697 results about "Hybrid data" patented technology

The invention provides a network traffic anomaly detection model training method and device and a readable storage medium, and the method comprises the steps: extracting a traffic statistical feature vector according to original network traffic data, and generating an initial mixed data set; generating a confrontation disturbance sample output enhanced feature matrix based on the initial mixed data set; constructing a self-adaptive feature fusion rule based on the enhanced feature matrix, embedding asset association degree parameters into an attention calculation layer of a feature encoder, and outputting encoding features fusing threat intelligence; inputting the coding features fused with the threat intelligence into a pre-constructed initial detection model, generating false report and missing report correction labels based on the suspicious traffic fragments, and outputting an adversarial sample correction data set; and performing adversarial training on the initial detection model through the adversarial sample correction data set to obtain an incremental detection model for network traffic anomaly detection. According to the invention, the detection precision, the anti-interference capability and the real-time defense response capability of the detection model to novel attacks can be improved.

The invention discloses a semi-supervised medical image segmentation method based on eye movement guide mixed data enhancement. The method comprises the following steps: collecting fixation position information of a doctor when reading a medical image through eye movement tracking equipment; using a teacher-student network model and adopting a two-stage training mode to train the model; the method comprises the following steps: taking GazeMix as a data enhancement strategy in training, extracting a fixation area in a labeled or unlabeled image, and pasting the fixation area to the background of other unlabeled images to generate a mixed image data pair; in the first stage, a GazeMix data enhancement strategy is combined to pre-train a teacher network; in the second stage, a student network is trained, GazeMix generates diversified data, and meanwhile, a teacher network is pre-trained for pseudo tag generation; multi-scale gaze perception is adopted, and model perception information is extracted on different scales and used for perception loss calculation; and the optimizer iteratively optimizes the student network parameters, and updates the teacher network by using an index moving average method.

The invention discloses an intelligent fusion analysis system for multi-source heterogeneous data, and the system comprises a dynamic data collection module which is used for carrying out the data collection, and carrying out the processing of a collected mixed data flow; the semantic alignment module is used for constructing a domain ontology knowledge graph according to a preset scene target and carrying out semantic alignment and coordinate alignment on the collected data; the self-adaptive fusion engine module is used for fusing the collected multi-source heterogeneous data; the trusted computing module integrates a secure multi-party computing protocol and a homomorphic encryption algorithm to realize that data is available and invisible; and the intelligent decision-making module constructs a state action reward model based on reinforcement learning according to a preset scene target, and performs analysis and decision-making by using historical data and data acquired in real time. According to the invention, the capability and effect of data processing and decision support are improved.

The invention discloses a substation equipment rare defect simulation and identification method and system and a storage medium. The method comprises the following steps: constructing an equipment reference feature library; marking dynamic features of rare defects in historical inspection according to a spatial-temporal feature enhancement algorithm, generating a knowledge graph, and constructing a dynamic defect learning library; the method comprises the following steps: learning space association and environmental factor influence of defects and equipment through a bimodal generation network, and generating initial defect data matched with a weak area of the equipment; generating high-credibility defect fusion data through physical constraint-intelligent detection double screening; constructing a three-dimensional mixed data set, and screening high-quality training samples through a dynamic defect evolution algorithm and hierarchical cognitive evaluation; and constructing a multi-algorithm collaborative fine tuning network by using a federated learning framework, simulating and labeling defect information, and outputting a multi-dimensional identification prediction report. The method aims at solving the problems that the model is insufficient in rare defect recognition precision and lack of evolution prediction ability, and high-quality simulation of rare defect samples and high-precision recognition of the model are achieved.

The invention discloses a low-altitude unmanned aerial vehicle trajectory tracking and monitoring method based on a 5G-A communication sensing integrated base station, and relates to the technical field of low-altitude traffic management and communication sensing fusion, and the method comprises the steps: firstly collecting multi-source data such as a communication sensing fusion signal, environment interference and unmanned aerial vehicle attributes, and carrying out the alignment and packaging of a unified timestamp and a coordinate system into a synchronous data frame; then, deep fusion and anti-interference processing are carried out on the data frames, noise is filtered out, and pure fusion data is generated; and furthermore, real-time track calculation and motion trend prediction are carried out on pure data by utilizing multi-base-station cooperative calculation and prediction. Based on this, through a multi-target feature recognition and clustering separation mechanism, independent individual trajectories are accurately stripped from a complex mixed data stream, and compliance verification and anomaly judgment are performed on the trajectories in combination with an airspace rule base. In this way, the problems of signal interference and multi-target aliasing in a complex environment can be effectively solved, and therefore high-precision global tracking of the low-altitude unmanned aerial vehicle and real-time monitoring of abnormal behaviors are achieved.

