186 results about "Data heterogeneity" patented technology

The invention relates to the technical field of petroleum and natural gas engineering, and discloses an oil and gas reservoir optimized mining method based on multi-modal data, which comprises the following steps: constructing an oil and gas reservoir multi-modal data acquisition system, seismic wave field data, logging interpretation data, production dynamic data, micro-seismic monitoring data, underground temperature and pressure time sequence data and shaft structure parameter data are obtained through the acquisition system; and performing space-time alignment processing on the acquired multi-modal data, establishing a unified geological coordinate system and a time reference, and eliminating data isomerism caused by different acquisition frequencies and spatial resolutions. A six-dimensional heterogeneous data acquisition system covering a seismic wave field, well logging interpretation, production dynamics, micro-seismic monitoring, an underground temperature and pressure time sequence and shaft structure parameters is constructed, so that the depiction precision of a reservoir porosity field, a permeability field, a saturation field and a pressure field is essentially improved.

The invention discloses a truss structure wind-induced dynamic response prediction method and system based on physical enhancement. The method comprises the following steps: carrying out feature extraction and alignment fusion on input data containing condition parameters and wind speed time sequence data by utilizing a long short-term memory network and a physical enhancement attention mechanism; extracting multi-scale features from the fusion features through expansion convolution, and performing weighted aggregation on the multi-scale features; the physical priori knowledge of structural vibration is fused into position coding and a self-attention mechanism so as to carry out response prediction; and integrating physical model information of the truss structure and a dynamic control equation into a loss function, and calculating physical information residual loss so as to improve the physical interpretability of a prediction result. According to the method, data heterogeneity can be eliminated, complementary information can be fused, the multi-scale characteristic of wind-induced response is coped with, the accuracy and efficiency of wind-induced dynamic response prediction of the truss structure are effectively improved, and the physical interpretability and generalization ability are enhanced.

The invention discloses a federal large model knowledge collaborative training method supporting multi-modal heterogeneous clients, which comprises the following steps: each client receives a model initialization parameter issued by a central server, and applies adaptive differential privacy noise to independently train a heterogeneous lightweight model based on local private data; updating the model to which the noise is applied and uploading a modal identifier of the model to a central server side; after model updating and modal identification of each client are received, based on a modal perception weighted consensus fusion mechanism, knowledge of each client is fused to update a global large model; and the central server side issues the updated presentation layer parameters of the global large model to the client side for initialization of the next round of local training. According to the method, on the premise that a public data set or specific task setting is not needed, comprehensive compatibility of data isomerism, client dynamic participation, model diversity and privacy protection requirements is achieved, and the adaptability, stability and knowledge utilization efficiency of large model federation training are remarkably improved.

The invention belongs to the technical field of asynchronous federated learning, and discloses a data isomerism-oriented knowledge alignment asynchronous federated learning method. According to the method, a data quality perception aggregation strategy is introduced, and a knowledge distillation mechanism based on the old degree is combined, so that a global model is subjected to balanced training on heterogeneous data of different devices, and the generalization ability of the model is improved. Meanwhile, a self-adaptive learning rate adjustment mechanism based on aggregation frequency and weight is designed, and it is ensured that contribution of different devices to the global model is more fair. According to the method, the training deviation in asynchronous federated learning is effectively relieved, the accuracy and stability of a global model are improved, and the method has a considerable application value for a real federated environment.

The invention discloses a temperature self-adaptive adjustment control method and system for an automobile sensor, and relates to the technical field of automobile sensors, and the temperature self-adaptive adjustment control method for the automobile sensor comprises the following steps: S1, obtaining data; s2, performing classification and label addition on the data to form a data set; s3, extracting core temperature characteristic parameters, and constructing a four-dimensional temperature mapping matrix; s4, introducing an ant colony algorithm to iteratively optimize a temperature dynamic weighting rule, and obtaining an adjustment target and a change trend; s5, generating a regulation control strategy, and constructing a temperature fusion regulation feedback model at the same time; s6, performing iterative verification on the model, and generating a sensor three-dimensional verification report; and S7, performing security judgment on the three-dimensional verification report of the sensor. According to the method, through multi-source sensing data acquisition and segmented labeling preprocessing, temperature local mutation features and long-time time sequence dependence features are captured, and temperature data heterogeneity of traditional single-model feature extraction is broken.

