Patents
Literature
47results about How to "Avoid precision loss" patented technology
Efficacy Topic
Property
Owner
Technical Advancement
Application Domain
Technology Topic
Technology Field Word
Patent Country/Region
Patent Type
Patent Status
Application Year
Inventor
Progressive knowledge distillation lightweight method, device and equipment based on pruning network
The invention relates to a progressive knowledge distillation lightweight method, device and equipment based on a pruning network. The method comprises the steps of obtaining an original model, a preset pruning layer number and a training sample in a target task scene; pruning a first layer of the original model according to a preset pruning proportion, taking the original model as a teacher model, taking the pruned model as a student model, training the teacher model and the student model by adopting a distillation training strategy according to the training sample, and determining a first lightweight model, and pruning the next layer on the basis of obtaining the lightweight model after pruning, and repeating the training steps until the number of pruned layers meets a set value. According to the method, the concept of progressive pruning is adopted, pruning is carried out gradually from the shallow layer of the model, learning is carried out based on the provided middle characteristic distillation loss and output distillation loss, knowledge in an original model can be better migrated, and therefore the precision loss of an inference result caused by model lightweight is avoided to the maximum extent.
Owner:NAT UNIV OF DEFENSE TECH
Physical information guided deep learning geoelectric model electromagnetic forward modeling method
ActiveCN119598808Aavoid precision lossDesign optimisation/simulationConstraint-based CADAlgorithmElectromagnetic response
The invention relates to the field of geophysics and artificial intelligence, and particularly discloses a physical information guided deep learning geoelectric model electromagnetic forward modeling method, which comprises the following steps: S1, preparing neural network input data; s2, building a full-connection deep neural network; s3, adding a physical information constraint, and adding a Helmholtz equation satisfied by an electromagnetic field of the geoelectric model and a boundary condition as the physical information constraint into a loss function; s4, carrying out network training, solving a partial derivative of network output to input by using automatic differential, and obtaining a trained full-connection deep neural network; and S5, result prediction is carried out, the space coordinates of any point are input into the training network with the stored neural network model parameters, and the electromagnetic response of the point in the geoelectric model is obtained. According to the method, the electromagnetic response at any position in the geoelectric model is solved, and an intelligent scheme is provided for calculation of the electromagnetic response in electromagnetic exploration.
Owner:CENT SOUTH UNIV
A low-depth sequencing population genotype filling calculation memory optimization method
ActiveCN113742070Bpadding implementationavoid precision loss
The application discloses a low-depth sequencing population genotype filling calculation memory optimization method, which comprises the following steps: carrying out block division on a single nucleotide polymorphism site (SNP), and setting a checkpoint value according to the block division; and carrying out calculation on a forward auxiliary variable and a backward auxiliary variable according to the set checkpoint value, and calculating a haplotype observation conditional probability according to the block division. The application has the advantages of simple principle, effective reduction of calculation memory consumption, improvement of genotype filling efficiency and the like.
Owner:GENETALKS BIO TECH (CHANGSHA) CO LTD
System and method for estimating rotor position and rotating speed of permanent magnet synchronous motor
PendingCN119561443AImprove filtering effectImprove estimation accuracyElectronic commutation motor controlElectric motor controlIntegratorLow-pass filter
The invention belongs to the field of permanent magnet synchronous motors, and relates to a system and method for estimating the rotor position and rotating speed of a permanent magnet synchronous motor. A second-order generalized integrator is used for replacing a low-pass filter in a traditional sliding mode observer algorithm, high-frequency switching noise and higher harmonics in a counter electromotive force signal are filtered out through the second-order generalized integrator, efficient filtering can be achieved, and phase delay cannot be caused; and meanwhile, the amplitude frequency self-adaptive frequency-locked loop is utilized to realize the self-adaption of the tracking frequency, so that the precision and the dynamic performance of the estimation system can be effectively improved, and the control performance of the permanent magnet synchronous motor is improved. The problem that the control performance of the system is poor due to the sliding mode buffeting effect and higher harmonic noise interference in the high-speed stage of the permanent magnet synchronous motor is solved.
