An improved dimensionality reduction algorithm for tunnel data based on KPCA

摘要

Dimensionality reduction of tunnel data is an important part of data mining in urban highway tunnel. It makes significant sense for tunnel maintenance, traffic safety management, traffic status analysis, and so on. However, the sample size is very large and the correlativity of data elements is complex, what makes these general methods, the principal component analysis (PCA) and kernel principal component analysis (KPCA), cannot achieve its dimensionality reduction work effectively. Therewith, we propose a two-step algorithm named as "KPCAT" which is based on KPCA for dimensionality reduction of tunnel data in this paper. The KPCA T consists of two procedures. At first, utilize the least square to compress samples by taking advantage of periodicity of tunnel data. And then, use the KPCA with an improved Gaussian radial basis kernel function to reduce the samples which have been compressed. We take the three methods, PCA, KPCA, and KPCA_T, to complete the dimensionality reduction experiment. Through the comparison experiment, the contrasting result proves the feasibility and effectiveness of KPCA T for dimensionality reduction of tunnel data with large size and complex correlativity.