Research On Monitoring Data Based On LCD And Manifold Learning

With the development of computer technology and artificial intelligence,feature extraction has become more and more important as an important link in the complete pattern recognition system.In the face of high complexity,non-stationary signal,selecting the appropriate feature recognition method is the key to mining effective information.Due to the existence of correlation and redundant information in the extracted high-dimensional feature space,the non-linear dimensionality reduction method of manifold learning reduction can reduce dimension and mine the intrinsic nature of data,achieving the visualization of data.Based on the fault diagnosis of train running gear and radar emitter signal recognition,this paper discusses the application of feature extraction and dimension reduction in signal processing area based on monitoring data.The main research is as follows:1.This paper proposes a feature extraction method combined local feature scale decomposition(LCD)with information entropy.Based on the bearing failure standard data set,this method extracts the multiple information entropy feature of ISC component obtained by LCD decomposition of the signal data to form fault feature vector.Simulation results verify the validity and feasibility of LCD information entropy feature in fault feature extraction.2.Aiming at the deficiency of LCD decomposition method,this paper proposes Ensemble Local Characteristic-scale Decomposition method(ELCD)using noise-assisted.Simulation of the improved algorithm can effectively suppress the aliasing phenomenon and has efficient algorithm efficiency.In view of the partial failure condition data of the lateral shock absorber,Extract ELCD multiple information entropy as feature vector.Since the original eigenvector contains a large amount of redundant information,the original high dimension vector is reduced dimension by using the LLTSA algorithm of manifold learning,and then use the Fisher ratio to evaluate the features before and after LLTSA dimension reduction.The experimenta results indicated that the features after dimension reduction has greater contribution to classification,that is,the dimensionality reduction algorithm can preserve the essential feature to the maximum extent,and using the feature analysis method combined with ELCD and LLTSA,lateral shock absorber partial fault condition recognition rate is higher.3.Aiming at the difference of the statistical variation of the instantaneous frequency variation caused by different modulation modes and noise of different radar emitter,a radar signal recognition method based on LCD and manifold learning is studied.this paper first calculates the time frequency curve of emitter signal by phase difference algorithm and builds hierarchical decision classifier model for modulation recognition to identify the modulation type of signal.Then,the radar signal is decomposed by ELCD and the Renyi entropy is extracted.The modulation recognition result,Renyi entropy and PDW parameters are composed of eigenvectors.Finally,using S-ISOMAP dimensionality reduction,SVM classification is used to identify the features after dimension reduction.Experimental results show that:1)The extracted features can effectively describe the pulse modulation rule of different signals,and the overall classification accuracy rate is 98.8%;2)The S-ISOMAP algorithm is used to reduce the dimension of the original feature vector and the classification of the obtained lower dimensional feature is better.