384 results about "Data Pooling" patented technology

In embodiments, systems and methods for improving machine-learning systems are disclosed. In embodiments, a system includes a data pool system that is configured to receive data from a plurality of different data sources and maintain a training data set that is used to train a specific machine-learning model based on the data from the plurality of different data sources. In embodiments, the system further includes a data scoring system that determines a data reliability score corresponding to the new data based on a set of intrinsic features of the new data and a data scoring model, wherein the data pool system selectively adds the new data to the training data set based on the reliability score of the new data. The system also includes a machine learning system that trains the specific machine-learning model based on the training data set.

The invention discloses a network security analysis early warning system based on artificial intelligence, and the system comprises a data collection layer which captures full flow based on DPI, aggregates firewall logs, terminal behaviors and threat intelligence, and constructs a structured data pool; through TLS fingerprint identification of AI driving, the encrypted traffic is penetrated, and a sampling strategy is dynamically adjusted in combination with reinforcement learning. The intelligent analysis layer is used for carrying out cross validation on known threats and abnormal behaviors; the time sequence CNN extracts encrypted traffic features, and a novel threat detector is rapidly generated by using historical attack fragments in combination with a meta-learning framework; sHAP value driving dynamic feature selection and optimization feature vector input; the decision-making early warning layer is used for fusing multi-source features through a Bayesian network and generating 0-100 score risk scores; a self-adaptive threshold module is combined to adjust a score threshold in real time, and a high-risk event is pushed; the collaborative response layer is used for triggering a preset decision tree, deploying a GAN dynamic honeypot to trap an attacker and reversely tracing; the Neo4j visually restores the attack path, and blocking is executed after the threat is confirmed by a progressive response mechanism.

The invention relates to the technical field of environmental governance and remediation, in particular to a combined contaminated site soil remediation data management method, which comprises the following steps of: firstly, integrating multi-source heterogeneous data through a unified space-time coordinate system to generate a fusion data pool, and then driving a pollution assessment model to generate a multi-dimensional space-time pollution map; and then intelligently matching a candidate repair process chain based on an associated knowledge base, performing iteration deduction through a dynamic multi-objective optimization engine to obtain an optimal repair scheme, converting the optimal repair scheme into a digital regulation and control instruction, and finally monitoring an execution state and environmental response data in real time and feeding back and updating a data pool to realize closed-loop optimization. According to the method, the data quality and the pollution evaluation accuracy are improved, the adaptability of the repair process and the cost performance of the scheme are optimized, the repair effect is ensured to be stable and reach the standard, and the efficiency and reliability of the repair project are improved.

The invention discloses a ground engineering construction quality supervision digital delivery system, which relates to the technical field of engineering supervision, and comprises the steps of generation and multi-source data checking, activity record structuring and anti-verification before hiding, space-time reference unification and point cloud fusion playback, rule checking and differential rectification, and object-level marshalling delivery and asset authority management. According to the method, a unified data pool, activities, events and evidence chains are constructed through object codes to achieve time-space consistent association, image, point cloud and detection consistency check is conducted before hiding, blocking difference and unified reference fusion point cloud are generated, indexes support three-dimensional playback, rule base batch check is conducted, and issuing rectification is conducted. And the object-level delivery list and fingerprints support cross-platform identification and auditing to form a traceable link.

The invention relates to the technical field of source network load storage intelligent scheduling, and discloses a source network load storage AI intelligent scheduling method and system, and the system comprises a data collection module, a fluctuation analysis module, a constraint calculation module, a load modeling module, an energy storage analysis module, a decision engine module, and a safety correction module. Output data of a photovoltaic power station, a wind power plant and a traditional power plant are collected in real time through an Internet of Things sensor, and a multi-source heterogeneous energy data pool is constructed; aligning the corrected source load data with the energy storage state and the power grid operation parameters according to a time sequence to generate a four-dimensional optimization combination matrix; matching an optimal scheduling algorithm through a bionic search strategy, and analyzing the data matrix through the deep reinforcement learning model to generate a three-section scheduling instruction; and finally, the power grid control system executes power generation adjustment, load regulation and control and energy storage charging and discharging instructions. According to the system, the edge computing gateway is adopted to realize data acquisition, the new energy consumption capability and the power grid stability are improved, and the risk of source-grid load-storage collaborative failure is reduced.

