59results about How to "With noise immunity" patented technology

The invention relates to a sound source positioning device, applying sound waves to position through the match of a plurality of directions, which is a three-dimensional space sound resource object positioning device. The sound resource positioning device is composed of a hearing transducer, a sound resource signal microprocessor and an upper computer system, wherein the hearing transducer is a microphone array of a four-dimension regular tetrahedron structure; four microphones are respectively positioned on the four top points of the regular tetrahedron structure; the sound resource signal microprocessor is composed of a sound signal amplifier module, an analysis latch module and a communication transmission module; the upper computer system is composed of a positioning calculation method and peripheral equipment; and the hardware parts of each part and each module are all connected with each other through wires. The sound resource positioning device realizes accurate positioning to sound resources at any position in a three-dimensional space, is constructed into an all-around hearing scene, has the advantages of simple structure and higher precision, and has an anti-noise ability.

The invention is a seismic wave impedance inversion method, determining amplitude and phase spectrums of pos-stack seismic data, and extracting seismic subwaves on a near-well seismic channel and determining amplitude and phase spectrums, spectrum widths and master frequencies of the seismic subwaves; picking up seismic stratigraphic horizons corresponding to geological stratigraphic horizon and checking and correcting as well as smoothing and interpolating the picked horizons; demarcating logging acoustic time difference and density curves, and demarcating depth-domain logging acoustic time difference and density curves into time-domain ones, consistent with horizons of stacked seismic data, generating in-well wave impedances; making wavelet odd analysis and wavelet characteristic extraction on the stacked seismic data; providing high resolution wave impedance parameters for reservoir description in the exploration, development and recovery of oil fields.

The invention relates to an inversion method for a stratum elasticity constant parameter of pre-stack seismic data in geophysical prospecting for petroleum. The method comprises the following steps of: acquiring seismic data, and processing the pre-stack seismic data; obtaining logging data and extracting angle gather seismic data to form all stacked data; performing horizon calibration on the seismic data and the logging data; inverting the seismic data which is stacked at different incident angles to obtain a wave impedance parameter; calculating a corresponding reflection coefficient; generating a logging and seismic data pair; calculating a weight function and a model parameter of a function mapping network model, an impedance parameter and an elasticity constant; and drawing an elasticity constant profile for reservoir lithologic identification, petroleum-gas prediction, oil-water interface determination and petroleum reservoir description. By the method, the elasticity constant can be calculated by the conventional inversion of the seismic data and the logging data; and the method has certain adaptability to the inversion of faults and pinch-out zones, a wide application range, high resolution, high calculation speed, high stability, high calculation accuracy and the characteristic of certain anti-noise property.

The invention discloses a method for observing the longitudinal motion state of an aircraft by using for a time varying measurement delay output signal aircraft. The longitudinal motion state observation method comprises four steps, i.e. firstly, acquiring a kinetic model of the aircraft in a pitching plane; secondly, designing an observer; thirdly, adjusting an observation parameter and checkingthe observation effect; and fourthly, finishing the design. According to the longitudinal motion state observation method, the current time state of the system is reconfigured according to a model ofcontaining certain time varying delayed output and system, and thus the observation of the system state is realized. The observation of a longitudinal motion kinetic model of the aircraft can be realized and the observation of other control objects can also be realized. Therefore, the longitudinal motion state observation method has better practical value and favorable application prospect in thetechnical field of flying control.

The invention relates to a clustering-based positioning method for a single-phase grounding fault section of a distribution network. The method comprises the following steps of: performing the determination of a fault time and the self-synchronization of the positioning signals of each measuring point of a fault line by time-frequency atomic decomposition; extracting each frequency component and the corresponding energy of fault components of an alpha mode component of a transient phase current after the fault by using the frequency dividing capability of the time-frequency atomic decomposition, constructing a normalized characteristic quantity of the energy and the frequency; and classifying the normalized characteristic quantity of the frequency and the energy by using a K-means clustering method, and selecting the fault line and the fault section. The method provided by the invention can accurately perform line selection and positioning on five kinds of typical grounding faults suchas the small resistance grounding of a phase voltage over peak, the small resistance grounding of a phase voltage zero crossing, the high resistance grounding of the phase voltage over peak, the highresistance grounding of the phase voltage zero crossing, and a bus fault. The method has strong adaptability. The results are still accurate and reliable under the influence of electromagnetic noise,arc fault, arc suppression coil compensation, and other factors.

