-offs of the eigenvalues of the transition section.Meanwhile,the implementation of the decor-relation and dimensionality reduction processing on the parameters can significantly improve the forecasting accuracy of the neural network.
维普
万方数据