50 results about "Transfer model" patented technology

The invention discloses a micro-seismic signal identification method based on transfer learning and BiLSTM-DCNN, belongs to the technical field of mining engineering micro-seismic monitoring and signal processing, and solves the problems of data scarcity in the initial stage of mine monitoring and low identification precision under the condition of small samples. Firstly, Mel spectrum feature extraction is carried out on a micro-seismic signal; the method comprises the following steps: constructing a BiLSTM-DCNN model comprising a bidirectional long-short term memory network and a deep convolutional neural network, and carrying out pre-training by using large-scale source mine data; and the model is adapted to target mine small sample data through transfer learning, and model parameters are finely adjusted, so that high-precision classification of the signals is realized. According to the method, the recognition accuracy and the model generalization ability of the micro-seismic signals under the small sample condition are remarkably improved, the test accuracy reaches 0.9444 and is improved by 80.85% compared with an unmigrated model, and the method is suitable for an intelligent early warning and safety monitoring system for mine dynamic disasters.

A defect detection method and a model transfer method. The model transfer method comprises: acquiring a source-domain image and a corresponding annotation, and a baseline model obtained from a source domain; acquiring target-domain images, which are fully annotated, partially annotated, or unannotated; inputting the annotation of the source-domain image into a denoiser to obtain a denoised annotation; inputting the unannotated target-domain images into an initial baseline model to obtain second target-domain prediction results, performing data augmentation on the unannotated target-domain images to obtain second target-domain augmented images, and using the second target-domain prediction results as pseudo-annotations of the second target-domain augmented images; and using the source-domain image and the corresponding denoised annotation, the annotated target-domain images and the corresponding annotations, and / or the second target-domain augmented images and the corresponding pseudo-annotations to train the baseline model, so as to obtain a transfer model. The method can complete training by using a small number of annotated target-domain images, thereby solving the problem of model transfer performance being poor in the case of insufficient target-domain images.

The invention provides a disease progress probability prediction method and device based on a Monte Carlo-Markov chain, and relates to the technical field of disease progress prediction. The method comprises the following steps: acquiring multi-modal historical follow-up visit data of a patient to be tested, and constructing a state space with transfer constraint; extracting and checking time series data according to the state space; performing Monte Carlo sampling on the time sequence data, establishing a non-homogeneous transfer model and forming an individualized non-homogeneous transfer kernel set; completing path interpolation and probability filling of missing fragments based on the transfer kernel set to obtain continuous complete state sequence data; a Markov chain Monte Carlo algorithm is used for parameter updating, an individualized transfer model subjected to posteriori updating is obtained, a calibration probability result is obtained, and dynamic prediction and uncertainty evaluation of individualized disease progression are achieved. According to the method, the problems of inaccurate state determination, single transfer modeling, incomplete observation and insufficient prediction calibration in the prior art are solved.

The invention discloses a multi-stage rotor assembly method based on four-in-one-axis cooperative regulation and control, and belongs to the technical field of aero-engine manufacturing. The four-in-one is a combination of a rotation axis, a geometric axis, an inertia main shaft and a mass center axis. The multi-stage rotor assembling method is realized through the following steps of S1, obtaining rotor size characteristics, geometric errors, mass center coordinates and the inertia main shaft; s2, establishing a geometric error representation model of the single-stage rotor; s3, establishing an error transfer model of multi-stage rotor assembly; s4, calculating end surface centroid and centroid coordinates of each stage of rotor after assembly; s5, fitting the rotation axis, the geometric axis and the mass center axis; s6, solving an inertia tensor matrix under the assembly coordinate system; s7, calculating an inertia main shaft inclination angle under the assembly coordinate system; and S8, four-axis-in-one assembling regulation and control are carried out. The problem that vibration exceeds the standard under the ultrahigh rotating speed due to low manufacturing precision of an aero-engine rotor can be solved.

This invention discloses a diffusion model-driven method for generating multilingual human motion, comprising: constructing a sample library; performing multilingual text translation and feature encoding using the Ali Tongyi Qianwen model; constructing a customized CondUNet1D motion denoising network with residual linear multi-head cross-attention, Dropout layer, and Rearrange layer optimization; and training using a joint strategy of exponential moving average and classifier-free guidance. During the inference phase, starting with pure noise, iterative denoising is achieved using a variant of the second-order DPMSolver++ sampler combined with training-free acceleration techniques. Simultaneously, foot slippage is identified and corrected using a vGRFs model transfer model. After secondary optimization of the posture using the diffusion model, a slippage-free motion sequence is obtained. Finally, a video is generated by a remote server and transmitted back to the local terminal. This invention achieves multilingual-driven, efficient, and highly realistic human motion generation.