Owner:CHINA AUTOMOTIVE ENG RES INST
A large language model quantization method based on channel arrangement and activation adaptive smoothing
PendingCN122114018Astable captureReduce accuracy fluctuationsBiological modelsLinguistic modelAlgorithm
A large language model quantization method based on channel arrangement and activation adaptive smoothing belongs to the technical field of large language model compression and efficient inference. In order to solve the problem that the existing post-training quantization method is difficult to simultaneously consider the difference of different channel activation outliers, the group mode quantization has large dynamic range in the group, and the low bit weight and high bit activation alignment efficiency is low, which leads to the problem of precision decline and inference efficiency limitation, the scheme calibrates the maximum activation amplitude of the input channel by statistical data, divides the significant and non significant activation outlier channels according to the percentile threshold, and respectively uses the logarithmic domain and power function domain smoothing strategy to generate channel smoothing scaling factor, and fuses it into the model parameter to extract the weight multidimensional feature to construct the channel quantization sensitivity index and execute the input channel rearrangement, and further generate the joint quantization scaling factor to realize the integer domain alignment inference. It is suitable for low bit compression deployment and efficient inference scene of large language model in resource limited environment.
Owner:GUIZHOU UNIV
Human activity recognition method based on multi-scale group information distillation and refinement network
The application discloses a human activity recognition method based on multi-scale group information distillation and refinement network, relates to the technical field of artificial intelligence and edge computing, and comprises the following steps: constructing an input tensor based on measurement data of an inertial measurement unit; inputting a trained activity type recognition model; and calculating, which comprises the following steps: processing the input tensor through a multi-scale feature extraction module to obtain multi-scale fusion features; processing the input features through a group information distillation module in a group information distillation stage to obtain residual enhancement features; processing the residual enhancement features through a feature aggregation module to obtain aggregated features; processing the aggregated features through a feature integration module to obtain a global feature vector; and processing the global feature vector through a classification module to obtain an activity type classification result. The method provided by the application can significantly improve the recognition accuracy under the premise that the parameter quantity and the calculation cost are controlled, can be directly deployed on a smart phone, and can provide an efficient and reliable solution for real-time medical monitoring and motion analysis.
Owner:WUHAN UNIV
Limited angle motor displacement precision testing device and testing method
PendingCN119509431AEnsure consistencyGuaranteed reliabilityMeasurement devicesAxial displacementClassical mechanics
The invention relates to a precision testing device and a precision testing method for displacement of a limited-angle motor. The device comprises a motor fixing device which is used for fixing a tested motor; the pressing device comprises a pressing piece capable of pressing the tested motor in the axial direction; the bayonet device comprises an axial driving assembly, a radial driving assembly connected with the axial driving assembly, an elastic pressing assembly connected with the radial driving assembly and a bayonet piece connected with the elastic pressing assembly, and the bayonet piece can be driven by the radial driving assembly to move in the radial direction so as to clamp an output shaft of the tested motor; the bayonet piece can be driven by the axial driving assembly to move in the axial direction to downwards press or upwards pull an output shaft of the tested motor, and the elastic pressing assembly is used for providing pressing force when the bayonet piece downwards presses or upwards pulls the output shaft; and the displacement acquisition device is used for acquiring the numerical value of the axial displacement. According to the invention, the test efficiency is improved, the consistency of each test condition can be ensured, and the reliability of a test result is ensured.
Owner:CHINA AERONAUTICAL CONTROL SYST RES INST
A method for training an AI large model based on deep learning
The application discloses a kind of based on deep learning AI big model training method of model, including the following steps: obtaining training data and model to be trained, construct attention acceleration mechanism and multi-stage steady training scheduling mechanism, execute first forward propagation;Stability monitoring is executed, and steady score sequence is obtained by frequency domain analysis;The attention acceleration configuration corresponding to current training stage is generated, and attention structure is updated;Forward propagation and back propagation are executed, and stage training result is obtained;Steady score sequence is recalculated to complete stage switching determination, and the scheduling instruction of next training cycle is generated, and stage training configuration is updated;Training cycle is repeatedly executed, and training completion model is obtained to meet termination condition.The application utilizes frequency domain steady score and layered attention acceleration training mechanism, constructs dynamic and can back multi-stage steady scheduling method, realizes that big model training is more stable, faster, more efficient.
