58 results about "Agglomerative hierarchical clustering" patented technology

A method and system for detecting and analyzing internet traffic from Internet of Things (IoT) devices is presented. A network telescope is employed to collect unsolicited data packets. The data packets are analyzed to determine whether they arise from a misconfigured device or from a malicious device (darknet sanitization) or from some other source. Traffic from misconfigured devices is filtered out. The data packets from malicious devices are analyzed and a classification model is trained to classify the data packets into originating from an IoT or from a non-IoT device. The classifier is then validated on a separate set of data. The data packets originating from malicious IoT devices are further analyzed by such techniques a clustering (e.g. agglomerative hierarchical clustering), geo-location analysis, methods of attack, and coordination of attacks.

The invention relates to the technical field of digital data processing, in particular to a circuit breaker fault detection method based on electrical parameter analysis, and the method comprises the steps: obtaining monitoring data, and constructing a parameter matrix; screening suspicious elements based on prediction deviation of elements in the parameter matrix and a mutation point detection result, and determining a first characteristic value according to a starting moment determined by the suspicious elements and a rule degree of data fluctuation in a fluctuation duration of the suspicious elements, and determining a fault interval by combining a consistency analysis result of irregular data fluctuation in parameter matrixes of different types of monitoring data, screening target data from residual terms, obtaining a plurality of clustering trees based on the target data by using an agglomerated hierarchical clustering algorithm, and determining a fault detection result of the circuit breaker based on differences among different clustering trees. According to the method, the residual component is used for replacing the initial data, so that the clustering effect of the clustering number on the monitoring data with different fault sensitivity degrees is improved while the clustering data size is reduced.

The invention relates to the technical field of data processing, in particular to a technological parameter monitoring method for silica gel foaming pillow production, which comprises the following steps of: acquiring a pressure data sequence and a temperature data sequence in a foaming cavity, acquiring the numerical value normal degree of each pressure datum according to the time sequence neighborhood segment of each pressure datum, and calculating the numerical value normal degree of each pressure datum according to the numerical value normal degree; according to the difference condition between each piece of pressure data and the temperature data in the temperature data sequence in the foaming cavity at the corresponding sampling moment, the sampling accuracy degree of each piece of pressure data is obtained; in the process of clustering the pressure data, according to the sampling accuracy, obtaining a merging distance between each group of clusters in the clustering process, inputting the merging distance between each group of clusters into an agglomerated hierarchical clustering algorithm, and gradually merging the clusters on the pressure data in an iteration mode by the algorithm to obtain a hierarchical clustering tree after clustering is completed; according to the hierarchical clustering tree, abnormal data are recognized, and the accuracy of abnormal data recognition is improved.

The invention relates to the technical field of atmospheric pollution treatment, and discloses an urban near-surface atmospheric pollution data analysis method, which comprises the following steps: acquiring atmospheric particulate concentration time sequence data and carrying out multi-time scale component separation to obtain a first feature sequence of the atmospheric particulate concentration time sequence data under different time scales; performing dynamic time warping processing on the first feature sequence among the different monitoring stations to obtain a second feature sequence for eliminating dynamic asynchronism among the different monitoring stations; performing agglomeration hierarchical clustering processing on the second feature sequence to obtain background atmospheric particulate concentration fluctuation features; according to the background atmospheric particulate concentration fluctuation characteristics, the urban near-surface pollution process of the atmospheric particulates is analyzed. According to the method, detailed analysis of urban near-surface atmospheric particulate matter concentration change characteristics on a minute-level time scale is realized, a reliable technical basis is provided for screening an exogenous particulate matter pollution process, and the efficiency and accuracy of atmospheric pollution analysis are improved.

The invention relates to the technical field of information, in particular to a sleep health dynamic assessment method and system based on user feedback driving and a medium. The method comprises the following steps: constructing an individual knowledge graph reflecting an individual health causal relationship by adopting a time sequence causal analysis method; constructing a group knowledge graph reflecting a group health causal relationship in combination with an agglomerated hierarchical clustering method and a hierarchical federated learning method; based on user feedback information and health factor data which are acquired in real time, triggering individual knowledge graph updating, and triggering group knowledge graph delay increment reconstruction; and dynamically fusing the weight of the updated double atlases based on the accumulated quantity of the user feedback information to obtain the comprehensive weight of the user health factor, and carrying out weighted calculation on the newly obtained health factor data to obtain a sleep health assessment score. Therefore, real-time feedback data is deeply utilized, dynamic, accurate and explainable sleep health assessment is realized, and the technical prejudice that groups and individuals are irreconcilable is broken through.

