152 results about "Screening algorithm" patented technology

The invention discloses an edge computing power sharing method, an edge computing power sharing device, edge computing power sharing equipment and a medium, and the edge computing power sharing method comprises the following steps: connecting edge nodes through a decentralized P2P network, constructing a network connection module, and realizing computing power resource sharing among the nodes; collecting and monitoring computing power resource state data of each edge node through a computing power resource management module, judging nodes in an idle state through an evaluation algorithm, and adding computing power resources of the nodes into a shared pool; the node matching module is used for screening out an optimal matching node from the shared pool by utilizing a preset screening algorithm according to task demand information released by a computing power demand side; and the distribution execution module sends the task instruction to the selected optimal node from the cloud, and after the task is executed, the node releases the computing power resource and performs expense settlement through the charging module. According to the method, the utilization efficiency of computing power resources is optimized through decentralized node management and an intelligent matching mechanism, so that edge computing resources are more easily integrated and utilized.

The invention provides an automatic deployment method for a robot system, and the method comprises the steps: collecting three-dimensional space data of a target environment through a multi-mode sensor array which comprises a laser radar, a binocular vision camera and an inertial measurement unit; fusing sensor data based on a space-time synchronization algorithm to generate an environment feature matrix; analyzing a user task instruction, extracting a task target and decomposing the task target into time, path priority and safety threshold parameters; constructing a multi-objective optimization model, and generating a candidate path set; adopting an improved Pareto leading edge screening algorithm to select an optimal path sequence; executing virtual verification and adjusting path deviation in the digital twin environment; and the physical execution mechanism is driven to implement deployment operation, and the local path re-planning module is triggered when the deviation is detected to exceed a threshold value. According to the invention, the accuracy, safety and efficiency of path planning can be improved, and the energy consumption and time cost are reduced.

The invention discloses an intelligent Chinese text error correction method integrating spelling and semantics, and relates to the technical field of data processing. Comprising the following steps: S1, according to an original Chinese text to be corrected and a correction intention instruction, obtaining an original candidate correction set of each agent through parallel operation of each agent; s2, processing the original candidate correction set according to the correction intention instruction, and obtaining the score of each original candidate correction item; and S3, comparing the original candidate correction items with the comprehensive evaluation scores through a diversity optimization screening algorithm, and determining a final candidate correction item set. According to the method, the candidate correction set is generated in parallel through multiple agents and is combined with the conflict resolution and multi-dimensional evaluation mechanism, so that the accuracy of error correction can be ensured, diversified expression selections can be provided, and the problems of single and rigid output of a traditional error correction system are avoided.

The invention discloses a large language model enhanced table question answering method based on thinking chain reasoning, and belongs to the technical field of natural language processing. The method comprises the steps that a form question and answer CoTSQL model based on a thinking chain and a framework extraction model are constructed, an example screening module generates an SQL framework architecture of a natural language question through the framework extraction model, and an example similar to a target question is selected by using a screening algorithm according to a framework category; mode linking is carried out in the mode processing module, key table information and mode relations are extracted, and a context environment is provided; the prompt construction module integrates the screened examples and the processed mode information into a CoTSQL style prompt, and inputs a large language model to generate a preliminary SQL query; and the SQL correction module detects and repairs illusion errors generated by the large language model through SQL repairs and a consistency strategy, and executes revised SQL query in a database to finally obtain an accurate table question and answer result. According to the method, the accuracy and robustness of SQL query are effectively improved.

