82 results about "Time clustering" patented technology

The invention discloses a multi-stage spatial-temporal clustering method and system based on a fused mahalanobis distance. The method comprises the following steps: acquiring a spatio-temporal data set, and determining a spatio-temporal neighbor relation of samples in the data set; calculating a space communication distance and a time decay distance of the sample; fusing the space communication distance and the time decay distance by using a mahalanobis distance to obtain a relative distance of the sample; selecting a class cluster center from the data set according to the local density of the sample and the relative distance; adopting a multi-stage distribution strategy to distribute non-class-cluster center samples to corresponding class clusters; wherein the multi-stage allocation strategy comprises an inevitable allocation stage based on space-time shared neighbor and a similarity allocation stage based on a weighted similarity matrix. According to the method, the key problems that an existing space-time clustering algorithm is insufficient in space-time attribute coupling processing and sensitive to distribution errors are solved, and the clustering accuracy, robustness and practicability in the fields of intelligent traffic analysis, seismic sequence recognition and the like are remarkably improved.

The invention discloses a time-space clustering-based instant retail intelligent distribution scheduling method, medium and system, and the method comprises the steps: obtaining the geographic information of a hierarchical geographic unit through an electronic fence, collecting the data of a to-be-distributed order, and calculating a distribution time window through a time window prediction model, so as to generate time-space clustering information; according to the method, orders with overlapped delivery time windows in the same geographic unit are aggregated into task packages with the same address, the task packages are dynamically allocated through a matching algorithm based on the real-time state of a rider, and algorithm parameters are updated according to delivery feedback. According to the invention, the accuracy of order aggregation and the rationality of task allocation are obviously improved, and the refined scheduling of transport capacity resources and the optimization and improvement of the overall distribution efficiency are realized.

The invention provides a traffic flow prediction method based on a graph convolutional network, and the method comprises the steps: training a space-time cluster prediction model through the traffic flow data of historical specific time steps of all traffic nodes in a traffic network graph, so as to predict the traffic flow of a future target time step, and specifically, carrying out the training of a time-space cluster prediction model through employing a convolutional layer, mapping the preprocessed traffic flow data to a feature space to obtain original features and an original feature map of the traffic flow data; extracting a spatial dependency relationship implied in the traffic flow data by using a self-attention-based space-time diagram cluster convolution module to obtain a corresponding feature representation; capturing a time dependency relationship of the traffic flow data by using a gating timing sequence convolutional network based on a gating mechanism and expansion convolution to obtain a corresponding feature representation; and predicting the traffic flow of a future target time step based on the feature representation of the spatial dependency relationship and the time dependency relationship of the historical specific time step. According to the method, external information related to time is introduced, so that the prediction accuracy is improved.

The invention discloses a distributed service deployment method and a distributed service deployment device based on Kubernetes. According to the method, resource object definition and service metadata are extracted by automatically analyzing a configuration file, and accurate acquisition and standardized processing of deployment information are realized; optimal deployment nodes are dynamically matched based on service characteristics and cluster real-time states, and resource scheduling rationality and service operation efficiency are improved; then automatically creating and starting a service instance at a target node according to an analysis result, and ensuring configuration consistency and standardization of a deployment process; and the service flow is controllably migrated to the new instance through the predefined version switching rule, so that smooth transition and risk controllability of version upgrading are realized. According to the method, the efficiency, the reliability and the version management capability of micro-service deployment are remarkably improved through full-process automatic closed loop from configuration analysis, intelligent scheduling to instance creation and flow switching.

The invention discloses a multi-sensor track fusion method, device and equipment, a storage medium and a computer product, and relates to the technical field of data processing, and the method comprises the steps: carrying out the time clustering processing and the space clustering processing of a plurality of tracks collected by a plurality of sensors in sequence based on the time information and the space distribution information of the tracks, and obtaining a multi-sensor trajectory cluster, and fusing all trajectories in the multi-sensor trajectory cluster to obtain a fused trajectory of the corresponding target object. According to the method, first clustering processing is carried out on a plurality of tracks collected by each sensor according to time information to form a preliminary track cluster, so that independent analysis processing of each cluster is realized, the complexity of subsequent calculation is reduced, and then the spatial distribution condition of each track is determined according to spatial distribution information, so that the accuracy of track distribution is improved. Therefore, the calculation data volume of sensor track fusion is further reduced, and the technical effect of improving the multi-sensor track fusion efficiency is achieved.