The invention provides an electrostatic damage protective device with a noise immunologic function, which comprises a detection circuit, an active drive circuit, a current discharge circuit and an alternating-current voltage division circuit, wherein the detection circuit comprises a capacitive device and a resistive device; the active drive circuit comprises a one-level or multi-level inverter, and can contain no inverter of the part of the circuit under a specific application; and the current discharge circuit comprises an N-type transistor or a P-type transistor. The electrostatic damage protective device also comprises the alternating-current voltage division circuit which is connected to two ends of the resistive device of the detection circuit in parallel. The electrostatic damage protective device pre-adjusts the input end voltage of the active drive circuit when the electrostatic damage protective device is influenced by electrostatic fluctuations by using the voltage division action of the alternating-current voltage division circuit under an alternating-current environment so that the current discharge circuit cannot be turned on by mistake, and the electrostatic damage protective device has the noise immunologic function.

A rapid early-warning system based on a phase-sensitive optical time-domain reflectometer includes a disturbance monitoring and positioning module, and a disturbance signal identifying module. The disturbance monitoring and positioning module: a disturbance monitoring part performs difference on the beat signal intensity of a phase-sensitive optical time-domain reflectometer, the rapid early-warning system based on the phase-sensitive optical time-domain reflectometer can monitor the difference of the beat signal intensity all the time, and once the difference of the beat signal intensity changes, it means that a disturbance signal generated by disturbance occurs; and a positioning module can determine position information of a disturbance point through the time difference of an optical time-domain reflectometer echo. The disturbance signal identifying module includes a pre-treatment module, a feature extraction module, a template training module, a pattern classification module, and a post-treatment module; template training means that templates of a plurality of types of signals can be constructed according to features of a series of different types of disturbance signals; the features of the disturbance signals and preset templates of the plurality of types of signals are matched, and the types of the signals can be determined according the matching degrees; and a warning measure or a no-warning measure can be preformed according to a result of pattern classification.

The invention provides a bottom-to-top focused information extraction method based on the vision focusing research results of psychology. The bottom-to-top focused information is formed by significances of the corresponding zone of each point of an image and the size of the zone automatically adapts to the complexity of local features. The new significance measurement standard comprehensively takes three characteristics, i.e. local complexity, statistical dissimilarity and primary vision features into account. The significant zones simultaneously appear significant in both feature space and scale space. The acquired bottom-to-top focused information is provided with rotation, translation and scaling invariance invariability and a certain anti-noise capability. A focusing model is developed from the algorithm and the application of the focusing model to a plurality of natural images demonstrates the effectiveness of the algorithm.

The invention discloses a method for detecting and recognizing the tying of a safety belt of a driver based on interval-maximizing and multi-sample dictionary learning. The method comprises the following steps: treating a driver-containing data-concentrated part of a windshield of a left half car as an image to be processed for input; acquiring a feature vector representation of each image; performing dictionary learning process for a training data set by an interval-maximizing and multi-example manner; respectively training an SVM (Support Vector Machine) multi-classifier for each type in the training data set after clustering to obtain a classifying model; coding the training data through the acquired dictionary; training the classifier through the coded feature vectors; detecting and recognizing an image to be recognized through the trained classifier, so as to determine whether the safety belt is tied from the image to be recognized. According to the recognizing method and system for the safety belt, the safety belt can be simply and easily detected and recognized, the popularization capacity is high, the detection and recognition accuracy is high, and the speed is fast; in addition, the influences of illumination and noise and other adverse factors can be effectively overcome.

The invention discloses a depth wavelet extreme learning machine-based multi-level inverter fault diagnosis method. The method comprises steps: the inverter AC-side voltage fault signals under variousworking states are acquired, acquired three-phase AC voltage is intercepted and combined to a piece of data, a total sample data set is obtained, the total sample data set is normalized, a training set, a verification set and a test set are generated, a deep feature extraction and fault diagnosis model is constructed the test set is used as to-be-tested samples, and the deep feature extraction and fault diagnosis model is used to diagnose a modular multi-level inverter fault. A wavelet extreme learning machine self-encoder formed by a wavelet base, an extreme learning machine and a self-encoder has the advantages of good feature extraction effects and quick training speed, and a sparse coefficient is added to the wavelet extreme learning machine self-encoder realization process, and the depth wavelet extreme learning machine has good anti-noise performance.