The invention discloses a reinforcement learning model training method and device based on dynamic data selection and electronic equipment, and the method comprises the steps: carrying out the screening from an original mathematical problem data pool according to preset data quality, diversity and difficulty evaluation indexes, and obtaining a candidate training set; calculating a data validity score of each training sample in the candidate training set; difficulty filtering is carried out on the candidate training set, and an adaptive training course is constructed; further filtering the candidate training set according to the solution reasoning process and the reward score of the candidate training set; dynamically selecting a data subset for the current training period by setting a temperature parameter attenuated along with the training process and a probability threshold based on the further filtered candidate training set and the data validity score; the selected data subsets are put into a reinforcement learning model for training, and the reinforcement learning model is trained in an optimized mode according to the correctness reward and the format reward. The model training period can be remarkably shortened, and the performance equivalent to or even better than that of full data training is achieved.

The invention relates to the technical field of three-dimensional modeling, in particular to a dynamic modeling method for a geological structure three-dimensional model. The method comprises the steps that multi-source data are acquired and preprocessed, a voxel semantic fusion algorithm based on variational optimization is introduced, after semantic information is extracted from the preprocessed multi-source data, the preprocessed multi-source data are mapped to a target three-dimensional space grid, and an optimal semantic fusion vector is obtained; based on the optimal semantic fusion vector, generating a standard voxel data pool, and constructing a geological structure three-dimensional model; monitoring data change, calculating the position of a newly added data point, combining the standard voxel data pool to obtain a space updating area, and modeling the space updating area to realize model updating; and after the modeling of the space updating region is completed, optimizing the boundary continuity. The problems that multi-source geological data cannot be directly used for structure construction and semantic fusion of a three-dimensional model, dynamic response to newly-added data is lacked, and geometric discontinuity and structural logic discontinuity exist at the boundary are solved.

The invention provides monitoring equipment and system for monitoring a reactor core of a nuclear power plant. The monitoring device comprises a data preprocessing module, a model updating module, a monitoring module, a shared data pool and a monitoring interface module. On one hand, on-line monitoring of a reactor core is realized by tracking actual operation data of a reactor, and the accuracy of a monitoring result is improved; and on the other hand, the reactor core three-dimensional power distribution reconstruction is carried out in combination with the actual measurement current, so that the monitoring result (such as the second reactor core physical parameter) obtained based on the reconstructed reactor core three-dimensional power distribution is more matched with the actual operation condition of the reactor core, and the accuracy is higher.

The invention relates to the technical field of meteorological data processing, in particular to a training data acquisition method and system for a power transmission line fault prediction model. The method comprises the following steps: constructing a multi-source heterogeneous original data pool containing severe convection meteorological data, power transmission line operation and fault data and geographic space data; performing time-space reference unified processing and grid division on the data; mapping the line paragraph to a grid unit, identifying a severe convection event and dividing an influence area through a density clustering algorithm based on a multi-parameter collaborative threshold value, calculating a grid exposure degree, and combining a strength grade weight and a terrain correction weight to calculate a space correlation strength; setting a differentiation time association window to calculate a time factor according to the severe convection dominant type, calculating association credibility based on the space association strength and the time factor, and screening high-credibility association results; and performing attribution labeling on the fault information to obtain a training data set. The problem of low training data quality is solved.

The invention discloses a traffic flow prediction method and device, a storage medium and electronic equipment. The method relates to the technical field of traffic flow analysis, and comprises the following steps: constructing a multi-source traffic data pool, and preprocessing historical traffic data in the multi-source traffic data pool to obtain a target multi-source traffic data pool; performing data enhancement processing on the historical traffic data in the target multi-source traffic data pool by adopting a pre-trained GANs model to obtain an enhanced sample set for training a target traffic flow prediction model; performing model training on an initial traffic flow prediction model by adopting a Dropout method based on the enhanced sample set to obtain a target traffic flow prediction model meeting a preset condition; and performing traffic flow prediction on multi-source traffic data acquired in real time by using the target traffic flow prediction model to obtain a traffic flow prediction result. According to the method, the accuracy of urban traffic flow prediction can be improved.