The invention belongs to the technical field of Internet data security, and discloses a multi-dimensional Internet data security fusion processing system, which is characterized in that an acquisition module depends on a three-dimensional decision model, combines reinforcement learning dynamic allocation tasks, only acquires threat associated key data in a low-computing-power and high-threat scene, and starts full-amount lightweight acquisition in a high-computing-power and low-threat scene; the data preprocessing module adds scene labels for data, filters logic and unifies formats according to scene design, and eliminates data isomerism. The identification module builds a basic threat library and a scene adaptation layer, adapts a new scene through transfer learning, calculates a risk value in combination with a four-dimensional quantitative model such as propagation probability, generates an attack link map, and effectively eliminates a monitoring blind area caused by edge computing power limitation; the decision-making module constructs a five-dimensional trust evaluation model of cooperation duration, compliance records and the like, calculates weights by using an analytic hierarchy process, establishes a compliance and trust linkage engine, synchronizes industry specifications in real time and adapts to scene adjustment rules.

The invention discloses an industrial internet attack and defense situation and risk early warning and sensing method, and particularly relates to the field of internet risk early warning and sensing, which comprises the following steps of: acquiring multi-source heterogeneous data, constructing a basic data set covering an attack, service and equipment ternary space, including attack characteristics, service influence and equipment control vectors, and solving the problems of data heterogeneity and dispersion; constructing a triple function based on the data set, respectively quantifying the attack comprehensive threat degree, the influence degree of the attack on the service and the malicious control risk of the equipment, and retaining the characteristics of each dimension; a dynamic network topology model is constructed, nodes, edges and edge weights are defined, an attack propagation path and influence intensity are described, and node state dynamic updating is achieved; and finally, a risk prediction model is constructed by fusing quantitative indexes and topological information, a global risk value is calculated, graded early warning is realized through triple dimensions, a corresponding response mechanism is matched, and the timeliness and effectiveness of industrial internet security protection are improved.

The invention provides a construction method of an ocean observation and exploration large model, and belongs to the technical field of large models.Multi-mode original data are collected by constructing a multi-source ocean data collection matrix, an environment change degree vector is established, preprocessing and noise reduction are conducted on the original data by adopting a nonlinear matrix mapping algorithm based on a Gaussian kernel function, and the large model is constructed. An ocean observation and exploration large model architecture of a liquid neural network structure is constructed, different branches are made to process input data of different dimensions by means of asymmetric design, a super sparse reconstruction matrix is established, and high-dimensional original data are reconstructed from low-dimensional observation by means of a compressed sensing reconstruction mechanism. And finally, a supervised training process is executed to optimize model parameters so as to complete the construction of an ocean observation and exploration large model, and the technical problem that high-precision fusion modeling of ocean multi-modal observation data is difficult to realize under the conditions of spatial-temporal distribution sparsity and data isomerism is solved.

The invention discloses a personalized federated graph learning method oriented to equipment resource isomerism, and aims to solve the defects in graph data isomerism, equipment resource adaptation and privacy protection in the prior art. According to the method, collaborative optimization is realized through a closed-loop process of local pre-training, embedding aggregation, personalized training, soft label collaboration and classifier distillation. Each client pre-trains a model based on a local graph data set and generates interlayer embedding, and uploads the model to a server after sampling and privacy enhancement; the server performs aggregation to form a public embedded data set and distributes the public embedded data set to the client to support distillation training and soft label generation; the soft labels are filtered and subjected to weighted aggregation to form global soft labels, and the client completes classifier knowledge distillation by combining the public embedded data set and the global soft labels, and iteratively optimizes the performance of the model. According to the method, the generalization ability and the resource utilization efficiency of the model are remarkably improved while the data privacy is guaranteed, and the method is suitable for distributed graph data training tasks of multiple scenes such as social networks.

The invention relates to the technical field of international marine scientific investigation collaborative technology and mutual identification, and discloses an international marine observation data fusion and achievement mutual identification system suitable for joint scientific investigation. Comprising a multi-source data standardization access module, a polar region adaptive intelligent fusion module, an international achievement mutual recognition standard and verification module, a hierarchical sharing and mutual recognition process management module, a collaborative iteration and operation and maintenance management module, a block chain full-process traceability system and an international joint collaborative operation and maintenance mechanism. According to the system, the problem of international data heterogeneity is solved through multi-source data standardization preprocessing, accurate data fusion under polar region / open sea complex working conditions is realized by utilizing an intelligent fusion technology, a mutual recognition system of bidirectional verification is established, and the traceability of the whole process is ensured by adopting a block chain. The system supports hierarchical sharing and dynamic iterative optimization, is cooperatively operated and maintained by two international parties, guarantees long-term adaptation to standard updating and equipment iteration of the two parties, and effectively improves the joint scientific investigation data sharing efficiency and the result mutual recognition credibility.