The invention discloses a large-scale rotary unit fault diagnosis method and device, and the method comprises the steps that an upper computer obtains user selection information, judges a diagnosis model needing to be called according to the user selection information, carries out the fault diagnosis through the called diagnosis model, and the diagnosis model comprises a knowledge graph inferencemodel and a random forest inference model; the knowledge graph model constructs a fault diagnosis knowledge graph of the large-scale rotary unit by utilizing the relationship among fault symptoms, a fault mechanism, fault reasons and solutions, and directly completes fault diagnosis from the fault symptoms according to real-time data acquired by a sensor; a decision tree model is built by the random forest reasoning model, and thus inputting real-time data acquired by a sensor into the decision tree model to determine a final diagnosis result; the invention has the advantages that the method is suitable for fault diagnosis of the large rotary unit, the diagnosis result is relatively accurate, and the diagnosis efficiency is high.

Disclosed is an inversion method of a waterland detector data seabed reflection coefficient. The method is characterized by comprising the following steps: exciting and acquiring water detector and land detector seismic data by use of a seismic artificial epicenter, and performing preprocessing; calculating Fourier transformation of water detector data and land detector data; calculating an upgoing wave field and upgoing wave field first-order and second-order delay data; calculating the upgoing wave field and upgoing wave field first-order and second-order delay time sequence data; calculating a seabed reflection coefficient characteristic equation coefficient; constructing and solving a seabed reflection coefficient characteristic equation; calculating upgoing wave data energy; determining an optimum seabed reflection coefficient; and drawing a seabed reflection coefficient profile and storing a seabed reflection coefficient in a seismic data channel head. According to the invention, merging processing on the water detector data and the land detector data is realized, seawater ringing multitime wave interference influences in seismic data are eliminated, the signal-to-noise ratio and the resolution of the seismic data are effectively improved, and a certain anti-noise capability is exhibited.

The invention discloses a method of fast phase correction processing of seismic data. The method adopts a Kaiser window Hilbert factor sequence, confirms a Hilbert seismic sequence and directly calculates and confirms a best phase correction factor the through estimation of phase information of residual wavelets, the phase correction processing the is carried out on seismic data records, the fact that influences of the residual wavelet phases are eliminated is achieved, so that the residual wavelet phases reach or are close to a zero phase so as to effectively improve the resolution ratio of the seismic data. The method of the fast phase correction processing of the seismic data only processes the phase of the seismic data wavelets, does not change amplitude spectrum of the data and has the advantages of being little in calculated amount, high in computation speed, good in stability and high in computational accuracy of the correction processing of the phase and has a certain anti-noise property.

The invention relates to a planar object detection method based on a continuous edge characteristic, and belongs to the technical field of machine vision image processing. The method comprises the steps including extracting the continuous edge characteristic, establishing a target contour model, extracting candidate objects based on a Hough transformation idea, screening the candidate objects and realizing accurate positioning. The method is suitable for an industrial scene, and can be used to establish a model of an unknown planar object rapidly and realize object detection. The method needs not to be adjusted for the specific object, and is suitable for different planar objects (as well as objects whose surfaces are planar), and the problem that the university of a present visual system is low is solved to certain extent. The detection precision is higher, the processing speed is high, and certain anti-noise capability is provided.

The method and system for speech identification is disclosed, which picking up vector sequence of characteristic parameter from speech signal; obtaining at least one candidate word after speech identification; segmenting speech signal into speech segmentations corresponding to verification unit; determining vector sequence of characteristic parameter of speech segmentation for verification; carrying out verification to generate score of verification. In the method, score of verification is calculated by using quasi-neural network for verification units. The quasi-neural network is perceptive from multiple layers. Score of verification is obtained by feedforward operation under comparison between vector sequence of characteristic parameter for verification and multiple layers of perception. Adding scores of verification of all speech segmentations obtains score of verification for word. Determining whether the candidate word is accepted or rejected based on defined threshold level.