This invention discloses a high-speed train bearing fault diagnosis method based on transfer learning in the field of rail transit safety monitoring technology. The method preprocesses source and target domain data separately, extracts multi-dimensional features, and uses the t-SNE algorithm for dimensionality reduction and filtering to construct source and target domain feature sets. Based on the source domain feature set, a Stacking ensemble model is constructed and trained as the source domain baseline fault diagnosis model. A multi-scale subspace alignment transfer model is constructed, projecting the source and target domain feature sets onto a shared subspace and performing geometric alignment to obtain domain-invariant features. Based on the aligned target domain features and combined with a multi-scale decision fusion strategy, fault classification is performed on the target domain data, outputting the fault diagnosis result. This method constructs a high-precision baseline model using labeled source domain data and utilizes transfer learning to align cross-domain features, achieving high-accuracy fault diagnosis in the unlabeled target domain.

The application discloses a kind of photoelectric detection model transfer sharing method and internet of things monitoring evaluation system of cloud service, the transfer sharing method specifically is: by calling temperature compensation model and spectrum transfer model, the spectrum information of agricultural product sample is corrected, calling detection model is calculated to the corrected spectrum information, obtains the detection result of agricultural product sample;Internet of things monitoring evaluation system utilizes spectrum transfer sharing method to carry out spectrum information correction, and then calls detection model to calculate, and detection result is returned to detection terminal in real time, realizes remote monitoring and evaluation of agricultural product quality.The application can realize that detection model is shared in different detection terminal, and has wide application prospect in agricultural product quality sampling evaluation.

An electronic component manufacturing system is configured to identify a machine learning model trained to generate analysis or prediction data for a first substrate processing domain associated with a type of substrate processing system. The system is further configured to obtain first trajectory data relating to the first domain to train the machine learning model. The system is further configured to a migration model for a second substrate processing domain associated with the type of substrate processing system. The migration model is generated based on the first trajectory data related to the first substrate processing domain and second trajectory data related to the second substrate processing domain. At least one of the machine learning model associated with the second substrate processing domain, or current trajectory data, using the migration model, is modified to enable the machine learning model to generate analysis or prediction data associated with the second substrate processing domain.

The invention relates to the technical field of compressors, and discloses a compressor running state recognition method and device, a medium and equipment. The method comprises the steps that the running state of the compressor is defined as double state spaces, and the double state spaces are the load state space and the health state space of the compressor; identifying the load state of the compressor and marking the health state of the compressor; establishing a running state space transfer model; performing parameter estimation based on an EM algorithm, and obtaining a parameter calculation result of the operation state space transfer model; compressor health state evaluation of the operation state space transfer model is obtained, and health indexes of the compressor are redefined to be in a continuous expression form; and the real-time load state, the health grading state and the health degree index of the compressor are obtained, and operation management personnel are prompted to conduct corresponding compressor health management operation. According to the method, the abnormality handling efficiency can be improved, and abnormal false alarms caused by compressor load switching and parameter fluctuation can be reduced.

The application discloses a kind of equipment life prediction methods based on particle filtering's LSTM model, it is related to system reliability field.The method proposes the RUL prediction method of fusion PF and LSTM, state transition model in PF filtering process is established using LSTM, the resampling method in PF is improved so that resampling can retain sampling gradient information for network model update.PF-LSTM model realized can give full play to LSTM on time series modeling and the state estimation capability of PF to nonlinear, non-stationary and non-Gaussian system.Original data is directly sent into network after standardization processing, and end-to-end learning ensures that the network only focuses on features related to the prediction target during the learning process, improving network sequence efficiency and accuracy.

