71 results about "Unsupervised clustering" patented technology

The method for automatic identification, classification and development trend analysis of net red villages based on multi-source data fusion and natural language processing comprises the following steps: UGC data is crawled from Xiaohongshu and Douyin through a distributed master-slave architecture, de-duplicated based on SimHash, and normalized in time and coding format; a text semantic fingerprint is generated, and multi-level semantic cache fingerprint matching is performed; for unassigned text, its complexity is calculated, and a large language model API is adaptively called to automatically complete and extract five-level administrative divisions; weights are determined based on the analytic hierarchy process, interaction indicators such as likes, comments, collections and forwards are integrated, and a comprehensive network heat index of the village is obtained; an external text mining tool is connected, and batch word frequency analysis, semantic network analysis and sentiment tendency evaluation are performed; a document-term matrix is constructed, TF-IDF weighting is performed, and unsupervised clustering algorithm is used for clustering analysis of village characteristics; cross-dimension analysis is performed on the clustering results, and a development portrait, advantage mining and operation suggestion warning are automatically generated in combination with the SWOT model.

The application provides an automatic driving rear-end collision and sudden state prediction method based on space-time occupation characteristics, and relates to the field of automatic driving.The application proposes a prediction network based on a space-time graph Transformer, solves the problem of lack of space-time occupation characteristics in automatic driving vehicle dangerous scene analysis.Through integration of multi-dimensional characteristics from a nuScenes data set, a rear-end near-collision scene knowledge graph RNSKG is constructed, and DBSCAN unsupervised clustering is used to classify dangerous states and emergency states.A space-time graph Transformer model coupled with a time sequence and a space attention mechanism is significantly better than a traditional model in predicting dangerous states and emergency states, verifying the effectiveness in capturing complex space-time interactions.

The application relates to a method and device for identifying a proxy signature, electronic equipment and a storage medium, and relates to the technical field of information security and pattern recognition. The method comprises the following steps: acquiring a set of signature images without labels; for each original signature image, performing standardization preprocessing on the original signature image; inputting each obtained target signature image into a feature extraction model trained through migration learning to extract a deep feature vector; freezing the parameters of a pre-trained bottom model and fine-tuning the parameters of a top model during training; using a density-based unsupervised clustering algorithm to cluster the target signature images according to the deep feature vectors; performing character recognition on each target signature image in each cluster to obtain signature texts; and if each signature text corresponding to a cluster contains at least two different names, it is determined that a proxy signature exists. The application realizes efficient, automatic and high-credibility identification of the specific illegal mode of one person proxying for multiple people without relying on supervised learning.

This invention discloses a remote sensing cloud simulation method and apparatus based on a classification-guided diffusion model, belonging to the field of remote sensing image cloud and fog simulation. The method includes: acquiring and preprocessing multi-temporal remote sensing images to construct a dataset containing registered pairs of cloudless and cloudy images; pre-training a feature extraction network based on cloud mask data and performing unsupervised clustering using the cloud features extracted by the network to achieve automated cloud classification; training a pre-constructed cloud simulation diffusion model using the cloudless image and cloud classification results as input, and the corresponding cloudy image as the ground truth supervision signal, to learn the mapping relationship from cloudless to cloudy states; and inputting the target cloudless image and a specified cloud category into the trained cloud simulation diffusion model to generate a simulated cloud-containing remote sensing image. This invention effectively improves the realism of generated clouds and fog, the naturalness of integration with ground features, and the controllability of generating specific cloud categories, providing sufficient and realistic training samples for cloud-related data-driven models.

The application relates to a model iteration method and device, electronic equipment and a storage medium, comprising: obtaining original sensor data collected by a vehicle terminal in a preset time window before and after a trigger time in response to a preset event trigger; clustering the obtained multiple groups of original sensor data using an unsupervised clustering algorithm, identifying an outlier cluster from the clustering result, and marking the original sensor data corresponding to the outlier cluster as long-tail candidate data; extracting images from the long-tail candidate data, using a cloud perception model and a vehicle terminal perception model to respectively infer the images, screening out images with inconsistent inference results of the two perception models as difficult example images, and labeling the difficult example images according to the inference result of the cloud perception model; optimizing and training the vehicle terminal perception model using the labeled difficult example images, and deploying the optimized perception model to the vehicle terminal. Thus, the problem of insufficient performance of the perception model in a long-tail scene is effectively solved.

