86 results about "Reduced model" patented technology

The application discloses a layout hotspot detection method based on geometric feature analysis, and the implementation steps are as follows: an under-sampling and over-sampling method is used to generate a layout sample set; the number of corner points, short-circuit sensitivity and open-circuit sensitivity of the layout sample are extracted to form a feature vector of the layout sample; an integrated learning model is trained by using a feature vector training set; and the feature vector test set is input into the trained integrated learning model to output a detection result. The application overcomes the problems in the prior art that the imbalance of layout category samples leads to reduced model detection precision, and that the recognition rate of a layout mode that has never been seen before is low and the false positive rate is high. Geometric features are extracted for photolithography hotspots in the layout, so that the application can maintain high detection precision and low false positive rate when detecting a layout that has never been seen before and a complex layout mode.

The application discloses a kind of automobile spoiler assembly performance simulation analysis method, system and storage medium, its method includes: importing spoiler assembly structure data in pre-processing software, calling abstract midplane unit to carry out abstract midplane processing to spoiler assembly structure data, obtain spoiler simplified model;In pre-processing software, the parameter counterweight assembly simulation processing of spoiler simplified model is carried out based on finite element grid, and the finite element simulation model of spoiler is obtained;Based on the connection working condition of spoiler finite element simulation model and the load force acting on the spoiler finite element simulation model, the modal performance, stiffness performance, strength performance and abnormal force resistance performance of spoiler assembly are analyzed in pre-processing software and post-processing software.The application replaces test stand test by simulation analysis, verifies automobile spoiler assembly performance, solves the problem that the performance of automobile spoiler assembly is verified by test in existing design stage, and leads to long research and development cycle, high research and development cost.

This application discloses a federated large language model training method, apparatus, and device. The method includes: constructing an initial mask matrix for each original model parameter in the initial large language model; splitting the initial mask matrix to obtain initial mask blocks based on the current pruning round; determining the current mask block corresponding to the initial mask block based on the mask reconstruction dataset, a predetermined sparsity rate, the original model parameter matrix corresponding to the initial mask block, and the inverse Hessian matrix of the original model parameter matrix, and determining the current model parameters corresponding to each non-zero, non-target element in the current mask block to obtain the current model parameter matrix; determining the pruned large language model based on the current model parameter matrix corresponding to each pruning round; and optimizing and training the pruned large language model based on a first optimized dataset to obtain the target large language model. This application can perform reasonable and accurate model pruning, simplify the model structure, and improve the model training speed.

This invention relates to the field of data analysis technology, and in particular to a query analysis method, apparatus, device, and storage medium based on a large language model. The method provides comprehensive and effective data support for enterprise decision-making. It achieves semantic fusion of multi-source heterogeneous data through a unified knowledge graph, breaking the limitations of traditional data silos and providing a comprehensive and accurate business knowledge foundation for the large language model, reducing model illusions from the source. It achieves differentiated analysis processing through task type classification, automatically scheduling different capability modules of the large language model according to user needs, balancing the efficiency of basic queries with the accuracy of in-depth analysis. It obtains the full business context through entity links, limiting the reasoning process of the large language model to the enterprise's real business scenarios, significantly improving the business relevance of the analysis results. It achieves in-depth diagnostic analysis through causal attribution, fully leveraging the logical reasoning capabilities of the large language model and overcoming the limitations of descriptive analysis.

The application discloses a kind of water turbine rotor dry mode model updating method and related device, comprising: forming numerical modal family;Equivalent boundary parameter and key connecting surface equivalent stiffness are set;Under the predetermined rotating speed set, calculate centrifugal prestress;On the basis of the centrifugal prestress, consider gyroscopic effect, sequentially solve each rotating speed condition, obtain the natural frequency and vibration mode branch varying with rotating speed;Identify test frequency, damping and vibration mode;With adjacent sector vibration mode phase difference consistency and vibration mode energy distribution as criterion, carry out node radial number determination to test vibration mode, and with the numerical modal family is matched one by one;With the deviation of test frequency and mode assurance criterion as comprehensive target, the equivalent boundary parameter and key connecting surface equivalent stiffness are carried out inversion updating, until meeting preset convergence criterion, complete the updating of water turbine rotor dry mode model, the method and related device can reduce model scale and improve solving convergence.