The invention discloses a grain defective product detection method combining fingerprint spectrum and finite element analysis, and particularly relates to the technical field of grain quality detection.The grain defective product detection method comprises the following steps that grain chemical and physical characteristics are extracted through a fingerprint spectrum and a finite element model, and key characteristic vectors are obtained through dimension reduction screening; on the basis of defective grain grade labels, mutual information is calculated, cross-modal analysis is carried out, and a strong dependency feature group is identified; feature sparsity and stability evaluation are combined, features are optimized by adopting a self-adaptive screening method, a food quality knowledge graph is finally constructed, and intelligent detection and quality evaluation of defective food are realized; according to the method, the accuracy of a detection model, namely a grain quality knowledge graph, is improved through feature sparsity and feature stability evaluation, an automatic grain defective product detection and recognition mechanism is constructed by utilizing a multi-modal information fusion method of fingerprint spectrum and finite element analysis and combining a mutual information analysis and dynamic feature screening algorithm, and the detection accuracy is improved. And the dependence on manual intervention and experience judgment is reduced.

The invention discloses an intelligent control method and system for saline-alkali soil slow-release fertilizer application, and the method comprises the steps: analyzing the space-time trend of soil salinity change and pH dynamic change through a convolutional neural network, judging the matching demand of the fertilizer release rate, and generating a target release curve; the method comprises the following steps: extracting a high-activity strain adapting to a high-salt environment from a salt-tolerant microbial library through a high-throughput screening algorithm, and determining optimal release time interval distribution of strain function activation by combining with the release characteristic of a dynamic release coated fertilizer; if the soil salinity change exceeds a preset threshold value, sensor monitoring data and a target release curve are integrated through a closed-loop control system, and an optimized fertilization amount scheme is obtained; a deep learning model is adopted to analyze the relation between the optimized fertilization amount scheme and soil deep salt distribution, the parameter requirement of depth accurate regulation and control is judged, and a layered fertilization depth control strategy is generated. Through real-time feedback and optimization iteration, accurate improvement and sustainable nutrient management of the saline-alkali soil are realized.

The invention discloses a task dynamic decomposition management system and method based on an AI framework, and relates to the technical field of task management, and the method comprises the steps: dividing task parameters into a plurality of dimension features, carrying out the fusion of the multi-dimension features through a space-time attention network, and constructing a three-dimensional feature tensor; recursively decomposing the three-dimensional feature tensor through an adaptive secondary screening algorithm, and outputting a decoupled modal set; predicting a resource load and quantifying a modal demand level based on a long short-term memory network, constructing a bipartite graph matching model, and adaptively allocating resource instances through dynamic matching; time performance and resource consumption data of task execution are collected in real time, a modal deviation value is calculated, if the modal deviation value exceeds a deviation threshold value, dynamic re-decomposition is triggered, resources are reallocated based on an updated modal set, and dynamic capture, efficient decomposition and accurate resource allocation of task features are achieved; therefore, the task execution efficiency and the resource utilization rate are improved.

The application belongs to the technical field of multimedia information processing, and discloses a kind of monitoring video key frame extraction method, system, medium, equipment and terminal, collect original video stream data, original video stream data is decomposed into image frame set;Image frame set obtained by decomposition is sampled, and the image frame result set obtained by sampling is filtered;The image frame set after filtering is adaptively clustered, and the result after clustering is collected to form a video abstract.In order to better utilize the memory space of storage medium and let the user better quickly browse the general content of original video stream, the application provides a key frame screening algorithm for original video stream by sampling, filtering and clustering.The key frame extraction method of the application screens out similar frames, redundant frames and fuzzy frames in original video stream through key frame screening algorithm for original video data, forms a video abstract storage of a section of original video, thereby greatly reduces the occupied storage space.

The invention provides an efficient annotation-driven hierarchical fault positioning method, which comprises the following steps of: guiding a large language model to intelligently analyze and generate semantic annotations of a text based on a positioning algorithm of large model annotations, and performing fine-grained software fault positioning from a file level to a function level and then to a position level by using the annotation-based hierarchical positioning method. And an efficient hierarchical progressive sorting and screening algorithm is used for ensuring the software fault positioning efficiency. According to the method, the defect positioning efficiency can be ensured while the positioning accuracy is ensured. The positioning algorithm based on large model annotation strategically balances the information density, provides enough context clues, and improves the model understanding ability; according to the annotation-based hierarchical positioning method, positioning is divided into file / function / position hierarchies, and annotations are added by using a large model cue word project in sequence, so that the positioning accuracy is improved; and an efficient hierarchical progressive sorting and screening algorithm is provided, so that the optimal balance between the performance and the cost is realized.