The present application relates to a real-time three-dimensional (3D) tumor tracking method and apparatus, and a computer device and a storage medium. The method comprises: acquiring a moving image and a reference image of a tumor site; performing hybrid data enhancement on the moving image and the reference image, so as to obtain a 3D CT image that represents each respiratory phase, and a segmented moving image; projecting the 3D CT image by using a ray casting algorithm, so as to generate a 2D DRR at a corresponding moment; and inputting the 2D DRR into a trained tumor tracking network, outputting a 3D deformation field of the 2D DRR by means of the tumor tracking network, and making the 3D deformation field act on the reference image, so as to generate a predicted segmented tumor image of the 2D DRR at the corresponding moment. In the embodiments of the present application, tumor tracking is performed on the basis of a single-angle X-ray image, such that the radiation dose is greatly reduced, high-precision tumor tracking performed at a low dose is realized, and the tumor tracking speed is significantly increased.

The invention discloses a method for constructing an image super-resolution generation model. The method comprises the following steps: constructing a mixed data set containing various types of images; preprocessing the images in the mixed data set to obtain initial low-quality images and corresponding image text descriptions; determining a first initial learning framework, and generating a corresponding high-quality reference image according to the preprocessed initial low-quality image; training the first initial learning framework to obtain a first-stage repair model; determining a second initial learning framework, and generating a high-resolution image according to the random noise-added original image, a high-quality reference image generated by the first-stage repair model and image text description prediction; and training the second initial learning framework to obtain a target second-stage image super-resolution generation model. According to the method, a two-stage network framework model is designed and developed based on a potential diffusion model, rich prior of the diffusion model is utilized, and the generalization ability of the model is greatly improved.

The invention provides a mixed data precision matrix multiplication and addition unit and a calculation method, the matrix multiplication and addition unit comprises a calculation unit, and the calculation unit comprises a format division module, a multiplication array module, an addition tree module, an accumulator module, a normalization module and a shift register module. The calculation unit converts the first input matrix and the second input matrix into input data in a middle floating point format; executing parallel multiplication operation on the input data to generate an intermediate product result; performing index alignment and accumulation on the intermediate product result to generate an intermediate accumulated value; accumulating the intermediate product result and the value of the third input matrix in a form of accumulating an intermediate accumulated value, and outputting an accumulated result; and converting an accumulation result into a normalized result and outputting the normalized result. The format division module supports various precisions and converts data with different widths into an intermediate floating point format, so that other hardware units can be reused, and the problems that hardware resources are complex and different model reasoning scenes are difficult to meet are solved.

A method may include: providing one or more inputs to a hybrid data generator, wherein one of the one or more inputs is based at least in part on a wellsite location, wherein the hybrid data generator comprises a large language model, and wherein the large language model is based at least in part on a machine learning algorithm; utilizing an information handling system to generate a cement job design based at least in part on the one or more inputs and the hybrid data generator; performing at least a portion of a cementing operation based at least in part on the cement job design; and collecting at least one measurement from at least one sensor during the cementing operation.

The invention discloses a fuzzy density electricity consumption anomaly detection method based on granular ball calculation, and belongs to the field of electric power data analysis, and the method comprises the steps: carrying out the normalization processing of electricity consumption data with mixed attributes, and obtaining standard data; dividing the hybrid power utilization data into pellets according to the data distribution to obtain a pellet set; constructing a particle-ball fuzzy information particle set based on the particle-ball center; calculating particle ball fuzzy density and attribute importance; according to the particle-ball fuzzy density and the attribute importance degree, constructing a particle-ball anomaly score; obtaining an abnormal score of the sample through abnormal score mapping; and finally, judging abnormal conditions of all sample points according to a set threshold value. According to the method, an existing anomaly detection model based on the finest granularity sample is changed, a particle ball generation method of the mixed data is provided, the particle ball fuzzy density is constructed by using the fuzzy information granularity theory so as to describe the polymerization degree of the particle balls, and unsupervised anomaly detection of the mixed power consumption data is realized.

