76 results about "Coherence analysis" patented technology

The invention relates to the technical field of intelligent medical instruments, and discloses a closed-loop electro-acupuncture therapeutic apparatus based on heart and brain coupling information feedback. The device comprises a forehead electroencephalogram signal acquisition module, a single-lead electrocardio acquisition module, a Bluetooth transmission module, a signal preprocessing module, an ECG and EEG feature extraction module, a dynamic coupling analysis module, an embedded XGBoost classifier and an electroacupuncture control module. According to the system, collected EEG and ECG signals are wirelessly transmitted through Bluetooth, HRV indexes and EEG frequency band power spectral density are extracted after preprocessing, and frequency domain coherence analysis is carried out to obtain heart and brain bidirectional coupling characteristics. The embedded XGBoost classifier outputs optimal electroacupuncture stimulation parameters based on the characteristics, and the electroacupuncture control module generates corresponding bidirectional pulse waves for stimulation. According to the therapeutic apparatus, closed-loop feedback control is achieved, therapeutic parameters can be dynamically optimized, the individuation and precision level is improved, and meanwhile safety is ensured through impedance monitoring and electrical isolation.

The invention relates to the technical field of bioinformatics and clinical medicine, and discloses a multi-omics data association analysis-based tumor early screening and typing early warning system, which comprises a data acquisition and preprocessing module for integrating standardized longitudinal multi-omics data; the dynamics and topology analysis module is used for generating topology fingerprints representing dynamic behaviors of the system through state space reconstruction and persistent coherence analysis; the causal inference and risk assessment module is used for calculating critical moderation indexes in parallel to synthesize risk indexes and constructing a dynamic causal network; and a collaborative diagnosis and report generation module. According to the system, risk indexes derived by critical moderation, topological fingerprints and a dynamic causal network are creatively combined, multi-modal information fusion is carried out through a collaborative diagnosis unit, and finally a comprehensive early warning report is generated. According to the invention, the accuracy and reliability of early risk early warning of tumors can be obviously improved, and a mechanism-level traceability basis is provided for clinical intervention.

The invention discloses a frozen soil temperature feature recognition algorithm and a moisture and electric field state prediction model, and the method comprises the steps: synchronously collecting temperature, moisture, resistivity and meteorological data through arranging a multi-depth sensor array, and constructing a multi-source data set with time-space alignment; utilizing continuous wavelet transform and cross-modal coherence analysis, and combining a hidden semi-Markov model to extract a depth-periodic characteristic profile; a time domain fusion Transform model embedded with physical constraints is adopted to realize frozen soil multi-parameter physical consistency prediction and uncertainty quantification; and finally, outputting a prediction result through a visual interface and setting an automatic alarm. The problem of cycle identification caused by weak and non-stable deep frozen soil signals is solved, the prediction accuracy and the physical rationality are remarkably improved, an end-to-end solution from data quality control to early warning is formed, and the reliability and the practicability of frozen soil engineering monitoring are greatly enhanced.

The invention discloses an AI-based gene data analysis system. The system comprises a plurality of omics data matrixes; local association pattern mining is performed on the multi-omics data matrix through a 1D-CNN one-dimensional convolutional neural network, a topological structure of a gene network is identified through continuous coherence analysis, dynamic weights are allocated to sequence features and a topological feature matrix by using a dynamic attention mechanism, and weighted multi-scale feature vectors are output; establishing a multi-modal fusion model based on a Transform architecture to fuse the multi-scale feature vectors, performing fine adjustment on the adaptive disease data set by using the general genome feature of a pre-training model, and outputting a fused feature vector; and inputting the fusion feature vector into an MLP multilayer perceptron for disease risk prediction, generating a disease risk prediction index in combination with an SHAP algorithm, and generating an auxiliary decision scheme according to the prediction index. And the accuracy and generalization ability of disease risk classification are effectively improved.