The invention relates to the technical field of image data processing, in particular to an automobile part electrophoresis machining surface detection method based on machine vision. The method comprises the following steps: dividing an electrophoresis machining surface image of the automobile part into a plurality of sub-blocks, and obtaining an initial clustering result of the electrophoresis machining surface image of the automobile part through an agglomerated hierarchical clustering method; according to the minimum bounding rectangle size of each edge in the sub-block and the number of edge pixel points in the sub-block, the primary fiber property of the sub-block is obtained; calculating fiber factors of the sub-blocks; and carrying out secondary cutting on the initial clustering result based on the fiber factor of each sub-block in the initial clustering result to obtain an electrophoresis processing surface detection result of the automobile part, so that the accuracy of the electrophoresis processing surface detection result of the automobile part is effectively improved.

The invention provides a key event discrimination method and system based on a hierarchical processing architecture, and the method comprises the steps: obtaining reported events from different channels, carrying out the data preprocessing of the reported events, obtaining processed event texts, and classifying the event texts; clustering the classified texts by adopting a semantic regularization agglomeration hierarchical clustering method; and according to a clustering result, utilizing an in-cluster entity keyword extraction and comparison mechanism to discriminate repeated events in the text to obtain a key event discrimination result. According to the key event identification method provided by the invention, a classification-clustering-discrimination hierarchical processing architecture is constructed, the technical problem existing in one-event multi-report screening is effectively solved, and finally, the screened repeated report events are taken as key events needing to be marked. Therefore, related managers can preferentially dispose the key events or give full attention to the key events.

The invention provides a fault diagnosis and positioning method and device based on an intelligent irrigation gateway, and relates to the technical field of agricultural irrigation, and the method comprises the steps: obtaining multi-source sensor data collected from an irrigation system controlled by the irrigation gateway, and carrying out the condensation hierarchical clustering, and obtaining a hierarchical clustering tree; determining a cut-off point in the hierarchical clustering tree according to a preset distance threshold, and cutting the hierarchical clustering tree to obtain a plurality of clusters; abnormal data identification is carried out on the multi-source sensor data in the cluster, and an abnormal fault point is determined; and tracing the abnormal fault point in the tree topology structure network of the irrigation system, and carrying out path planning to obtain a shortest path for tracing the abnormal fault point so as to position the abnormal fault point in the irrigation system. Through the method and the device, the defects of low fault positioning precision and efficiency and influence on the maintenance efficiency of the irrigation system caused by possible misjudgment or missed judgment in fault detection of the irrigation system and difficulty in timely fault positioning are overcome.

The invention relates to a transformer area distributed photovoltaic power ultra-short-term prediction method and system, and belongs to the technical field of distributed photovoltaic power prediction. On the basis of an agglomerate hierarchical clustering algorithm, transformer area photovoltaic in a region is clustered into a sub-region set with homogenized output characteristics, decoupling representation of spatial-temporal characteristics in the sub-regions is realized in combination with a GraphSAGE graph neural network and a Transform encoder model, and finally, a photovoltaic power prediction result of each transformer area in the sub-regions is synchronously output on the basis of the composite spatial-temporal characteristics. Furthermore, a proper migration modeling strategy is constructed based on shared features among the sub-regions, rapid lightweight modeling of each sub-region prediction model is realized, and finally, a region total power prediction result is obtained through spatial aggregation of sub-region prediction power.

The invention provides a TCN-LSTM-AM-based hydropower output prediction method, and the method comprises the steps: constructing a TCN-LSTM-AM network which employs a causal convolution and cavity convolution structure, extracts the local fine features and long-distance dependence relation of meteorological-water level-output data in a time sequence, and outputs a multi-scale feature sequence; the LSTM processes the feature sequence output by the TCN through a gating mechanism of a forgetting gate, an input gate and an output gate, adaptively retains historical key information, and outputs a hidden state sequence containing time sequence dynamics; the AM dynamically allocates attention weights to a hidden state sequence output by the LSTM, focuses on key time step features of prediction, and generates a final prediction value; classifying meteorological data based on Euclidean distance by adopting an agglomerated hierarchical clustering algorithm, dividing similar hydroelectric output mode clusters under different climate conditions, and training a TCN-LSTM-AM network for each cluster to enable the model to adapt to output laws under different climate conditions; and the trained TCN-LSTM-AM network is used to generate a next-day hydropower output prediction value.

