52 results about "Recursive algorithms" patented technology

The invention discloses an intelligent power prediction method considering dynamic load change, and relates to the technical field of power grid load prediction, and the method comprises the steps: collecting original load data, carrying out the preprocessing, constructing a VMD constraint optimization model, carrying out the four-stage improvement of optimization parameters through employing an improved dung beetle optimization algorithm, so as to generate IMF components, reconstructing the IMF component by calculating a sample entropy to obtain a low-frequency component and a high-frequency component; establishing a Kalman filtering state space model based on the low-frequency component, and decomposing the low-frequency component into a residual component and a pseudo trend component through a Kalman filtering recursive algorithm; external features are obtained, the high-frequency component, the residual component and the pseudo trend component are aligned and spliced with the external features, multi-component collaborative prediction is carried out through a local-global interactive attention mechanism, and a final load prediction result is obtained; and generating a power demand visualization chart based on the final load prediction result. And reliable decision support is provided for power dispatching and energy management.

The invention provides a self-adaptive regulation and control method for a hydrogen production system based on real-time parameter feedback, and relates to the technical field of process control, the method comprises the following steps: collecting an input instruction signal and a controlled state variable signal of the hydrogen production system, and comparing a fluctuation energy index with a preset identification dead zone threshold; if the fluctuation energy index is lower than the identification dead zone threshold value, the current controller parameter is kept unchanged; if the fluctuation energy index is higher than the identification dead zone threshold value, activating an online parameter identification step; calculating a dynamic confidence interval of process model parameters; judging whether the process model parameters estimated in real time fall into the dynamic confidence interval or not; if the process model parameters fall into the dynamic confidence interval, converting the process model parameters estimated in real time into control parameters of a PID (Proportion Integration Differentiation) controller, and updating the control parameters into a control loop of the hydrogen production system; and if the process model parameter exceeds the dynamic confidence interval, resetting an internal state vector of the recursive algorithm in the online parameter identification step by using a boundary value.

The invention belongs to the technical field of industrial automation control, and relates to a seasoning filling PID parameter adaptive optimization control method and system. The method comprises the steps that the real-time temperature of fluid in a filling pipeline, the static pressure of a valve inlet, the instantaneous flow and a valve opening instruction are collected, and moving average filtering preprocessing is carried out; constructing a self-adaptive forgetting factor calculation model based on the viscosity-temperature characteristics, and calculating a current optimal forgetting factor according to the fluid temperature change rate, the flow prediction error and the fluctuation condition of the flow data; a recursive algorithm is used for identifying the system process gain online, and the proportional gain of a PID controller is calculated in real time according to the identified process gain and the ratio of the hydrostatic pressure to the standard reference pressure; and the gain is input into a controller, and a final valve opening instruction is generated in combination with integral and differential terms. According to the method, the dynamic adjustment of the algorithm memory length can be realized, the viscosity change can be quickly adapted, excessive filling and slow response are eliminated, and high-precision filling under complex working conditions is ensured.

The invention relates to the technical field of agricultural environmental engineering and control, and discloses an agricultural non-point source pollution multistage control planning system and method based on a discussion hall architecture, and the system comprises an organization-level module, an adaptation-level module, a coordination-level module, an operation control-level module, a controlled-level module, a comprehensive integration discussion hall module, and an open external interface module. The organization level uses deep learning and causal inference to generate a full-watershed strategy; the discussion hall module fuses expert experience conversion constraints and carries out case migration; the adaptation level utilizes a recursive algorithm to dynamically correct model parameters; the coordination level adopts a hybrid optimization algorithm to solve a multi-objective optimization problem; the operation control level is combined with a feedforward feedback mechanism to generate an instruction; the controlled level executes actions and collects feedback; and the interface module is responsible for security encryption and verification of data interaction. According to the method, the expert intelligence of the discussion hall and the five-level hierarchical architecture are fused, qualitative and quantitative combination and full-basin dynamic collaborative optimization are realized, and the management and control precision and the response speed are improved.