The invention discloses a heterogeneous data migration method and device, equipment and a storage medium, and relates to the technical field of data migration, and the method comprises the steps: obtaining a heterogeneous mapping rule between a source database and a target database; creating a migration transition table in the target database according to the heterogeneous mapping rule, and synchronizing the to-be-migrated data in the source database to the migration transition table to generate isomorphic mirror image data; performing batch screening on the isomorphic mirror image data based on the target migration factor to generate a batch migration list; and performing format conversion on the batch migration list according to a heterogeneous mapping rule to obtain target format data, and writing the target format data into a target database. The data heterogeneous difference is shielded through the migration transition table, the format conversion process is postposed, and the precision and stability of format conversion are improved; and the risk of single migration is reduced through gradual batch migration, so that the risk of switching between a new system and an old system is reduced. The heterogeneous data migration complexity can be reduced, and the migration efficiency can be integrally improved.

The invention discloses a water conservancy construction progress prediction system based on multi-source data fusion, and relates to the technical field of artificial intelligence and engineering management, and the system comprises a multi-source sensing collection module which carries out the hierarchical adaptive sampling, quality evaluation and space-time alignment of multiple types of data of a construction site; the data comprises project progress, meteorological environment, geology and hydrology, energy consumption, water consumption and personnel state, and the sampling frequency is dynamically adjusted through an adaptive sampling function according to the prediction error. According to the invention, through hierarchical adaptive sampling and data quality evaluation of the multi-source sensing acquisition module, comprehensive, real-time and high-quality acquisition of construction site data is realized, the problems of data isomerism and uneven quality are effectively solved, the reliability of data input is improved, and the data input efficiency is improved. Through working condition semantic driving and physical consistency constraint fusion of the feature fusion modeling module, internal association among multi-source data can be deeply mined, and fusion features with better interpretation and robustness are generated.

The invention relates to the technical field of health information, in particular to a dietary inflammation tendency analysis system based on a nutrition database, which comprises a nutrition data acquisition module, a user health data acquisition module, an inflammation tendency analysis engine and a personalized diet suggestion generation module, nutrient components and user health indexes are mapped to a unified inflammation semantic space through a multi-dimensional feature embedding technology, a dietary inflammation tendency index fusing food components, metabolic states and inflammation markers is dynamically calculated based on a deep learning model, and personalized analysis is realized in combination with cloud platform data synchronization and machine learning optimization. According to the scheme, accurate quantification and dynamic evaluation of the individualized dietary inflammation risk can be realized, and the problem of insufficient analysis precision caused by data isomerism and static rules in a traditional method is solved.

Entropy based federated learning is disclosed. In federated learning, a model is trained at multiple clients using corresponding local data. An entropy associated with the local training is determined and provided, along with a model update, to a central server. The central server selects specific clients to participate in the current aggregation operation based on the entropy values. This minimizes the number of drifted and noisy clients that are included in the aggregation operation. The model updates of the selected clients are aggregated and a new or updated global model is generated. To aid in accounting for data heterogeneity, the clients may be grouped and a new or updated global model may be generated for each of the groups using model updates from corresponding selected clients.

The invention belongs to the technical field of smart agriculture and Internet of Things, discloses a smart agriculture-oriented heterogeneous Internet of Things data fusion and multi-model collaborative optimization method, and aims to solve the problems of high data isomerism, low fusion precision, single regulation and control strategy and insufficient adaptive ability of the traditional agricultural Internet of Things. The method comprises the steps of collecting multi-dimensional agricultural data through a distributed heterogeneous Internet of Things terminal, and generating a standardized feature data set through multi-level data fusion processing; constructing a three-stage model collaborative architecture of'spatio-temporal feature extraction-growth demand prediction-multi-objective optimization ', and dynamically adjusting model parameters in combination with a crop growth cycle; and realizing model adaptive iteration based on reinforcement learning, and outputting an optimization regulation strategy considering the crop yield, the resource utilization rate and the low-carbon target. Through heterogeneous data deep fusion and multi-model collaborative decision making, accurate matching of the agricultural environment and crop growth is realized, the method is suitable for various intelligent agricultural scenes such as greenhouses and field planting, and the intelligence and sustainability of agricultural production are remarkably improved.