The invention discloses a method for predicting the residual life of an existing asphalt pavement structure, particularly relates to the technical field of computer simulation and data processing, and is used for solving the problem that the existing prediction method cannot efficiently represent the non-uniformity of an existing pavement material and the dynamic environment load influence on the premise of ensuring the precision. Existing damage data of a road surface are acquired to construct a simulation model reflecting a real initial state, material fatigue parameters are acquired in combination with cross-scale analysis, a dynamic environment boundary is driven by utilizing climate time sequence data, and then time-frequency coherence analysis is performed on a damage mode and a climate load to identify a high-threat working condition, so that the high-threat working condition is identified. And finally, preferentially operating high-fidelity thermal-mechanical coupling numerical simulation aiming at a high-threat working condition, and mapping damage evolution data output by simulation to a macroscopic performance degradation process so as to determine the residual fatigue life.

The invention discloses a fuzzy theory fused tongue diagnosis characteristic quantitative analysis method and system, and relates to the technical field of traditional Chinese medicine tongue diagnos.The method comprises the steps that a micro piezoelectric sensor array and a hyperspectral imaging module are integrated, tongue surface Young modulus distribution data and hyperspectral reflectivity data are synchronously collected, and a multi-mode tongue picture fusion data set is generated; further performing crack region segmentation and super-resolution reconstruction through a neural network model, outputting a crack network diagram and a tongue papilla form entropy, generating topological fingerprints in combination with continuous homology analysis, identifying a mechanical gradient mutation region and extracting a metabolic index, constructing a mechanical-metabolic associated feature set, and obtaining a mechanical-metabolic associated feature set; topological fingerprints and mechanical-metabolic characteristics are fused in a three-dimensional space-time coordinate system, a dynamic intuitionistic fuzzy decision field is constructed, finally, tongue picture pathology evolution risk indexes and a visual early warning report are output through a pathology prediction model, and the system provides quantifiable decision support for'prevention before illness' in traditional Chinese medicine.

The invention discloses a defect repeated alarm screening method and system based on a twin network and topology analysis, and relates to the technical field of intelligent substations. The method comprises the steps of extracting image pair features through a twin network and generating matching potential energy; constructing an anti-environmental interference dynamic reference curved surface by using historical data; constructing a weighted coupling graph based on the potential energy residual error and the spatio-temporal context, and screening potential repeated alarm clusters through belief propagation; performing time sequence alignment and multi-dimensional consistency evaluation on intra-cluster alarms, and eliminating low-contribution noise nodes in combination with Shapley value game analysis; and finally, identifying a stable defect chain through topology persistent coherence analysis, selecting main alarms and screening out repeated alarms. According to the method, the problem of misjudgment caused by illumination change, equipment aging and complex interference is effectively solved, the recognition accuracy and the automation level are greatly improved, and the operation and maintenance rechecking workload is remarkably reduced.

The invention discloses a high-precision identification method and system for a noise source of an asynchronous motor, and relates to the technical field of asynchronous motor detection, and the method comprises the following steps: constructing a finite element model of a to-be-analyzed asynchronous motor, dividing the surface into a plurality of detection sub-regions, and carrying out the operation simulation of the finite element motor, analyzing vibration speed and sound pressure signal data of each sub-region, synchronously acquiring actual signals, calculating relative differences with simulation data, evaluating detection priorities by combining historical fault times, forming a detection sequence according to the priorities, and sequentially performing sound-vibration coherence analysis and coherence function calculation to obtain a sound-vibration coherence detection result; according to the method, noise frequency bands are classified into structural noise and air noise, order tracking analysis is performed on the structural noise obtained through classification, an amplitude stability index and an amplitude change gradient are calculated, automatic classification and identification of electromagnetic noise and mechanical noise are realized based on a preset threshold value, and the efficiency and accuracy of fault processing are remarkably improved.