This invention provides an integrated method for pruning and controlling large-scale models of unmanned aerial vehicles (UAVs), belonging to the field of model lightweighting technology. The proposed method significantly reduces model size and computational load, greatly improves inference speed and control response performance, and enables the deployment of complex neural network controllers on resource-constrained airborne platforms. Simultaneously, the lightweight model maintains or approaches the control accuracy and stability of the original uncompressed model. This method has wide applicability and strong scalability, covering various scenarios from basic flight control to complex mission control, providing a feasible, efficient, and cost-effective solution for applying large models to airborne controllers of UAVs, and possesses high engineering practical value.

To address the problems of semantic structure destruction, geometric feature distortion, and difficulty in controlling multi-level errors in the LOD simplification process of existing 3D models, this application discloses a static LOD processing method, device, and storage medium for urban digital twin scenes, belonging to the field of computer graphics and 3D modeling technology. The method includes: semantic recognition and region division of the urban building 3D mesh model to construct a semantically labeled mesh; establishing a simplification control model based on the semantically labeled mesh and determining simplification constraints; performing mesh simplification processing under constraints; preserving geometric features of the simplification results; evaluating errors in each LOD level model and performing adaptive correction; and replacing semantic components that cannot be preserved to generate multi-level LOD model data. This application reduces model complexity while achieving the coordinated preservation of semantic consistency and geometric accuracy, making it suitable for efficient rendering and visualization applications of urban-level 3D scenes.

The application discloses a resource management API constraint extraction method and system based on a large language model, and belongs to the technical field of software automation analysis. In order to solve the illusion problem caused by directly using a large model to analyze API constraints, the application extracts function declarations and resource management sentences from API documents; constructs a multi-dimensional prompt template to obtain API allocation judgment results and reasoning processes; extracts core predicates of reasoning text and performs consistency verification to correct illusions; structures function declarations and extracts resource object types in combination with semantics; searches for paired release functions by taking the object types as constraints, and outputs API pairs; and finally performs resource misuse detection based on the API pairs through static analysis. The application can significantly reduce model output errors, improve release function searching accuracy, has strong anti-illusion ability and high automation degree.

This application provides a knowledge distillation-based method for optimizing unmanned delivery vehicle routes, relating to the fields of combinatorial optimization and intelligent logistics scheduling. The method includes: constructing a problem instance with capacity constraints and safety preferences, and a comprehensive objective function; inputting node features into a machine learning model, constructing path solutions through autoregression and saving intermediate solutions, iteratively outputting candidate solutions; inputting candidate solutions into an adaptive large neighborhood search module, and outputting high-quality solutions and generating a new dataset after destruction repair, acceptance criterion judgment, and weight updates; extracting the action sequence and probability distribution of the teacher model as supervision information; constructing a lightweight student model and using a shared matrix to predict node selection; constructing a distillation loss and jointly training to obtain the student model; and obtaining the final model through secondary training on the new dataset. This application reduces model complexity and inference overhead, improving path solution quality and computational efficiency in small-sample scenarios.

This application relates to the fields of computer vision and road defect detection technology, and proposes a lightweight road defect detection method for embedded deployment. Using YOLOv10n as a baseline, this method introduces a C2f_GBK structure that integrates the low-cost feature generation mechanism of G-GhostNet into the backbone feature extraction network to reduce the number of model parameters and computational cost. A pyramid sparse Transformer module (PST) is introduced into the feature fusion network to enhance the expressive power of multi-scale road defect features and compensate for information loss caused by lightweighting. A DySample dynamic upsampling strategy is introduced at the upsampling node to improve feature reconstruction quality and reduce redundant computation. Through these improvements, the method effectively solves the problems of existing road defect detection models having large parameter counts, high computational overhead, difficulty in deployment on resource-constrained devices such as vehicle terminals, and the tendency for detection accuracy to decrease during lightweighting. It significantly reduces model complexity while maintaining high detection accuracy, making it suitable for real-time road defect detection and embedded deployment.