The invention provides an ERP system dynamic configuration method and device and a medium, and relates to the technical field of resource configuration management.The method comprises the steps that a real-time order list in an ERP system is obtained through dynamic monitoring, and a preset clustering strategy is called to obtain a real-time clustering result; performing return analysis on the first clustering cluster to obtain a first cluster return index, and obtaining initial cluster weight distribution; obtaining a first cluster prediction return index of a predetermined period through a return prediction model; when the first index ratio accords with a predetermined index ratio constraint, target cluster weight distribution is obtained through adjustment; obtaining a resource database; generating a target resource allocation decision in combination with the target cluster weight allocation; and carrying out dynamic configuration processing on the ERP system. According to the resource allocation management method and device, the problem of resource utilization efficiency of existing resource allocation management can be solved, the hysteresis of resource allocation management is effectively avoided, and the technical effect of improving scientificity and effectiveness of preposed resource allocation processing of the ERP system is achieved.

A bot detection service is enhanced by providing improved threat scoring using dynamic clustering of telemetry data received from a server. The detection approach leverages the notion that bot traffic is statistically anomalous compared to empirical models of human traffic. In particular, bots have more similar and repeating network patterns (i.e., lists of field values derived from client telemetry) that lend themselves to tighter clustering than random human traffic. The technique herein leverages this notion by providing a bot detection service with a network pattern clustering algorithm that differentiates bot-versus-human traffic by searching for similar and repeating network patterns that cluster together by virtue of having lesser variation (as compared to human traffic) in their field value sets.

The invention relates to the technical field of distributed deep learning, and discloses a multi-target distributed deep learning container scheduling method and system, and a storage medium, and the method comprises the steps: carrying out the modeling of a communication structure between GPUs in nodes, the communication performance in the nodes, and the communication performance between the nodes through employing a graph neural network, generating an embedded vector representing node interaction characteristics, and obtaining a cluster state; inputting the cluster state into the deep Q network, defining a mapping relation from a container to a node through an action space, and optimizing a deep Q network decision based on a reward function so as to minimize communication overhead, balance node loads and train the deep Q network; and deploying the deep Q network model after offline training to a scheduler, and generating a container scheduling scheme according to a real-time cluster state. According to the scheduling method provided by the invention, the communication bottleneck in distributed deep learning job training is effectively relieved while the scheduling real-time performance is ensured.

The invention provides a method and system for dispatching heterogeneous GPUs through cross-cluster management software, and relates to the technical field of GPU resource dispatching.The method comprises the steps that physical topology data, hardware specification data and historical task records of GPU nodes are obtained through the cross-cluster management software, an association degree matrix between GPU units is established, and a topology grouping set is generated; calculating a communication efficiency index based on topology detection analysis, constructing a topological graph model, and determining an optimal communication path of a multi-card task; generating a perception scheduling strategy in combination with a multi-objective optimization algorithm, and selecting an optimal GPU combination from the grouping set; in the task execution process, a scheduling strategy is dynamically adjusted according to the real-time cluster state and task efficiency data, continuous adaptation of a physical topological structure and task requirements is kept, and finally efficient collaborative scheduling of cross-cluster heterogeneous GPU resources is achieved. According to the method, the scheduling efficiency of heterogeneous GPU resources in a distributed computing scene and the overall performance of the system are improved.

The invention relates to the field of battery packs, in particular to a method and a system for judging a voltage difference fault of a battery pack. The method comprises the following steps: acquiring a voltage value of any battery module in a battery pack at a target moment, constructing graph structure data for the battery pack, and clustering the graph structure data by utilizing spectral clustering to obtain a plurality of clusters; obtaining an adjacent cluster, matching the adjacent cluster with the target moment cluster, and determining a value of a judgment function; calculating a pressure difference abnormal value of the battery module according to the calculated pressure difference variable quantity; and obtaining the accumulated value of the abnormal voltage difference value of any battery module in the preset sampling period, and taking the maximum value of the accumulated value in the battery pack as the abnormality degree of the battery pack to complete the fault judgment of the battery pack. According to the technical scheme of the invention, the precision of the voltage difference fault judgment result of the battery pack can be improved, and the operation safety of the battery pack is ensured.