The invention discloses a federated learning-based cross-regional flood risk prediction management system and method, and relates to the technical field of flood risk prediction, and the system comprises a cross-regional data collection module which collects multi-source flood observation data in a distributed manner and adds a credibility label; the coherence analysis clustering module is used for calculating coherence correlation coefficients among the monitoring stations based on the weighted Pearson's correlation coefficients and carrying out topological consistent clustering by combining physical connectivity and relation stability; the joint prediction analysis module is used for training a flood risk prediction model by adopting a federal learning mode based on a clustering result, fusing a statistical channel and a physical channel to carry out joint prediction, and judging and optimizing simulation input through a dominant boundary; and the decision write-back module is used for generating multi-level early warning and targeted scheduling suggestions according to the predicted value and writing back execution data to realize closed-loop optimization. The problem of insufficient flood prediction precision under cross-regional data isomerism and privacy protection is solved, and accurate, efficient and safe cross-regional flood risk prediction management is realized.

The invention discloses a transformer winding vibration abnormity diagnosis and early warning method based on lightweight edge calculation, and relates to the technical field of transformer monitoring. According to the method, vibration signals and magnetic flux signals of a transformer winding are collected in real time at edge nodes, and a basic data set is obtained after preprocessing; vibration and magnetic flux features are extracted to construct a basic diagnosis feature vector, a three-dimensional tensor is constructed through directional coherence analysis, a boundary enhancement structure map is generated, and a structure enhancement fusion feature vector is formed after fusion. The method comprises the following steps of: realizing abnormal grade and trend prediction by a lightweight Transform diagnosis model based on knowledge distillation and parameter compression optimization, and triggering model adaptive updating by combining risk index modeling and a dynamic risk threshold value. And finally, calculating an abnormal level posterior probability through a Bayesian updating rule, and outputting multi-level early warning. The method has the advantages of high precision, low calculation overhead and self-adaptive capability, and is suitable for online monitoring and intelligent operation and maintenance of the state of the transformer.

This invention belongs to the field of wind power equipment condition monitoring and fault diagnosis technology, specifically relating to an intelligent diagnostic device and method for blade icing based on acoustic-vibration feature fusion. It includes: a multimodal sensing unit for acquiring vibration response signals and aeroacoustic signals; a signal conditioning and synchronous acquisition unit connected to the multimodal sensing unit for conditioning and synchronous analog-to-digital conversion of the vibration response signals and aeroacoustic signals to obtain synchronized digital vibration and digital sound signals; and an edge computing and diagnostic unit connected to the signal conditioning and synchronous acquisition unit. This invention synchronously acquires and deeply fuses vibration signals reflecting the blade's structural state with specific frequency band sound signals reflecting the aerodynamic state of the blade surface, constructing an intelligent diagnostic system based on acoustic-vibration coherence analysis and machine learning models. This achieves high-precision, robust online diagnosis and early warning of blade icing conditions.

The invention belongs to the technical field of millimeter-wave radar target detection and signal processing, and particularly discloses a millimeter-wave radar multipath target suppression method in a tunnel environment, and the method comprises the steps: building a geometric position incidence relation between a real target and each-order multipath target in a distance-Doppler graph through employing a millimeter-wave radar echo signal in a tunnel; on the basis, coherence analysis and energy consistency verification are sequentially carried out on candidate signal point sets meeting geometric constraint conditions, so that real target echoes and multi-path interference signals are effectively distinguished, and the judgment accuracy of the multi-path signals is remarkably improved. Meanwhile, after a real target and a multipath signal are distinguished, a multipath signal space distribution estimation graph is constructed by utilizing the identified space position and amplitude information of a multipath signal point, and multipath signal components are eliminated from an original distance-Doppler graph according to the graph and a subtraction proportion, so that multipath interference is directly suppressed in a signal processing stage.

The invention relates to the technical field of bridge health monitoring, and discloses a real-time anomaly prediction and fault diagnosis method and system for a bridge component, and the method comprises the steps: collecting the real-time data and historical data of the bridge component, carrying out the modal parameter recognition of the real-time data, so as to obtain a modal parameter set of the bridge component, constructing a three-dimensional feature tensor of the bridge component based on the real-time data; and performing multi-head self-attention mechanism operation on the three-dimensional feature tensor to obtain a space-time correlation feature tensor of the three-dimensional feature tensor, performing correlation analysis on the space-time correlation feature tensor based on the historical data to obtain a real-time space-time feature tensor of the three-dimensional feature tensor, and storing the real-time space-time feature tensor of the three-dimensional feature tensor. And performing frequency domain spectral coherence analysis on the real-time spatial-temporal feature tensor and the modal parameter set to obtain a spectral coherence coefficient of the bridge component, and the method can perform accurate prediction and diagnosis on the abnormal event of the bridge component.