The invention relates to a monitoring data recursive calculation method and application thereof in a CCUS project. An estimation error used for describing the difference between an estimation value and a true value and a measurement error used for describing the difference between a measurement value and the true value are introduced; the comprehensive error is obtained by dividing the previous estimation error by the sum of the previous estimation error and the current measurement error; an initial estimation value and an initial estimation error are randomly given, then an initial comprehensive error is estimated, then a subsequent estimation value is calculated through a recursive algorithm, and the initial comprehensive error is updated to be used for calculation of the subsequent estimation value. After a plurality of rounds of measurement and calculation, an estimated value is closer to a real value, and all measurement data in the earlier stage does not need to be backtracked, so that the problems of a large amount of repeated redundancy and large error of monitoring data are effectively reduced; and the method provided by the invention is used for realizing carbon dioxide leakage monitoring in a CCUS project.

The embodiment of the invention provides a category and library table dependency relationship determination method and device, equipment, a medium and a product. The method comprises the following steps: for any data access layer bean in a target project, running a first execution code corresponding to the data access layer bean; according to an SQL statement generated during the operation of the first execution code and a method name for calling the SQL statement, determining a first dependency relationship between a method contained in the data access layer bean and library table information; for any control layer bean in the target project, detecting dependency classes of all hierarchies corresponding to the control layer bean through a recursive algorithm, and when the dependency classes are data access layers bean, determining methods included in the data access layers bean; and determining the dependency relationship between the classes and the library table information in the target project according to the dependency classes of all hierarchies corresponding to each control layer bean, the methods contained in the data access layer bean and the first dependency relationship corresponding to each data access layer bean, so that the overall efficiency of analysis can be improved, and the probability of misplacement missing and misplacement can be reduced.

The invention relates to the field of data processing, and provides a medical bed allocation method, device and equipment, and the method comprises the steps: building a house type combination mathematical model; according to the mathematical model, an asynchronous recursive algorithm is adopted to search the house type combination, and an initial house type combination is determined; receiving a filtering condition input by a user in real time, and performing dynamic real-time filtering on the initial house type combination according to the filtering condition; and generating a room type combination record according to the room type combination result meeting the filtering condition, and visually displaying the room type combination record. The problems that in the prior art, medical bed combination distribution calculation efficiency is low, and design constraints cannot be dynamically responded are solved, and efficient, real-time and interactive compliance bed scheme automatic generation and display are achieved.

The invention discloses a function enhancement method based on an activiti workflow engine, and the function enhancement method comprises the following steps: inheriting a DeFaultActivityBehaviorFactoryMappingConfiguer class, and carrying out the monitoring enhancement on the overall situation of the activiti; a core command mechanism of the Activiti is rewritten, so that the functions of randomly returning and randomly skipping the process which is not supported by the original are realized; a set of pluggable form construction factory mechanism is constructed; through an algorithm and data processing, a complex BPMN flow chart is converted into a linked list view which is easy to understand by business personnel, and future path prediction is provided; by customizing a multi-instance behavior of Activiti, real-time statistics of a passing number and a rejection number is added for a countersigning node, and countersigning judgment logic is realized. A future approval path prediction function based on an approval history linked list view and a recursive algorithm is realized, and understanding and decision support of business personnel on a process are remarkably improved.

The invention discloses a method and a device for realizing reliability evaluation of an aero-engine, a double-framework model comprising a structural BN and a dependent BN is constructed based on a model framework such as BN, a recursive algorithm uses the condition independence of the BN to reduce the complexity to a pseudo polynomial level, and the calculation efficiency is remarkably improved. In addition, large deviation caused by a traditional independent hypothesis is avoided, and the evaluation precision is improved. According to the embodiment of the invention, any layered structure and cross-module dependence are supported, key risk path analysis can be output, and technical support is provided for efficient implementation of aero-engine design optimization, predictive maintenance and airworthiness certification.

The application relates to a method, device and equipment for generating a query SQL based on hive metadata, the method comprising the following steps: obtaining metadata information, storing the metadata information to a preset mysql database after marking the metadata information, and obtaining initial metadata; determining field lists of different tables based on the initial metadata, connecting the field lists of different tables, generating a data table association relationship, storing the data table association relationship to a preset subject domain, storing the preset subject domain to the preset mysql database; obtaining a target data relationship in the data model, generating an SQL template for querying the target data relationship by using a recursive algorithm, and developing the SQL template by modification. The application solves the problem of low development efficiency caused by the fact that the data is extracted by means of a big data component due to the writing of a complex SQL statement, the relationship between metadata and data tables needs to be maintained, and an operator needs to write a complex SQL according to the relationship between tables.