The application discloses a kind of methods for automatically analyzing and evaluating report, it is related to computer software and information processing technical field, including: based on graph neural network and visual-textual dual modal feature extractor, realize the adaptive analysis of complex non-standard structure report version, and the document is parsed into two-dimensional data table containing different classification dimensions;Based on BERT vectorization, etc. Adaptive clustering identifies the cluster structure of uneven semantic distribution, adopts the agglomerative hierarchical clustering to construct tree-shaped multi-granularity semantic hierarchy, realizes semantic aggregation;The application realizes the text content of non-standard form, the identification and processing of complex semantics;Overcome the defects of low efficiency, easy to make mistakes, subjective influence and only read specific format or specific location of text content, lack of flexibility and unable to handle complex semantics in prior art manual input.

The invention discloses a low-voltage transformer area user phase identification method based on multiple measurement data, which comprehensively utilizes two measurement data of voltage and power to carry out integrated similarity measurement, and solves the problems of a single voltage data method under the condition of three-phase voltage balance and a single power data method under the condition of low power consumption of users. According to the method, firstly, the redundancy of original data is reduced through voltage data screening and power feature extraction, then, agglomerated hierarchical clustering is carried out on users by using a screening voltage sequence to generate in-phase user clusters, and finally, secondary clustering is carried out on the users by using integrated similarity combining voltage and power similarity to generate in-phase user clusters. The phase attribution discrimination of the transformer area users is realized, and the method has relatively high phase identification accuracy and stability.

The invention discloses a spectrum map construction method based on dual-scale fusion, relates to the technical field of spectrum map construction, and aims to improve the adaptability and characterization accuracy of a spectrum map to a complex environment by using agglomerated hierarchical clustering on sampling points; according to the method, a Voronoi diagram of random sampling points is constructed by taking a global region to be interpolated as a boundary, the Voronoi diagram is divided according to an optimal clustering result, space division and the clustering result are combined, the advantages of the Voronoi diagram in space division are fully utilized, and improvement of local precision and overall spatial resolution of a spectrum map is facilitated; the Kriging interpolation method combining the region and the global is adopted, and the local characteristics of different regions and the overall trend of the global can be fully considered, so that the interpolation error in a complex environment is reduced.

The invention discloses a heat dissipation performance evaluation method and system for a backlight source. The backlight module is powered on in a first test period, a module infrared image is obtained, color features are extracted based on a color histogram, and backlight sub-regions are divided by adopting agglomeration hierarchical clustering; in the second test period, the sub-regions and the whole are electrified respectively, RGB color features and temperature differences of two tests are extracted, color feature differences and temperature differences are calculated, the heat dissipation index of each region is evaluated, and the index can comprehensively reflect the heat dissipation performance of different regions and effectively guide the backlight module to carry out structure adjustment and production optimization; and the production test efficiency and the performance evaluation accuracy are improved.

The invention relates to an urban inland inundation modeling method considering spatio-temporal heterogeneity. The urban inland inundation modeling method comprises the following steps: constructing a rainfall time sequence and a multi-scale spatio-temporal grid with spatio-temporal heterogeneity; data preprocessing: dividing a research area into initial response units based on topographic features, land utilization types and drainage system distribution; measuring the waterlogging response process similarity among the response units by adopting a dynamic time warping method, and performing time sequence clustering on a division result according to the waterlogging response process similarity; further clustering and determining an optimal partition number by adopting an agglomerated hierarchical clustering method, and introducing a spatial adjacency constraint to ensure spatial connectivity; different hydrological parameters are given to different response units; and utilizing the multi-scale rainfall space-time grid data and the response unit division result to drive modeling of an urban inland inundation earth surface-pipe network coupling model. According to the method, the precision of urban waterlogging modeling is improved by constructing the multi-scale rainfall grid and dividing the urban waterlogging response units, and the modeling and simulation method of an existing urban waterlogging model is expanded.