The invention discloses a port bulk cargo storage location intelligent distribution method and system, and belongs to the technical field of port storage location management and intelligent distribution. The method comprises the following steps: collecting historical data related to port cargo scheduling, and defining a storage location distribution demand; designing constraint conditions for a storage location distribution process; constructing a multi-objective optimization model based on a genetic algorithm according to the constraint condition; generating an initial feasible solution meeting all hard constraints by adopting a greedy pre-screening algorithm; taking the initial feasible solution as the input of a multi-objective optimization model, and solving the optimal solution of storage location allocation by using an improved NSGA-II algorithm; and according to an optimization result, distributing the goods to the optimal storage location. By adopting the technical scheme of the invention, the problems of low storage location utilization rate, load imbalance and high delivery cost in port multi-type cargo scheduling can be effectively solved, so that the port resource utilization rate and the operation efficiency are improved.

The invention relates to an elastic resource adjustment method in distributed data parallel training. The elastic resource adjustment method comprises the following steps: step 1, monitoring operation performance in real time by using a performance monitoring module; 2, performing comprehensive consideration according to preset adaptability function evaluation and a redistribution punishment mechanism, and performing comprehensive evaluation on the resource condition of the current cluster; when it is judged that the performance of the cluster is seriously influenced when rescheduling is not carried out through an adaptive function, a rescheduling process is rapidly started; 3, according to continuous monitoring and evaluation on the adaptability function, judging which elastic operation needs to be executed on which operation; meanwhile, selecting the most suitable node from the available node pool by using an intelligent node screening algorithm according to an elastic adjustment target; 4, executing an elastic scheduling strategy, and actually carrying out elastic adjustment operation on resources; for different elastic operations of different jobs, dynamic adjustment and optimal configuration of resources are realized through an incremental communication topology reconstruction mechanism and a gradient cache and interpolation compensation mechanism.

The invention belongs to the technical field of artificial intelligence algorithm and bioinformatics crossing, and relates to a drug interaction prediction method based on multi-modal molecular characterization. According to the method, through the edge perception GCNII architecture and the Hop2Token multi-hop coding mechanism, effective modeling of atomic-level and bond-level local environments and a cross-substructure high-order dependency relationship in drug molecules is realized, and the accuracy and robustness of drug interaction prediction are improved. According to the method, Mol2Vec and MolT5 cross-modal molecular characterization is integrated, fusion of molecular overall semantics and substructure grammar semantic association is achieved, and the generalization ability of the model to complex molecules and unknown medicine combinations is remarkably improved. According to the method, the dynamic feature screening algorithm driven by the SHAP value of the artificial intelligence technology is adopted for biological verification, the feature redundancy problem is effectively solved, the molecular biological information analysis processing calculation efficiency and the model transparency are improved, and the traceability of the prediction process is guaranteed.

The invention discloses a solar radiation forecasting method fusing a decomposition algorithm and a GraphCast model, and relates to the technical field of solar radiation forecasting. Comprising the following steps: acquiring meteorological variable data in an ERA5 reanalysis open source database; preprocessing the data to obtain meteorological variable data conforming to the input format of the GraphCast model; a GraphCast model is constructed to carry out weather forecast; acquiring actually measured meteorological variable data and solar radiation data of a certain region site; through an XGBoost feature screening algorithm, meteorological variables with high importance relative to solar radiation are screened out; preprocessing the screened meteorological variables by using a VMD decomposition algorithm, and constructing a training set and a test set by using the preprocessed data; taking actually measured solar radiation as a label of the model, and performing training optimization on the RF model by using data of the training set; and on the basis of a GraphCast coupling VMD decomposition algorithm hybrid RF model, solar radiation is predicted according to meteorological data of a test set. According to the invention, solar radiation forecasting with higher precision can be achieved.