This invention relates to the field of artificial intelligence technology, specifically to an AI-based medical information consultation method and apparatus. The method includes: acquiring multimodal medical consultation input from a user; preprocessing and extracting features from the multimodal input, and fusing them to generate a unified multimodal feature representation; based on the feature representation, performing topic indexing in a pre-constructed medical knowledge graph, and extracting a set of highly relevant medical knowledge using unsupervised clustering analysis and a dynamic threshold filtering mechanism; inputting the filtered medical knowledge set and multimodal features into a medical large-scale language model for logical reasoning, and running a multi-turn dialogue intent tracking algorithm during the reasoning process to complete missing entities; intercepting high-risk behaviors and correcting compliance issues in the preliminary response generated by the large-scale language model, and outputting a safe and professional medical information consultation response. This invention significantly improves the cross-modal perception capability, logical reasoning accuracy, and clinical application legitimacy of medical AI.

This invention relates to the field of numerical simulation and computational technology for seismic wave propagation, and discloses a method and apparatus for generating seismic ground motions based on target spectra and Green's functions. It aims to address the difficulty in achieving a balance between physical consistency, computational efficiency, and spectral compatibility in existing methods. The scheme mainly includes: preprocessing environmental noise records and extracting empirical Green's functions through unsupervised clustering; constructing a spatially continuous propagation tensor by performing depth correction and gradient interpolation based on surface wave eigenfunctions; discretizing the kinematically finite fault model into sub-source units and generating long-period seismic ground motions including path effects using propagation tensor convolution; optimizing source parameters through Bayesian inversion using the GMPE target spectrum as a constraint; and finally generating high-frequency components and fusing them with the long-period waveform to obtain a broadband seismic ground motion time history. This invention achieves a unification of physical propagation mechanisms and statistical spectral constraints, significantly improving computational efficiency while ensuring spectral compatibility, and is particularly suitable for basin areas.

The application discloses a landslide susceptibility prediction method combining terrain combined features and partition explainable analysis, and relates to the technical field of geological disaster evaluation. Slope, plane curvature and profile curvature factors are extracted from a digital elevation model of a research area, and after standardization, Gaussian mixture model is used for unsupervised clustering. The class labels obtained by clustering are taken as terrain combined features and are coded, and are combined with landslide influence factors to form an input feature set. The input feature set is input into an XGBoost model for training and the hyperparameters are optimized through grid search. Based on the trained model, the SHAP method is used to calculate the global feature contribution degree, and the contribution degrees of various factors are calculated according to terrain combination types, and the spatial differences of the factors under different terrain units are analyzed. The terrain types obtained by clustering are used as explicit features to participate in model training, and the spatial heterogeneity of the factors is revealed through grouped contribution degree analysis, which provides a new technical means for landslide susceptibility evaluation.

This invention provides a computing resource scheduling method and system, belonging to the field of computing resource scheduling technology. The scheduling method includes: collecting multi-source heterogeneous data to generate a full-dimensional fused dataset; using a pre-defined unsupervised learning technique to transform the full-dimensional fused dataset into a standardized unsupervised feature vector set; unifying the dispersed multi-dimensional potential energy components and using a pre-defined unsupervised clustering algorithm to perform comprehensive calculations of potential energy values; selecting unsupervised auxiliary causal variables and causal relationships to generate a preliminary system causal graph, and using a pre-defined quantization method to calculate the causal effect strength of the preliminary system causal graph; transforming the target task of the target computing facility into quantifiable sub-objectives, and using the do-calculus operation of the system causal graph model to generate multiple sets of candidate parameter tuning schemes. This invention achieves a balance between global optimization and local efficiency, proactively avoids potential risks, dynamically balances multiple objectives and continuously evolves, and maximizes overall benefits.

The embodiment of the present disclosure provides a method, device and equipment for fusing multi-source virtual-real heterogeneous data. The method comprises: in response to obtaining multi-source heterogeneous virtual-real data to be fused, performing feature extraction on the multi-source heterogeneous virtual-real data to obtain preliminary features; wherein the multi-source heterogeneous virtual-real data comprises structured data and unstructured data; grouping the preliminary features by using an unsupervised clustering algorithm to form data groups; determining a data cleaning rule set matched with the features of the data groups by using a preset mapping relationship, and cleaning the data groups by using the data cleaning rule set to obtain cleaned data groups; and fusing the cleaned data groups by using a preset data weaving rule to generate fusion data conforming to a unified data interface form. The method of the embodiment of the present disclosure can significantly improve the automation degree, processing efficiency and result quality of data fusion.