This invention discloses a data communication scheduling system for intelligent manufacturing production lines based on edge computing, belonging to the field of industrial IoT edge computing data scheduling technology. The system synchronously acquires the real-time load current discrete sequence of the target device node and the real-time signal-to-noise ratio sequence of the main communication path, extracting the second-order derivative zero-crossing discrete component and the first-order attenuation discrete component of the envelope. A cross-correlation algorithm is used to perform coherence analysis to derive a cross-domain healthy coupling index. Using this index as the independent variable, a weighted mapping of the real-time packet loss rate is performed through an exponential gain function to generate a virtual impedance factor. It determines whether the index exceeds a threshold to trigger backup gateway context pre-synchronization and calculates the positive offset. Finally, a normalization function is used to calculate the traffic allocation weight and configure path scheduling labels to achieve dynamic scheduling of multi-path traffic splitting actuators. This solution achieves a priori prediction of industrial communication link quality and refined multi-path traffic splitting scheduling.

The invention discloses a wavelet packet space coherence-based evaluation method for the curative effect of microacupotomy treatment on upper limb spasm. The method comprises the following steps of S1, collecting and preprocessing electroencephalogram signals and electromyographic signals of a post-stroke upper limb spasm testee; s2, performing time decomposition and frequency decomposition on the preprocessed electroencephalogram signal and the preprocessed electromyographic signal by utilizing wavelet packet transformation, and respectively generating an electroencephalogram time-frequency sub-band signal and an electromyographic time-frequency sub-band signal; s3, based on spatial regularization coherence analysis, establishing linear and nonlinear relationships between the electroencephalogram time-frequency sub-band signals and the myoelectricity time-frequency sub-band signals; s4, constructing a WPCOH algorithm in combination with wavelet packet transformation and spatial regular coherence analysis, and capturing a dynamic coupling relationship between the electroencephalogram time-frequency sub-band signal and the myoelectricity time-frequency sub-band signal at different time points and frequency bands; s5, extracting key features of the WPCOH algorithm according to the dynamic coupling relationship; the treatment effect is evaluated by comparing the changes of the key features before and after the microacupotomy treatment.

The invention relates to the technical field of artificial intelligence, and particularly provides a sentence-level illusion detection method, system and equipment for a long text and a storage medium, and the method comprises the following steps: carrying out data cleaning and standardization on the input long text, dynamically determining a sentence boundary based on semantic coherence analysis, and dividing the sentence boundary into a sentence sequence; a hierarchical context encoder is then used to generate a context enhanced semantic representation for each sentence fusing local and global information. On the basis of the representation, feature vectors are extracted from three dimensions of semantic anomaly in sentences, context coherence mutation between sentences and overall topic path deviation. And finally, inputting the feature vectors into a classifier to obtain the illusion probability of each sentence, and calculating a confidence score. According to the method, the defects of weak context modeling and insufficient coherence analysis in a long text in a traditional method are overcome, and the credibility and practicability of the generated content are remarkably improved.

The invention relates to the technical field of signal processing, and discloses an anti-electromagnetic interference processing method and system for a sensor, and the method comprises the steps: carrying out the multi-stage electromagnetic shielding processing of an original mixed signal, and obtaining an initial anti-interference signal; based on the high-order statistical characteristics of the initial anti-interference signal, performing adaptive blind source separation on the initial anti-interference signal to obtain a vibration response signal and an electromagnetic interference reference signal; performing time-frequency domain transformation on the vibration response signal, and identifying a dominant vibration mode component from the transformed vibration response signal; performing coherence analysis on the dominant vibration mode component and the electromagnetic interference reference signal to obtain a coherence coefficient matrix; based on the coherence coefficient matrix, performing cancellation filtering on the dominant vibration mode component and the electromagnetic interference reference signal to obtain an anti-interference vibration signal; performing phase delay compensation on the anti-interference vibration signal to obtain a final anti-interference vibration signal; the anti-electromagnetic interference efficiency of the sensor can be improved.