The application discloses a multi-stage interference transfer modeling-based electromagnetic compatibility design method and system for power intelligent equipment, and belongs to the technical field of electromagnetic interference. The method comprises the following steps: synchronously collecting external port excitation and internal loop node interference data under multiple working conditions, and establishing a transfer matrix from the port to the loop; synchronously collecting loop node and chip key pin response data, fitting a transfer matrix from the loop to the pin, and cascading to form a full-link interference transfer model; positioning an electromagnetic sensitive point by calculating the sensitivity S of each level node in a preset time window i ; iteratively updating associated electromagnetic compatibility design parameters, repeating the modeling and evaluation process until all nodes Si meet the threshold requirement; and finally outputting multi-stage transfer model parameters, sensitive point positioning results and final design parameter configuration. Through full-link quantitative modeling and closed-loop iterative optimization, the application solves the blindness of traditional empirical design, and significantly improves the electromagnetic compatibility design efficiency and the on-site reliability of the equipment.

The invention discloses a load transfer switch control method and related equipment, and the method comprises the steps: obtaining the operation parameters of a power distribution network line according to a preset data structure based on a power distribution network line topological graph of a target power distribution network; determining a target function for the power distribution network line according to a rule of lowest transmission loss in a transfer process, and constructing an initial transfer model based on the target function and a preset constraint condition; through a second-order cone optimization auxiliary variable, converting the initial wheeling model to obtain a target wheeling model; and substituting the operation parameters into the target transfer model to obtain a to-be-solved model, and performing second-order cone optimization solution on the to-be-solved model to obtain the control method for the load transfer switch.

The invention provides a steel production energy consumption data processing method and device, a medium and electronic equipment. The method comprises the steps that production information and energy metering information of all working procedures in the steel production process are obtained, the production information comprises product identifiers and corresponding production time periods, and the energy metering information comprises energy medium types and corresponding consumption; performing time-space synchronization processing on the production information and the energy metering information, and establishing an association mapping relation between the production information and the energy metering information; calculating single-process energy consumption of each piece of secondary product in each process based on the association mapping relation and the material flow characteristics of each process; and according to the single-process energy consumption and the process circulation logic of steel production, constructing an energy consumption circulation model, and calculating the accumulated energy consumption of each product through the energy consumption circulation model. According to the invention, the accuracy of steel production energy consumption data processing can be improved.

This invention provides a controllable intelligent regeneration method for architectural contexts based on the Flux large-scale model ecosystem. The method involves inputting a design model drawing into the ComfyUI interface and reading its dimensions; determining the modeling accuracy of the design model drawing and matching it with the corresponding basic shape control algorithm; inputting an architectural context reference image and generating architectural context prompts; setting a style transfer model sequence and connecting it to a Flux large-scale model-based image generation module to obtain an architectural context rendering image; determining whether the architectural context rendering image conforms to the design intent; if it does, it is directly output; otherwise, precise modification and regeneration of the rendering image are completed through text control; finally, the architectural context rendering image is enlarged and restored to generate a high-resolution architectural context rendering image, which is then output. This invention solves problems such as insufficient intelligence in the image generation process, the need for manually written prompts, reliance on manual selection for modification results, and deficiencies in the quality and content matching of architectural context rendering images.

The application discloses a blower bearing fault diagnosis method based on a multi-source feature fusion transfer model. The method synchronously collects multi-source operation data by deploying a sensor network on a target blower and extracts fusion features, and simultaneously trains a basic diagnosis model by using complete fault data of a laboratory benchmark blower. When deployed, the similarity of the feature distribution of the target blower and the benchmark blower is quantitatively evaluated to determine the feasibility of transfer, and the model is safely fine-tuned and lightened based on the screened source domain knowledge. Finally, the optimized model is deployed to the edge side of the target blower to realize real-time diagnosis. The method effectively reduces the dependence on the historical fault data of a new blower, realizes rapid, safe deployment and precise self-adaptation of the diagnosis model, improves the timeliness of fault early warning, and guarantees the safe operation of the blower.

This invention provides a method and system for cross-domain data transfer in large-scale model applications, belonging to the fields of artificial intelligence and blockchain technology. The system includes a large-scale model, a blockchain system, and domain databases of system participants. In terms of system architecture, the method and system for cross-domain data transfer in large-scale model applications provided by this invention, by combining the trusted foundation of blockchain technology, extends the traditional single-user, single-point application of large-scale artificial intelligence models to multi-user, multi-domain shared use scenarios. Furthermore, through the cross-support of data from different domains in a distributed network, it significantly improves the effectiveness of traditional RAG-assisted generation based on local data. In terms of operational mechanism, it proposes a data transfer model based on blockchain and smart contracts for the data element domain, realizing on-demand access to data from different domains while ensuring the ownership, revenue rights, security, and privacy of the data of each participant.