The application discloses a rehabilitation action recognition and classification method, system, device and storage medium, the method comprising: collecting a three-dimensional coordinate sequence of a human skeleton point corresponding to a target rehabilitation action performed by a user; performing normalization and time alignment processing on the three-dimensional coordinate sequence to generate a standardized posture matrix; extracting principal component features for representing action patterns from the standardized posture matrix; performing unsupervised clustering analysis on the principal component features, and dividing a plurality of action pattern categories according to a clustering analysis result; performing similarity matching on the principal component features and features corresponding to the plurality of different action pattern categories in a typical action pattern database; and determining a specific action pattern category to which the target rehabilitation action belongs according to a similarity matching result. The application improves the objectivity, repeatability and clinical interpretability of rehabilitation action recognition and classification, and expands the applicable scenarios.

The present application relates to a kind of kitchen garbage disposal equipment stirring motor torque compensation control method, to solve the problem of parameter adjustment lag and insufficient stability when coping with complex load change in traditional method. By high-frequency synchronization to obtain multi-source sensing signal, adaptive time window segmentation and feature cleaning are used, multi-dimensional statistical feature vector is extracted, and unsupervised clustering is used to determine the working condition category and generate the working condition-optimal PID parameter mapping table. Combined with lightweight decision tree, the working condition is quickly classified and the optimal parameters are matched, and the parameter soft switching mechanism is used to smooth the working condition transition, and the control effect is automatically optimized. This scheme significantly improves the load adaptability and operating stability of the stirring motor, reduces the difficulty of control parameter maintenance, and is suitable for garbage disposal equipment with dynamic and dramatic changes in load.

A method for determining the association and continuity combination of top-side-rock structural surfaces based on unsupervised learning is proposed. This method involves investigating the structural surfaces of the roof and side-rock masses in underground engineering projects to obtain geometric and statistical information. A unified encoding is performed to construct a joint feature set of roof-side-rock structural surfaces. An unsupervised density-based clustering algorithm is used for initial cluster analysis of the joint feature set. A swarm intelligence-based Lüperfox optimization algorithm is introduced to optimize key parameters of the DBSCAN clustering algorithm. A Latin hypercube sampling method is used to improve the population initialization process of the Lüperfox optimization algorithm, resulting in better DBSCAN hyperparameters and optimized structural surface clustering results. The method determines the structural surface system affiliation of structural surfaces in different locations, quantitatively calculates the spatial continuity and combination relationship of structural surfaces, and outputs the spatial association determination results and continuity levels of the roof-side-rock structural surfaces. This method enables intelligent identification and quantitative analysis of the spatial relationship between roof and side-rock rock structural surfaces.

The application discloses a lung cancer gene mutation prediction method based on unsupervised clustering double-stage attention multi-instance learning, and relates to the technical field of pathological image analysis and gene detection. H&E staining pathological whole section images of non-small cell lung cancer patients and corresponding gene mutation data are collected to construct a data set; the images are preprocessed by using the OTSU method, segmented into blocks and high-dimensional feature vectors are extracted; the block features are grouped into cluster feature sets through unsupervised clustering; a double-stage attention mechanism composed of intra-cluster and inter-cluster is used to hierarchically aggregate and generate global features; finally, a classification model is used to output mutation positive / negative prediction results. The application groups the features through unsupervised clustering and structures the features, combines double-stage attention to strengthen key information, does not need complex manual annotation, adapts to various driver gene mutation prediction requirements, effectively deals with tumor heterogeneity and feature sparsity, improves prediction accuracy and generalization ability, and provides low-cost and efficient targeted therapy preliminary screening technical support for clinics.

This invention discloses a method for ultra-dense wireless audio terminal networking and multi-source data fusion, belonging to the field of ultra-dense networking and audio processing technology. The method involves the terminal collecting sound field temporal characteristics, wireless channel status, and location information; performing distributed time-slot collaborative scheduling based on sound field spatial correlation, allocating orthogonal short time slots to suppress co-channel interference; implementing unsupervised clustering of multi-source data and generating fusion weights through a Dirichlet process variational autoencoder; dynamically adjusting point density and transmission power using the Navier-Stokes equation, and optimizing relay links by combining small-world networks and quantum heuristics; compensating for high-frequency audio gaps using a conditional generative adversarial network, and uploading the data after joint encoding of the source and channel; and employing a clonal selection algorithm to achieve self-healing of network anomalies. This invention can reduce interference and redundant transmission, improving audio transmission quality and system robustness in ultra-dense scenarios.