The invention relates to the technical field of communication, discloses an electromagnetic spectrum abnormal signal monitoring method and system based on double acquisition and comparison, and aims to solve the problems of asynchronous double-channel sampling and high abnormal signal omission ratio and false alarm rate caused by clock drift and physical path delay in the prior art. According to the method, a hardware-level self-adaptive clock synchronization compensation circuit is constructed, clock path delay is quantized and offset in real time before analog-to-digital conversion, and a first sampling clock and a second sampling clock which are accurately synchronized are generated; a dual-channel ADC is used for synchronously collecting the same radio frequency signal, and abnormal signals are judged and identified through spectrum coherence analysis and differential spectrum threshold. The system comprises a main clock unit, a self-adaptive synchronous compensation circuit, a dual-channel ADC module and a data processing comparison module. The picosecond-level synchronization precision is achieved, external time service is not needed, the missing detection and false alarm rate is remarkably reduced, and the method is suitable for high-reliability frequency spectrum monitoring in the complex electromagnetic environment.

The invention relates to the technical field of big data analysis, in particular to a financial risk early warning system driven by big data. The system comprises a data acquisition module, a feature topology construction module, a continuous coherence analysis module, a topology risk quantification module and a risk early warning module. The data acquisition module acquires time series financial data from a plurality of enterprise data sources; the feature topology construction module maps financial indexes into a high-dimensional space point cloud and constructs a simple complex; the continuous coherence analysis module generates a continuous coherence bar code through multi-scale coherence group calculation; the topological risk quantification module calculates the weighted life cycle of the bar code and converts the weighted life cycle into a topological invariant; and the risk early warning module generates an early warning signal by comparing the difference between the current topology invariant and the historical mode. According to the method, deep structure features in financial data can be effectively captured, and the accuracy and timeliness of risk early warning are improved.

The application belongs to the technical field of millimeter wave radar target detection and signal processing, and specifically discloses a millimeter wave radar multipath target suppression method in a tunnel environment. The method establishes a geometric position correlation relationship between a real target and each order of multipath targets in a range-Doppler diagram by using millimeter wave radar echo signals in the tunnel. On this basis, a candidate signal point set meeting a geometric constraint condition is subjected to coherence analysis and energy consistency verification in sequence, so that the real target echo and the multipath interference signal are effectively distinguished, and the discrimination accuracy of the multipath signal is significantly improved. At the same time, after distinguishing the real target and the multipath signal, the spatial position and amplitude information of the identified multipath signal point are used to construct a multipath signal space distribution estimation diagram. According to the diagram and in accordance with a reduction ratio, the multipath signal component is removed from the original range-Doppler diagram, so that the multipath interference is directly suppressed in the signal processing stage.

The application discloses a kind of tensor field direction constraint weighted fitting fracture prediction method and system.It includes: the original seismic data is carried out dip orientation filter processing, obtains filtered seismic data body;The filtered seismic data body is carried out eigenvalue coherence analysis, obtains coherence volume, and the coherence volume is carried out coherence enhancement processing, obtains enhanced coherence volume;Based on enhanced coherence volume, using tensor direction constraint vector ant colony tracking algorithm, automatically track fracture strike, obtain fracture tracking result;Based on fracture tracking result, construct gradient structure tensor field, use the direction information of gradient structure tensor field to carry out anisotropic Gaussian weighted fitting, obtain the quantitative description result of fracture.The application is through with tensor field direction information as unified constraint throughout fracture prediction whole process, realizes the quantitative description of small scale, low density fracture, and the coincidence rate of prediction result and disclosed fracture area reaches more than 80%, provides reliable basis for mine safety exploitation and reservoir evaluation.