The invention discloses a method for detecting heart rate, respiration and oxyhemoglobin saturation based on micro-vibration images, and belongs to the technical field of physiological detection and image processing. A camera collects facial micro-vibration videos of a testee in real time, uses an ROI algorithm to lock a facial key area, extracts RGB micro-vibration change signals to replace red light and near-red light signals, and determines whether the micro-vibration videos are abnormal or not; performing wavelet transform five-layer decomposition, three rounds of peak valley screening and abnormal value elimination on the RGB microvibration time sequence signal, performing fine tuning training by using a yov11 model, and calculating to obtain heart rate, respiration and blood oxygen physiological indexes. According to the invention, non-contact detection is realized through video acquisition and signal processing of micro-vibration of a face area, and the comfort level and the use convenience of a testee are improved; interference caused by motion artifacts, baseline drift and expression changes is eliminated through multi-scale wavelet denoising; the three-wheel peak-valley screening algorithm deeply excavates the features of the face micro-vibration signals, and the recognition accuracy of the signal peak-valley pairs is improved; and an abnormity elimination and self-feedback parameter adjustment mechanism improves the calculation precision.

The invention provides an online car-hailing ride-sharing dynamic matching method and system, relates to the technical field of dynamic matching, and aims to reasonably divide a dispersed ride-sharing request set through a ride-sharing request clustering algorithm according to an initial position and a destination of a request so as to effectively reduce the subsequent processing scale and improve the matching efficiency. By means of a matching vehicle screening algorithm, based on a clustering result and considering constraint conditions such as the number of vehicle seats, the driving direction and the passenger waiting time, a vehicle set capable of providing service near each request set is screened out, and the invalid matching condition is avoided; on the basis of master-slave pricing matching, driver and passenger matching is constructed into a multi-round iterative dynamic process. In each round of iteration, online car-hailing adjusts charging standards according to matching conditions, passengers reselect cars according to new prices and service information, continuously adjust strategies and gradually approach to the optimal matching state, finally, passenger utility maximization is achieved, meanwhile, it is ensured that income appeals of drivers are reasonably met, and optimal balance of benefits of the drivers and the passengers is achieved.

The invention discloses a multi-station multi-target angle correlation positioning method, which comprises the following steps of: classifying angle measurement sets from different aircrafts to obtain a plurality of groups of angle measurement value sets of different aircrafts possibly from the same target, and providing a reliable measurement set for a coupling weight least square angle positioning algorithm; secondly, establishing a group of overdetermined nonlinear angle measurement equations including position estimation errors and angle measurement errors of the aircraft, and proposing a coupling error weight least square angle positioning algorithm; and finally, each target angle measurement set obtained through classification by using an angle correlation screening algorithm is used as input of a coupling error weight least square angle positioning algorithm, a multi-target position estimation result is obtained, and a multi-target positioning process is completed. According to the method, the calculation complexity of the angle measurement set association process in the multi-target angle positioning process can be greatly reduced, the measurement association operation time is effectively shortened, and the multi-target angle positioning real-time performance and the positioning precision are improved.

The invention belongs to the technical field of surface enhanced Raman spectroscopy, and relates to an AFB1 detection method based on SERS mapping and data fusion. The method comprises the following steps: firstly, acquiring an SERS (Surface Enhanced Raman Scattering) spectrum and a mapping image of an AFB1 standard sample solution; carrying out pretreatment on the SERS spectrum; then, carrying out feature variable extraction on the preprocessed SERS spectral data by using a variable screening algorithm, and obtaining RGB values from corresponding mapping images based on coordinates of screened feature variables; and selecting a variable combination with the lowest interactive verification root mean square error, fusing the variable combination with the RGB value of the mapping image at the feature variable, and constructing a prediction model, so that the rapid detection of the AFB1 concentration of the unknown sample can be realized. Compared with a traditional detection technology, the stability and precision of a detection result are remarkably improved through a spectrum-image multi-dimensional data fusion strategy, and the method has good application prospects in the technical fields of food safety, environment monitoring and the like.