The invention relates to the technical field of control engineering, and discloses a control system and method for high-temperature gas heating, and the system comprises a high-temperature gas heating physical entity, a sensing detection group, and an advanced control unit. The advanced control unit comprises the following modules which work cooperatively: a data acquisition and preprocessing module which is used for acquiring system output (a gas outlet temperature measurement value), control input and measurable disturbance data acquired by the sensing detection group, and constructing a data regression vector reflecting dynamic historical information of the system; through a dynamic characteristic matrix self-correction module, a dynamic characteristic matrix representing a system input and output mapping relation is constructed and updated in real time by using a recursive algorithm with a forgetting factor; the method effectively solves the problems of strong nonlinearity and time-varying characteristics caused by dramatic changes of gas physical properties and heating element characteristics along with temperature in a high-temperature gas heating process, and remarkably improves the adaptive capacity of the model in a full working condition range.

The invention provides a ship-bridge collision detection method and system, computer equipment and a storage medium, and belongs to the field of bridge structure health monitoring and safety alarming, and the method comprises the steps: collecting the displacement response of a bridge structure key position under ship collision disturbance; the method comprises the following steps: firstly, expanding acquired response data in a frequency domain by using discrete wavelet transform, extracting a low-frequency approximate signal to carry out co-integration test to obtain a multi-scale co-integration equation, and carrying out online real-time estimation on a co-integration coefficient in each scale equation by using a Kalman filtering recursive algorithm; when the co-integration coefficient changes and exceeds a safety limit, automatic alarm is realized; alarm results of co-integration coefficients of multiple scales are synthesized, and when 75% of co-integration combined alarm is set to be exceeded, a ship and bridge collision alarm decision is made. According to the method, the Kalman filtering algorithm is combined, so that the change of the co-integration coefficient can be predicted on line, and the actual engineering requirements are met. On the basis, multi-scale wavelet decomposition and co-integration improve the robustness and precision of alarm, a plurality of alarm results are provided for comprehensive decision making, and scientific basis and technical support are provided for bridge structure health monitoring and accident prevention and control.

The invention discloses an unmanned aerial vehicle formation flight performance collaborative path planning method and system based on an artificial path potential energy field, and belongs to the technical field of collaborative path planning. The method comprises the steps of 1, extracting feature points of an image, and adjusting the number of the feature points to a target value by using a recursive algorithm; step 2, distributing pattern feature points for each unmanned aerial vehicle based on the target numerical value of the image obtained in the step 1; and step 3, based on the pattern feature points distributed in the step 2, performing cooperative path planning for the unmanned aerial vehicle group flying from the starting point to the ending point, and realizing unmanned aerial vehicle formation flight performance cooperative path planning based on the artificial path potential energy field. The method is used for solving the problems that an existing unmanned aerial vehicle formation collides during flight performance and the path length of the unmanned aerial vehicle reaching a specific position point is not optimal.

The application discloses a user residence point identification method and device, electronic equipment and storage medium. Target signaling is added to each user in first user signaling data obtained, and second user signaling data is obtained. The target signaling is signaling with maximum time stamp for start time and end time and invalid values for longitude and latitude. Recursive calculation is performed on the second user signaling data through a recursive algorithm. A plurality of residence segments corresponding to each user are identified according to a distance threshold and a time threshold. The signaling length of each signaling in each residence segment is used to assign a corresponding weight to each signaling, and the offset calibration residence point of each residence segment is calculated. Finally, the offset calibration residence point of each residence segment and the second user signaling data are judged through the recursive algorithm, and the corresponding residence point of the plurality of residence segments of each user is output. Thus, the accuracy of identifying the residence state of the user can be improved.

This application relates to the field of HVAC equipment control technology, and discloses a multi-protocol compatibility and edge computing method for HVAC equipment digital controllers. The method includes: parsing protocol bytecode to extract spatiotemporal metadata and establishing an instruction distribution table; executing instruction parsing data, calculating weights based on time consumption deviations and updating status words; filtering and writing to a buffer according to weights, or generating loss alarms by counting the number of discards; calculating the rolling variance representing fluctuations using a recursive algorithm based on an effective dataset; determining the trigger freeze signal based on comprehensive parameters, locking the buffer, and generating a fault snapshot. This invention achieves decoupling between the device communication protocol and the controller's underlying firmware by loading protocol bytecode containing spatiotemporal metadata and using a virtual machine engine for interpretation and execution. This allows the edge controller to adapt to HVAC equipment with different physical layer response characteristics and timing requirements without modifying the underlying code or recompiling the firmware, only by updating the protocol bytecode file.