The invention discloses a defect classification and segmentation method and system based on unsupervised and weak supervised combination, and the method comprises the steps: employing a training data set only containing a defect-free sample, extracting the feature representation of the defect-free sample through a feature extractor, and storing the feature representation in a feature memory library; comparing the mixed data set without labels including defect and defect-free samples with the features in the feature memory bank to determine false classification labels; designing a class activation graph-based weak supervision network for extracting features in the image, a semantic segmentation network and a loss function thereof, and training the semantic segmentation network by using a training data set of a defect-free sample and pseudo-classification label data; according to the method, the advantages of weak supervision and unsupervised learning algorithms are combined, and efficient defect detection and positioning can be realized.

The invention discloses a semi-supervised learning flow field prediction method based on a Gaussian mixture model. The method comprises the steps of obtaining a flow field calculation data set through a CFD calculation method; dividing the data set into a marked flow field data set, an unmarked flow field data set and a verification set; according to the marked flow field data set, constructing a flow field prediction model based on deep learning; training the flow field prediction model to obtain pseudo-labeled flow field data; respectively training the mixed data set by using the flow field prediction model and a pre-constructed decoder model; obtaining a simple sample in the pseudo-labeled flow field data through a bipartite Gaussian mixture model based on a design loss function; adding the simple samples into the flow field prediction model in an equal proportion until the training is completed; and verifying the flow field prediction effect by using the verification set to obtain a modeling flow field. By effectively utilizing the unmarked data, the prediction capability of the model under the condition of processing an unknown flow field is greatly enhanced while the prediction precision is maintained.

The invention belongs to the technical field of data transmission and processing, and discloses a zero-copy-based multi-thread high-throughput data stream channel separation method and system, and the method comprises the steps: directly writing a mixed data stream into a front-end buffer region configured as an annular structure through a data receiving module by adopting direct memory access; then, a multi-thread processing module dynamically allocates a plurality of processing threads from a thread pool to separate channel data in parallel, each thread adopts a zero copy algorithm based on pointer offset, positions the channel data in a memory, creates pointer reference and associates the channel data to a corresponding rear-end buffer area, and logic separation is achieved without physical copy; and finally, the data storage module efficiently writes the separated data into persistent storage in an asynchronous I / O mode. According to the method, zero-copy, multi-thread parallel and two-stage dynamic buffering strategies are combined, the data separation efficiency is remarkably improved, CPU occupation and memory bandwidth are greatly reduced, and the real-time performance and stability of high-throughput data processing are guaranteed.

The invention provides a high-voltage circuit breaker voiceprint denoising method based on data enhancement and a storage medium, and the method comprises the steps: processing a collected original voiceprint data sequence, extracting stable and effective Mel-frequency cepstrum coefficient features, and constructing a two-dimensional feature matrix; a parallel mixed data enhancement strategy is adopted to generate a positive sample pair, and an encoder is trained in combination with a contrast learning mechanism, so that the representation robustness of the model under different voiceprint change conditions is improved. The method comprises the following steps: decomposing an original signal containing noise fringes into a plurality of modal components by using variational modal decomposition, extracting low-frequency effective components, introducing Gaussian white noise, constructing a corrosion target signal as a decoder training target, learning through a denoising automatic encoder, and finally outputting a denoised voiceprint feature signal. The method has the advantages of high robustness, high noise suppression capability, excellent feature expression capability and the like, is suitable for the field of online monitoring and intelligent diagnosis of the state of high-voltage circuit breaker equipment, and has good application prospect and engineering value.

The invention relates to the technical field of project intelligent management and control, and discloses a multi-agent power grid project intelligent monitoring, management and control and evaluation system and method, and the system comprises a mixed data resource library, a multi-agent system, an index calculation module and a process control module. The mixed data resource library stores structured business data and unstructured documents; the multi-agent system comprises an information extraction agent, an index design agent, a code generation and treatment agent and the like, works cooperatively, and converts a high-level natural language management and control rule into an executable code; the index calculation module is responsible for executing codes according to a scheduling strategy and calculating quantitative indexes; and the process control module automatically identifies risks and executes management and control operations such as process locking according to the index result and a preset threshold value. According to the invention, the complex business logic can be automated and coded, the accuracy and timeliness of risk identification are improved, and refined and prospective intelligent management and control of the power grid project are realized.

