68 results about "Layered model" patented technology

The invention discloses a humanoid robot adaptive anti-interference recovery method based on reinforcement learning enhanced hierarchical model prediction control, and the method comprises the steps: receiving a user instruction, a robot state and a model prediction control parameter through a reinforcement learning decision module, and outputting a first-layer MPC parameter and a second-layer MPC parameter which are subjected to strategy enhancement after training; the first-layer MPC is combined with dynamic constraints, the multi-step foothold sequence is optimized and updated through the first-layer MPC parameters with the enhanced output strategy and event triggering signals, the second-layer MPC is combined with the first-layer model to predict and control the optimized robot state, the updated multi-step foothold sequence and the second-layer MPC parameters with the enhanced output strategy are optimized, and the multi-step foothold sequence is optimized. The invention relates to a contact wrench for optimizing a multi-step foothold sequence. According to the method, the walking disturbance recovery agility and the self-adaptive balance capability of the humanoid robot in complex terrains and external disturbance can be improved.

The invention relates to an energy storage electric power spot market simulation clearing method based on hierarchical model control prediction, and the method comprises the following steps: S1, building an electric power spot market global simplified clearing model containing energy storage, and taking the model as an upper prediction control model; s2, establishing an electric power spot market detailed clearing model containing energy storage as a lower execution model; s3, dividing long-time simulation into M solving periods; in each solving period, global simplification clearing optimization is firstly utilized to obtain an energy storage final state SOC constraint value of each remaining period; on the basis of the energy storage final state SOC constraint value of the current period, market simulation clearing of the current period is completed by using a detailed clearing model; and finally, updating the system state according to the current clearing result, and rolling to advance to the next cycle until clearing of all cycles is completed. According to the method, dynamic global optimization of the energy storage SOC constraint is realized through a hierarchical model prediction control mechanism, and the calculation efficiency and the market simulation clearing effect are effectively balanced.

The invention discloses a transformer design method and system based on a hierarchical genetic algorithm and a medium, and belongs to the field of transformer design. The method comprises the steps that a structural hierarchical model of an amorphous alloy three-dimensional roll iron core transformer is constructed based on transformer parameters; inputting the structure layering model into a preset multi-physics field coupling simulation model, calculating iron core no-load loss power and winding load loss power, and determining a simulation temperature field in combination with the amorphous alloy iron core thermal conductivity and a dynamically calibrated boundary convective heat transfer coefficient; according to the sequence of the three-dimensional roll iron core layer, the winding layer and the oil duct layer, hierarchical genetic algorithm optimization is executed to obtain an optimization scheme, iterative simulation is conducted on the optimization scheme until an obtained simulation result meets a first preset condition, and the amorphous alloy three-dimensional roll iron core transformer is constructed based on the determined target structure parameters. And the design rationality and the operation safety of the amorphous alloy three-dimensional roll iron core transformer can be improved.

The invention discloses an image recognition method based on end, edge and cloud layered federated learning of knowledge distillation, which is characterized in that original data or full-quantity model parameters are replaced by transmitting embedded vectors, a synthetic sample is combined to serve as a knowledge carrier, the communication traffic is reduced by more than 60% compared with the traditional federated learning, and the image recognition method is adaptive to mass equipment concurrent scenes. In the training process, original data are only locally converted into vectors and never leave end equipment, so that the risk of data leakage is avoided from the source, and the requirements of privacy protection laws and regulations are met. By introducing a knowledge distillation method and hierarchical model deployment (end side light weight and cloud side high performance), equipment with different computing power can participate in cooperative training, and the problem of computing power resource mismatching is solved. Meanwhile, a Logs-based sample screening mechanism and multi-dimensional structured loss (and) are innovatively combined, and compared with existing schemes such as FedAvg and FedAgg, the method has the advantages that the test accuracy is remarkably improved, and particularly, the advantages are more outstanding in a high data heterogeneous scene.

The application relates to an Internet of Things abnormal data detection method based on improved feature selection and a hierarchical model, mainly comprising the following steps: collecting Internet of Things data, performing data missing value, abnormal value and Z-score standardization data preprocessing; removing redundant features from the preprocessed data by using an improved correlation feature selection algorithm, screening important features, and performing data dimension reduction; dividing the feature-selected data into a training set and a test set; further dividing the training set into a sub-training set and two validation sets; inputting the sub-training set into a multilayer hierarchical model for training, saving the trained model; and obtaining a final training model. Finally, the test set is used to test the effect of the overall model, and a detection result is obtained. The application effectively selects features of Internet of Things data, reasonably divides data and fuses hierarchical models, and the detection effect of the Internet of Things abnormal data is obviously improved.