Owner:BEIJING LIYANG ZHIGUANG TECH CO LTD
A debugging fixture and machine tool mechanism for an automatic tool changer
ActiveCN224587572Ureduce dependenceEliminate the steps of "disassembly - installation of tooling - reassembly"
This utility model relates to the field of CNC machine tool technology, and provides a debugging fixture and machine tool mechanism for an automatic tool changer. The debugging fixture includes a tool position debugging fixture, a debugging ring fixture, and a central axis fixture. The tool position debugging fixture is installed on the tool holder mounting position of the tool magazine to be debugged, and has a central hole coaxial with the central hole of the tool holder. The debugging ring fixture has the same shape and size as the tool holder and is installed on the jaws of the robotic arm used for tool changing on the machine tool, with the debugging ring fixture coaxial with the clamping jaws of the robotic arm. The central axis fixture is installed on the central hole, with both ends extending along a first direction, and one end of the central axis fixture being coaxially and movably inserted into the debugging ring fixture. This application simplifies the debugging operation process and improves debugging accuracy and efficiency.
Owner:LANGFANG JINGDIAO MACHINE TOOL MFG
AC real-time parameter rapid correction method
PendingCN121978602Aavoid occupyingFast operationElectrical measurementsMicrocontrollerAC - Alternating current
The invention discloses an alternating current real-time parameter rapid correction method, which relates to the technical field of alternating current parameter correction, and comprises the following steps of: 1, establishing an alternating current real-time parameter coordinate system; 2, selecting a plurality of standard calibration points in the coordinate system; 3, selecting a plurality of to-be-corrected points in the coordinate system, and calculating a broken line slope K corresponding to each to-be-corrected point; and 4, directly updating the to-be-corrected point through the broken line slope K to obtain corrected data. According to the invention, a curing operation can be carried out before real-time measurement is started, and a large amount of resources occupied by division operation of a single-chip microcomputer are avoided, so that the operation speed is improved, and the accuracy of a measurement result is ensured. The method is suitable for measurement of real-time electrical parameters of alternating current and other basic measurement fields, and has a wide application prospect.
Owner:BENGBU POWER SUPPLY COMPANY STATE GRID ANHUI ELECTRIC POWER +1
A gear welding device and welding method for a press
ActiveCN121423773BRealize fully automatic transferEliminate offsetProgramme-controlled manipulatorWelding/cutting auxillary devicesMachine visionGear wheel
This invention relates to a gear welding device and method for a press, belonging to the field of press manufacturing. The technical solution is as follows: A gear welding device for a press includes a work platform with a ground rail. A transfer trolley is movably mounted on the ground rail to carry the gear to be welded. Positioners are mounted on both sides of the ground rail, each equipped with a rotatable gripper. A gantry frame is also mounted on the work platform, spanning the ground rail. A welding robot is movably mounted on the gantry frame, its movement direction being horizontal and perpendicular to the ground rail. The welding robot is equipped with a machine vision module. This solution, through the positioner grippers, can precisely achieve 180° rotation of the gear, providing a stable posture adjustment basis for double-sided welding. The welding robot, combined with the machine vision module, enables automated and high-precision acquisition of weld information, providing data support for intelligent welding path planning.
Owner:JIER MACHINE TOOL GROUP
Hardware-aware model quantization method, system, medium, product and terminal
The application provides a hardware perception-based model quantization method, system, medium, product and terminal, a hybrid execution mechanism is constructed between a high-level software framework and a bottom layer hardware behavior model, only core calculation operators in a quantization calculation graph of an initial quantization model which need to be executed on target hardware are handed over to a hardware behavior model for bit precision simulation, real calculation characteristics of the target hardware are simulated in a quantization process, and other auxiliary, non-computation-intensive operators are reserved in the high-level software framework for execution, so that the simulation verification efficiency is improved, the simulation precision and the execution efficiency are considered, and the goal that the quantized model runs more efficiently and has lower precision loss on the target hardware is achieved.
Owner:SHANGHAI GUANGYU XINCHEN TECHNOLOGY CO LTD
Positioning block for preventing drill bit from rotating, drill bit and machining method
ActiveCN119609196Aprevent rotationImprove the limit effectTransportation and packagingTrepanning drillsInvoluteIndustrial engineering
The invention relates to the technical field of machining equipment, and discloses a positioning block for preventing a drill bit from rotating, the drill bit and a machining method.The positioning block comprises a head-changing drill bit module, and a boss structure is arranged at the top of the head-changing drill bit module; the boss structure is a hexahedron structure and comprises a bottom surface, a top surface, two side surfaces and two end surfaces; wherein the bottom surface of the boss structure is arranged at the top of the head-changing drill bit module; the top face is of a plane structure, the two side faces are first stress faces, and the two end faces are second stress faces. The first stress surfaces are a pair of planes which are centrosymmetric to each other, and the second stress surfaces are a pair of involute curved surfaces which are centrosymmetric to each other. Relative rotation between the cutter body and the head-changing drill bit module is effectively prevented in the cutting process, blockage is avoided, and therefore the situation that chip removal is affected is avoided, and the stability of the machining quality is guaranteed.
