Method Of Structural Characteristic Parameter Preprocessing Based On SVM

The subject represented in this paper stems from the National NaturalScience Foundation of China and China Academy of Engineering Physics MutualFunds under Grant (NSAF)“Research on the identification of the mechanicalstructural characteristic parameters and its prediction based on kernel function”(NO.11176027).This project is mainly to predict the health status of mechanicalstructure mechanical structure about weapon equipments in long-term storage.This paper is a part of the NSAF, which is study on the methods of structureperformance preprocessing.The main tasks are summarized as follows.(1) This paper dealed with machine learning, kernel function and the kernelfunction construction. It focuses on the kernel function construction aboutcomplex Gaussian wavelet kernel function. After that, this kernel machine modelbased on SVM was built. The experimental results verify its correctness in fieldof mechanical structure performance preprocessing．(2) This paper studies the steps of mechanical structure performancepreprocessing: wavelet packet de-noise, data normalization, feature extractiontechnologies using PCA, multi-classification algorithms based on SVM. Afterthat, this paper provides reference about the necessity of the existence aboutevery step.(3) This paper focuses on the one-versus-one multi-classification algorithm,the secondary segmentation method of the one-versus-one multi-classificationalgorithm based on SVM was first proposed. The experimental results verify thatthe improved algorithm increase the classification accuracy about SVM.(4) The research under the background of NSAF, This paper has designedand built the experiment platform，which it can simulate the health status of themechanical structure performance in the real environment. We could obtain thedata of the mechanical structure performance from the simulation experimentplatform.And then,these data could be preprocessing using the improved kerneland the improved one-versus-one multi-classification algorithm,which has beenproposed in this paper.After that, the experimental results,which from the threerespects including dfferent kernel function,dfferent multi-classification algorithm and the regression forecasting,verifies the correctness of the research method inthis paper.