The invention discloses a mass flow controller active-disturbance-rejection control method and system based on multi-interference-source cooperative suppression, and belongs to the field of general control or regulation systems.The method comprises the steps that a physical transmission model containing multiple interference sources such as inlet pressure disturbance and temperature change is established; a decoupling expansion state observer containing three independent channels is constructed, and interference of different frequency bands is separated through a decoupling compensation matrix; designing an ESO bandwidth adaptive law based on a disturbance spectrum and a tracking error, and adopting a bimodal adjustment and smooth transition strategy; an interference estimation value is obtained through an observer and serves as a feed-forward compensation amount, and a control instruction is generated by combining the feed-forward compensation amount with self-adaptive non-linear state error feedback; data are collected in real time, model parameters are corrected online, and a system is correspondingly provided with a model construction module, an observer construction module and the like. According to the invention, multi-interference cooperative suppression is realized, interference immunity and stability are balanced, and control precision and stability under complex working conditions are improved.

This application belongs to the field of twisting equipment control technology, and particularly relates to an adaptive control system for the operating state of twisting equipment. The system includes: acquiring operating signals from the twisting equipment; constructing a state observation sequence using a sliding window and extracting phase lag, vibration envelope skewness, tension gradient, spectral energy difference, sign change density, and multi-channel correlation disorder; dividing the operating state to establish a discrete state set; statistically analyzing the dwell length of each discrete state to construct a semi-Markov transfer model; combining the winding diameter range and tension gradient with piecewise correction to obtain a time-varying state transfer model; adjusting the prediction time domain length based on the expected remaining dwell time of each discrete state; predicting future state paths and solving the objective function to obtain the control increment; and outputting the control increment to the actuator to complete the operating state control. This application can improve the adjustment accuracy and reliability of the operating state control of twisting equipment.

This invention discloses a model-based method for learning robot operation skill parameters, including a single-skill policy learning module and a task-skill parameter learning module. The single-skill policy learning module includes an object-oriented representation model and a policy model conditioned on semantic goals. The task-skill parameter learning module includes an implicit state transition model learning module and an online skill parameter planning module. The single semantic skill policy learning module of this invention combines representation learning and reinforcement learning, integrating the reasoning transformation from visual input to the logical representation of the current system state with task-oriented robot policy learning to generate action parameters that satisfy the target logical state. Using reinforcement learning avoids extensive model design and processing, as well as the need for expert teaching data. Furthermore, by predefining the action sequences of skills based on an operation knowledge base, the search space for actions is reduced, lowering the exploration burden on the agent.

A 3D digital transfer management cloud platform, management method, equipment, and medium include: a project management module for creating new projects and using a directory template to allocate project data uploaded by each participant at each stage to the corresponding directory structure; a deliverables transfer module for performing data quality review and optimization processing on the digital deliverables to be transferred selected in the project management module, and transferring the reviewed and optimized digital deliverables to the digital asset management module; a digital asset management module for receiving digital deliverables from the deliverables transfer module, performing asset coding, classification, and attribute information entry, and storing them in a 3D design general model library; and a data display module for obtaining the transferred models from the digital asset management module, performing 3D model browsing, and model-image linkage. The deliverables transfer module of this application supports data transfer, quality review, and optimization processing to ensure the accuracy, completeness, and usability of the transferred data.

The invention discloses a multi-source-domain collaborative optimization UHPC anchoring area performance transfer learning prediction method and device, and relates to the technical field of high-performance concrete structure performance prediction. The method comprises the steps that a source domain data set and a target domain data set are acquired and preprocessed, and the preprocessed source domain data set and the preprocessed target domain data set are acquired; constructing a pre-training model, and training the pre-training model by adopting the preprocessed data set to obtain an optimal pre-training model; a target migration model is initialized, a shared feature extractor of the optimal pre-training model is frozen and migrated to the target migration model, optimization is carried out through source domain adversarial training and target domain fine tuning, and a trained target migration model is obtained; and inputting to-be-predicted design parameters of the anchoring area of the coarse aggregate ultra-high performance concrete into the trained target migration model, and outputting a performance prediction result of the anchoring area. By adopting the method, the performance prediction precision and generalization of the new material anchoring area can be improved.

