Abstract:In order to improve the scientificity of power material subdivision and the rationality of demand forecasting, this paper constructs the power material subdivision model based on NPCA-SOFM algorithm with the material demand characteristic as the breakthrough point. Firstly, the non-linear principal component analysis (NPCA) is used to reduce the dimensionality of the index and the loss of information caused by the standardization of indicators. Afterwards, we use the SOFM neural network algorithm to cluster the principal components after dimension reduction. Finally, the validity of the method is verified with an example. The results show that the clustering performance of NPCA-SOFM neural network algorithm is superior to PCA-SOFM and SOFM algorithm alone, and the dimension reduction effect is more obvious, which can provide reference value for intensive management of electric material and enterprise operation decision.