The invention relates to an unmanned aerial vehicle track decision-making method based on turbulence prediction, and belongs to the technical field of unmanned aerial vehicle navigation and weather prediction. Aiming at the problems of difficulty in monitoring and predicting low-altitude turbulence and high air route planning risk, the defects of data heterogeneity, insufficient numerical weather forecast resolution and the like exist in the prior art; according to the method, a three-dimensional turbulence intensity field is generated through multi-source meteorological observation data fusion to serve as an observation benchmark, a diagnosis model is constructed in combination with numerical weather forecast data, linear calibration is carried out, or short-term prediction is generated by adopting an observation extrapolation model when forecast data is lacked; further, the turbulence field is mapped into a weighted graph, the height, the climbing rate and the airspace constraint are combined, the optimal track is solved by using a path search algorithm, and the accumulated turbulence cost is minimized; the method is clear in structure, full-process optimization from data fusion to decision making is achieved through multi-model complementation, and the method is suitable for the fields of low-altitude logistics, urban air travel and the like.

The invention relates to the technical field of data processing, and discloses a diagnosis abnormal data identification method based on machine learning, and the method comprises the steps: S1, collecting the multi-modal diagnosis data of a patient, including structured medical record data and unstructured free text description data; s2, carrying out data preprocessing and standardized conversion processing on the multi-modal doctor-seeing data of the patient to generate standardized doctor-seeing feature data; multi-modal medical data are integrated and standardized, a unified feature representation system is established, consistency and comparability of different types of medical data in anomaly detection are ensured, meanwhile, real-time anomaly detection is performed on doctor-seeing feature data by adopting an unsupervised machine learning algorithm, potential anomaly modes in the data are identified, and the anomaly detection accuracy is improved. The comprehensiveness and the accuracy of abnormal doctor-seeing data identification are improved, and detection errors caused by data isomerism are reduced.

The invention provides a time-domain-crossing non-intrusive load monitoring method fusing federated learning and Mama, and the method comprises the steps: dividing UK-DALE data into a time-domain-crossing client data set according to a time period, and carrying out the normalization preprocessing at a local side; the method comprises the following steps of: constructing a non-intrusive load monitoring model based on a Mama structure by utilizing Mama state space modeling and long sequence dependence capture capability; under a federated learning framework, each client executes local training and designs a diversity correction strategy to inhibit the influence of data heterogeneity on model convergence, and meanwhile, global model parameters are optimized through weighted aggregation. According to the method, the problems of privacy leakage risk, insufficient model generalization ability, overweight calculation burden and the like existing in traditional load monitoring are effectively solved.

The invention belongs to the technical field of wind power plant fire assessment, and discloses a multi-modal fusion wind power plant fire hazard multi-source data space-time synchronous assessment method and system, and the method is characterized in that a reference calibration module builds a digital twin simulation, double closed-loop dynamic calibration and credibility quantification three-in-one mechanism; the time synchronization adopts a triple strategy of GPS time service, local clock compensation and transmission delay prediction, and the transmission delay is compensated in advance in combination with an LSTM network; the space calibration depends on a three-dimensional digital twin model of the fan, and mounting deviation and vibration drift are corrected through visual identification and coordinate matching; the quality grading module is used for constructing a three-dimensional quality model, distributing weights according to data quality grading, and reducing evaluation deviation caused by data heterogeneity; the feature fusion module adopts a spatial-temporal feature, modal feature and quality weight cross fusion mechanism; and capturing data time sequence association through an overlapped time window, and mining spatial association in combination with a digital twin spatial topological relation.

This invention discloses a short-term power load forecasting method based on KNN federated distillation learning and Seq2Seq. The method involves acquiring historical power load data for a preset time period, inputting it into a trained local model on an edge computing device, and obtaining power load data for a future preset time period. This method trains local models from multiple edge computing devices together, employing federated learning and knowledge distillation during training. The local models use an encoder to extract spatiotemporal features and a decoder to convert these features into output for prediction. The KNN algorithm selects the parameters of the most similar local models from each edge computing device and aggregates them as the parameters of the teacher model. Knowledge distillation is achieved by transmitting knowledge between edge computing devices, rather than simply using a global model to update local models, thus reducing the impact of data heterogeneity. Experimental results show that the proposed S3TKFDL has better performance and robustness compared to the baseline model.