The invention provides a multistage rotor assembly method based on cooperative regulation and control of a geometric axis and a mass center axis, and belongs to the technical field of rotor assembly. In order to solve the problems that rotors need to be assembled / disassembled repeatedly in the existing rotor assembly process, the efficiency is low, and collaborative optimization of geometric accuracy and dynamic performance is difficult to achieve fundamentally, the method comprises the steps that mass center information of a k-level rotor under global coordinates is obtained; according to the direction and position information of the k-stage rotor in the global coordinate system, calculating the total gravitational torque of the multi-stage rotor assembly; rotor pose information is analyzed, and a multistage rotor geometric error transfer model is established; based on a rotor geometric error transfer model, the multi-stage rotor centroid coaxiality under the rotary coordinate system after assembly is obtained; establishing a multi-objective optimization equation of the coaxiality and the total gravitational torque; and solving the multi-objective optimization equation by using a non-dominated sorting genetic algorithm fused with a simulated annealing mechanism to obtain an optimal assembly phase group, and assembling the multistage rotor based on the optimal assembly phase group.

The application discloses a kind of based on transfer learning and BiLSTM-DCNN's microseismic signal identification method, belongs to mining engineering microseismic monitoring and signal processing technical field, solve the problem of low identification precision under the condition of small sample with data scarcity in the initial stage of mine monitoring. First, the microseismic signal is extracted for Mel spectrum feature;BiLSTM-DCNN model containing bidirectional long short-term memory network and deep convolutional neural network is constructed, and large-scale source mine data is used for pre-training;Then, through transfer learning, the model is adapted to the target mine small sample data, and the model parameters are fine-tuned to achieve high-precision classification of signals. The application significantly improves the recognition accuracy of microseismic signals and the model generalization ability under the condition of small sample, and the test accuracy reaches 0.9444, which is improved by 80.85% compared with the non-transfer model, and is suitable for mine dynamic disaster intelligent early warning and safety monitoring system.

This invention provides a method for analyzing the oscillation transfer effect in a network-type virtual synchronous generator system, belonging to the field of virtual synchronous generator control. It establishes a small-signal model of the GFM-VSG system; based on this model, it establishes coupling transfer functions between various electrical quantities and an oscillation transfer model for the GFM-VSG system, where each electrical quantity includes active power, reactive power, frequency, and voltage; based on the amplitude-frequency characteristics of the oscillation transfer model, it evaluates whether the oscillation transfer effect between electrical quantities is an amplification or suppression effect; and based on the gain ratio of the oscillation transfer between electrical quantities, it determines the magnitude of the oscillation amplitude of the corresponding electrical quantities under different operating conditions. The invention also provides a system capable of assessing the strength of the oscillation transfer effect between electrical quantities, identifying whether the oscillation effect exacerbates or suppresses system oscillation, and using the gain ratio of the oscillation transfer between electrical quantities as a basis for determining the trend of oscillation amplitude changes in different electrical quantities.

The application provides a ground motion spectrum acceleration prediction method, a transfer learning method, equipment and a medium. A global model is obtained through first training by using basic knowledge data set, and the first training is evaluated by using regional knowledge data set. Then, a transfer model is obtained through second training by using the regional knowledge data set, and the second training is evaluated by using the regional knowledge data set. The establishment method of the transfer model retains the data characteristics of the regional database, learns the difference between the global data set and the regional data set, thereby ensuring the prediction accuracy of the model, and additionally enhancing the generalization ability of the model.

The invention relates to the technical field of communication, in particular to a mobility load balancing processing method and system based on weight adaptive adjustment. Comprising the following steps: generating a plurality of communication cells according to a communication structure, and constructing an association transfer model; obtaining load monitoring packets of each communication cell according to a preset time node, and generating a load abnormal value of each communication cell according to all the load monitoring packets; setting a plurality of risk cells according to all the load abnormal values, and generating an adjustment sub-strategy of each risk cell according to a preset load adjustment model; the method comprises the following steps: analyzing historical execution records of each communication cell, setting a plurality of load adjustment models, setting initial weight parameters of each load adjustment scene to ensure that the weight is adjusted in the direction of improving the overall communication quality, and setting an iteration sub-model of each load adjustment scene to improve the overall communication quality. The system oscillation caused by the weight parameter optimization process is avoided, and the weight self-adaption efficiency is improved.