Owner:CHENGDU TOOL RES INST
Fast extraction of scattering center parameters of SAR target based on depth unfolding
ActiveCN120275968Breduce inference timeImprove reasoning efficiencyNeural learning methodsComplex mathematical operationsComputational physicsTesting Methods
The application provides a SAR target scattering center parameter fast extraction method, and mainly solves the problems of slow speed and low precision of a traditional SAR target scattering center parameter extraction method; firstly, the application models a scattering center parameter solving problem based on a sparse representation theory; then, a SAR target scattering center parameter extraction problem model based on a semi-quadratic splitting method is constructed, a deep unfolding network is constructed according to an optimization process based on the semi-quadratic splitting method, and the SAR target scattering center parameter is extracted by using the deep unfolding network; compared with a traditional method, the method constructs a deep unfolding network based on the semi-quadratic splitting method and a scattering center model, and realizes efficient and interpretable SAR target scattering center parameter extraction.
Owner:NORTHWESTERN POLYTECHNICAL UNIV
Spectral confocal probe pose calibration method and device based on plane constraint
ActiveCN121761761BImprove computing efficiencyImprove calculation accuracyUsing optical meansOptical axisStandard plane
The application provides a plane constraint-based spectral confocal probe pose calibration method and device, and belongs to the field of precision measurement and instrument technology.The method comprises the following steps: firstly, at least three non-coplanar plane standard parts are measured by using a contact probe, and through coordinate conversion and fitting, equations of the standard planes in a machine coordinate system are accurately established; then, the same plane parts are measured by using a spectral confocal probe, and the obtained measurement point coordinates and distance values are combined with the known plane equations to construct a linear over-determined equation set about the probe pose parameters (including an optical axis direction vector and a space position vector); finally, the equation set is solved by using a least square method to obtain accurate pose parameters.The application converts a nonlinear problem into a linear solution, avoids iterative calculation, and significantly improves the calibration efficiency and accuracy.Meanwhile, the use of plane constraint allows the measurement points to be uniformly distributed, effectively reduces the equation set ill-conditioning, and enhances the robustness of the calibration process.
Owner:HUAZHONG UNIV OF SCI & TECH
Bionic manipulator
The utility model discloses a kind of bionic manipulator, it is related to manipulator technical field, bionic manipulator includes palm, finger and first driving device, second driving device and third driving device, finger is articulated by end knuckle, middle knuckle, tip knuckle in turn, end knuckle is movably connected with palm.The first driving device drives end knuckle flexion and extension relative to palm;Second driving device drives middle knuckle flexion and extension relative to end knuckle, and drives tip knuckle movement in parallel;Third driving device drives end knuckle (together with entire finger) relative to palm lateral swing.Three active degrees of freedom (flexion and extension, flexion and extension, side swing) transmission path is independent, realizes complete mechanical decoupling.
Owner:SUZHOU DONGSHAN PRECISION MANUFACTURING CO LTD +1
Prefabricated concrete sheet member quality detection method based on lightweight Transform and GPR data
PendingCN121837279AReduce computational complexitysmall amount of calculationImage enhancementImage analysisFeature DimensionAlgorithm
The invention relates to the technical field of nondestructive testing and deep learning, in particular to a prefabricated concrete sheet member quality detection method based on lightweight Transform and GPR data. Comprising the steps of GPR image data input and preprocessing, lightweight convolutional neural network feature extraction, feature dimension reduction, position coding and serialization, Transform encoder global feature modeling, Transform decoder target query and detection, target classification and bounding box regression, post-processing and result output and the like. According to the method, the lightweight backbone network is adopted, the parameter quantity and the calculation quantity are reduced through the deep separable convolution, the inverse residual structure and the SE attention mechanism, and meanwhile, the detection precision of small targets and dense targets is improved; the problems that in the prior art, detection precision and calculation efficiency are difficult to consider at the same time, and a lightweight model is insufficient in small target and dense target detection capacity are solved.