This disclosure provides a method, apparatus, and device for regulating model demanders based on federated learning. The method includes: receiving data uploaded by model demanders in a federated learning architecture and determining a range of the number of model demanders; generating a data regulation task based on the range of the number of model demanders to create multiple model demanders, and generating corresponding configuration information for each model demander; setting corresponding behavior strategies for each model demander based on their task requests; establishing communication channels between the model demanders and other participants when executing the data regulation task, and generating a demander object corresponding to each model demander based on the configuration information and behavior strategies, and executing the data regulation task based on the demander object. This disclosure simplifies the model demander creation process, reduces labor costs, and improves the efficiency of model demander creation in simulation platforms.

A model construction method of a photovoltaic power generation unit, characterized in that the method comprises the following steps: step 1, model simulation is performed on a grid-connected inverter of the photovoltaic power generation unit, the simulation model comprises a basic model and a simplified model, wherein the simplified model is used to simplify a half-bridge structure of the grid-connected inverter into a three-phase power supply in the simulation process; step 2, the simulation model of the grid-connected inverter is input into power system simulation software, and a running disturbance is added to the simulation model to obtain an adjustment effect of the simulation model on the running disturbance; and step 3, based on the adjustment effect of the simulation model and the network scale to be analyzed, the basic model or the simplified model is selected to be used. The method is simple, the improvement of the simulation model is accurate and effective, and fully meets the demand of model calculation efficiency under large-scale simulation.

This invention provides a lightweight, edge-side LSTM-based integrated flood forecasting method based on ONNX. It acquires flow data and processes missing and outlier values, performs sliding window sampling and standardization, and constructs and trains an LSTM flood forecasting model based on the processed sample data. The trained LSTM flood forecasting model is then converted into an ONNX model and optimized through computational graph optimization, constant folding, quantization optimization, and hardware adaptation optimization to obtain a lightweight flood forecasting model. This lightweight flood forecasting model is then deployed on edge devices to perform local rolling inference on real-time collected flow monitoring data and output flood forecast results. This invention enables closed-loop operation of data acquisition, local inference, and forecast output on the edge, reducing model size, memory usage, and power consumption, and improving the real-time performance and reliability of flood forecasting in remote areas and scenarios with weak networks.

This invention discloses a static vector model generation method based on distillation, comprising: obtaining all independent lexical units in the basic model's word segmenter; inputting each independent lexical unit individually into the basic model for forward propagation to extract representations and constructing an initial embedding matrix; applying principal component analysis to the initial embedding matrix for dimensionality reduction; and using a smooth inverse frequency weighting algorithm based on Zipf's law, employing vocabulary ranking as a proxy variable to calculate lexical probabilities, and performing weighted optimization on the dimensionality-reduced embedding matrix to generate the target static vector model. This invention eliminates the dependence of traditional model distillation on massive amounts of data and GPUs, simplifies complex Transformer nonlinear inference to an extremely fast table lookup operation through computational collapse, and retains most of the deep semantic understanding capabilities of the original model while reducing model size and increasing CPU inference speed.

This invention discloses a five-finger dexterity hand grasping detection method based on soft-mask region representation and multi-task learning. This method decomposes high-dimensional continuous grasping parameters into three sub-tasks: grasping quality prediction, grasping width regression, and grasping gesture classification. It constructs a soft-mask multi-color grasping region representation to generate pixel-level grasping quality, width, and gesture labels. A grasping-oriented channel-space-geometric attention mechanism is designed to construct a lightweight multi-task generative grasping detection network. Taking RGB-D images as input, it outputs grasping quality maps, width maps, and gesture maps in parallel. An adaptive weighted loss function based on effective region constraints is used for training to suppress background interference and dynamically balance multi-task learning. During inference, the grasping center is located by searching for peaks in the quality map, and the corresponding width and gesture are read to achieve single-target or multi-target grasping detection. This invention improves the accuracy, real-time performance, and robustness of grasping detection in multi-object scenes while reducing model complexity.

This invention relates to the field of IEGS probabilistic energy flow technology, and particularly to an IEGS probabilistic energy flow monitoring method based on a sparse arbitrary chaotic polynomial model, comprising: S1, establishing a deterministic energy flow model for an integrated energy system (IEGS); S2, modeling random input variables and configuring corresponding non-Gaussian probability density functions; S3, generating P multidimensional orthogonal aPC basis functions; S4, generating M0 initial collocation points; S5, transforming the problem of solving the sparse aPC coefficient vector A into an l1-l2 norm minimization optimization model; solving for the sparse aPC coefficient vector A under the current collocation point set; and constructing a sparse aPC surrogate model; S6, verification; outputting the final sparse aPC surrogate model; S7, analytically calculating the statistical moments of key output quantities and reconstructing the probability density functions of key output quantities; S8, evaluating the safety margin of IEGS operation. This method can effectively reduce model dimensionality and improve computational efficiency by combining sparse recovery techniques while retaining the good adaptability of the aPC method to non-Gaussian inputs.