The application discloses a system control method and device based on model predictive control and reinforcement learning, comprising: acquiring a training data subset of a controlled system and a total state vector at a current moment, the historical state vector at the current moment comprising a control input vector and a state output vector before the current moment, and a plurality of training data subsets being obtained by unsupervised clustering of training data based on historical state vectors at historical moments; determining an MPC intelligent agent of a local working condition corresponding to a training data subset with the highest similarity according to the similarity between the historical state vector at the current moment and the historical state vector at the historical moment; and generating a control input vector at the current moment according to parameters of the MPC intelligent agent and the total state vector and sending the control input vector to the controlled system, wherein the parameters of the MPC intelligent agent are obtained by training a linear prediction model and reinforcement learning based on the training data subset corresponding to the local working condition. According to the application, the complexity of system control can be reduced, and the interpretability and safety can be improved.

The application discloses a kind of neighborhood collaborative flow field sequence mapping transmission line icing galloping detection method, it belongs to power grid disaster prevention technical field.The application solves the problems of large computing power consumption, poor adaptability and low detection accuracy caused by environmental representation distortion of existing methods.In the multi-end collaborative data domain, the high-dimensional space characteristic cosine similarity is calculated by sliding window traversal, and the representative sequence of airflow distortion is extracted.Through this way, the fluid mechanics simulation is replaced, and the hardware computing power consumption of large-scale three-dimensional fluid calculation is reduced.The complex microtopography of transmission line neighborhood and the physical response of light icing are deeply mapped and seamlessly complementary, the environmental induction mechanism is maximized, the environmental representation is not distorted, and high-accuracy transmission line icing galloping early warning is realized.According to the entity object set and unsupervised clustering algorithm, transmission line icing galloping early warning under different neighborhood microtopography can be output, which has strong adaptability.The application can be applied to icing galloping detection.

The invention provides a coal type replacement path generation method based on hidden coal blending strategy identification, and relates to the technical field of intelligent coal type replacement. The coal type replacement path generation method based on recessive coal blending strategy recognition comprises the steps that historical successful coal blending schemes are collected through data collection and preprocessing; an unsupervised clustering algorithm is adopted, and a hidden coal blending strategy is mined from the multi-dimensional historical coal blending data; carrying out feasibility verification on the coal blending replacement scheme based on coal blending attribute balance analysis; carrying out refined quantitative analysis on the feasible coal blending replacement scheme; and outputting a quantized coal type replacement report. According to the method, all complex formula modes which are proved to be successful in history are objectively concluded and summarized, view limitation and subjective prejudice of manual searching are overcome, the discovery process of the replacement opportunities is more comprehensive and systematic, and implicit knowledge of enterprises is dominated and structured.

This invention discloses a data annotation method and system based on user behavior and attention tracking. It synchronously and in real-time collects multi-source behavioral signals from doctors, including mouse, keyboard, and eye movements. Combined with identity and interface metadata, the data undergoes standardization, anomaly removal, and short-term behavioral unit segmentation. Unsupervised clustering is used to extract individual behavioral micro-patterns, and a behavioral profile library is constructed. Through multimodal temporal modeling and an adaptive spatiotemporal attention mechanism, behavioral features, interface regions, and report text are deeply integrated to output multi-level correlation probabilities, achieving high-precision automatic labeling of content and image regions.

A method for assessing the icing status of wind turbine blades based on dynamic weighted Mahalanobis distance and Gaussian mixture model clustering includes the following steps: acquiring a wind turbine blade icing fault diagnosis dataset; calculating the dynamic weighted Mahalanobis distance between iced samples and normal samples using normal sample data as a benchmark; mapping the dynamic weighted Mahalanobis distance to a health index of the wind turbine blade's operating status; and using the health index and Gaussian mixture model clustering to quantitatively analyze the icing status of the wind turbine blades, thereby deriving a quantitative classification of the severity of the icing status samples. This method utilizes weighted dynamic Mahalanobis distance to map high-dimensional features to a one-dimensional health index, and then achieves objective classification through unsupervised clustering. It not only provides a new, quantifiable, and interpretable status assessment tool for wind turbine blade icing faults but also offers a new paradigm for predictive health management of complex equipment, possessing significant theoretical and engineering application value.