Owner:CHINA RAILWAY JINAN GRP CO LTD +1
Cutting device for LED light bar processing
The utility model relates to LED lamp strip processing technical field, and disclose a cutting device for LED lamp strip processing, including processing platform, motor, round knife cutting device and shunt anti -adhesion device, the processing platform top side wall is equipped with motor, motor and processing platform side wall pass through the fixed connection of adaptation bolt, and motor output shaft end is provided with round knife cutting device, round knife cutting device one side axle end is welded with motor output shaft end, and round knife cutting device is placed in the processing platform top, and processing platform side wall is provided with shunt anti -adhesion device, the opposite side wall of shunt anti -adhesion device passes through the fixed connection of adaptation screw and processing platform side wall, and shunt anti -adhesion device and round knife cutting device axial identity, this cutting device for LED lamp strip processing simple structure can freely adjust the cutting width size, and cooperate shunt anti -adhesion device to guarantee that the LED lamp strip after cutting will not appear the situation of bonding, and the processing quality is improved greatly.
Owner:SHENZHEN SEPITEK TECH CO LTD
White light interference detection device for on-machine detection
ActiveCN224216051UReal-time reflection of real three-dimensional topographyImprove production inspection efficiencyUsing optical meansNumerical controlWhite light interferometry
The utility model provides a white light interference detection device for on-machine detection, comprising an adjusting mechanism which comprises a transverse adjusting table sliding left and right, a longitudinal adjusting table sliding up and down, and a two-dimensional radian adjusting table used for adjusting the horizontal inclination and angle, the transverse adjusting table is fixedly installed on the front side of the longitudinal adjusting table, and the longitudinal adjusting table is fixedly connected with the two-dimensional radian adjusting table through a connecting plate. The optical detection device is fixedly installed above the two-dimensional radian adjusting table, one side of the tool rest connecting piece is fixedly connected with the front side of the transverse adjusting table, and the other side of the tool rest connecting piece is fixedly connected with a tool rest of the numerical control machine tool. According to the utility model, the detection device is directly arranged in a processing environment, and the position of the detection device is adjusted through the adjusting mechanism, so that on-machine detection in production is realized, the processing fixing position of a product is not influenced by detection, parts do not need to be disassembled, and the production detection efficiency is improved.
Owner:XIAN ENKE MICRO-NANO PHOTOELECTRIC TECH CO LTD
Vision-based wind blade damage defect detection method and system
ActiveCN121686077BRapid positioningRealize dynamic adjustmentImage analysisCharacter and pattern recognitionPattern recognitionImage resolution
The application provides a visual-based wind blade damage defect detection method and system, and relates to the field of image recognition.The method comprises the following steps: constructing a damage correlation knowledge graph based on the working environment characteristics and damage defects of a plurality of sample wind blades; constructing and training an improved YOLOv11 detection network, wherein the improved YOLOv11 detection network comprises an adaptive input unit, and the adaptive input unit is used for adjusting the resolution of an image based on the image features of the image and the damage correlation knowledge graph; constructing damage detection paths corresponding to various working environment types based on the damage correlation knowledge graph and the improved YOLOv11 detection network; determining a damage detection path based on the working environment characteristics of a wind blade to be detected and the damage detection paths corresponding to various working environment types; and performing damage defect detection on the wind blade to be detected based on the damage detection path and the improved YOLOv11 detection network, thereby achieving the advantage of automatic detection of wind blade damage.
Owner:VOCATIONAL & TECH COLLEGE OF INNER MONGOLIA AGRI UNIV
A method for online identification of dynamic loads based on interference observers
ActiveCN117786382BSolve the bottleneck problem of being unable to go onlineAccurate real-time identificationSimulationState space equation
This invention provides an online dynamic load identification method based on a disturbance observer. First, a structural state-space equation is established based on the mass matrix, damping matrix, and stiffness matrix of the linearly loaded structure. Then, the state-space equation of the loaded structure is compared and combined with the model equations in the control domain to design and determine the form and parameters of the disturbance observer characterized by the state-space equation parameters of the system model. This observer is then applied to the real-time online identification of external dynamic loads on the structure. By observing the difference between the actual output response and the estimated output response of the system in real time, the observer gain k is gradually adjusted until the estimated value approximates the actual input dynamic load. This method eliminates the need for complex design and extensive calculations, enabling accurate real-time identification of random dynamic loads on single-point / multi-point linearly loaded structures. It solves the problems of large solution errors and slow computation caused by matrix inversion operations in existing dynamic load identification technologies.