The invention discloses a construction method of a fish-to-radionuclide concentration dynamic transfer model, and the method comprises the following steps: building a water body-fish two-chamber model, and obtaining a differential equation for describing the dynamic transfer of radionuclides in a water body and a fish body in a concentration process; transforming the differential equation to obtain a parameter equation that the radionuclide specific activity in the fish body library changes along with time; aiming at the obtained parameter equation, setting a fish body nuclide concentration coefficient, an uptake half-life period and a water body density in an equilibrium state, and combining an initial condition that the initial specific activity is 0 to obtain an equation set and carrying out simultaneous solution; the dynamic transfer model of the radionuclide in the fish body in the concentration process is obtained by substituting the water body density, the specific activity of the radionuclide in the water body, the nuclide concentration coefficient of the fish body and the specific numerical value of the uptake half-life period into a formula, and the dynamic transfer model of the radionuclide in the fish body in the final concentration process is obtained. By means of the method, the dynamic transfer behavior of fishes to radionuclides in water can be simulated, the dynamic transfer rule of fishes to radionuclides is explored, and therefore the influence of nuclear power liquid effluent discharge on the water environment is further evaluated.

This invention provides a strategy generation method, apparatus, and electronic device for integrated energy systems, applicable to the field of energy dispatching technology. The method includes: processing historical load information from the user end using a dynamic trading adjustment model to generate initial resource trading information for the distribution network; processing the initial resource trading information, initial electricity load transfer information, and a predetermined proportion of load that can be shifted using a dynamic load transfer model to generate a power consumption strategy for the user end; processing initial environmental information and photovoltaic power generation equipment parameters using an objective function to generate the expected output power of the photovoltaic power generation equipment; determining the distribution strategy of the distribution network based on the power consumption strategy, expected output power, and energy storage status of the energy storage system; determining the energy storage strategy of the energy storage system based on the initial resource trading information; and generating a target strategy by iteratively updating the initial resource trading information, power consumption strategy, and energy storage strategy using a reinforcement learning algorithm.

The invention provides a method for constructing a machine-die-material-part precision transfer model of a medium plate fine blanking component, and belongs to the technical field of fine blanking machining. The method comprises the steps that a compliance matrix and a load vector of a target machine tool are obtained; performing multiplication calculation on the compliance matrix and the load vector to obtain first error information; a second error, a nominal clearance, a cutting edge geometric parameter and a cutting edge abrasion loss of the target mold are obtained; based on the first error information and the second error information, the effective gap and the dislocation amount delta of the target mold are determined; based on the effective gap, the dislocation amount and the material parameter information, the material springback amount of the target board is determined; the yield strength, the elastic modulus, the nominal thickness t and the thickness deviation of the target plate are obtained; according to a constructed material springback value prediction model s, beta0, beta1 and beta2 are model coefficients of the material springback value model and are effective gaps, and delta is a dislocation value; and according to the constructed multi-level precision transfer model, the weight coefficients calibrated for the dimensional deviation, gamma 1, gamma 2 and gamma 3 of the target fine blanking component are effective gaps. The method does not need numerical simulation, is small in calculation amount, can be applied online, and is suitable for multi-dimensional precision prediction and process parameter optimization of the medium-thickness plate fine blanking part.

The application discloses a kind of RIS auxiliary UAV-ISAC system security energy efficiency optimization methods, comprising: using unmanned aerial vehicle UAV subsystem acquisition individual link channel information data;According to channel posterior distribution and channel transfer model, obtain channel posterior covariance;Multi-agent PPO decision network is constructed, and there is deep reinforcement learning model DRL in multi-agent PPO decision network, and multi-agent PPO decision network generates the beamforming matrix of artificial noise AN, and generates total transmission signal;Optimize the control parameter of multi-agent PPO decision network, output convergent multi-agent PPO decision network, to optimize the communication rate, energy efficiency and security performance of system.The present application introduces the mechanism of bayesian channel estimation, improves the robustness of system under the imperfect scene of CSI (channel state information), ensures the stable performance of channel uncertainty RMSE and SEE.