This invention discloses a method and apparatus for constructing first-arrival samples based on compressed sensing and unsupervised learning. The method includes: preprocessing raw seismic data; preprocessing includes normalization, flattening, and / or denoising; clustering the preprocessed data using an unsupervised clustering algorithm and principal component analysis to obtain multiple categories of first-arrival sample clusters; for each category of first-arrival sample cluster, generating a random sampling matrix using compressed sensing theory, randomly projecting each first-arrival sample in the cluster to obtain a corresponding low-dimensional compressed representation, and generating random sample numbers based on the low-dimensional compressed representation; randomly selecting first-arrival samples according to a preset ratio based on the random sample numbers, and labeling the selected first-arrival samples to obtain a first-arrival sample set. By integrating clustering and selection, sample construction is achieved. Principal component analysis clustering classification makes the clustering results more stable, and selection after compressed projection ensures uniform coverage and reduces reliance on manual intervention.

The application relates to the technical field of intelligent management and energy consumption prediction of new energy automobile batteries, in particular to an electric vehicle driving range prediction method based on energy consumption decoupling and double-dimension scene matching, which comprises the following steps: acquiring real-time feature vectors of vehicle operation in a low-temperature domain state and preprocessing, adopting a sliding window method based on power gradient to recombine features; defining the boundary of a scene and calibrating a label based on unsupervised clustering, and determining the scene label of a real-time feature vector based on supervised classification; adopting a regression model to perform heterogeneous model prediction scheduling to obtain a predicted driving range; the application logically recombines discrete fragmented data according to power loss gradient, introduces multi-dimensional energy loss features to construct a regression prediction system, defines the boundary of a scene by adopting an unsupervised clustering algorithm, and combines a supervised real-time classification logic to solve the problem of low prediction accuracy of the driving range of an electric vehicle in a low-temperature environment in the prior art.

This invention discloses a device and method for measuring the volume of liquid in a transparent container. The measuring device includes a light source, a container placement unit, an image acquisition device, a signal processing unit, and a pixel-object mapping calibration unit. The method for measuring using the aforementioned measuring device includes: acquiring an original image through illumination by the light source and the image acquisition device; generating an image feature histogram after preprocessing; dividing the histogram data into an image portion containing liquid (without liquid above and liquid below) using an unsupervised clustering method; performing image boundary analysis on each portion to obtain the center position data of the transparent container; performing image segmentation processing on the image portion containing liquid below to determine the image boundaries of the meniscus, pure liquid, and the bottom of the transparent container; finally, slicing and stitching the images to obtain a binary image; then generating a three-dimensional model and determining the volume of the liquid to be measured based on the pixel-object mapping relationship. This invention greatly improves the accuracy and convenience of measurement.

This invention discloses an array ACFM detection method and system for detecting wide weld seams. The method includes: S1, establishing an AC electromagnetic field by attaching the array detection probe to the surface of the wide weld seam to be tested, and scanning along the weld seam direction; S2, synchronously acquiring multi-channel magnetic field disturbance signals using a magnetic field sensor array; S3, preprocessing and demodulating the multi-channel magnetic field disturbance signals to extract in-phase and quadrature components; S4, calculating the amplitude and phase to construct a joint feature vector; S5, constructing a similarity matrix and a degree matrix to obtain a Laplace matrix, and using an unsupervised clustering algorithm for defect classification; S6, calculating the confidence index of the classification result and comparing it with a preset threshold to determine whether crack defects exist. This invention achieves accurate determination of crack defects in wide weld seams by extracting the joint feature vector and spatial correlation features of multi-channel magnetic field signals and performing unsupervised clustering using the Laplace matrix, and then calculating the confidence index based on the classification results.

The present application relates to a fraud risk prevention and control method and system, wherein the method inputs multi-source heterogeneous data into a pre-trained channel risk assessment model for channel risk assessment, splices the generated channel risk assessment result with the individual feature vector of the user, inputs the obtained risk portrait feature into a Gaussian mixture model for unsupervised clustering to obtain K potential fraud scenarios and a probability vector of each user belonging to each potential fraud scenario. Based on the real-time context feature vector of all users, the risk portrait feature and the probability vector of each user belonging to each potential fraud scenario, fraud risk assessment is carried out, a heterogeneous evidence graph is constructed for fraud reasoning to generate a fraud risk list, and the risk prevention and control feedback data is used for reverse optimization to realize dynamic risk prevention and control. Thus, the present application can quickly respond to the changing new fraud scenarios while realizing the source perception of fraud, and realize accurate fraud risk prevention and control.