The invention belongs to the technical field of data processing, and particularly discloses an intelligent design method and device for long and short legs of an iron tower, electronic equipment, a medium and a program product. Generating a feasible solution space containing a tower leg length candidate set and a center pile adjustment height candidate set; the height of each candidate center pile is adjusted, a height difference matrix is constructed, the optimal tower leg length is determined through a conditional screening algorithm, and a leg matching scheme is formed; an error absolute value, an error variance and a center pile adjustment height are extracted as evaluation indexes, and after a comprehensive score is calculated through normalization and weighting, a scheme with the minimum score is selected. According to the method, automation and precision of iron tower long and short leg design are achieved, and design efficiency and scheme reliability are improved.

The invention relates to the technical field of inertial sensor state data analysis, and discloses a frequency domain processing-based inertial sensor data analysis method, which realizes frequency domain analysis on actual error composition of an inertial sensor, thereby improving the accuracy of effective system performance analysis of the inertial sensor. The method specifically comprises the following steps: S1, acquiring an original output signal by adopting an MEMS inertial measurement unit of a sensor; s2, decomposing a time domain signal in the original output signal into a solid mode equation and a residual term through a mode decomposition algorithm; extracting a solid-state model equation by using a screening algorithm through empirical mode decomposition, and updating a residual error; selecting a center frequency by using signal characteristics and priori knowledge through variational mode decomposition, extracting a solid-state mode equation at the center frequency, and modeling as a function about amplitude phase frequency; s3, performing Hilbert-Huang transform on the components of the solid mode equation to generate a Hilbert spectrum in time-frequency domain joint distribution; and S4, constructing a sensor performance evaluation model based on the Hilbert spectrum.

The invention discloses a farmland navigation path extraction method based on an affiliation filtering strategy, and the method comprises the following steps: building a corn crop row data set in various growth environments, marking a corn plant and the root of the corn plant through a marking tool, and carrying out the data enhancement of the data set; the plants and the roots are detected through an RS-YOLOv8 lightweight model; based on an affiliation filtering algorithm of a corn plant and a root detection frame, screening the root detection frame overlapped with the plant detection frame, and removing an isolated false detection frame; extracting a region of interest (ROI) through a region screening algorithm according to the filtered detection frame; and extracting a center point at the bottom of the root detection frame as a feature point, dividing crop ridges by adopting a clustering algorithm, fitting crop row lines in the ROI through a least square method, and extracting a navigation path. According to the technical scheme, the high-quality feature points which are not easily interfered by the environment are extracted through the root detection frame, and then the navigation path is fitted, so that the reliability of the route information is ensured.

The invention provides a calibration method for real-time dynamic blood glucose monitoring. The calibration method comprises the steps that 1, weak current signals generated by a glucose sensitive electrode are obtained regularly; 2, filtering the detected weak current signal by adopting a sliding window double-layer screening algorithm; 3, monitoring the body temperature of the human body, compensating the temperature based on the filtered weak current value, and estimating the glucose concentration after temperature compensation; 4, monitoring the PH value of the human blood, compensating the PH value according to the filtered weak current value, and estimating the glucose concentration after PH value compensation; and 5, combining the glucose concentrations after temperature and pH value compensation, and carrying out filtering processing on the combined glucose concentrations by adopting a parameter compensation filtering algorithm to obtain a final glucose concentration estimation result.