The invention relates to the technical field of artificial intelligence, and discloses an information security model auxiliary decision-making method and system based on an intelligent knowledge graph, and the method comprises the steps: constructing a multi-modal security data pool of a to-be-decided scene, training a joint extraction model of the to-be-decided scene, and extracting data examples and example relationships of the multi-modal security data pool; constructing a security knowledge graph of the scene to be decided; constructing an attack tactics-technology-process ontology layer of the scene to be decided to calculate a potential attack path of the scene to be decided, and calculating an attack path occurrence probability and an attack influence range of the potential attack path; marking a strategy decision point of the security knowledge graph, and analyzing a Top-K strategy of the strategy decision point by using a preset PPO algorithm; and constructing an attack chain analysis interface of the Top-K policy in the security knowledge graph to generate a policy optimization parameter of the Top-K policy, and executing the information security model aided decision of the scene to be decided based on the policy optimization parameter. According to the invention, the efficiency of security decision-making of to-be-decided scene information can be improved.

The invention discloses a health monitoring system of an old-age care robot, which relates to the technical field of health monitoring, and comprises the steps of constructing an indoor multi-modal data pool, generating an encrypted feature vector, establishing a steady-state physiological behavior baseline according to a historical encrypted feature vector, and outputting a health risk level and an intervention instruction set. According to the invention, all-time and all-scene health monitoring is realized by integrating a multi-mode sensor, old people do not need to wear equipment, feature compression and homomorphic encryption technologies are adopted, data security is ensured, a threshold value can be updated in real time, and the method is suitable for being applied to the health monitoring of the old people. The health state of the old people is accurately monitored, health risk prediction and personalized intervention are carried out in combination with analysis of behaviors such as falling and static abnormity, the prompt mode is dynamically adjusted according to the ability of the old people, the data processing efficiency is optimized through the edge cloud collaborative architecture, the data privacy and stability are guaranteed, and the life quality and safety are improved.

The invention discloses a power industry language model training data dynamic selection method, system and device, and the method comprises the steps: obtaining original scene corpus data based on a power industry power grid dispatching and fault handling application scene aimed at language model training, and constructing a candidate data pool and a scene verification data set; performing parallel or pipelined multi-dimensional quality evaluation on the candidate data in the candidate data pool, dynamically adjusting the weight corresponding to each dimension in the multi-dimensional quality evaluation based on an adjustment strategy, and calculating to obtain a comprehensive quality score; a failure case library is constructed, in the language model training process, the failure case library is updated regularly, semantic relevance between candidate data and the failure case library is quantified, and batch construction of training data is carried out based on the semantic relevance; and based on the batch construction result, performing language model training feedback circulation. According to the invention, the overall efficiency and utilization rate of training data are effectively improved.

In various embodiments, a system for detecting and preventing equipment failure using a plurality of agents includes an interface configured to receive information about a state of the equipment from a plurality of sources. The system includes a data pool including the information from the interface. The system further includes a plurality of agents configured to access data in the data pool, where the plurality of agents are trained to assess a condition of the equipment based on the data. The system includes an orchestrator configured to: evaluate an assessment of a first agent; access the second agent for further assessment in response to the evaluation of the assessment of the first agent indicating access of a second agent; and cause a modification to be made to the equipment.

The invention discloses a structure construction deviation automatic identification and safety evaluation method, which comprises the following steps: acquiring multi-modal original signals such as space measurement, environment, equipment operation and images from multiple positions of a structure construction site, and establishing a trend drift feature distribution model through time sequence alignment, normalization, noise reduction and feature extraction; then through combination with generative artificial intelligence, virtual samples of historical and extreme working conditions are expanded, and construction of a mixed working condition data pool is realized; through distribution drift analysis and multi-target parameter optimization, a deviation detection threshold value is dynamically and adaptively adjusted, parameters and a judgment result are pushed to a digital twinning and safety control platform in real time, and meanwhile, whole-process log and performance archiving are carried out, so that the sensitivity, adaptability and system safety of structure construction deviation detection are improved, and the construction safety is improved. And the method has relatively high data fusion efficiency and algorithm generalization ability.

The invention relates to the field of geographic information mutual analysis, and discloses a multi-source dynamic collaborative geographic information mutual analysis method and device, equipment and a medium. The method comprises the following steps: collecting time-synchronized static geographic information, dynamic GPS positioning data and a plurality of pairs of starting and ending point unstructured text data sets, and constructing a dynamic spatio-temporal data pool; high-quality spatio-temporal data is obtained through dual anomaly verification and Kalman filtering cleaning correction; optimizing an address longitude and latitude mapping table in combination with a three-level administrative range and a dynamic high-frequency word; a non-standard address is analyzed into a structured text through a BERT-CRF model, and a precise coordinate and a standardized text are obtained through a hierarchical dynamic matching algorithm. By adopting the method, meter-scale positioning is realized, the non-standard address resolution accuracy is greatly improved, the duplicate name address mismatching rate is reduced, a dynamic scene is adapted, the resolution efficiency and reliability are considered, and an efficient technical support is provided for bidirectional accurate mutual resolution of geographic coordinates and text addresses.