The invention belongs to the field of intelligent medical treatment. The invention provides a two-channel feature screening method for cognitive impairment recognition modeling and a modeling method thereof, and the method comprises the steps: taking multi-site cognitive impairment screening data as input data, and carrying out the preprocessing of the input data; constructing a traditional robust feature screening channel and an LLM knowledge enhancement screening channel, and performing feature screening on the preprocessed input data; integrating dual-channel feature screening results, performing bias perception joint score optimization, and obtaining a feature set; and performing semantic alignment and version normalization on the feature set to obtain a high-quality feature set. And applying the high-quality feature set to a classifier for cognitive impairment recognition modeling, and generating a clinical decision support result. According to the parallel feature screening method based on combination of a large language model and traditional feature screening, the problem of data heterogeneity bias in cognitive impairment recognition is systematically solved by constructing a dual-channel collaborative feature evaluation architecture, and organic unification of statistical robustness and clinical interpretability is achieved.

The invention discloses a model anti-poisoning attack method based on heterogeneous federal learning, a medium and equipment, and the method comprises the steps: carrying out the recoverability analysis of received client model updating data, synthesizing an agent data set, and carrying out the simulation training, and calculating the similarity between the original model update data and the simulation model update data as a recoverability score. Dividing a temporary benign candidate group and a potential malicious group based on recoverability score clustering, and respectively generating a benign frequency domain prototype and a malicious frequency domain prototype; for each model update data, the similarity between the frequency domain features thereof and the two prototypes is calculated to obtain a normal form score. And finally, synthesizing the recoverability score and the normal form score, screening out a final benign update data set from all updates, and carrying out aggregation to generate an updated global model. According to the method, interference caused by data isomerism is effectively overcome, strategic poisoning attacks can be accurately recognized and filtered, and the robustness and safety of a federated learning system are remarkably improved.

The invention relates to the technical field of federated learning, and provides a federated learning method, system and device, a program product and a storage medium. The method comprises the steps that a server trains a to-be-trained global model based on local data and a scaling factor of the server to obtain the global model; generating a plurality of different sub-models based on the variation direction and variation amplitude of the global model; and distributing a plurality of different sub-models to a plurality of clients, and returning to the step of training the to-be-trained global model based on the local data and the scaling factor to obtain the global model until a training ending condition is met, thereby obtaining a final global model. According to the method and the device, the problems that the communication overhead is increased, the accuracy and the convergence speed of the global model are reduced and the high-quality data of the server cannot be fully utilized can be solved while the influence of the data isomerism of the server side on the generalization ability of the global model is effectively relieved.

The application discloses a cross-bit federal collaborative modeling method under data and model dual heterogeneity. The application enables the edge terminal of the high-bit level model to connect the knowledge transmission between different bit models and reduce quantization error. For the bit heterogeneity scene, an edge local validation set assisted same level model assistance optimization strategy is designed, and the large and small edge terminals communicate with each other at the same level to cooperatively optimize and specifically reduce quantization loss. For the model drift problem under data heterogeneity, a model pseudo-update strategy is additionally designed on the edge terminal of the high-bit model, data and computing power are indirectly shared, deep fusion of similar edge terminals is realized, and the stability of model optimization is improved. In order to reduce the complexity of model screening, a Contact Map indicating potential candidate models is designed and stored in the cloud. The application solves the joint optimization problem of cross-bit models in the edge terminal data heterogeneity under the condition of protecting data privacy, and has superiority in model accuracy and training stability.

The application relates to a model parameter updating method and device, computer equipment and a storage medium. The method comprises the following steps: receiving a parameter updating instruction for a data distribution prediction model shared by each federal learning client sent by a federal learning server; wherein the parameter updating instruction carries shared parameters and specific parameters of the data distribution prediction model; in response to the parameter updating instruction, updating the shared parameters and the specific parameters to obtain updated shared parameters and updated specific parameters; sending the updated shared parameters to the federal learning server and retaining the updated specific parameters locally; receiving aggregated shared parameters sent by the federal learning server after the federal learning server aggregates and processes the shared parameters; and determining corresponding target shared parameters and target specific parameters based on the aggregated shared parameters and the updated specific parameters. The method can solve the data heterogeneity problem and improve the accuracy of the data distribution prediction model.