The application provides a high-precision rainfall prediction method based on SFDL sequence feature decomposition and a dendritic learning model, through SFDL sequence feature decomposition technology, original rainfall time series data is accurately disassembled into a trend item, a seasonal item and a residual item, long-term change law, periodic fluctuation and irregular disturbance characteristics are captured respectively, hierarchical modeling of complex sequences is realized, for the residual item after decomposition, an improved dendritic learning model DNM is introduced, the bionic structure and multiplication unit characteristics of the model can efficiently fit nonlinear relationships, meanwhile, model parameters are optimized by combining a BP algorithm, the prediction ability for chaotic data is improved, the SFDL decomposition accuracy is optimized by local weighted regression LOESS, the original feature correlation of data is reserved, and the high-dimensional mapping effect of time series data is improved by adopting phase space reconstruction PSR preprocessing; through modular design, a decomposition-prediction-fusion process is integrated, the model structure is simplified, the calculation complexity is reduced, and the real-time prediction demand and the engineering deployment feasibility are considered.

This invention proposes a method for constructing a transmission line insulator defect detection model based on edge technology. The method includes the following steps: S1: Edge device adaptation analysis and requirement definition; S2: Insulator defect dataset construction and optimization; S3: Lightweight detection model construction; S4: Model training and parameter optimization; S5: Edge device deployment and compatibility adaptation; S6: Dynamic model update and maintenance. By simplifying the model structure, employing depthwise separable convolution, model pruning, and quantization optimization, the constructed detection model's size does not exceed 100MB. By introducing SE attention mechanism and knowledge distillation technology, the model's detection accuracy is no less than 95%; the model inference speed is no less than 25 frames / second, enabling real-time detection during UAV inspections. This constructs a lightweight, high-precision insulator defect detection model adapted to UAV edge devices, achieving real-time, autonomous detection of insulator defects and providing technical support for transmission line inspections.

The simulation method comprises the following steps: cutting simulation is performed by establishing a simplified finite element model of a tool-workpiece to obtain a cutting force; an original model of the piston is established based on the geometric characteristics of the piston; the oil groove part, the weight reduction cavity part and the non-bearing fillet part of the original model are removed to determine a simplified model of the piston; the turning area is meshed based on a structured grid, and the non-turning area is meshed based on the simplified grid; a piston grid model is established according to the meshing results of the turning area and the meshing results of the non-turning area; the cutting force is applied to the piston grid model, and the turning process is simulated by using the birth-death element method to determine the residual stress distribution of the piston along the turning path and the deformation amount of the piston. The piston turning process can be simulated efficiently.

The application belongs to the technical field of arc simulation of low-voltage circuit breakers in power systems, and relates to a three-dimensional simulation method and system for low-voltage air switch arc containing arc extinguishing sharp peaks. The system obtains a three-dimensional simplified model of a low-voltage circuit breaker through a model acquisition module; performs local encryption processing on an arc movement path, a grid and a sheath layer area through a grid generation module; obtains a three-dimensional distribution of each parameter of the arc changing with time through an arc simulation module; corrects a sheath layer pressure drop model at an arc extinguishing stage according to experimental data through a sheath layer correction module; and compares and verifies simulation results obtained according to the corrected sheath layer pressure drop model with experimental data through a simulation result verification module, so as to ensure that the calculation results are close to actual arc characteristics. The application can effectively overcome the deficiency that the traditional simulation method is only applicable to the stable arc burning stage, and significantly improves the accuracy of arc voltage calculation at the arc extinguishing stage.