The invention provides a virtual power plant resource optimization scheduling method and system considering source load uncertainty. The method specifically comprises the following steps: establishing a wind and light related output scene by using a coupling function theory, an average influence value method and an agglomerated hierarchical clustering algorithm; establishing a virtual power plant flexibility demand model considering dual uncertainty of wind and light output and load demand; establishing a load response model and a flexibility supply model of the multi-element flexibility resources of the virtual power plant; establishing a probability mass function and a cumulative distribution function based on the flexibility demand of the source load historical data; establishing a flexibility quantitative index and a flexibility supply excitation mathematical model which simultaneously consider predictable and unpredictable flexibility requirements; and establishing a master-slave game model of the virtual power plant operator and the load aggregator, and solving the clearing electricity price of the virtual power plant to obtain a flexible resource response strategy. According to the method, by optimizing flexible resource configuration, renewable energy consumption is improved, and economic and low-carbon operation of the virtual power plant is achieved.

The invention discloses a method and system for determining a temperature representative value, and relates to the field of statistics, and the method comprises the steps: carrying out the fitting replacement of abnormal temperature data in one year through employing a cubic interpolation method, carrying out the smoothing processing through employing Gaussian filtering, and carrying out the calculation of a temperature representative value; the temperature difference is calculated, a difference distance matrix of the temperature change characteristics of different months is obtained by using adaptive dynamic time warping (Adaptive DTW) and a mean Euclidean distance, a comprehensive distance matrix is calculated by combining weight parameters with the Adaptive DTW and the Euclidean distance, cohesive hierarchical clustering analysis is performed on the temperature change characteristics, and a plurality of temperature change modes are obtained; a typical summer temperature month and a typical winter temperature month are determined according to the plurality of temperature change modes, a daily extreme temperature difference is obtained from the typical summer temperature month and the typical winter temperature month, and probability distribution of the daily extreme temperature difference in different return periods is obtained by using a generalized extreme value distribution model for the daily extreme temperature difference; according to the probability distribution result, temperature action representative values of different return periods are obtained. According to the invention, the temperature action representative value can be accurately obtained.

The present invention belongs to the technical field of aviation airborne system safety analysis, and specifically relates to a method and system for generating FMES data for airborne system safety analysis. The method comprises: S1: acquiring FMEA data of the airborne system, performing data processing, and extracting data features using the BERT model; S2: using sinusoidal similarity to measure the FEMA structured data feature vectors and calculate the similarity of the FEMA structured data features; S3: obtaining the FEMA structured data feature similarity matrix and performing a first agglomerative hierarchical clustering analysis; S4: performing a second clustering analysis on the tree-like clusters of the FEMA structured data features based on fault tree analysis (FTA); and S5: generating structured Failure Mode and Effect Summary (FMES) data based on the second clustering results. The present invention proposes a complete framework for automatically generating FMES data, and improves the quality of the automatic generation through secondary clustering.

The invention discloses an electronic medical record distributed sharing model based on a hybrid alliance chain and a stackelberg game, a hybrid alliance chain architecture is constructed, and credible and safe sharing of EMRs data is realized through cooperation of multiple participants. Then, decision conflicts possibly faced by participants in the sharing process are analyzed by applying a Stackelberg game theory, and a corresponding incentive mechanism is designed, so that each party can obtain a reasonable return while sharing data. In addition, the decomposition characteristic and the rapid convergence of the alternating direction multiplier method are utilized, and the profit maximization of the EMRs data provider and the effectiveness maximization of the EMRs data consumer are realized in a distributed mode. And finally, clustering the consensus nodes in the alliance chain network by using a condensation hierarchical clustering algorithm to improve the traditional PBFT consensus algorithm, improve the efficiency of EMRs data sharing in the alliance chain network, and further reduce the existence of data islands.

The invention discloses a differentiated regulation and control method for irregularity of a large-span bridge track of a high-speed railway. The method comprises the following steps: integrating track irregularity historical detection data at different environment temperatures into multi-stage track irregularity historical detection data with environment temperature labels; establishing a multivariate empirical wavelet transform method, and obtaining a plurality of intrinsic mode components; defining the intrinsic mode component as an intrinsic mode, extracting multi-dimensional physical feature vectors of all the intrinsic modes to construct feature vectors, and performing unsupervised classification on all the intrinsic modes by using an agglomerated hierarchical clustering algorithm; component category division is carried out according to a classification result, and different differentiation regulation and control strategies are made according to different component categories; and when the orbit adjustment amount required by the classification exceeds the limit, establishing an optimization model containing the adjustment amount physical constraint and the smoothness expectation target, and solving the optimal orbit adjustment amount considering the current state and the future temperature adaptability.