The application discloses a modular defect detection method based on data heterogeneous meta-learning, which solves the problems of few samples, cross tasks and data heterogeneity in industrial surface defect detection. The method adopts a modular architecture, including a reusable feature backbone, a switchable neck module including a classification neck and a positioning neck, and a task-specific head including a classification head, a detection head and a segmentation head, which has high flexibility. Through a two-stage learning strategy: first, learn general features in the ImageNet-1k pre-training backbone, and then optimize shared parameters through meta-learning on multiple heterogeneous defect datasets, so that the model quickly adapts to different defect tasks. Compared with the prior art, the method significantly improves the detection accuracy and robustness under the condition of few samples, reduces the dependence on large-scale labeled data, and can efficiently process various surface defect detection tasks.

The application relates to a non-IID federated learning model backdoor robustness evaluation method based on a generative adversarial network, which comprises the following steps: selecting a client in federated learning as a test node, taking a downloaded server-side global model as a discriminator, and designing a generator locally to form a generative adversarial network model; a backdoor attack target category is specified, a class representative sample is reconstructed by using the generator in each round of global training, and the target category is marked as pre-poisoning data participating in training; a supplementary data set is generated locally; a source category of the backdoor attack is specified, a backdoor trigger is optimized, the supplementary data set is used for class-specific backdoor training, the discriminator is updated and uploaded to the server, and the global model is updated; the non-IID degree of data and the number of specified source categories are adjusted, and the robustness of the federated learning global model to the backdoor attack is observed. Compared with the prior art, the application can verify the effect of the backdoor attack on the federated learning model under different degrees of data heterogeneity.

The invention belongs to the field of aircraft test data processing, and particularly relates to a multi-aircraft-type comprehensive data processing system which comprises a hardware layer, a data middle layer, a data conversion layer and a control layer. The data intermediate layer sends the received data to the data conversion layer according to a specific time interval, the data conversion layer determines the position, time information and model of a corresponding function node according to associated information, and data acquisition information belonging to the same time node is sent to the control layer; and the control layer processes the data of different models respectively to generate control instructions. According to the method, hardware of different models and at different positions is decoupled, unified data collection is achieved, then different data are processed according to different time and models in a message queue mode, the problems of data isomerism, synchronism and correlation analysis in a multi-model comprehensive test are solved, and the test efficiency is improved. And a universal data processing framework is provided for manned-unmanned cooperation and new energy multi-model tests in the future.

This invention discloses a method and system for dynamic evaluation of investment benefits in distribution network infrastructure projects based on multi-source data fusion, belonging to the field of distribution network evaluation technology. This invention eliminates data heterogeneity and quality issues through multi-source data preprocessing, providing a reliable data foundation for evaluation; it focuses on core influencing factors through key feature screening, reducing redundant information interference; and it accurately quantifies the nonlinear relationship and causal path between key features and comprehensive benefits using grey relational analysis and Bayesian network models, improving the scientific rigor and adaptability of the evaluation model. Simultaneously, it verifies the rationality of the evaluation logic by combining additive explanatory value analysis, enabling decision-makers to clearly understand the driving factors of investment benefits. This effectively solves the defects in dynamic evaluation of investment benefits in distribution network infrastructure projects caused by weak data foundation, ambiguity of core factors, and inaccurate characterization of correlations, leading to insufficient evaluation accuracy. It achieves a leap from static evaluation to dynamic decision support, significantly improving the accuracy of evaluation results.

The present application relates to a kind of importance sampling-based heterogeneous federal model adjustment method, belong to the field of computing model, method includes: S1, by similar perception layer analysis, the importance of layer is quantitatively analyzed, the importance score is obtained by analyzing each layer of central server, so as to obtain the importance distribution for layer;S2, probability sampling is carried out based on the hierarchical importance distribution obtained in S1, generate sub-model extraction binary mask suitable for heterogeneous client, to improve the matching degree of model structure and resource condition;S3, each client extracts sub-model by binary mask and completes local deployment;S4, introduce TSDL and SDL to carry out local training, and upload and aggregate lora parameter;S5, repeat the steps of S2-S4 until convergence.The present application introduces TSDL, which can alleviate the problem of uneven aggregation caused by local aggregation;Introducing SDL can alleviate the problem of data heterogeneity.