The application provides a logging data layering method based on a T-KAN structure, relates to the geological field, and comprises the following steps: obtaining logging curve data of multiple wells and performing pretreatment; performing windowing on the logging curve data and constructing corresponding labels to obtain training samples; constructing a layering model based on a Transformer encoder; training the layering model by using supervised learning and in combination with the training samples; inputting the pretreated and windowed logging curve data of a new well into the trained layering model to output a point-by-point prediction sequence; performing smoothing filtering and merging layer boundary candidate points on the point-by-point prediction sequence; alternately determining layer tops and layer bottoms in a depth sequence to divide continuous stratum intervals and output top boundary depths and bottom boundary depths of the intervals. The application realizes high automation and intelligence of a logging layering process and greatly reduces the dependence on artificial experience.

Techniques for adaptive multi-level recommendation include processing input features to generate pre-processed input features, processing the input features and the pre-processed input features using a plurality of expert models to generate expert outputs, mixing the input features and the expert outputs to generate mixed expert outputs, processing the mixed expert outputs using a first model of the hierarchical model to generate intermediate outputs, and processing the mixed expert outputs and the intermediate outputs using a second model of the hierarchical model to generate a final output.

The invention relates to the technical field of soil improvement, and discloses a soil improvement effect real-time monitoring method and system and a storage medium. The method comprises the following steps: collecting multiple depth data samples in the vertical direction of soil, and constructing a layered soil characteristic data set; identifying the position of the high-heterogeneous section based on the data set; injecting a tracer agent at the position and imaging in real time to obtain a permeation path image sequence; predicting a diffusion velocity and a retardation point position by using a random forest model, dynamically adjusting injection parameters to eliminate retardation, and obtaining permeation path data; tracking and quantifying the action change of the modifier in each soil layer through time sequence analysis to obtain an action effect quantitative index; the index is compared with an initial data set, a hierarchical model is optimized, and a hierarchical monitoring framework is constructed; and integrating multiple depth data samples to verify the framework, and generating an evaluation report of the soil improvement effect. According to the invention, the accurate characterization of the vertical layering characteristic of the soil and the scientific and quantitative evaluation of the improvement effect are realized, and the use efficiency of the modifier is improved.

The application discloses a CTC system localization hardware performance monitoring data visualization implementation method, solves the problems of poor adaptability of existing general monitoring schemes to localization hardware and railway business scenes, disconnection of data and business, weak visualization analysis and difficult special network deployment. The method ensures the improvement of data accuracy through automatic matching and multi-dimensional verification of hardware types, differentiated packet loss completion and business correlation abnormal value filtering; constructs a device-line-business three-dimensional layered model, combines railway operation diagrams for differentiated down-sampling and weighted statistics, and automatically associates business event labels to realize deep binding of performance data and business logic; through multi-index linkage visualization, pixel density detection intelligent labeling and multi-dimensional interaction filtering, the performance bottleneck positioning efficiency is improved; through three-level dependency list management, resource path mapping and static link encapsulation, an executable file that can completely run offline is generated, and the stable operation rate of the localization terminal is ensured.

The application discloses a kind of highway survey radar measurement systems based on electromagnetic wave, through quantum magnetic force gradient tensor analysis and space coordinate calibration, synthetic aperture radar echo feature extraction module cooperation, using quantum magnetic force navigation model and synthetic aperture radar algorithm, realize multi-source data space-time registration and weighted fusion, construct highway geology layered model and inversion geological parameter, after grid processing of fusion data, generate three-dimensional visual model.The system overcomes the defects of traditional technology multi-source data fusion deficiency and low precision of geological parameter inversion, improves the precision, efficiency and reliability of highway survey, provides accurate geological information and visual survey results for highway engineering design and construction, has important application value in the field of highway survey.

An Artificial Intelligence (AI) based computational system is used to non-invasively estimate the presence of a range of aneuploidies and mosaicism in an image of embryo prior to implantation. Aneuploidies and mosaicism with similar risks of adverse outcomes are grouped and training images are labelled with their group. Separate AI models are trained for each group using the same training dataset and the separate models are then combined, such as by using an Ensemble or Distillation approach to develop a model that can identify a wide range of aneuploidy and mosaicism risks. The AI model for a group is generated by training multiple models including binary models, hierarchical layered models and a multi-class model. In particular the hierarchical layered models are generated by assigning quality labels to images. At each layer the training set is partitioned in the best quality images and other images. The model at that layer is trained on the best quality images, and the other images are passed down to the next layer and the process repeated (so the remaining images are separated into next best quality images and other images). The final model can then be used to non-invasively identify aneuploidy and mosaicism and associated risk of adverse outcomes from an image of an embryo prior to implantation.