The invention relates to the technical field of centrifugal pump control, and discloses an intelligent control multi-working-condition ultrahigh-lift single-stage centrifugal pump control method which comprises the steps that attention-guided dimension reduction fusion and multi-scale feature extraction are conducted on multi-source working condition data, and multi-scale operation features are obtained; time evolution and clustering processing are conducted on the multi-scale operation characteristics through a time clustering algorithm based on density, and the current working condition of the ultrahigh-lift single-stage centrifugal pump is obtained; and taking the current working condition, the performance parameters and the performance score of the ultrahigh-lift single-stage centrifugal pump as state vectors, and selecting an optimal control vector from a control space by adopting a reinforcement learning model. By fusing multi-source data and introducing time evolution modeling and reinforcement learning optimization strategies, intelligent sensing and self-adaptive control of the ultra-high-lift single-stage centrifugal pump under multiple working conditions and the whole life cycle are achieved, the problems that response to complex operation states is slow and adjustment lags in traditional rule driving control are solved, and the service life of the ultra-high-lift single-stage centrifugal pump is prolonged. The energy efficiency ratio, the operation stability and the equipment service life are improved.

The invention discloses an expressway traffic flow prediction method based on a spatial-temporal clustering generative adversarial network under an emergency, belongs to the field of expressway traffic flow prediction, and remarkably improves the precision of expressway traffic flow prediction under the emergency. The method comprises the following steps: step 1, acquiring traffic flow data from a traffic detection system, and constructing a traffic network graph structure; 2, integrating geographical location information, traffic function similarity and delay propagation characteristics to construct a comprehensive clustering algorithm, and realizing the comprehensive clustering algorithm; 3, designing a generator network based on a space-time attention encoder and a discriminator network structure fusing a bidirectional long-short-term memory network and graph convolution based on the clustering result and the feature representation in the step 2, and constructing a complete prediction model; 4, performing prediction model training based on the composite loss function; and step 5, applying the trained prediction model to real-time data prediction to realize traffic flow prediction under emergencies.

The invention relates to the technical field of power monitoring, in particular to a concentrator with an electrical equipment state monitoring function. Obtaining an unbalance degree according to the state characteristic values among the three phases; clustering is carried out according to the unbalance degree of all historical moments, and distribution intensity is obtained according to interval characteristics of different moments in a moment cluster; obtaining a data similarity according to the unbalance degree in the time cluster; obtaining reference credibility according to the moment quantity characteristics, the distribution intensity and the data similarity in the moment cluster; according to the difference characteristic of the unbalance degree corresponding to the current moment and the reference moment cluster and the reference credibility of the reference moment cluster, obtaining the abnormity degree; and obtaining an anomaly trend value according to the anomaly degree of the adjacent historical time period of the current moment. The early warning index of the current moment is obtained according to the abnormity degree and the abnormity trend value, and early warning is carried out, so that the monitoring accuracy of the concentrator on the power utilization state of the electrical equipment is improved.

The invention relates to the technical field of intelligent traffic, and discloses a vehicle passing track auditing analysis system and method based on spatial-temporal feature clustering, and the system comprises a topology construction module which is configured to construct a highway networking charging attribute graph based on nodes and edge sets, and form a topology network of a node reachable relationship; the feature processing module is configured to perform standardization processing on the spatio-temporal features of the historical passing tracks and eliminate dimensional differences; the clustering analysis module is configured to perform adaptive clustering based on the standardized spatio-temporal characteristics and determine division of a time cluster and a space cluster; and the trajectory reasoning module is configured to generate candidate paths in combination with the space-time cluster characteristics, calculate a comprehensive score through space-time similarity calculation and weight distribution, and output an optimal reasoning path. The method corresponds to the system. According to the invention, the efficiency of the networking charging auditing work is greatly improved, and the national networking charging order is effectively maintained.

The invention discloses a federated learning method based on one-time clustering and inter-cluster prototype alignment, and the method comprises the steps: firstly, each client calculates and uploads a feature signature based on a distributed isomorphic proxy model; and the server performs one-time clustering on the clients based on the feature signatures. In the subsequent iteration training, the server distributes the corresponding cluster level agent model and the global class prototype to each cluster client; each client maintains a heterogeneous privacy model and an isomorphic proxy model, and performs joint training; after training, the client uploads the updated proxy model and a newly generated local class prototype; and the server aggregates the proxy model in each cluster, aggregates the local class prototypes of all the clients to update the global class prototype, and continuously iterates until the model converges. By using the method, huge communication and calculation overhead caused by repeated clustering in the training process is avoided. The method can be widely applied to the technical field of distributed machine learning.