The invention discloses a cross-border e-commerce service data management method, and relates to the technical field of service data management, and the method comprises the steps: obtaining commodity service data from a plurality of overseas e-commerce platforms, and caching the commodity service data to an edge data pool according to the attributes of a data source; based on source attributes, updating time and historical accuracy of the data sources and supervision rating of judicial jurisdictions, comprehensive credible weight values of the data sources are calculated, and main data channels are screened out according to the weight values; inputting commodity service data in the main data channel into a multi-language semantic encoder, generating semantic vectors, calculating cross-language semantic alignment, and comparing logic consistency of key structured fields; generating a consistency score according to a comparison result of the semantic alignment degree and the structured field, when the score is lower than a preset threshold value, marking as abnormal and entering a rechecking process, and otherwise, entering a compliance review stage; and calling the regional compliance rule base to carry out regulation matching on the data passing the consistency verification.

The invention discloses a credible industrial data space construction method, and particularly relates to the technical field of industrial data credible management and sharing. Collecting identity feature data and behavior data of an industrial subject, and establishing a behavior data pool; generating a behavior view map based on the behavior ontology model; constructing a behavior consensus index model, and evaluating historical behavior credibility; performing semantic alignment on the behavior credibility index and the current access intention, constructing a credibility mapping function in combination with a multi-source decision factor, and generating a credibility rating value of the access behavior; a one-time pass token is generated or access is denied according to a rating result, behavior feedback is used for model optimization and graph evolution, and finally an extensible trusted industrial data space is constructed; according to the method, dynamic trust judgment, behavior semantic understanding and cross-domain cooperative control of industrial data access are realized, and the credibility, the adaptive capability and the structure treatment capability of a data space are improved.

The invention relates to the technical field of land reclamation planning, and discloses a land reclamation collaborative planning system based on GIS and BIM fusion. The system comprises a basic data acquisition module which is used for acquiring a historical land GIS spatial data set, a historical BIM component data set and a historical improvement project data set of a target area; the multi-level integration module is used for carrying out space-time correlation integration on the data set to generate a land reclamation multi-source fusion data pool; the strategy generation module is used for performing multi-factor dynamic indexing in the data pool based on a preset renovation target and generating an initial collaborative planning strategy set; the strategy adjusting module is used for dynamically adjusting the initial strategy through a dual-channel regulation and control mechanism and outputting a target collaborative planning strategy; the execution feedback module drives planning execution and collects a real-time land state feedback data set; and the data pool updating module is used for comparing the difference factors of the feedback data and the preset target and updating the multi-source fusion data pool. The system improves the dynamics and accuracy of land reclamation planning.

The invention provides an intelligent ocean data security sharing and grading disaster early warning system based on artificial intelligence federal learning, and is applied to the technical field of data processing. According to the method, federal feature extraction and cross-domain association analysis processing are performed on basic data pool data of an intelligent ocean data security sharing and grading disaster early warning system, and cross-domain data association fusion model data is generated; processing the data of the cross-domain data association fusion model, and generating data security sharing and grading disaster early warning collaborative strategy parameters based on federal deep learning; processing the data security sharing and grading disaster early warning collaborative strategy parameters to generate a dynamic security sharing and early warning adjustment scheme; processing the dynamic safety sharing and early warning adjustment scheme and sea disaster historical data, and generating a data feature and marine disaster mapping model based on a federal intelligent diagnosis algorithm; and processing the target sea area monitoring information to generate intelligent ocean data security sharing state assessment and graded disaster early warning information.