The present application relates to a kind of table area distributed photovoltaic power ultra-short term prediction method and system, belong to distributed photovoltaic power prediction technical field.Based on the condensed hierarchical clustering algorithm, the table area photovoltaic in region is clustered into the homogeneous sub-region set of output characteristics, combined with GraphSAGE graph neural network and Transformer encoder model, the decoupling representation of space-time characteristics in sub-region is realized, based on composite space-time characteristics, finally synchronously output the photovoltaic power prediction result of each table area in sub-region, further based on the shared feature between sub-region, construct suitable migration modeling strategy, realize the fast lightweight modeling of each sub-region prediction model, finally, the spatial aggregation of sub-region prediction power obtains regional total power prediction result.

The invention provides an intelligent insurance evaluation monitoring and early warning system based on a logistics transportation whole process, and relates to the technical field of logistics monitoring. Comprising the steps of collecting transportation data in a logistics transportation process; preprocessing the transportation data; performing unsupervised clustering on the preprocessed logistics data by using an agglomerated hierarchical clustering algorithm, and dynamically dividing transportation risk types to obtain risk cluster labels; constructing an intelligent prediction model fusing a light GBM model and a neural network model, and obtaining and outputting a risk probability distribution threshold within a future t moment; and judging whether the real-time detection data exceeds a preset risk probability distribution threshold, and if so, giving an alarm and reminding. According to the invention, through the intelligent prediction model fusing the light GBM model and the neural network model, the risk probability distribution threshold in the future t moment is predicted, the time and computing resources required by model training are reduced, and the efficiency of the whole logistics risk prediction work is improved.

The application provides a key event identification method and system based on a hierarchical processing architecture. The method according to the application comprises: acquiring reported events from different channels, performing data preprocessing on the reported events to obtain processed event texts, and classifying the event texts; adopting a semantic regularization condensed hierarchical clustering method to cluster the classified texts; identifying repeated events in the texts according to the clustering results by using a cluster-in-entity keyword extraction and comparison mechanism to obtain key event identification results. That is, the key event identification method proposed in the application constructs a hierarchical processing architecture of "classification-clustering-identification", effectively solves the technical problems existing in the screening of one event reported by multiple persons, and finally screens out repeated reported events as key events that need to be marked, so that relevant management personnel can give priority to the disposal or sufficient attention to these key events.

The application discloses a shortest path divide-and-conquer search method based on a condensation level, improves the operation efficiency of an algorithm by introducing network layering, limiting a search level, divide-and-conquer search and the like, wherein the network layering is adopted to divide network nodes according to a condensation level clustering method, a plurality of level subgraphs are obtained, and the complexity of the search is reduced; the search range is limited within a certain level by limiting the search level, and invalid calculation in the search process is avoided; meanwhile, in order to further improve the search efficiency, the divide-and-conquer search is adopted to divide the whole search process into a plurality of subtasks and perform simultaneously, and the whole search method can accurately and efficiently perform the shortest path search in an edge computing network.

The invention discloses a rural power distribution network adjusting method and system considering load damage and node coupling degree. The method comprises the steps of collecting voltage, current, injection active power and injection reactive power of each load node of a power distribution network; based on the dynamic impedance matrix, obtaining an electrical distance between the load nodes, and according to the electrical distance between the load nodes, carrying out partitioning on the load nodes by adopting an agglomerated hierarchical clustering method; in any partition, a reactive power compensation node selection model is established according to the impedance mode margin and the node comprehensive coupling degree, and reactive power compensation nodes in the partition are determined according to the model; based on the least square method, a fitting model of the reactive compensation capacity in the partition and the impedance modulus margin at the reactive compensation node is established, and when the minimum value of the impedance modulus margin at the reactive compensation node is smaller than an impedance modulus margin threshold value, the impedance modulus margin threshold value is used as a benchmark, and based on the fitting model, the impedance modulus margin is calculated; and determining the reactive compensation capacity configured by the reactive compensation node in the partition. According to the invention, the problems of uneven load, large power loss and unstable voltage of the rural power distribution network are solved.