The application provides a kind of electroencephalogram feature extraction method and device based on frequency domain analysis.The method first adjusts signal according to the characteristics of electroencephalogram and noise environment through dynamic filtering, then accurately captures the time-varying characteristics of the signal through energy distribution extraction on the frequency-time plane, and applies frequency domain coherence analysis technology to predict the functional connection between brain regions, while combining nonlinear dynamics theory to extract frequency domain nonlinear features of the signal.Finally, the extracted frequency domain features are screened by feature importance scoring method, and the target feature set is obtained to identify and classify electroencephalogram based on it.The application introduces dynamic filtering, frequency domain coherence analysis and nonlinear dynamics theory, solves the problems of insufficient non-stationary processing of electroencephalogram, ignoring time-varying characteristics and low accuracy and computational efficiency of feature extraction in the prior art.

This invention belongs to the field of power system monitoring technology and discloses a method, system, equipment, and medium for analyzing the icing state of transmission lines based on point cloud modeling and deep learning, thereby overcoming the limitations of existing methods in time series processing. The method includes: dividing the time series point cloud of the transmission line into multiple time subsequences; generating a topological persistence graph for each subsequence through voxelization and persistent coherence analysis; calculating the dynamic topological distance between subsequences based on the persistence graph; extracting topological features to construct a feature matrix, and fusing topological distance, feature space, and feature distribution three views to construct a similarity matrix; using multi-view joint learning to obtain a consensus subspace and performing subsequence clustering within this space to obtain cluster labels; combining physical context information and predefined rules to assign physical pattern labels to the clustering results, generating an icing state classification report and visualizing it. This invention achieves temporal topological quantitative analysis and highly robust state identification of icing morphology.

The invention relates to the field of ultrasonic detection, in particular to an ultrasonic echo feature extraction and obstacle recognition system based on deep learning, which comprises a signal acquisition and preprocessing module, a topology perception spectrum peak detection module, a deep learning feature fusion and recognition module and a system control and adaptability adjustment module, the core innovation of the system is that a topology perception spectrum peak detection module constructs a multi-scale topology space through a continuous coherence analysis unit and extracts persistent topology features, and a direction sensitive gradient calculation unit calculates topology gradients in horizontal, vertical and oblique directions. The topological weight mask generation unit generates a feature weight mask based on topological saliency evaluation, the system organically combines the topology theory and deep learning, normal form transformation from signal form recognition to structure recognition is achieved, the system improves the recognition rate in a low signal-to-noise ratio environment, and the system response delay is smaller than 20 ms.

The invention discloses a power equipment maintenance analysis system and method based on the Internet of Things, and relates to the technical field of power equipment state monitoring, and the method comprises the steps: collecting a historical infrared thermal image sequence of a transformer in a normal operation state, and extracting temperature data of a sleeve and a winding region to form a historical temperature sequence; coherence coefficients between the sequences are calculated through power spectrum density estimation and amplitude square coherence analysis, and a historical coherence coefficient set is established; determining a dynamic threshold interval through statistical analysis; infrared thermal images are collected in real time, a real-time coherence coefficient is calculated, and when the coherence coefficient exceeds a threshold value, automatic early warning is carried out; the system comprises four core modules, namely a historical data acquisition and processing module, a historical coherence analysis module, a real-time monitoring and analysis module and an early warning judgment and output module. According to the method, a coherence analysis technology is applied, a dynamic early warning mechanism is established by monitoring correlation characteristics of temperature changes of a transformer bushing and a winding area, and accurate recognition and intelligent early warning of early faults of the transformer are achieved.

The application provides a method and device for diagnosing abnormal gas-solid flow in a slanted pipe, electronic equipment and a storage medium, wherein the method comprises: acquiring pressure pulsations of fluids at three places, i.e., upstream, midstream and downstream, in the slanted pipe in real time, and performing data retrieval; determining a frequency domain signal of the pressure pulsations according to a time domain signal of the pressure pulsations retrieved by the data retrieval; performing coherence analysis on the pressure pulsations of the fluids at the upstream and midstream and the midstream and downstream in the slanted pipe based on the frequency domain signal of the pressure pulsations, and obtaining a first coherence coefficient and a second coherence coefficient respectively; determining whether wave transmission occurs in the slanted pipe according to a size relationship of the first coherence coefficient, the second coherence coefficient and a preset first fixed value; and if the wave transmission occurs in the slanted pipe, performing causality judgment based on the time domain signals of the pressure pulsations of the fluids at the three places, i.e., the upstream, the midstream and the downstream, in the slanted pipe, and determining a flow direction of the wave in the slanted pipe.