The invention relates to the technical field of GPU resource management, and discloses a GPU resource pool dynamic management system and method based on a container platform, a scheduling demand is obtained by analyzing a user resource request, and resource matching and scheduling decision making are performed based on the demand and a real-time cluster resource state. And issuing and executing a resource pool management instruction according to the decision, and finally aggregating operation results to synchronously update the cluster state. The method aims at solving the problems that in the prior art, a GPU resource pool management mode is rigid, a scheduling strategy is solidified, and heterogeneous resources are difficult to manage in a unified mode. The GPU resource pool can be dynamically created, updated or released as required by constructing a closed-loop management process which is driven by a strategy and from request analysis to state synchronization, so that unified abstraction and scheduling of heterogeneous GPU resources are realized, and the flexibility of resource management, the system response efficiency and the adaptability to diversified business scenes are remarkably improved.

The invention relates to a DUET photovoltaic power generation power prediction method and system based on space-time bi-clustering and dynamic gating fusion, and belongs to the technical field of new energy power system prediction. For the problems of low prediction precision and poor robustness caused by meteorological dynamic disturbance of photovoltaic power, a time clustering module TCM is adopted to extract potential distribution and a linear mode of historical power data, and a time drift error is solved; key control variables such as scattered radiation and temperature are screened through a channel clustering module (CCM) in combination with causal sparse constraint and a frequency domain attention mechanism, and periodic features are enhanced; a bidirectional space-time cross attention and adaptive gating technology of a fusion module FM is utilized to dynamically integrate a time sequence self-correlation characteristic and a cross-channel joint influence. According to the method, the prediction precision in sudden weather change is remarkably improved, the generalization ability and the real-time performance of the model are enhanced, and reliable decision support is provided for power grid dispatching.

The invention discloses a distributed storage method and device based on a time sequence database, and relates to the technical field of data storage. Obtaining original time sequence data and carrying out time clustering segmentation to obtain a time sequence data fragment set; performing data merging on the time sequence data fragments in the time sequence data fragment set to obtain a merged data fragment set; determining a storage position of each merged data fragment in the merged data fragment set; constructing a time query index for the data in each merged data segment; and for the data in each merged data segment, obtaining a storage address of the data, enabling the storage address to correspond to the time query index of the data, and storing the data in the storage address. According to the method, data of a time sequence is clustered and then segmented, then the segmented data is merged, storage fragmentation is reduced, finally, a storage address and a time query index are associated by constructing a query tree, efficient storage is still met in the face of mass data, and the data management efficiency is improved.

The invention discloses an optimal scheduling method for a virtual power plant, and relates to the technical field of power regulation, and the method comprises the steps: obtaining multi-modal data of a detection region, calculating a regional load fluctuation index and a resource mobility index, and generating a dynamic feature data set; performing real-time clustering division on the power grid nodes through a clustering algorithm based on the dynamic feature data set to generate a dynamic sub-region set; if the sub-region feature fluctuation exceeds a boundary stability threshold value, triggering boundary redivision; according to the dynamic sub-region set, comparing a building power consumption predicted value with the available load of the charging pile in real time to calculate a demand difference, and generating an elastic scheduling instruction through multi-objective optimization; uncertain factors are monitored in real time, a risk index is calculated, and the priority and the resource allocation strategy of the elastic scheduling instruction are dynamically adjusted based on the risk index; and generating a risk avoidance scheduling scheme through a robust optimization algorithm.

The present application relates to the technical field of cluster resource scheduling, in particular to a dynamic computing resource elastic scheduling method for a heterogeneous server cluster, comprising: receiving an upper-layer application computing task description and analyzing resource demand characteristics and dependency constraints; pre-selecting initial candidate server nodes in a cluster global resource state space through an improved deep reinforcement learning algorithm with a dynamically adjustable reward function; constructing a multi-objective constraint optimization model in combination with node real-time load and task demand; calculating an adaptation score and scheduling the task to an optimal node for execution; and collecting task execution progress in real time and comparing it with resource demand to dynamically adjust reinforcement learning algorithm parameters. This method makes node pre-selection more in line with task demand, keeps scheduling decisions consistent with real-time cluster and task states, optimizes cluster load allocation states, and enhances the dynamic adaptation capability of the scheduling process.