The invention discloses a mixed data encryption method and system based on multiple algorithm cores, and the method comprises the steps: receiving a mixed encryption request based on a first preset path, carrying out the encryption type analysis and data address analysis of the mixed encryption request, and determining a plurality of target encryption algorithm cores corresponding to an encryption type analysis result; sending a starting instruction to a plurality of target encryption algorithm cores based on a second preset path, and determining a multi-core encryption strategy corresponding to the hybrid encryption request; acquiring to-be-encrypted data based on a first preset path, a first preset data handling engine and the data address resolution result, and calling a plurality of target encryption algorithm cores to perform hybrid encryption processing on the to-be-encrypted data according to a multi-core encryption strategy to obtain a hybrid data encryption result, the delay of data encryption can be reduced, and the flexibility of a data encryption mode and the security of an encryption result can be improved.

The invention discloses a multi-language emotion recognition method based on semantic comprehension, which comprises the following steps: realizing cross-language emotion analysis through a three-stage innovation framework, and separating semantic and emotion features by adopting improved relative position coding; the method comprises the following steps of: constructing an anchor point vector by utilizing a Hofstep culture dimension, and realizing culture sensitive vector space alignment through a loss function; the emotional features are dynamically adjusted in combination with the power distance and personal difference, and the cross-culture adaptability is enhanced in cooperation with an emotional intensity quantification formula. In a SemEval-2023 task, the accuracy of the system is 21.6% ahead of the baseline at 87.3%, the recognition rate of the small language reaches 78.2%, and the culture misjudgment rate is reduced by 43%. And through a multi-language parallel corpus and a mixed data enhancement strategy, the reasoning speed is increased by three times while the model parameter quantity is reduced by 40%, and an efficient solution is provided for a cross-culture NLP task.

The invention discloses a back door defense method based on forgetting learning and implicit hypergradient. The method comprises the steps of assuming a back door label, generating a possible back door trigger corresponding to the label and performing parameter search, obtaining a static back door trigger, adding disturbance, enabling an optimizer to optimize the back door label in a direction deviating from a correct label, and adapting to a dynamic back door trigger. Carrying out probability superposition by using the balanced sample; according to the parameters of the static trigger and the dynamic trigger and the prediction probability accumulative distribution, calculating a comprehensive score of the backdoor tag, and calculating a target tag; information such as model features is extracted as a training set, and the trained backdoor detection model is used for model backdoor detection; constructing a mixed data set, and performing data enhancement on the mixed data set; and performing forgetting training and multi-stage implicit super-gradient training on the classification model by using the enhanced data set, and deleting the back door of the classification model. According to the method, the backdoor is effectively deleted from the neural network model, so that the safety and reliability of the model are ensured.

A method and system for solving a computational problem using a hybrid data processor comprising a classical computer and quantum computer the method comprising receiving or determining information on one or more quantum circuits defining operations of a parameterized quantum model of a target function, the target function representing an approximate solution to the computational problem, preferably one or more differential equations and one or more associated boundary conditions, the one or more quantum circuits comprising one or more quantum feature maps for encoding one or more classical input features associated with the target function into the Hilbert space of the quantum register, wherein the one or more feature maps include one or more unitary operators defining a time evolution over a Hamiltonian, preferably a generator Hamiltonian, applied to quantum elements of the quantum register, the Hamiltonian evolution being parameterized by a first set of variational parameters and the one or more classical input features; and, determining the target function based on the parameterized quantum model and a loss function associated with the computational problem, the determining including varying the first set of variational parameters to determine a set of eigenfrequencies of the Hamiltonian which is optimized for the computational problem, the set of eigenfrequencies defining quantum modes, which the parameterized quantum.

The invention provides a building energy consumption data generation and parallel prediction method and device based on a generative adversarial network, and a medium, and belongs to the technical field of building energy consumption prediction. The method comprises the following steps: collecting actual building energy consumption data as a training sample, including time sequence data and physical attributes of a building; constructing and training a generative adversarial network model; inputting the actual building energy consumption data into the generative adversarial network model to generate enhanced building energy consumption data; the actual building energy consumption data and the enhanced building energy consumption data are mixed to obtain mixed data, the mixed data are used for training a machine learning model, a building energy consumption parallel prediction model is constructed, and a future building energy consumption value is predicted; and evaluating the performance of the prediction model on the test set, and adjusting the model structure or increasing the number of training rounds to improve the prediction precision. According to the method, the problem of insufficient building energy consumption data is solved, the precision and reliability of building energy consumption prediction are remarkably improved, and powerful support is provided for building energy consumption management.