The invention relates to a sound source positioning device, applying sound waves to position through the match of a plurality of directions, which is a three-dimensional space sound resource object positioning device. The sound resource positioning device is composed of a hearing transducer, a sound resource signal microprocessor and an upper computer system, wherein the hearing transducer is a microphone array of a four-dimension regular tetrahedron structure; four microphones are respectively positioned on the four top points of the regular tetrahedron structure; the sound resource signal microprocessor is composed of a sound signal amplifier module, an analysis latch module and a communication transmission module; the upper computer system is composed of a positioning calculation method and peripheral equipment; and the hardware parts of each part and each module are all connected with each other through wires. The sound resource positioning device realizes accurate positioning to sound resources at any position in a three-dimensional space, is constructed into an all-around hearing scene, has the advantages of simple structure and higher precision, and has an anti-noise ability.

The invention discloses a white light interference micro-profile restoration method based on cross-correlation computation. The method comprises the following steps of reading a group of white light interference fringe images and system parameters, wherein the system parameters include center wavelength, spectrum width and scanning step length; for any one pixel point on the surface of a sample to be tested, the gray scale value change sequence of the pixel point in the group of white light interference images is taken and is recorded as a signal to be tested; then, a reference signal is generated according to system parameters and white light interference principles; a correlation function of the signal to be tested and the reference signal is determined; the obtained correlation function is subjected to peak detection to obtain the delay quantity of the signal to be tested and the reference signal; the delay quantity and the scanning step length are multiplied to obtain the relative height between the signal to be tested and the reference signal; further, the relative height of each pixel point is determined; the micro-profile of the surface of a sample to be tested is obtained through restoration. The white light interference micro-profile restoration method has the characteristics that the principle is single; the detection precision is high; the anti-noising capability is high, and the like.

The invention provides an HVDC power transmission line fault intelligent identification method. The method comprises the following steps: collecting fault current traveling waves after faults of a rectification side and an inversion side of an HVDC power transmission line; establishing internal and external fault feature vectors of a multi-scale S transformation fluctuation index region and fault pole feature vectors of multi-scale S transformation energy and ratios; establishing a combined feature sample vector capable of simultaneously reflecting the internal and external fault features and the fault pole features; establishing a random forest fault intelligent identification model; and identifying whether the current HVDC power transmission line suffers from internal and external fault features of the region or not by using the random forest fault intelligent identification model. According to the invention, fluctuation indexes are used for reflecting internal and external fault characteristics of the region, and energy and ratios are used for reflecting fault pole characteristics, a combined characteristic sample set is used as an input vector of a random forest intelligent algorithm, internal and external fault judgment of the region and fault pole selection are achieved through the same network, and threshold value setting is not needed.

The invention provides a method for evaluating full-polarized SAR isolation degree based on distributed targets. The method obtains a ground feature, reversely adds crosstalk to the actually measured scattering matrix of the selected ground feature, obtains the estimated value of the SAR crosstalk by using a correlation coefficient between channels, so as to reflect the isolation degree of the polarized channel of the polarized SAR. The method mainly solves the estimation of the polarization isolation degree in the full-polarized SAR quality evaluation, and comprises main steps of: selecting a ground feature, dividing the actually measured scattering matrix of the selected ground feature, namely data M to be evaluated, into N data blocks MN with the same size; reversely adding crosstalk [delta] to the data blocks MN, and computing the correlation coefficients [rho]HHHV, [rho] HHVH, [rho]VVHV, [rho]VVVH between the co-polarized channel and the cross-polarized channel under different [rho]; acquiring the added crosstalk values which minimize the correlation coefficients [rho]HHHV, [rho] HHVH, [rho]VVHV, [rho]VVVH for each data block; subjecting the 4N crosstalk values to statistical averaging to obtain the crosstalk estimated value, and taking the negative value of the crosstalk estimated value as the polarized channel isolation degree.

The invention provides a differential equation model-based gene regulation and control network construction method, and relates to the field of biotechnology and complex network science, and comprisesthe following steps: measuring gene expression in a cell cycle and under a specific state condition at different time points, observing change conditions of the gene expression at each time point orin the cell cycle, and finding a regulatory gene and a target gene; Acquiring time sequence data of gene expression, and discretizing the time sequence data into a plurality of expression levels by adopting a discretization method; utilizing An iterative algorithm of an ordinary differential equation, making corresponding assumptions according to the modeling purpose and problem analysis, mining the interaction relation between genes from large-scale gene expression data, and giving a topological structure model of the gene regulation and control network; Setting time delay of a gene regulation and control network; Graphically displaying the gene regulation and control network and carrying out structure analysis; And carrying out experimental verification on the gene regulation and controlnetwork of the target gene regulation and control relationship. Through data initialization processing, the defects of high complexity and large calculation amount are weakened.