The invention discloses a multi-source load management system based on dynamic regulation and control, and the system comprises a space-time clustering feature calculation module which employs a multivariable recombination formula to calculate a regulation and control task recombination value, combines a regulation and control region, a time period, and a load space-time position, generates space-time features of each load through a space-time clustering feature formula, and stores the space-time features of each load; quantifying feature significance based on a distribution probability formula, and generating a feature sequence according to a probability descending order; the load regulation and control task matching module is used for dividing loads into a plurality of sets through feature distance calculation, calculating comprehensive features of the sets, calculating matching values of tasks and the sets by adopting a multi-dimensional matching formula, and selecting an optimal matching set; and the load regulation and control task execution module executes a regulation and control task on the load in the matching set according to the priority of the feature sequence, ensures that the high-probability feature load responds preferentially, and supports a dynamic adjustment strategy. According to the invention, stable and efficient operation of a power system can be guaranteed, the intelligent management level is improved, and fine management of complex multi-source loads is realized.

According to the method, the complexity of a remote rural distribution environment is considered, the high dispersion characteristic of order space-time distribution is considered, the differentiation characteristic of multi-vehicle-type transport capacity resources is also considered, and the order combination distribution method based on space-time clustering in a rural common distribution mode is provided. In the first stage, customer geographic positions and order time window constraints are comprehensively considered, reachability-guaranteed clustering is formed to realize order combination, and an initial path feasible solution is constructed based on a space-time clustering result; in the second stage, a simulated annealing framework is designed to be fused into adaptive large neighborhood search, a clustering result is dynamically adjusted according to multi-vehicle-type differential time-space characteristics, and an initial solution is improved; on-demand combination of orders, accurate path planning and multi-vehicle collaborative distribution are realized, and logistics enterprises are helped to cope with distribution challenges of scattered orders, fuzzy time windows and complex road networks in remote rural areas with lower cost on the premise of ensuring distribution timeliness and service quality.

The invention discloses a submarine cable risk area dynamic division method and system based on spatio-temporal clustering and dynamic weight analysis, and the method comprises the steps: multi-source data collection and preprocessing: collecting multi-source spatio-temporal data related to submarine cable risks, and carrying out the cleaning, normalization, abnormal value elimination and missing value filling of the multi-source spatio-temporal data, performing space-time alignment on the data by using Kalman filtering and smooth interpolation; spatio-temporal feature extraction and three-dimensional spatio-temporal cube model construction: extracting spatio-temporal features from the acquired multi-source spatio-temporal data, and dynamically constructing a three-dimensional spatio-temporal cube model of submarine cable risks based on the spatio-temporal features; streaming density clustering analysis: carrying out streaming clustering analysis on data in the three-dimensional space-time cube model by adopting a clustering algorithm, introducing a dynamic weight distribution mechanism, and adjusting the weight of a dynamic risk index; and risk assessment and region dynamic division: forming a risk index based on a clustering result to carry out submarine cable risk assessment so as to dynamically divide risk region grades.

The application discloses a meteorological-vegetation drought transmission characteristic identification method and system based on a three-dimensional clustering algorithm, which comprises the following steps: S1, identifying meteorological drought events and vegetation drought events in a research period based on a three-dimensional clustering algorithm; S2, determining characteristic variables of the meteorological drought events and the vegetation drought events; S3, proposing a drought transmission identification rule based on three-dimensional space-time clustering, and matching meteorological-vegetation drought event pairs; S4, determining types of the meteorological-vegetation drought event pairs; and S5, extracting characteristic variables of a meteorological drought to vegetation drought transmission process, and clarifying the meteorological-vegetation drought transmission process. The method and system can quantitatively identify characteristic changes in a meteorological-vegetation drought transmission process, reveal transmission characteristics of meteorological drought to vegetation drought, clarify the internal relationship, response characteristics and transformation mechanism between meteorological drought and vegetation drought, and provide scientific and technological support for accurately predicting and preventing vegetation drought.

The invention provides a link flooding attack defense method and system based on a data plane programmable SDN (Software Defined Network), and relates to the technical field of network security defense. The method comprises the following steps: carrying out real-time clustering analysis on a Traceroute data packet passing through a switch by adopting an online clustering algorithm based on a weighted Manhattan distance, positioning a suspicious link and sending alarm information to a controller; establishing a link attribute graph sequence by collecting link state data based on the alarm information, and performing abnormal link identification by using a space-time double-attention detection model based on GAT-Seq2Seq to obtain an attack path; and identifying a link flooding attack flow by adopting an attack flow identification method based on reinforcement learning, and discarding the link flooding attack flow at a source end switch of an attack path. According to the method, the link flooding attack defense system with high accuracy, low resource overhead and quick response capability is constructed, so that efficient defense of attacks is realized.