The invention discloses a new energy station operation index multi-dimensional comparative analysis display method and system, and belongs to the field of new energy station operation management and data analysis. The method comprises the following steps: fusing and standardizing multi-source heterogeneous data, collecting and cleaning multi-station data, and classifying and storing the multi-station data as a standardized data pool; rasterized multi-dimensional indexes are calculated, grids are divided, data are mapped, and grid-level and hierarchical multi-dimensional indexes are generated; performing dynamic weight and TOPSIS evaluation, constructing an evaluation model to calculate the score of a single station, extracting a core index intersection, and transversely comparing the ranking of the stations; the evaluation result is subjected to instantiation storage and hierarchical storage to guarantee data query and tracing; and performing multi-dimensional visual display, displaying the data by multiple terminals, and generating a decision report. The system comprises corresponding function modules. According to the method, the problems of data fragmentation, evaluation system deficiency, non-visual display and the like are solved, multi-station data integration, quantitative evaluation and visual decision making are realized, and the operation management efficiency is improved.

The invention relates to the technical field of power grid exception handling, and discloses a power grid exception event handling method and system based on a knowledge graph. The method comprises the following steps: collecting time sequence data flow and space coordinate information of a power grid abnormal event in real time through multi-source sensing equipment, and fusing equipment operation parameters to construct a dynamic event data pool; performing graph structure conversion on the dynamic event data pool based on a knowledge graph topological connection relationship, calculating event node propagation path intensity and generating an event propagation network model; key abnormal indexes are separated out from the model by adopting a multi-modal feature extraction technology, risk probability calculation is completed in combination with an adaptive learning algorithm, and an abnormal risk distribution diagram is output; and classifying event levels according to the graph, and automatically generating a processing instruction sequence. According to the method, omnibearing data perception and accurate analysis of the abnormal events of the power grid are realized, an event propagation rule is clearly presented, risk assessment is enabled to be closer to an actual operation state, and accurate guidance is provided for abnormal disposal.

The invention discloses a green smart port rating method based on big data, particularly relates to the field of port green rating, and is used for solving the problem of lack of analysis of influence of collaborative scheduling on carbon emission in existing port green rating. By collecting multi-source real-time data, a multi-modal data pool of a transfer full chain is constructed, carbon emission traceability across transport entities is achieved, and a seamless carbon track chain is generated; a multi-agent strategy matrix is adopted to simulate a port and transportation cooperation scene, carbon emission dynamic influence data under different decision combinations are extracted, and cooperation efficiency indexes such as a supply chain carbon intensity variation coefficient and an anchoring ground idling index are calculated; establishing a mapping relation between decision behaviors and collaborative efficiency through a regression model, performing green collaborative prediction on future job scheduling, and generating a port green collaborative rating; a dynamic, traceable and predictable green evaluation mechanism is realized, a quantitative reference is provided for port scheduling strategy optimization, and the port low-carbon cooperative treatment capability is improved.

In various embodiments, a system for detecting and preventing equipment failure using a plurality of agents includes an interface configured to receive information about a state of the equipment from a plurality of sources. The system includes a data pool including the information from the interface. The system further includes a plurality of agents configured to access data in the data pool, where the plurality of agents are trained to assess a condition of the equipment based on the data. The system includes an orchestrator configured to: evaluate an assessment of a first agent; access the second agent for further assessment in response to the evaluation of the assessment of the first agent indicating access of a second agent; and cause a modification to be made to the equipment.

The invention relates to the technical field of sanitation vehicle management, and discloses a sanitation vehicle life cycle management system and method, and the system comprises a data fusion module, a butt joint design software interface, a production assembly database, a vehicle-mounted sensor group, a disassembly evaluation terminal, and a construction data pool. Associating design parameters, real-time assembly tolerance records, use state flow and recovery damage data; the twinborn modeling module is used for creating a digital twinborn body of a VIN index based on the data pool, loading an assembly process standard to the multi-physics field model, and driving the model to output a key component state index and an assembly error prediction value; a dynamic calibration module; a decision generation module; and an execution control module. According to the method, multi-source heterogeneous data is associated through a unified space-time reference, a VIN-indexed digital twin is created, an assembly process standard is loaded to a multi-physics field coupling model, a key component state index and an assembly error prediction value are driven to be output in real time, and accurate dynamic mapping from a physical world to a virtual model is achieved.

