To address the issue that principle component analysis (PCA)shows a poor ability to acquire the characteristic of nonlinearity data, a kernel principle component analysis(KPCA) temperature point optimization method is proposed. Firstly, nonlinearity mapping function is introduced to map the input temperature data into the characteristic space,and a Gaussian radial basis is selected to be a kernel function. Secondly, inner product operation in characteristic space is transformedinto kernel function operation in input space, eigenvalues and kernel eigenvectors are found. Finally, a comprehensive independent variable is formed. According to an experiment conducted on a CNC machine center,and comparedwith the PCA model, RMSE and Maximum residual error reduces by 36% and 29%, respectively.KPCA can preferably acquire the characteristic of temperature data, and the prediction ability of KPCA model has an obvious improvement.