The invention discloses a real estate data analysis method and device based on multi-dimensional data fusion, and the method comprises the steps: carrying out the recursive analysis of to-be-analyzed real estate data through a recursive algorithm when the to-be-analyzed real estate data is received, and obtaining column elements; screening out cell values corresponding to the column elements from the to-be-analyzed real estate data, binding the column elements with the cell values corresponding to the column elements, and displaying on a preset interface; in response to the triggered operation of the measurement unit selection function in the preset interface, displaying a table interface; and displaying the chart interface in response to the triggered operation of the switching function in the table interface. Real estate data are analyzed through a recursive algorithm, column elements and cell values are bound, and meanwhile user-defined configuration of various chart types is provided. Through the optimization, the data processing efficiency and the display speed are improved, a user can more intuitively analyze large-scale complex data, and the performance bottleneck and interaction problems existing when multi-dimensional data are processed are effectively solved.

The invention relates to the technical field of building engineering construction, in particular to a simulated assembling method and system for fabricated building components. According to the method, the differentiated historical vibration retention coefficient and amplitude conversion coefficient are calculated by obtaining the geometric parameters of each steel bar, the cumulative vibration state value is updated by using the recursive algorithm in combination with the real-time vibration signal, and then the dynamic swing coverage radius is obtained through overlay analysis. And finally, the radius is compared with the effective guide radius of the sleeve in real time, a setting control instruction is automatically generated, accurate quantification and safety decision making of the non-synchronous swing risk of the reinforcing steel bar in the blind area are achieved, and the construction safety and the automation level are improved.

The invention discloses a power station equipment and measurement data association method based on a BiLSTM-CRF algorithm, and relates to the technical field of power station equipment data processing.The method comprises the steps that multi-source data of power station equipment is obtained in an acquisition and input mode, preprocessing including abnormal value and missing value processing, format standardization and space-time alignment is carried out, and the data of the power station equipment is obtained; obtaining a standardized intermediate data set; inputting the standardized intermediate data set into a power enhancement language model, identifying an equipment entity, a measurement type and a semantic relationship, and outputting an entity tag sequence; according to the method, a four-dimensional knowledge graph is constructed on the basis of an association relationship between entity tag sequence identification equipment and measurement points, and time complexity is optimized from a square level to a linear logarithm level through an optimization recursive algorithm, so that automatic association of primary equipment, secondary equipment and the measurement points of a plant station is realized. According to the invention, the problems of long time consumption and easy mismatching and mismatching omission caused by dependence on manual configuration of power station equipment and measurement data association are solved.

The invention discloses a risk processing method and device for financial transaction behaviors, equipment and a medium, and relates to the technical field of financial science and technology, and the method comprises the steps: obtaining risk transaction metadata through a preset task node, and setting a task fragment number threshold value in a target configuration file, performing data splitting on the risk transaction metadata based on the task fragment quantity threshold value by utilizing a recursive algorithm to determine a target data fragment set; determining a target risk amount, judging whether the target risk amount is smaller than a preset risk amount threshold, and determining a target priority of the target data fragment based on a judgment result; dividing the target data fragments into a preset queue according to the target priorities of the target data fragments, determining a target idle thread from a preset thread pool, calling the target idle thread through a preset thread scheduler, and executing a preset risk processing task on the target data fragments in the preset queue. In this way, the bottleneck problem of data processing efficiency is solved, and the real-time generation rate and storage throughput of early warning are improved.

The invention provides a database connection selection rate calculation method and a related product, and relates to the field of database optimization. The method comprises the following steps: grouping connection operations in connection query according to data tables participating in connection; traversing each group, and respectively matching a plurality of columns of de-duplication values of the connection columns in the connection tables at the two ends in each group; using a recursive algorithm to calculate the selection rate of connection conditions of each layer in the connection operation layer by layer according to the multi-column deduplication value and the single-column deduplication value of each connection column; multiplying the calculated selection rates of the connection conditions of each layer to obtain a grouping selection rate; and calculating the ratio of the grouping selection rate to the Cartesian product row number of the connection operation to obtain a final result of the connection selection rate. According to the scheme, the multi-column deduplication values of the connection columns of the data tables at the two ends are matched, and the multi-column correlation distribution characteristics are included into computational logic, so that the one-sidedness of single-column statistics is made up, and the accuracy of selection rate estimation in the multi-column connection scene is improved.