This invention discloses a BIM-based construction method for large-span post-tensioned bonded prestressed beams, comprising: acquiring vibration signals of the working reinforcement and reference signals of the doubling reinforcement; inputting the doubling reference signals into an adaptive filtering algorithm with phase-preserving constraints to perform noise cancellation processing on the vibration signals of the working reinforcement, generating purified pressure wave signals; performing multi-scale decomposition on the purified signals, and extracting the propagation delay of the direct wave based on coherence analysis between components, generating a phase-delay feature matrix; calculating the real-time stress distribution field based on the geometric parameters of a pre-configured BIM 3D model, and synchronously updating it to the BIM 3D model. This invention solves the phase distortion problem under noise interference through physical reference channels and phase-preserving filtering, and improves the accuracy of time delay extraction through multi-path separation technology, achieving high-precision and robust real-time monitoring of prestress.

The invention provides a sound source positioning method and system based on a multi-microphone array, and the method comprises the steps: carrying out the preprocessing of a first sound source signal collected by a microphone array, and outputting a second sound source signal and an acoustic physical parameter; performing time-frequency decomposition on the second sound source signal to obtain a plurality of first time-frequency regions; performing acoustic coherence analysis on the first time-frequency region according to the acoustic physical parameters to obtain a time-frequency coherence matrix; detecting a second time-frequency region dominated by a single source from the time-frequency coherence matrix based on sound wave multipath and diffraction characteristics, and constructing an output single-source time-frequency point index according to the second time-frequency region; generating a plurality of direction-of-arrival estimation values according to a preset frequency domain graph signal and the single-source time-frequency point index; and performing clustering and weighted fusion on the plurality of direction-of-arrival estimated values to obtain direction-of-arrival information of the sound source.

The application discloses a rail vehicle complex sound source decoupling separation method, comprising the following steps: collecting static noise outside a preset distance of the whole rail vehicle; calculating the sound power of the whole rail vehicle static noise source according to the static noise; testing the noise and vibration of the motor and the gear box; carrying out coherence analysis on the vibration and noise of the gear box and the motor, obtaining the frequency spectrum value of the coherence coefficient of the motor and the gear box, and obtaining the sound pressure level of the gear box and the motor noise respectively; combining the dynamic passing noise test result outside the vehicle, the sound power of the whole rail vehicle static noise source, and the sound pressure level of the gear box and the motor noise, and obtaining the sound pressure level of the wheel-rail noise through a whole vehicle noise Vaone calculation model; respectively calculating the sound power of the whole rail vehicle static noise source, the wheel-rail noise, and the passing noise outside the vehicle when the noise of the motor and the gear box acts alone, and obtaining the contribution degree of various noise sources to the noise outside the vehicle. The application can improve the abnormal vibration noise tracing analysis efficiency after the whole vehicle is assembled.

The invention relates to the technical field of data processing, and discloses a large model desensitization and compliance verification method and system for private data development, and the method comprises the steps: carrying out the context coherence analysis of original private data, and obtaining structured representation data; performing multi-granularity privacy risk assessment on the structured representation data to obtain risk level features of the structured representation data; performing dynamic reversible privacy transformation on the structured representation data to obtain desensitized representation data of the structured representation data; performing cross-domain rule compliance verification on the desensitization characterization data to obtain a compliance verification report of the desensitization characterization data; based on the compliance verification report, performing accurate statistical characteristic compensation on the desensitization characterization data to obtain a training sample set of the desensitization characterization data; performing importance weighted coding on samples in the training sample set to obtain an input sequence adaptive to a large model architecture; according to the method, the efficiency of large model desensitization and compliance verification for developing private data can be improved.