The invention discloses a voice recognition method based on a CLDNN+CTC acoustic model, and the method comprises the steps: 1, obtaining a real-time voice signal, carrying out the feature extraction of the voice signal, and obtaining a frame-by-frame acoustic feature sequence; 2, taking the acoustic characteristic sequence as the input of a CLDNN+CTC acoustic model, and outputting a phoneme sequence; and 3, establishing a decoding model for converting the phoneme sequence into the character sequence, taking the phoneme sequence as the input of the decoding model, and outputting the character sequence through the decoding model. The method is a voice recognition method based on two-stage end-to-end (seq2seq), and comprises an end-to-end model of a 'voice-phoneme sequence' and an end-to-endmodel of a 'phoneme sequence-character sequence', the two models are different from the end-to-end model of an existing 'voice-character sequence', the two models do not need super-large-scale corpustraining, the advantages of the two parts can be complementary, and a language model can make up for the deficiency of an acoustic model in a noise environment.

The invention discloses a power distribution network power quality disturbance detection method based on EWT and MFDE, and the method comprises the steps: extracting a PQ disturbance signal in an active power distribution network system, and carrying out the preprocessing; using empirical wavelet decomposition (EWT) to perform modal decomposition on the preprocessed PQ disturbance signal to obtain an intrinsic mode function (BLIMF) containing feature information; using the intrinsic mode function BLIMF containing feature information as the input of a multi-scale oscillation dispersion entropyMFDE algorithm, employing the multi-scale oscillation dispersion entropy MFDE algorithm to carry out the dispersion entropy calculation of the intrinsic mode functions BLIMF obtained through mode decomposition, and obtaining the multi-dimensional entropy vector of the PQ disturbance signal under each intrinsic mode function BLIMF through calculation; carrying out PCA dimension reduction accordingto the obtained entropy vector, and then taking the result as an input quantity of an SVM algorithm; and carrying out PQ disturbance signal identification on the active power distribution network system containing the distributed energy. According to the method, each disturbance in the composite power quality disturbance can be accurately detected and classified, the classification is accurate, and the method has certain anti-noise capability.

The invention discloses a noisy image edge detection method based on a Scharr operator. The method comprises the following steps: step 1, inputting a noisy image; 2, improving a median filtering valuing method to obtain a denoised image; 3, equalizing the image to obtain a gradient enhanced image; step 4, improving the Scharr edge detection operator to obtain improved edge detection operators of the 45-degree direction template and the 135-degree direction template; step 5, determining edge points to obtain a preliminary edge image; step 6, fusing the edge images to obtain a fused image; and step 7, outputting a final edge detection image. According to the invention, the Scharr operator is improved; a two-direction template of a traditional Scharr operator is expanded to four directions, the image gradient is enhanced through image histogram equalization, the anti-noise capability and the edge detection capability of an image are enhanced through image noise reduction, a proper threshold value is found through image binarization, and the edge detection effect of the image is further improved.

The invention discloses a correction method for eliminating seismic data remaining wavelet phase. Pure phase correction factor is used for approaching a part of phase which is changed slowly with frequency, the pure phase correction factor is estimated, effects of the remaining wavelet phase are eliminated, phase correction process is carried out on seismic data record, the effects on the remaining wavelet phase are eliminated to enable the remaining wavelet phase to reach or approach zero phase so as to effectively improve seismic data resolution ratio. According to the correction method, only the phase of the seismic data remaining wavelet is processed, the amplitude spectrum of the data is not changed. Optimum phase correction factor a can be directly calculated and determined. The correction method is small in calculating quantity, fast in calculating speed, good in stability and high in calculating accuracy, and has a certain anti-noise property. The seismic data resolution ratio is improved.

The invention discloses a voice recognition method based on a DBLSTM+CTC acoustic model, and the method comprises the steps: 1, obtaining a real-time voice signal, carrying out the feature extractionof the voice signal, and obtaining a frame-by-frame acoustic feature sequence; 2, taking the acoustic characteristic sequence as the input of a DBLSTM+CTC acoustic model, and outputting a phoneme sequence; and 3, establishing a decoding model for converting the phoneme sequence into the character sequence, taking the phoneme sequence as the input of a decoding model, and outputting the character sequence through the decoding model. The method is a voice recognition method based on two-stage end-to-end (seq2seq). The method comprises an end-to-end model of a 'voice-phoneme sequence' and an end-to-end model of a 'phoneme sequence-character sequence', the two models are different from the end-to-end model of an existing 'voice-character sequence', the two models do not need super-large-scalecorpus training, the advantages of the two parts can be complementary, and a language model can make up for the deficiency of an acoustic model in a noise environment.