The application discloses a kind of for the management system of product bill of materials of virtual prototype, comprising: recursive index module, for using recursive algorithm to describe the hierarchical and grouping relationship between product modules, form recursive system index data model for product;Recursive system index data model contains a series of pointers to product system engineering process data;Integrated interface module is used to describe the interface and interoperation relationship between adjacent product modules, and based on object model, software interface for interoperation between prototype products under virtualization environment is provided.The application makes the data generated in system engineering process form an organic whole, decomposes the system according to engineering habits, indexes all kinds of process data and controls its state, improves the interoperation of subsystem product data, test data and other data, fully utilizes all kinds of process data and result data, and guarantees the direct consistency of final product and system engineering process data.

The invention discloses a credit risk prediction method and system based on a decision tree and a recursive algorithm. The method comprises the steps of 1, collecting a user basic data set based on user historical behaviors; 2, preprocessing the user basic data set to obtain feature data, and storing the feature data for later use; 3, randomly selecting a plurality of variables, constructing a decision tree by combining the selected variables and the feature data, and calculating the risk identification efficiency of the decision tree; 4, repeating the operation of the step 3 by using a recursive algorithm to perform variable mining until a preset recursive cycle number is reached, traversing all the decision trees, selecting the decision tree with the highest risk identification efficiency, and taking a variable combination corresponding to the decision tree as an optimal variable combination; and step 5, risk identification effect evaluation is carried out based on the optimal variable combination, and a risk prediction result is obtained. The method has the advantages of high stability, high efficiency and high interpretability.

The application provides a method and system for eliminating misoperation caused by a large number, which comprises the following steps: obtaining a new sampling value in each sampling interruption, calculating by using a recursive algorithm to obtain a recursive result of a current sampling point; triggering a care task, performing non-recursive calculation on the current sampling point corresponding to the triggering of the care task to obtain a non-recursive result of the current sampling point; comparing the recursive result and the non-recursive result, if the same, ending the care task, waiting for the triggering of the care task again to perform the next round of verification, if different, locking the protection function and clearing the data of the sampling related cache area, and then performing the next round of verification. The method combining the cyclic verification with the protection start trigger verification can accurately and timely identify the cumulative abnormality and the sampling abnormality, effectively solves the calculation precision problem and the misoperation problem caused by the cumulative error and the abnormal sampling, guarantees the safety and reliability of the operation of the relay protection device, and has good engineering practical value.

The application discloses a kind of long-acting time-varying monitoring methods for dam reinforcement based on Bayes real-time updating, comprising: S1. ARMA model is constructed to simulate dam state and determine model order;S2. using prior distribution and likelihood function, using Bayes method to calculate and obtain the posterior density function of each parameter of dam in running state;S3. using last posterior distribution as prior distribution in real-time calculation, repeat the S2 step, in the calculation process, least square method is used to realize model parameter estimation, variable forgetting factor is introduced, and recursive algorithm is used, different forgetting factor is used to past monitoring data, realize parameter adaptive adjustment.The application can adapt to the time-varying characteristics of dam working behavior, real-time fusion historical data and new monitoring information, more accurately and reasonably reflect the nonlinear time-varying relationship between influence quantity and effect quantity, so as to fully embody the typical time-varying characteristics of dam working behavior.

The invention relates to a computer-implemented method (100) for optimizing at least one parameter of at least one function for estimating an open-circuit voltage (Vocv) of at least one electrical energy storage device (12), wherein the function has at least one parameter vector, wherein at least one component of the parameter vector represents the open-circuit voltage (Vocv), and wherein at least one recursive least-squares algorithm is used (104) for estimating the at least one parameter vector, characterized in that a forgetting factor of the recursive algorithm is set to the value 1 (114) if at least one component of the parameter vector is at least temporarily unobservable. The method (100) increases the robustness of the recursive algorithm against situations in which no statements can be made about modeled parameters.