A hybrid, redundant, fail-safe architecture provides a unified fail-operational framework for autonomous agents operating across physical and virtual domains. The system employs a multi-modal data source suite, an adaptive hybrid data fusion module, and an intelligent decision-making module. A novel closed-loop interaction enables a health monitoring module that detects an incipient fault in a data source by monitoring ancillary performance metrics. Upon detection, the module generates a fault signature, including a quantitative prognostic estimate of a future failure time, and transmits it to an adaptive data fusion module. The fusion module proactively reconfigures its state estimation algorithm by decreasing reliance on the degrading data source in proportion to the prognostic estimate. This preemptive compensation ensures the system maintains a high-integrity environmental model and achieves true fail-operational continuity. The architecture is applicable to numerous embodiments providing a universal solution for proactive fault management and system resilience.

The invention discloses a temperature prediction method based on a graph neural network and a related device, and belongs to the technical field of cloud computing system data center thermal management, and the method comprises the steps: S1, collecting simulation environment basic information, and building a simulation model and simulation data of a data center machine room; s2, expanding the simulation data to obtain generated data, and merging the simulation data and the generated data to form a mixed data set; s3, constructing a graph topological structure of the data center machine room and a thermal prediction model of the data center machine room, and training the thermal prediction model of the data center machine room to obtain a trained thermal prediction model of the data center machine room; and S4, deploying the trained data center machine room heat prediction model to the data center, and inputting the real-time server power, the air conditioner set temperature and the indoor temperature into the trained data center machine room heat prediction model to obtain the server return air temperature. Not only can a complex heat transfer rule be captured, but also different space structures and dynamic changes can be adapted.

The invention provides a big data processing method and system for realizing mixed data analysis, and the method comprises the steps: receiving real-time data streams from different sources, and enabling the real-time data streams to comprise structured data and unstructured data; based on a data fusion logic in a pre-constructed dynamic data fusion framework, the structured data and the unstructured data are fused to obtain fused data, and the data fusion logic comprises a multi-level fusion logic and a dynamic weight adjustment mechanism; carrying out abstraction processing on the fused data to obtain a data view; based on the data view, defining an analysis task, and executing the analysis task to obtain an analysis result; according to the technical scheme provided by the invention, efficient, real-time and flexible mixed data analysis is realized, and the overall performance and support capability of big data processing are improved.

A method and system for solving a computational problem using a hybrid data processor comprising a classical computer and quantum computer the method comprising receiving or determining information on one or more quantum circuits defining operations of a parameterized quantum model of a target function, the target function representing an approximate solution to the computational problem, preferably one or more differential equations and one or more associated boundary conditions, the one or more quantum circuits comprising one or more quantum feature maps for encoding one or more classical input features associated with the target function into the Hilbert space of the quantum register, wherein the one or more feature maps include one or more unitary operators defining a time evolution over a Hamiltonian, preferably a generator Hamiltonian, applied to quantum elements of the quantum register, the Hamiltonian evolution being parameterized by a first set of variational parameters and the one or more classical input features; and, determining the target function based on the parameterized quantum model and a loss function associated with the computational problem, the determining including varying the first set of variational parameters to determine a set of eigenfrequencies of the Hamiltonian which is optimized for the computational problem, the set of eigenfrequencies defining quantum modes, which the parameterized quantum.

The invention relates to a digital pre-distortion method and system based on a hierarchical iterative neural network, and belongs to the field of digital pre-distortion of power amplifiers. According to the method, the pre-distortion process is decoupled into two stages of off-line generalization training and on-line specialization optimization. In the offline stage, a mixed data set containing multiple modulation modes and bandwidths is adopted to train an offline neural network, and a basic model with high generalization ability is constructed; in the online stage, aiming at a signal with a specific modulation mode and bandwidth, a single configuration data set is utilized to train an online neural network, and the accurate compensation capability of the model on a real-time input signal is enhanced. In practical application, a baseband signal is firstly subjected to targeted preprocessing through the online neural network, then is subjected to deep optimization through the offline neural network, and finally is input into the power amplifier PA. According to the method, through a generalization-specialization double-layer structure, the dynamic response capability and compensation precision of a digital pre-distortion system to a complex signal environment are remarkably improved.