This invention discloses an active distribution network integrated optimization scheduling method and system based on SVM-L2O, belonging to the field of distribution network optimization scheduling technology. The method includes: solving a multi-timescale integrated scheduling model based on acquired historical wind power output and load data to obtain unit start-up and shutdown strategies and continuous output decision variables, and constructing a training dataset; training a support vector machine for the start-up and shutdown strategies of each unit in each time period to obtain a surrogate model predicting the unit start-up and shutdown strategies; removing start-up and shutdown related constraints from the model based on the generated start-up and shutdown strategies to obtain a simplified model; constructing an L2O neural network, using Lagrange multipliers to incorporate the constraints of the simplified model as penalty terms into the loss function, training the L2O neural network to obtain a surrogate model predicting continuous output decision variables; acquiring wind power output and load data in real time, generating optimal scheduling decisions through the two surrogate models, and realizing integrated and efficient solution of multi-timescale distribution network scheduling.

The application discloses a kind of based on multilayer perceptron's graph classification model construction method and graph classification method, the former includes: the classification task of required application scene is acquired to several graph data and constitutes training set;At least part of graph data of training set has class label;Using training set, selected graph neural network is trained, and the graph neural network that training is completed is used as teacher network and saves the classification result corresponding to graph neural network;Selected multilayer perceptron is used as student model, based on training set, teacher network and corresponding classification result, student model is trained using the method of knowledge distillation, and the student model that training is completed is used as graph classification model for the remaining graph data classification of required application scene.The graph classification model of the application combines the respective advantages of multilayer perceptron and graph neural network, without the dependency of graph in reasoning process, ensure higher accuracy while can greatly reduce model calculation complexity, improve reasoning speed, can be used for time limited engineering deployment.

A solid-liquid bundled rocket structure simplified modeling method, comprising: establishing a rocket structure three-dimensional finite element model, the rocket structure three-dimensional finite element model comprising a main structure model and a secondary structure model of a rocket body; for the main structure model of the rocket body, according to the type of the main structure model of the rocket body, simplifying the type of the main structure model of the rocket body into a low-dimensional model; for the secondary structure model of the rocket body, according to the type of the secondary structure model of the rocket body, simplifying the type of the secondary structure model of the rocket body into a particle element model; simplifying each main structure model and each secondary structure model of the rocket body, and connecting the obtained low-dimensional model and particle element model according to the corresponding relationship of the rocket structure three-dimensional finite element model to obtain a rocket structure simplified model.

The present application relates to the technical field of active suspension system control, and particularly relates to a gain scheduling pulse control method for vehicle suspension under network delay. The method first establishes a dynamic equation and a state space model of vehicle active suspension containing network delay, and constructs a linear time-invariant state space model based on dynamic integral quadratic constraint; a corresponding gain scheduling pulse controller is designed, an equivalent form of the suspension state space model is established through an uncertain operator, and a pulse feedback interconnection system is derived; a stability condition of the control system is given under the dynamic integral quadratic constraint framework, a controller design method is proposed, a gain matrix is solved, and software simulation and hardware-in-the-loop experiment verification are performed. The method effectively simplifies the model processing flow and reduces the calculation complexity, can realize the stability control of the vehicle active suspension under network delay, and at the same time, takes into account the safety constraints of suspension displacement and actuator control force, and improves the ride comfort and control stability of the vehicle.

The application relates to a kind of transient load under the function gradient porous structure of multiphase filling dynamics topology optimization method and system, belong to the topology optimization design field of function gradient porous structure.The technical problems to be solved by the application are: how to fully exploit design space, construct the efficient function gradient porous structure of multiphase filling dynamics topology optimization method, realize the optimal anti-vibration design of porous filling structure with high-quality transient dynamic performance.Technical points: based on characteristic orthogonal decomposition method, the reduced transient dynamic finite element equation of function gradient porous multiphase filling structure is constructed;the material interpolation model and volume constraint model of function gradient porous multiphase filling structure are constructed, and the transient dynamic multiphase filling topology optimization reduced model with the minimum dynamic compliance as the optimization objective is proposed;the reduced sensitivity time-domain accompanying equation is constructed, and the sensitivity of dynamic compliance and volume constraint is efficiently calculated;based on ADMM-MMA hybrid optimization algorithm, the optimal topology design of function gradient porous multiphase filling structure is efficiently solved and obtained.The application can maximize the optimal anti-vibration performance of function gradient porous filling structure, and is especially suitable for the structure lightweight anti-vibration design of large engineering equipment.

