Electronic Devices Based On Spectrum Features Anomaly Identification Technology Research

With the continuous research on electromagnetic spectrum,researchers found that electronic devices will release electromagnetic signals to the surrounding environment during the working process,and these electromagnetic signals are related to the composition or working principle of the devices,therefore,the method of working state identification based on electromagnetic information is derived.This method has the characteristics of relatively simple data collection and good generalization.However,because the traditional electronic equipment anomaly identification scheme is mainly through the setting of hard threshold by feature comparison method to identify the abnormal state of the equipment,the conditions are too harsh and can not make full use of all the spectrum data,resulting in low efficiency and poor accuracy of the identification of abnormal operating conditions.Based on this,this thesis proposes a scheme of electronic device anomaly identification based on spectrum features.By constructing the spectrum feature space of electronic equipment,the problem of equipment abnormality identification is transformed into the problem of classification of feature samples in the spectrum feature space,and combined with the relevant algorithms of machine learning,the abnormality identification model of electronic equipment is constructed with the electromagnetic emission spectrum of various states as the original data to complete the detection of equipment abnormal state.In the state identification of electronic equipment,the identification scheme proposed in this paper can efficiently and accurately identify whether the equipment is in abnormal working state and give reasonable speculation on the type of abnormality when it is abnormal,which is of great value to both practical engineering and scientific research.The main work of this thesis is as follows.The first part focuses on the electromagnetic emission characteristics of common electronic devices,and analyzes the typical spectrum features that can reflect the composition or working principle of the devices.On this basis,by mapping the electromagnetic emission spectrum of the equipment into a high-dimensional space,the problem of electronic equipment anomaly identification is transformed into a classification problem of sample points on the feature space,and the overall scheme of electronic equipment anomaly identification with support vector machine as the identification function is established by combining relevant algorithms of machine learning.In the second part,we firstly analyze the performance of each type of electromagnetic emission features on the spectrum by combining the typical electromagnetic emission features of electronic devices,and design the corresponding extraction scheme for peak,envelope and signal energy.Secondly,the extracted spectrum features are combined with the spectrum segmentation process and spectrum feature quantization to determine the dimensionality of the spectrum feature space and the significance of parameters in each dimension,to construct the spectrum feature space of electronic devices,and to complete the mapping from the electromagnetic emission spectrum to the sample points in the feature space.The third part analyzes the problems of sample points in the spectrum feature space and designs an optimization strategy for the feature samples.To address the problem of unbalance in the number ratio among sample points in the spectrum feature space,a SMOTE-Tomek Link-based data augmentation scheme is proposed to expand the sample points representing the abnormal operating state of the equipment,which effectively alleviates the sample unbalance problem existing in the spectrum feature space;to address the problem of useless or redundant information in some feature dimensions caused by the previous use of spectrum segmentation processing.The PCA-LDA-based feature parameter screening scheme is proposed to reduce the computational load of the system by reasonably reducing the dimensionality while retaining useful information,and to make the reduced dimensional data more conducive to the training of the recognition model;finally,in order to obtain an anomaly recognition model with excellent generalization capability for electronic devices,the F1-Score is selected as the model performance index and an automated parameter tuning scheme based on genetic algorithm is designed to effectively reduce over-parameter tuning.Finally,the F1-Score is selected as the model performance index and an automated parameter tuning scheme based on genetic algorithm is designed to effectively reduce the time consumption of hyperparameter tuning.The fourth part investigates the actual needs of electronic equipment anomaly identification,and applies the electronic equipment anomaly identification scheme to build an anomaly identification model specifically for blowers,and designs controlled experiments to illustrate the effectiveness of the data augmentation algorithm and feature parameter screening design,and then verifies the reliability of the overall scheme design.