The present application relates to the technical field of electromagnetic sounding, in particular to a multi-step variation AB-Hy encoding electromagnetic sounding one-dimensional inversion method and device, for the measured AB-Hy encoding electromagnetic sounding full area apparent resistivity observation value, the method presets the feasible solution space of one-dimensional horizontal layered model, randomly generates a population, forward calculation of the full area apparent resistivity of each individual in the population, calculates the fitness value of each individual in the population according to the full area apparent resistivity calculation value and observation value of each individual;Select the individual with the highest current fitness value as the optimal individual, and perform iterative calculation of multi-step differential evolution with the optimal individual participating in the population individual;The individual with the highest fitness value obtained is taken as the inversion result of the observation data. AB-Hy electromagnetic sounding one-dimensional inversion based on multi-step differential evolution algorithm is realized, the geoelectric model parameters corresponding to the observation data are obtained, and the basis for geological interpretation is provided. The method has fast optimization convergence speed and high accuracy of inversion result.

This invention provides a method, apparatus, and electronic device for modeling maritime rescue processes. The method includes: constructing a hierarchical model architecture driven by discrete event system specifications for the maritime rescue process, dividing the maritime rescue system model into a discrete event system layer and an object model layer; constructing a generalized rescue platform model and modular sub-models; the modular sub-models include a controller model, a maneuvering model, and a sensor model; constructing a complete rescue system simulation architecture based on the generalized rescue platform model; and performing repeated simulations and performance analysis on the complete rescue system simulation architecture to obtain a performance analysis report. This invention achieves accurate representation of the rescue system and solves the problems of poor accuracy and completeness in existing related technologies for rescue modeling.

The invention belongs to the technical field of saturated sandy soil dynamic analysis, and particularly relates to a saturated sandy soil pore pressure evaluation method based on weak permeable interlayer permeability. Establishing a layered model composed of saturated sand and the weak permeable interlayer; selecting a dynamic constitutive model suitable for cyclic loading conditions; applying a self-weight load to the dynamic constitutive model to generate an initial stress field of the dynamic constitutive model; based on a finite difference theory, constructing a dynamic fluid-solid coupling analysis model of a saturated sand dynamic response control equation and a pore water seepage control equation; simulating pore water pressure response of the saturated sandy soil under different power working conditions; and quantitatively evaluating the influence of the permeability of the weakly permeable interlayer on the pore pressure accumulation, dissipation and liquefaction response of the saturated sand. According to the method, the dynamic numerical model considering the interlayer permeation difference is established, and the influence of the weakly permeable interlayer on the pore water pressure generation, diffusion and dissipation process is systematically analyzed, so that the pore pressure response of the saturated sand is accurately evaluated.

The application relates to a kind of robot end effector multiscale clamping force control device and method based on DIC, wherein the method comprises the following steps: S1, obtaining multiscale signal, collecting multiscale speckle image: S2, processing multiscale data and extracting features, image is processed by power platform multi-thread;S3, mapping multiscale physical quantity and constructing model, power platform is associated with multiscale feature and parameter and builds layered model;S4, control multiscale closed loop and adjust real-time, power platform uses layered adaptive algorithm to adjust PWM parameter.The application can consider non-contact sensing, multiscale modeling and real-time control, and break the force control bottleneck of robot soft body end effector.

The invention discloses a power distribution network planning standard file knowledge graph construction method and system based on an ALBERT-Transformer-Attention-CRF model, and the method comprises the steps: constructing a three-layer structure knowledge unit model for a standard file, defining the composition, entity, relation and attribute of each layer of knowledge unit, and giving an instance indication of the knowledge unit; the method comprises the following steps: firstly, dividing knowledge units, extracting knowledge of a standard core element layer, extracting data of a standard file layer by adopting a rule-based method, finally designing an ALBERT-hierarchical Transform-hierarchical Attention-CRF model to realize an entity recognition task of a standard article, and extracting an article relationship; and performing knowledge storage and application, and storing the constructed knowledge graph. According to the method, efficient management and effective utilization of power distribution network planning standard file knowledge are realized, and a new thought is provided for power distribution network planning standardized intelligent construction.