The application discloses a kind of micro-grid system's network type inverter adaptive control method and system, belong to micro-grid operation control technical field, method includes: obtaining the transient security analytical constraint condition when micro-grid system endures big disturbance;Build the upper agent optimization model with the objective function of minimizing the virtual inertia and virtual damping of network type inverter, and after solving current optimal control parameter is input to lower full-order system verification model and is carried out dynamic simulation evaluation, whether micro-grid system meets preset safety condition is judged: if meet, then output final optimal control parameter, otherwise, then according to simulation data updates transient security analytical constraint condition, and is solved again to upper agent optimization model, until preset safety condition is met.The application can effectively isolate the prediction error caused by simplified model, realize the online adaptive correction and optimal configuration of control parameter, improve the response stability and rapidity of micro-grid system.

This invention relates to the field of wind farm operation status identification technology, and more particularly to a method for identifying the operation status of offshore direct-drive wind turbines based on an improved one-dimensional convolutional neural network. The method compresses state labels through feeder iso-modeling, reducing model complexity. An improved one-dimensional convolutional neural network model is constructed, integrating dilated convolution and average pooling to enhance multi-scale feature extraction capabilities, thereby improving the accuracy on both the training and test sets. Global average pooling layers replace fully connected layers, reducing the number of parameters and avoiding overfitting. This effectively solves the gradient problem and insufficient multi-scale feature capture issues inherent in traditional one-dimensional convolutional neural network models when handling complex wind farm operation status identification. The model achieves high accuracy on both the training and test sets, and maintains high accuracy even under different Gaussian noise levels, demonstrating good robustness and generalization ability.

This invention discloses a hyperspectral image classification method based on multi-scale self-supervised contrastive learning. The method comprises a network structured as follows: a contrastive learning feature extraction module suppresses unimportant features in image patches and filters out useful features; when the number of labels is small, a self-supervised mode is used to learn features; a deep and shallow multi-scale feature learning module constructs hierarchical scales to extract multi-scale information from hyperspectral images; a spatial-spectral reconstruction module constructs spectral and spatial branches, using separation, filtering, and reconstruction to combine deep and shallow features, reducing model parameters and complexity, and enhancing feature representation capabilities; the network output is then average-pooled and fed into a classifier to obtain the result. This invention acquires rich spatial-spectral features through the multi-scale feature learning module, spatial-spectral reconstruction module, and contrastive learning feature extraction module, broadening the application scenarios for images with few labels, accelerating network convergence, and thus improving classification accuracy.

This invention discloses a comprehensive assessment method for the risk index of crack flow hazards on rhythmic beaches, comprising: simulating the evolution of beach topography using a topographic linear instability model based on set initial equilibrium beach geomorphological characteristic parameters and incident wave parameters to generate rhythmic beach topography; performing crack flow simulation calculations using a simplified model based on nonlinear gentle slope equations based on the rhythmic beach topography and incident wave parameters to obtain the horizontal velocity field at the still water level; calculating the time-averaged Lagrange crack flow velocity distribution based on the horizontal velocity field to determine the crack flow risk area and crack flow velocity magnitude; obtaining the float drift trajectory based on numerical buoy simulation deployed in the nearshore area and calculating the float retention rate to characterize the crack flow cycle type; and quantitatively calculating the crack flow hazard risk index using a comprehensive assessment formula based on the crack flow velocity magnitude, the range of the crack flow risk area, and the float retention rate.

The application provides a target area atmospheric OH radical parameterization characterization method, system, equipment and storage medium, and the method comprises the following steps: S1, obtaining initial data of a target area based on an air quality model according to model input data containing a range of the target area, wherein the initial data comprises OH radicals and characteristic factors; S2, establishing an initial model of the relationship between the OH radicals and the characteristic factors based on machine learning; S3, performing feature alignment on the characteristic factors and ground observation data of the target area by using a transfer learning method, and obtaining an optimization model based on the initial model; S4, screening the characteristic factors in the optimization model to obtain key characteristic factors, and establishing an OH radical simplified model based on the key characteristic factors. The characterization method realizes rapid, accurate and simplified characterization of atmospheric OH radicals, and the required cost and difficulty are significantly lower than direct field observation, and the accuracy is higher than numerical simulation.