The invention provides a multi-task learning model, a training method, electronic equipment and a computer storage medium, and relates to the technical field of multi-task learning. The method comprises the following steps: dividing data according to timeliness to obtain processed data; wherein the processed data comprises real-time layer data, short-term layer data and long-term layer data; inputting the processed real-time layer data into a real-time network for training to obtain a real-time feature vector and a corresponding real-time clustering center; inputting the processed short-term layer data into a short-term network for training to obtain a short-term feature vector and a corresponding short-term clustering center; inputting the processed long-term layer data into a long-term network for training to obtain a long-term feature vector and a corresponding long-term clustering center; and optimizing the real-time network, the short-term network and the long-term network through the loss function. And the generalization ability and the prediction precision of the multi-task learning model on the timeliness difference data are improved.

The invention belongs to the technical field of underwater multi-target positioning and real-time clustering of unmanned surface vehicles, and particularly relates to a target positioning and real-time clustering method based on foresight sonar detection, which can realize simultaneous positioning and clustering of underwater multi-target information, realize real-time detection and positioning of underwater multiple targets during navigation of an unmanned surface vehicle, and improve the positioning accuracy of the underwater multiple targets. And the underwater target positioning precision is ensured through a clustering method. By fusing the high-precision space-time reference provided by the unmanned ship inertial navigation system and the distance and azimuth information of the multiple targets obtained by foresight sonar detection, the geographic coordinates of the multiple underwater targets can be calculated at a time. Compared with a traditional underwater acoustic positioning mode in which a special transponder needs to be arranged for a single target, the method greatly improves the efficiency of multi-target detection and the integrity of information acquisition, and provides a key technical support for comprehensively mastering the underwater situation.

The invention provides a dirty page brushing method and a related product. The dirty page disk brushing method is applied to any node in a database cluster, and comprises the following steps: responding to a global check point process started by the database cluster; participating in the collection of redo sites by the database cluster; node-level dirty page disc brushing is carried out for multiple times; synchronizing dirty page disk brushing progress of the buffer area with other nodes; and performing one-time cluster-level dirty page disk brushing with other nodes. The method has the advantage that the dirty page brushing efficiency during the global check point period can be improved.

The invention relates to the technical field of data processing, and provides an efficient space-time clustering method based on GPS trajectory data, and the method comprises the steps: obtaining target GPS trajectory data; carrying out data cleaning on the target GPS trajectory data, wherein the data cleaning comprises abnormal value deletion, trajectory noise removal and trajectory smoothing processing; matching the target GPS trajectory data points with map data to generate at least one target three-dimensional trajectory; performing trajectory segmentation on the target three-dimensional trajectory data by adopting a compression linear reference technology to obtain a representation result of each target three-dimensional trajectory in a two-dimensional CLR space; determining the longest common subsequence of each pair of tracks based on the representation result of each target three-dimensional track in the two-dimensional CLR space, and calculating a similarity matrix based on the longest common subsequence of each pair of tracks; and carrying out track clustering based on the similarity matrix. Through the above mode, the compressed linear reference technology is utilized, the representation form of the trajectory is simplified, and the calculation efficiency and accuracy of the trajectory similarity are enhanced.

The invention provides a track segmentation labeling method and related equipment, and relates to the technical field of track data processing. Original track data of a moving target is acquired; clustering the original trajectory data by adopting a space-time clustering method to obtain a plurality of clusters, and segmenting trajectories in all the clusters in combination with an adaptive sliding window to obtain a segmentation result; carrying out manual annotation on the segmentation result to obtain a manual annotation result, taking the manual annotation result as priori knowledge, and carrying out feature extraction on the segmentation result to obtain multi-dimensional spatial-temporal features; performing automatic label inference on unlabeled segments in the segmentation result based on the multi-dimensional spatio-temporal features to obtain an inference result, and calculating the confidence of the inference result; the inference result is optimized according to the confidence coefficient, and a trajectory labeling result of the moving target is obtained; and the labeling efficiency is remarkably improved while the labeling accuracy is ensured.