The invention discloses an interactive lattice radiotherapy planning method and system based on a functional image, and relates to the technical field of radiotherapy, and the method comprises the steps: obtaining a radiotherapy positioning CT image of a patient and a functional metabolism image after the space registration with the radiotherapy positioning CT image; based on the radiotherapy positioning CT image and a preset safety boundary parameter, determining a candidate area of lattice layout in the tumor target area; for each candidate position in the candidate region, adaptively calculating lattice parameters for the position based on the corresponding SUV value of the position in the functional metabolism image; determining a final lattice center set from the candidate positions through an iterative screening algorithm on the basis of the calculated lattice parameters of the candidate positions and preset geometric constraints and biological constraints, and generating a lattice target region set on the basis of the final lattice centers and the lattice diameters corresponding to the final lattice centers; and outputting the lattice target region set for dose calculation of a radiotherapy planning system. Individualized dosage improvement really based on tumor internal heterogeneity is achieved.

The invention relates to a batch material transfer method, system and equipment based on a workbin robot and a medium, and the method comprises the steps: obtaining the original code scanning data of a plurality of workbin robots for a target workbin, updating the original code scanning data based on a verification screening algorithm, and obtaining the target code scanning data; acquiring transfer sensing data of a target material box in a transfer time period based on the material box robot, determining a first transfer parameter of the target material box, and determining a second transfer parameter of the target material box according to the target code scanning data; and according to the first transfer parameter and the second transfer parameter, based on an accuracy prediction neural network, an accuracy parameter of the material transfer information in the target material box is determined, and the accuracy parameter is used for representing the accuracy degree of the material transfer process in the production plan.

The application provides a dynamic motion trajectory extraction method based on Kalman filtering and repeated data screening, which comprises the following steps: installing IMU on a plurality of key motion joints of a dynamic living body, generating an initial trajectory data set through data collected by the IMU; setting a state space model based on the measurement data in the initial trajectory data set and Euler angles, and performing noise compensation on the initial trajectory data set through a Kalman filter; optimizing the initial trajectory data set after Kalman filtering through a repeated data screening (RDS) algorithm to obtain an optimized trajectory data set; and calculating the joint axis direction through the optimized trajectory data set, and constructing the dynamic motion trajectory of the dynamic living body according to the calculated joint axis direction and the angular velocity data after Kalman filtering. The method combines the Kalman filtering algorithm and the repeated data screening algorithm to extract the trajectory, and significantly improves the robustness and accuracy of the dynamic motion trajectory extraction of the living body.

The invention provides a Web-based single-cell RNA sequencing data intelligent analysis system and method. The method comprises the following steps: receiving a cell group through a Web interface; performing differential gene analysis on the single-cell RNA sequencing data contained in the cell group to obtain an original differential gene list, and filtering the original differential gene list by adopting a multi-threshold screening algorithm to obtain a target differential gene list; performing species automatic identification processing on the target differential gene list, calling a target local gene set database based on a result of the species automatic identification processing, and performing parallel enrichment analysis independent of network connection on the target differential gene list according to the target local gene set database to obtain a gene enrichment analysis result; and generating an interactive chart by adopting an intelligent label anti-overlapping algorithm so as to visualize the interactive chart. According to the method, the problems of high operation threshold, low batch analysis efficiency, unstable result and poor interactivity in the prior art are solved.

The invention discloses a human body infrared spectrum intelligent screening feedback device and immune enhancement and anti-inflammatory application thereof, and relates to the technical field of human body feedback devices. The human body infrared spectrum intelligent screening and feedback device comprises an infrared detection unit, at least one biomarker sensor, a central processing unit and an infrared feedback unit. According to the human body infrared spectrum intelligent screening and feedback device, the infrared detection unit and the biomarker sensor are integrated, multi-source information is subjected to fusion analysis through the central processing unit, comprehensive judgment of the human body immune or inflammation state is achieved, the device dynamically calls a matched infrared spectrum screening algorithm according to the judgment result, and the human body infrared spectrum intelligent screening and feedback device is suitable for being applied to human body health. The infrared feedback unit is driven to carry out targeted irradiation, real-time sensing analysis and dynamic screening are achieved, the information is accurately fed back to uncomfortable body parts, and the problems that in the prior art, diagnosis and treatment are disjointed, and intervention parameters are solidified are effectively solved.