The invention discloses a manufacturing technology for a building module unit. The manufacturing technology comprises the following steps of (1) light-weight slurry preparing, (2) base material preparation, (3) frame welding, (4) transverse beam arranging, (5) reinforced steel mesh arranging, (6) light-weight slurry filling, and (7) surface processing. The manufacturing technology is simple, the building module unit obtained through the manufacturing technology has the advantages of being light in weight, capable of saving energy and environmentally friendly; filling is conducted by adopting prepared light-weight slurry, a house is firmer, and the energy-saving environment-friendly effect is achieved better; meanwhile, by adding multiple kinds of powder, the light weight, heat preservation, fireproof, waterproof and noise prevention functions are achieved; and by adding landscaping waste fibers, energy saving and environment protection are achieved, the production period is short, andthe manufacturing technology has the wide market prospects.

The invention discloses a converged communication scheduling system supporting routine inspection and an implementation method thereof, the converged communication scheduling system supporting routineinspection comprises a scheduling station and terminal equipment, wherein the terminal equipment comprises a telephone set arranged on a scheduling site, and a mobile phone, and a camera device carried by a worker the scheduling station comprises a call module and a notification module; the call module comprises an active call module and a passive call module; the active call module is used for calling one or more equipment codes at the same time; an active pushing module further comprises an active pushing module, an alarm module and a forced insertion module; the passive call module is usedfor receiving incoming calls of one or more pieces of terminal equipment; the passive call module comprises a call forwarding module and an on-duty call setting module; the terminal equipment comprises an routine inspection module which is used for selecting and / or shooting a picture according to the need of the worker, receiving character inputs related to the picture, and uploading the pictureand characters to a scheduling server. According to the invention, the general converged communication scheduling system which is comprehensive in function and low in system expenditure and supports routine inspection can be realized.

The invention discloses a foreground target detection method in a complex scene of a water conservancy project. The method comprises the following steps: S1, defining a monitoring area and carrying out image noise reduction preprocessing; S2, performing image background modeling and graphic operation processing; and S3, adopting the accumulated foreground image. The method has the advantages thatcompared with traditional manual inspection and pure video monitoring, the method can effectively utilize a large number of monitoring devices and monitoring video images, the labor cost is reduced, and the overall safety protection automation level of a water conservancy project site is improved; and 2, the accumulated foreground image is adopted, so that the influence of environmental noise on aforeground target detection result is reduced to a certain extent, false alarms are reduced, and the detection efficiency is improved.

The invention provides a power transmission line fault type discrimination method and system based on a capsule network, and the method comprises the steps: obtaining a large amount of PMU (Phasor Measurement Unit) data, and carrying out the imaging of the PMU data, and generating a radar map; constructing a power transmission line fault type discrimination model based on a capsule network, dividing a training set and a test set, inputting the training set into the model for training, and extracting change features of graphical PMU data; inputting a test set into the trained model for testing, and classifying and outputting test results in the form of a confusion matrix; therefore, the fault type of the power transmission line is discriminated. Logical reasoning does not need to be carried out, a large amount of knowledge for describing protection system behaviors does not need to be introduced, and only the features corresponding to different fault types need to be extracted for the graphical PMU data. Modeling and model modification processes are simple and easy to operate, PMU data are graphical, computing resources are saved, the model training test speed is high, and the diagnosis accuracy is relatively high.

The invention discloses a power distribution network high-resistance grounding fault identification method and device, a terminal and a medium. The method comprises the steps: determining the number of target decomposition layers through wavelet transformation according to the sampling frequency of zero-sequence current data in combination with the corresponding relation between preset sampling frequency and the number of target decomposition layers. According to the invention, feature extraction and feature screening are achieved, screened fault features are classified through a semi-supervised learning algorithm classifier, high-resistance grounding faults and common disturbance events can be more reliably distinguished under the condition of a small number of labeled training samples, and a certain anti-noise capability is achieved, so that effective recognition of the high-resistance grounding faults is achieved, and the technical problem of high difficulty in high-resistance grounding fault identification in the prior art is solved.