The invention discloses an intelligent disaster avoidance path planning method for a mine. The method comprises the following steps: acquiring environmental parameters, disaster information, personnel distribution data and risk avoidance facility state data of the mine in parallel; when it is continuously detected that parameters exceed a safety threshold for multiple times, disaster early warning is triggered, a disaster event identifier is generated, and a dynamic data pool is formed by delimiting associated data in an influence area and a convergence area according to disaster information by taking an event as a center; disaster diffusion simulation and personnel evacuation trajectory simulation are executed based on the dynamic data pool, and a disaster diffusion simulation result and a personnel evacuation trajectory simulation result are fused to form a dynamic obstacle map; and obtaining a plurality of paths according to the personnel distribution data and the risk avoiding facility state data, and obtaining a disaster avoiding path based on the dynamic obstacle map by adopting a multi-objective optimization algorithm according to the path length, the safety redundancy and the group evacuation efficiency. According to the method, the dynamic planning of the mine disaster avoidance path is realized, and the path safety and the evacuation efficiency are remarkably improved.

The invention discloses an improvement potential quantification method for carbon emission in a mining life cycle process, and relates to the technical field of mine carbon emission management, and the method comprises the steps: collecting real-time sensor data, equipment operation logs, process parameters and unstructured documents of each stage in the full life cycle of a gold mine, and constructing a dynamically updated multi-source data pool; based on the multi-source data pool, carbon emission entities and a carbon emission entity semantic relationship are extracted, and a carbon footprint knowledge graph is constructed; performing multi-objective optimization on a collaborative decision result, solving a non-dominated solution set, screening an optimal equilibrium solution, and generating an optimization strategy sequence and corresponding expected carbon emission; and inputting the optimization strategy sequence into a mine production scheduling process for execution, monitoring actual carbon emission data in real time, comparing the actual carbon emission data with expected carbon emission, and correcting the dynamic carbon emission reference model through a comparison result. According to the method, the carbon footprint knowledge graph is constructed, so that the semantic understanding and correlation analysis of the whole-process carbon emission of the mine from exploration to pit closing are realized.

The invention discloses an order issuing control method and device based on a rule engine and electronic equipment, and the method comprises the steps: extracting effective order information through automatic recognition of order features; then, the orders are classified according to a preset rule, that is, the orders pushed by an upstream system are stored in different data pools based on rule matching through three data pools of rejection issuing, instant issuing and planned issuing; finally, the orders can be issued in order according to the pushing rules of different data pools. Therefore, compared with the traditional method of directly issuing the orders to the downstream system, the method has the advantages that the classified pushing of the orders is realized, so that the flexibility of issuing the orders is improved, the ordered issuing of the orders is realized, the operation pressure of the downstream system is reduced, and the operation stability of the system is ensured; meanwhile, the whole process does not need manual intervention, the efficiency is improved, and the error rate is reduced; therefore, the method is very suitable for large-scale application and popularization.

The invention discloses a power transmission line typical defect image processing evaluation method, system and device and a storage medium, and the method comprises the steps: collecting power transmission line inspection image data and line operation environment parameter data, and carrying out the convergence of the data, and constructing a multi-source heterogeneous data pool; based on an improved multi-head self-attention mechanism, performing feature extraction and fusion on the power transmission line inspection image data in the multi-source heterogeneous data pool to form an enhanced feature vector; a multi-dimensional image anomaly detection model is preset to perform preliminary detection of an image anomaly region on the enhanced feature vector, and a feature association network is constructed based on a preliminary detection result; and performing quantitative scoring and grade judgment on the defect features based on the feature association network. The method can improve the accuracy and efficiency of the typical defect image processing evaluation of the power transmission line, and guarantees the safe and stable operation of the power transmission line.

The invention relates to the technical field of automatic driving, and particularly discloses an off-line evaluation and iterative optimization method and system for an automatic driving system, and the method comprises the steps: recording a task-level data packet, and binding a reproducible fingerprint for the task-level data packet; in the off-line stage, a consistent input data stream is reconstructed, and the consistent input data stream is input into a current online running old model in parallel to be compared with a to-be-evaluated candidate new model for evaluation; according to a comparison evaluation result, extracting a scene fragment data packet based on a preset multi-dimensional evaluation index and a triggering rule; the method comprises the steps that scene fragment data packets are obtained, failure type labels are automatically generated for the scene fragment data packets, bucket dividing and weighted sampling are carried out according to scene features, and scene fragment data are returned to a model training data pool, full-link data packaging is carried out with task-level data packets as the granularity, reproducible fingerprints are bound for all the data packets, and therefore the scene fragment data can be obtained. Therefore, the offline playback, the A / B evaluation result and the backflow training sample can be subjected to consistent indexing and tracing through the fingerprint.