The embodiment of the present disclosure provides a personalized learning path recommendation method based on a fractal recursive structure, comprising: acquiring learning behavior data of a user and constructing a multi-level knowledge network graph; using a fractal recursive algorithm to assign weights to the knowledge network graph and expand the structure; generating an initial learning sequence according to the weight difference of the nesting relationship between knowledge nodes; and optimizing and adjusting the arrangement order of the knowledge points in the learning sequence based on a dynamic regulation strategy. Through the scheme of the embodiment of the present disclosure, the problem of how to regulate the learning sequence of the recommended path according to the weight difference of the multi-level nesting relationship between knowledge points to solve the problem of low knowledge transfer efficiency can be solved.

The invention provides an intelligent substation equipment state sensing and early warning system and method, and relates to the technical field of industrial early warning, and the method comprises the steps: obtaining a multi-physics field heterogeneous feature set comprising an electrical quantity, a network quantity and a physics field quantity based on a substation operation full time sequence; fitting the historical state of the equipment based on a long-short term memory neural network model, and outputting a reference aging trend term; calculating an event-driven catastrophe item under operation vibration by using a damage model based on a physical mechanism, and calculating a difference value between the event-driven catastrophe item and the reference aging trend item to obtain a transient disturbance deviation degree; constructing an orthogonal fusion state vector containing the reference aging trend term and the transient disturbance deviation degree, and performing state updating on a time axis by adopting a recursive algorithm; and based on the updated orthogonal fusion state vector, calculating a dynamic safety margin index and residual hidden danger-free operation time of the equipment. According to the invention, the problem of systematic missing report of the model to the vibration hidden danger is solved.

The application discloses a kind of UKF-based trajectory prediction method of projectile, system, equipment and storage medium, the method includes: through observation model, current time observation position vector under machine coordinate system is converted into current time global position vector under global coordinate system;Current time global position vector is determined based on the current time global position vector through uncontrolled projectile dynamics model current time global state quantity of projectile;According to current time global state quantity, the current time state estimation vector of projectile is constructed;According to current time state estimation vector and next time observation position vector, next time state estimation vector is calculated through UKF recursive algorithm;According to next time state estimation vector, the trajectory prediction of projectile is realized through projectile dynamics discrete state equation.The present application does not need to rely on prior information of the dynamics parameters such as mass of projectile, solves the numerical instability problem caused by large span of parameter value range through logarithmic space mapping strategy, and improves the parameter estimation accuracy.

This invention relates to the field of wind turbine technology, and more particularly to a wind turbine main control system and load collaborative optimization method based on full-condition adaptive identification. The system includes a full-condition adaptive identification unit, a multi-objective collaborative optimization unit, a main control execution unit, a load online monitoring unit, and a data interaction unit. The full-condition adaptive identification unit collects multi-source data on wind conditions, turbine operating status, and structural response in real time, and uses an adaptive recursive algorithm to determine wind characteristics. This invention provides a wind turbine main control system and load collaborative optimization method based on full-condition adaptive identification. This invention can identify wind characteristics, turbine electromechanical parameters, and structural modal parameters online in real time, constructing an adaptive weighted multi-objective optimization system to dynamically balance power generation, load, and fatigue damage. It possesses proactive emergency control capabilities under extreme conditions, significantly improving the control accuracy and robustness of the turbine under all operating conditions, effectively reducing load and fatigue damage to key components, and extending the service life of the turbine.

The invention provides a document difficult negative sample generation method, device and equipment, and the method comprises the steps: reconstructing a sampling document into a three-level index structure based on a recursive algorithm and a large language model, and obtaining a sample database, the three-level index structure comprising an abstract vector, a text block and an abstract; a discriminator of a GAN (Generative Adversarial Network) judges a first probability that the multiple pieces of sample data correctly answer the to-be-processed question based on the abstracts and the text blocks of the multiple pieces of sample data, and takes the sample data of which the first probability is lower than a threshold value as candidate negative samples; the candidate negative samples with the similarity between the abstracts and the answers higher than a threshold value serve as sampling negative samples; a GAN generator generates difficult negative samples based on the multiple sampling negative samples, a loss function comprising semantic loss items and logic loss items is adopted to train a GAN model, the similarity between the difficult negative samples and answers reaches an extreme value, and logic is inconsistent; and generating document difficult negative samples based on the difficult negative samples.