This invention discloses an unsupervised clustering and attribution method and system for dynamic deception-enhanced network attacks. The invention identifies suspicious behavioral clusters by real-time collection of raw network logs and unsupervised incremental clustering. When the suspiciousness exceeds a threshold, deception resources matching the behavioral pattern are automatically and dynamically deployed to capture deep interaction sequences of the attacker, generating enhanced data. The features of the raw logs and the enhanced data are spatiotemporally aligned and fused to generate an intent feature vector representing the attacker's tactics, techniques, and processes. Based on this vector, refined clustering and attacker attribution are performed. This invention solves the problems of inaccurate clustering intent identification and low attribution accuracy caused by the lack of deep interaction data in existing technologies, significantly improving the ability to discover unknown threats and the accuracy of attack attribution.

The application discloses a kind of structural information enhanced multimodal heterogeneous data fusion representation method, belong to data processing technical field.Method includes: by text, image, audio and video obtain multimodal heterogeneous data, after executing pre-processing operation to multimodal heterogeneous data, feature extraction and conversion operation are executed, obtain the multimodal feature matrix of uniform feature space and using graph structure enhancement technology constructs graph structure with stability and explainability;Structural entropy regular discriminant representation learning framework is constructed, structure information optimization framework and soft allocation mechanism are designed, hypergraph structure is constructed, and the structural entropy of hypergraph structure is calculated;Based on hypergraph structure entropy, guide multimodal unsupervised clustering.The present application is based on structural entropy, constructs hypergraph structure and introduces soft allocation mechanism, has the characteristic of automatically learning data structure information, improves the processing efficiency of unstructured data, effectively breaks through the traditional restriction condition, provides new solution.

This invention relates to an online partial discharge detection device and monitoring system. The discharge detection device includes a sensor module for acquiring partial discharge signals; a data acquisition system connected to the sensor module, which includes a multiplexer for multi-channel signal selection and switching, a signal conditioning circuit including a low-noise amplifier and bypass control circuit, an analog-to-digital converter, and an FPGA data acquisition module for receiving digital sampled data and power frequency synchronization signals; a data processing system, which includes an FPGA data processing module with data buffering, trigger control, phase detection, waveform management, and digital filtering functions, and an embedded processor communicating with the FPGA for feature extraction, unsupervised clustering, and supervised classification; and a communication system for transmitting the identification results to a remote monitoring platform. This invention can meet the application requirements of large-scale online monitoring of power equipment.

This invention provides a backdoor defense method for federated learning based on gradient consistency and dual unsupervised clustering, comprising: obtaining gradient updates submitted by multiple clients participating in federated learning in the current training round; for each client, predicting the gradient update of the current round based on historical global model parameters and the client's historical gradient updates; calculating the difference between the actual gradient update and the predicted gradient update of each client to generate a malicious score for the client; adaptively determining the optimal number of clusters based on the malicious scores of all clients using the Gap statistic; performing dual unsupervised clustering detection with a coarse-to-fine approach based on the optimal number of clusters: first, removing clients with high malicious scores using K-means clustering, and then removing residual malicious clients from the remaining clients using DBSCAN clustering, thus selecting benign clients; and performing federated learning global model aggregation based on the gradient updates of the selected benign clients.

This invention relates to the field of power monitoring, specifically to a method and system for real-time monitoring of load fluctuations in power metering. The method includes the following steps: synchronously accessing three types of data: load time series, scenario attributes, and power grid operating conditions, and outputting a compliance feature sequence; classifying scenarios based on historical standardized data, filtering normal samples through unsupervised clustering, constructing baselines for each scenario through time series decomposition, fitting and calculating initial anomaly judgment boundaries, and forming a scenario-based baseline library; real-time matching of the scenario corresponding to the current operating condition and loading the baseline, updating the baseline incrementally in unsupervised manner based on normal samples within a sliding window, dynamically calibrating the anomaly judgment boundaries, and setting hard constraints for boundary adjustment; establishing a small-sample fault prototype library, and performing real-time sequence anomaly pattern matching based on the small-sample fault prototype library, making a judgment based on a dual-dimensional fusion of boundary exceedance and pattern matching results, and outputting a graded alarm. This application achieves accurate and efficient real-time monitoring of load fluctuations in power metering.