The present invention discloses a method for using a medical smart bed cabinet, comprising: obtaining relevant information data of people using the bed cabinet in a fixed area, and classifying and storing the relevant data in a database; collecting feature data required by a sorting and classification model, and sorting and classifying the feature data using the sorting and classification model; performing feature matching based on the sorted and classified data with the classified data stored in the database. If the matching degree is higher than a preset value, the match is successful, and accompanying personnel are assigned. When an emergency event occurs in the fixed area, a screening algorithm is used to quickly screen out vacant bed cabinets and corresponding accompanying personnel, thereby realizing the intelligent application of medical bed cabinets. The present invention combines smart bed cabinets with accompanying services to achieve the rational allocation and full utilization of hospital bed resources, improve the quality of accompanying services, and reduce patient expenses.

The invention provides a flow behavior recognition method and device, equipment and a storage medium, and relates to the field of big data, and the method comprises the steps: mapping user data used for representing time-space behavior characteristics of a user into a bitmap sequence according to a geographic space dimension and time resolution precision; generating migration behavior data based on the user data, wherein the migration behavior data is used for representing migration behaviors of the user from a starting point to an end point in a time period; based on task information, the bitmap sequence is processed through a bitmap sequence addition algorithm and a residence duration interval screening algorithm, a processed bitmap sequence is obtained, and the task information comprises a source area, a target area, a time period and residence duration; and based on the processed bitmap sequence and the migration behavior data, identifying a user conforming to the task information from a plurality of users corresponding to the user data. And the flexibility of user area flow behavior identification is improved.

The invention discloses a transformer fault prediction method based on a DeepSeek large model. The method comprises the following steps: acquiring transformer fault related original data; the method comprises the following steps of: dividing transformer fault related original data into a training set and a verification set by adopting a layered random sampling method, layering an original training data set for the training set according to categories of fault and normal samples, and dividing the original training data set into a secondary sampling subset and a single sampling subset; an improved SMOTE algorithm is adopted for the secondary sampling subset to increase fault sample data; screening normal sample data for the single sampling subset by adopting a clustering-based screening algorithm; evaluating the data set balance degree of the secondary sampling subset and the single sampling subset; combining the subsets after secondary sampling and the subsets after single sampling to form a final training data set; identifying deep features in the data by using a DeepSeek model, and extracting feature information with high correlation on transformer fault prediction; and inputting and integrating the deep features into a plurality of learning models to predict and obtain a final prediction result.

The invention discloses a military crane chassis fault reproduction and practical training system based on working condition simulation, and relates to the technical field of military equipment practical training, and the system comprises an entity chassis building unit which selects and uses entity parts of subsystems of a military automobile crane to be arranged and installed according to a real automobile, retains a key relation, presets an interface, defines a safety threshold value and the like; the dual-mode fault generation unit constructs an initial library, and an intelligent agent is embedded to generate a fault; the multi-dimensional data acquisition unit responds to the trigger signal to acquire multiple types of data; the fault diagnosis practical operation unit is used for trainees to operate; the training evaluation and fault iteration unit generates a report and feeds back optimization suggestions; according to the invention, a dual-mode fault generation unit is constructed, an initial fault library and an intelligent agent exploration capability are integrated, a screening algorithm is matched, and multiple fault scenes are provided; a complete practical training system is built, multi-dimensional data acquisition is achieved, practical operation and evaluation iteration are standardized, the practical training quality is improved through closed-loop design, and the military crane chassis fault practical training requirements are met.