Owner:NORTHWESTERN POLYTECHNICAL UNIV
Foundation bolt positioning device and positioning method based on X / Y-direction adjustable positioning piece
PendingCN121992951AFlexible and independent adjustmentPrecise independent adjustmentBuilding material handlingControl theoryAnchor bolt
Owner:五矿二十三冶建设集团有限公司
A novel random phase difference real-time mapping system
This invention discloses a novel real-time random phase difference mapping system, comprising: a real-time random phase difference measurement unit for capturing the random phase difference between an external trigger signal and a system clock signal, and outputting a start node signal START and an end node signal STOP; and a time amplitude mapping unit, including a capacitor initial voltage circuit, a capacitor charging circuit, and a voltage holding sub-circuit. The START signal initiates capacitor charging, and the STOP signal locks the capacitor terminal voltage, linearly mapping the phase difference to voltage amplitude. The voltage holding sub-circuit operates in analog mode when the holding time does not exceed a threshold; after exceeding the threshold, it switches to digital drive mode, with the digital-to-analog converter output replacing the capacitor voltage, fundamentally eliminating voltage drops caused by analog leakage and achieving long-term drift-free, high-precision holding. The measurement range is extended by cascading multiple internal clock triggers, effectively eliminating the metastable measurement blind zone of boundary phase differences; and the range is flexibly configured through an adjustable reference voltage source and an adjustable constant current source.
Owner:UNIV OF ELECTRONICS SCI & TECH OF CHINA
A method and system for numerical simulation of immersed boundaries based on grid pruning method
ActiveCN119203386BGuaranteed accuracyImprove computing efficiencyGeometric CADSustainable transportation
The present invention provides an immersed boundary numerical simulation method and system based on a grid pruning method, relates to the field of computational fluid dynamics, and solves the problems of grid generation and computational accuracy in numerical simulation of flow fields with complex geometric shapes; the method comprises: setting a computational domain grid, identifying and determining the interface between a fluid and a geometric object; the computational domain grid forms a cut grid under the action of the interface, and the basic parameters of the cut grid are determined by calculation; the control equation of a standard grid that is not cut by the surface of the geometric object in the computational domain grid is established; on the basis of the form of the control equation of the standard grid, the control equation of the interface grid that is cut by the surface of the geometric object in the computational domain grid is determined; based on the control equations of the standard grid and the interface grid, the flow field is numerically simulated in an iterative solution manner; the present invention can effectively reduce the amount of grids and the technical difficulty of numerical simulation, and can use a high-order method to obtain high-order accuracy.
Owner:CALCULATION AERODYNAMICS INST CHINA AERODYNAMICS RES & DEV CENT
Multimode fiber dynamic disturbance imaging reconstruction method based on parallel neural network
PendingCN121861134Aavoid precision lossquality improvement2D-image generationBiological modelsReconstruction methodImaging data
The invention discloses a multimode fiber dynamic disturbance imaging reconstruction method based on a parallel neural network, and the method comprises the following steps: employing an LED as an illumination light source, and constructing a single-excitation wide-view-field multimode fiber (MMF) imaging system; under the condition that the disturbance range is 0-200 mm, image data of the output end of the multimode optical fiber are collected. By designing a parallel neural network structure, the network can perform efficient feature extraction and accurate image reconstruction on an optical fiber transmission image under dynamic disturbance, and particularly realizes high-fidelity image restoration under a small sample condition. The parallel neural network comprises a plurality of parallel sub-networks, each sub-network specially learns image mapping features under different disturbance conditions, and finally feature fusion is performed at an output end to obtain a stable image reconstruction result. According to the invention, the method has the remarkable advantages of high reconstruction precision, strong anti-disturbance capability and high training efficiency, and provides a novel and effective implementation way for high-fidelity imaging of the multimode optical fiber in a dynamic environment.