The application discloses a kind of dynamic blockchain sharding method and system based on composite clustering algorithm, the method includes: based on blockchain sharding system, determine transaction frequency matrix and sharding allocation matrix, construct the target optimization transaction model of blockchain sharding system;Based on the target optimization transaction model of blockchain sharding system, preliminary global division processing is carried out through agglomerative hierarchical clustering algorithm;The preliminary blockchain sharding result is divided and processed by DBSCAN clustering algorithm, and the blockchain sharding result after secondary division is obtained;Based on the blockchain sharding result after secondary division, carry out noise account division processing, obtain the final blockchain sharding result.The application can preferentially aggregate high-frequency associated accounts and classify low-frequency accounts, minimizing the proportion of cross-shard transactions and achieving load balancing.The application is a kind of dynamic blockchain sharding method and system based on composite clustering algorithm, which can be widely applied in the field of blockchain transaction technology.

The invention relates to the technical field of electrical parameter measurement and industrial equipment fault diagnosis, in particular to a UV curing machine fault detection method based on data analysis, and the method comprises the steps: continuously collecting voltage data points of a UV curing machine, and calculating the voltage data points of the UV curing machine according to the difference between the voltage data points and the mean value of surrounding data segments; and determining a suspected noise factor according to the difference between the voltage data point and the adjacent voltage data point, and calculating the noise probability in combination with the suspected noise factor and the fluctuation possibility. In recursive aggregation of the agglomerate hierarchical clustering algorithm, every two clusters are used as a group, score values are calculated according to noise probabilities and distances of data points in the group, a group of data points on the basis of single-link aggregation is determined according to the score values so as to optimize the agglomerate hierarchical clustering algorithm, and anomaly detection is performed on voltage data points through the optimized algorithm. And determining whether the UV curing machine generates a voltage fault or not. According to the method, the accuracy of fault detection of the UV curing machine is improved.

The application discloses a temperature representative value determination method and system, and relates to the field of statistics, which comprises the following steps: using a cubic interpolation method to fit and replace abnormal temperature data in a year, and using Gaussian filtering to perform smoothing processing; calculating a temperature difference and using an adaptive dynamic time warping (Adaptive DTW) and a mean Euclidean distance to obtain a difference distance matrix of temperature change characteristics of different months; using a weight parameter to combine the Adaptive DTW and the Euclidean distance, calculating a comprehensive distance matrix, performing a condensed hierarchical cluster analysis on the temperature change characteristics, and obtaining a plurality of temperature change modes; determining a typical summer temperature month and a typical winter temperature month according to the plurality of temperature change modes, and obtaining a daily extreme temperature difference therefrom; using a generalized extreme value distribution model to obtain a probability distribution of the daily extreme temperature difference in different return periods; and obtaining temperature action representative values in different return periods according to the probability distribution result. The application can accurately obtain temperature action representative values.

The invention discloses a routing control method based on a node neighbor energy peak value and aggregation hierarchical clustering, which can reduce network energy consumption so as to prolong a network life cycle. The method comprises the following specific steps: firstly, clustering sensor nodes through an agglomerated hierarchical clustering algorithm; then, parameters about the radius of the cluster are set in the cluster, the value of the total residual energy of all nodes within a certain distance is solved, and the node corresponding to the maximum value is taken as a cluster head; and finally, selecting the most suitable routing node to which the next hop arrives by adopting a formula screening method according to the distance, and transmitting the data to the base station by multiple jumps. According to the method, the network life cycle is prolonged to a certain extent. In the invention, the cluster head is selected according to the energy of the nearby (within a certain range) points, so that the nodes with high nearby node density and very low nearby energy are not selected, and the communication efficiency is improved.

The application discloses a label flipping attack method for a machine learning model and a defense method thereof. In the label flipping attack method, firstly, training data to be attacked is selected by performing condensed hierarchical clustering on the training data, label flipping attack is performed on the selected training data, and then a classification model is trained by using a obtained contaminated data set to implement the attack; in view of the above attack, a small pure set and the contaminated training data are used, the TrAdaBoost method is used to update the weight of the contaminated data, the attacked data is identified according to the updated weight, the data without a label or with a wrong label is re-labeled, a new training set including the contaminated training set and the small pure set is obtained, and the classification model is re-trained by using the new training set to improve the model performance; the label flipping attack can be quickly and accurately performed on the training data vulnerable to contamination, and the suspicious contaminated data can be found, and the label flipping attack can be prevented by disinfecting the contaminated data.