The invention discloses a superconducting cable aging state evaluation system and method based on mixed data, and the system comprises a data analysis module, an aging evaluation module, a physicochemical association module and a report generation module which are in communication connection. Wherein the data analysis module comprises a simulation sub-module, a data acquisition sub-module and a data processing sub-module connected with the data acquisition sub-module, and in the aging state evaluation process of the aging evaluation module, superconducting cable system model parameters are output and calibrated through a physicochemical association model; and then an aging state evaluation fitting function is output through the data processing sub-module, and an aging state evaluation report is automatically generated by the report generation module. According to the method, the operation state of the superconducting cable is evaluated by analyzing the mixed data, aging analysis of the superconducting cable system under multiple working conditions is carried out, and decision support is provided for system maintenance.

The invention relates to the technical field of deep learning, and discloses a deep learning model training data enhancement method, which comprises the following steps: data preprocessing: preprocessing original training data, including image size adjustment, normalization and other operations, to ensure the consistency and stability of input data; gAN data enhancement: using a pre-trained generative adversarial network to generate an image sample similar to original data distribution; enhancing neural style conversion data: combining the content of the original image with images of different styles by using a neural style conversion technology to generate an image sample with a new style; and mixed data enhancement: mixing the image generated by the GAN and the image generated by the neural style conversion to form a final enhanced data set. The deep learning model training data enhancement method aims to combine the generation capability of the GAN and the style migration characteristic of neural style conversion, and can generate enhanced data with diversity and richness.

The invention discloses an energy robot-oriented model training method, and relates to the technical field of energy digitization, and the method comprises the steps: synchronizing a business database to a local metadata knowledge base, and carrying out the business feature labeling of a metadata table in the metadata knowledge base according to the business logic; establishing a business problem knowledge base based on the business requirements; constructing a semantic map model of the robot based on the metadata knowledge base; generating mixed data features based on the labeled metadata knowledge base and the business problem knowledge base, and performing multi-model cooperative training of robot intention recognition, intelligent dialogue and intelligent table splicing according to the mixed data features in combination with a semantic map model; performing distributed incremental training on each model of the robot in response to the user session log, and performing model migration according to containerization to complete deployment of the robot; the intention recognition ability and data retrieval and query efficiency of the robot on the text input by the user are improved, then the question-answer response efficiency of the robot is improved, and efficient business question-answer is achieved.

The invention relates to the technical field of satellite communication networks and on-board buses, in particular to a data deterministic transmission scheduling method for a satellite-borne time-sensitive network. The method comprises the following steps: constructing a satellite-borne network architecture, connecting each terminal and a bus gateway by a TSN management node to form a unified data bus, distributing priority and monitoring link state; performing feature modeling on the data frames to define six-tuple parameters; designing a hierarchical queue architecture comprising a key task queue, a load data queue and a background traffic queue; a dynamic shaping mechanism is adopted, corresponding shaping devices are arranged for different queues, and parameters are set; the invention further provides an anti-disturbance scheduling strategy which comprises dynamic priority improvement and elastic bandwidth allocation. And finally, performing output scheduling according to a strict priority and a weighted fair queue rule. According to the method, by constructing a unified TSN bus and integrating heterogeneous subsystems, low-delay and low-jitter transmission of key task data is realized, the bandwidth utilization rate of scientific load data is improved, the network reconstruction time is shortened, the anti-interference capability is enhanced, and satellite-borne mixed data stream transmission in a high-dynamic and multi-task scene is effectively optimized.

The invention provides an abnormal sound detection method based on multi-scale time-frequency feature perception, and relates to the technical field of acoustic detection for industrial machine state monitoring, and the method comprises the steps: inputting an original sound signal, and carrying out the dual-branch feature extraction to generate a time domain coding spectrogram and a logarithmic Mel spectrogram; executing hybrid data enhancement; splicing the enhanced spectrogram and inputting the enhanced spectrogram into a dense encoder for compression modeling; extracting features through a two-stage multi-scale time-frequency sensing network, processing along a time dimension in the first stage, and processing along a frequency dimension in the second stage; inputting the high-order features into a lightweight classifier to output an abnormal score; and comparing the abnormal score with the gamma distribution threshold to judge abnormity. According to the method, the weak anomaly detection rate and the cross-equipment stability are remarkably improved, and the perception and discrimination capability of the model on the multi-scale time-frequency characteristics is effectively enhanced.