Owner:SHANGHAI INST OF OPTICAL INSTR +1
Air compressor modeling method and system based on data fitting driving
This invention discloses a data-driven modeling method and system for air compressors, comprising: acquiring operating data of an air compressor under various operating conditions; preprocessing the operating data to obtain cleaned operating data; dividing the cleaned operating data into time windows, performing multi-order polynomial regression fitting within each time window to generate an initial fitting parameter set corresponding to each time window; constructing a fitting parameter matrix based on all initial fitting parameter sets, performing matrix decomposition operations, and extracting principal component features of the fitting parameters; establishing an air compressor performance degradation trend model based on the principal component features to describe the performance decay law of the air compressor under different operating durations; and dynamically modeling based on the performance degradation trend model and combined with real-time acquired operating data to generate a real-time performance mapping model. This achieves automatic extraction of performance degradation features from historical data and dynamic compensation modeling.
Owner:HUBEI TEWATT POWER TECH
Power transmission line icing probability prediction method based on icing identification and quantile regression
PendingCN119557851AImprove forecast accuracyavoid precision lossBiological modelsInformation technology support systemQuantile regressionCluster algorithm
A power transmission line icing probability prediction method based on icing identification and quantile regression belongs to the technical field of power transmission line icing thickness probability prediction, and comprises the following steps: step S10, obtaining historical icing data, meteorological data and topographic data of a power transmission line in a prediction region to obtain key features; step S20, dividing the icing modes of the power transmission line into N types of labels by means of a k-means clustering algorithm, and obtaining sub-data sets under different icing mode labels; step S30, constructing a power transmission line icing mode identification model by means of a CNN model, and outputting a label of an icing mode at a to-be-predicted moment T; and S40, constructing an icing thickness probability prediction model by means of combination of quantile regression and an LSTM neural network, outputting an icing thickness probability distribution interval of the power transmission line at a to-be-predicted moment T, and obtaining icing thickness probability prediction results in different icing modes. Accurate icing thickness probability prediction can be provided, and the method is of great significance to power grid transmission line anti-icing work, power grid safe operation and operation and maintenance.
Owner:SUPER HIGH VOLTAGE BRANCH OF STATE GRID JIBEI ELECTRIC POWER CO LTD +2
一种保持物理一致性的神经网络分层差异化量化方法及装置
PendingCN122414277ARealize refined modelingavoid precision loss
Owner:HANGZHOU INTERNATIONAL INNOVATION INSTITUTE OF BEIHANG UNIVERSITY
Task-oriented energy efficiency optimization method and device for dynamic position deployment of unmanned aerial vehicle
PendingCN122363246AResolve task contention issuesImprove global utilizationQuality of serviceVehicle dynamics
The application provides a task energy efficiency optimization-oriented unmanned aerial vehicle dynamic position deployment control method and device, and relates to the technical field of unmanned aerial vehicle communication control.The application constructs an unmanned aerial vehicle mobile edge computing system model, collects a current environment state at each discrete time step to generate a local observation vector, and generates a deterministic action vector; then, a distributed matching game with a ground mobile device is performed to complete task association; then, position updating is performed, a multi-target reward function is calculated, and centralized network model training is performed to iteratively update a Critic evaluation network and an Actor strategy network; when the model training reaches a preset requirement, the training is terminated, and a task energy efficiency optimal unmanned aerial vehicle dynamic position deployment control strategy is obtained. The application can realize deep coupling optimization of physical trajectories and offloading decisions, effectively solve user association conflicts and uneven computing load problems in a dynamic scene, and significantly improve system comprehensive energy efficiency while guaranteeing service quality.
Owner:PUTIAN UNIV
Motor imagery classification method and system based on weight quantization sharing SNN
PendingCN121980345Aimprove performanceLow inference costBiological modelsData setSpiking neural network
The invention discloses a motor imagery classification method and system based on a weight quantization sharing SNN, and the method comprises the steps: firstly carrying out the preprocessing of an original electroencephalogram data set, and selecting a neural network architecture; secondly, a spiking neural network SNN is constructed based on a neural network architecture, and spiking neurons adopt SRIF to reduce errors; and constructing an ANN based on a neural network architecture, training the ANN by using the preprocessed electroencephalogram data set, and obtaining pre-training weight and activation value distribution. And then carrying out quantization processing on the pre-training weight and the activation value through quantization perception training, and constructing a quantization ANN. And finally, based on the SNN and the quantized ANN, carrying out joint training by sharing weights, and outputting a classification result through the SNN after joint training to realize motor imagery classification. The SNN obtained through conversion can have the advantages of being high in performance and low in reasoning cost, and motor imagery classification tasks can be accurately and rapidly achieved.
Owner:HANGZHOU DIANZI UNIV