The invention discloses an event-driven data hierarchical model statistical method, which relates to the technical field of data processing, and adopts a self-adaptive processing mechanism capable of sensing context characteristics of a data stream, and a system performs data hierarchical model statistical analysis according to real-time throughput and data attributes. According to the method, an event-driven mode, a timed task scheduling mode and a micro-batch processing mode are flexibly and automatically switched, so that the high throughput demand of historical data backtracking is effectively considered while low-delay real-time calculation is ensured, and the optimal balance between the resource utilization rate and the calculation efficiency is realized. And the user retention and active indexes are processed by introducing a bitmap-based combined storage structure and displacement arithmetic logic, so that the storage space occupation is greatly compressed, the complex set operation is converted into efficient binary bit operation, and the query performance of cross-cycle statistics is remarkably improved. In addition, a multi-level cache mechanism is utilized to optimize a streaming association process, and the access pressure on an external dimension database is effectively reduced.

The invention belongs to the technical field of artificial intelligence, and particularly relates to an artificial intelligence model service method based on edge computing, which comprises the following steps: S1, edge-cloud collaborative task allocation; s2, hierarchical model deployment and parallel acceleration; s3, adaptive processing of multi-modal feature driving is carried out; s4, dynamic resource scheduling and priority control; s5, credibility evaluation and result fusion; s6, hardware-level acceleration and energy efficiency optimization are carried out; according to the method, edge-cloud cooperative task allocation is adopted, a lightweight model is executed at an edge end according to data features, a complex model is processed at the cloud end, a large amount of data is prevented from being transmitted to the cloud end, meanwhile, communication between edge nodes adopts a lightweight protocol, the overhead is reduced, data layer transmission is processed through compressed sensing, and 5GURLLC slice rapid transmission is combined, so that the data transmission efficiency is improved. Therefore, the network bandwidth pressure is reduced, and the transmission delay is reduced.

The invention discloses a firepower attack target distribution method and system based on a multi-constraint dynamic pointer network, and the method comprises the steps: firstly building a hierarchical model which comprises a platform layer, a target layer and a rule layer, carrying out the modeling of a data relation in collected data through employing a hyperedge based on the hierarchical model, and forming a time-varying hypergraph which is updated along with time steps; according to the method, the functions of integrally representing the capacity, the time window, the collaborative grouping, the area limitation and the fair coverage constraint in the same graph space by adopting the time-varying hypergraph and the advancement rule and timely distributing are realized; a primitive-dual online adjustment mechanism is introduced in the aspect of multi-target tradeoff to automatically allocate constraint budget for a Lagrange multiplier according to real-time violation degree, and load variance fairness and regularization are superposed, so that interpretable dynamic balance can be obtained for resource utilization, time window satisfaction, collaborative consistency and coverage balance, a security boundary can be provided, and the real-time performance of the multi-target tradeoff can be improved. And secondary selection of upper-layer strategies or players in the set can be facilitated, and the method is suitable for being widely popularized and used.

The invention discloses a photovoltaic power generation prediction method based on the combination of regional features and weather features, and relates to the technical field of photovoltaic power generation prediction, and the method comprises the following steps: constructing a regional feature hierarchical model which comprises a regional feature model, a seasonal feature model and a weather feature model, respectively quantifying the influence of regional parameters, seasonal changes and subdivided weather types on photovoltaic output; according to the method, the integrity, the accuracy and the consistency of the data are improved through hierarchical acquisition and preprocessing of the regional data, and a high-quality data basis is provided for model construction; according to the regional characteristic hierarchical modeling, influence factors are deeply analyzed from three dimensions of region, season and weather, and the accuracy and reliability of photovoltaic output prediction under different geographical conditions, season changes and weather conditions are remarkably improved; typical weather scene setting and simulation ensure that the model keeps high prediction precision under complex and changeable weather conditions.

The invention provides a logging data layering method based on a T-KAN structure, and relates to the field of geology, and the method comprises the steps: obtaining logging curve data of a plurality of wells, and carrying out the preprocessing; windowing the logging curve data and constructing a corresponding label to obtain a training sample; a hierarchical model is constructed on the basis of a Transform encoder; training the hierarchical model by adopting supervised learning in combination with a training sample; preprocessing and windowing logging curve data of a new well, inputting the logging curve data into the trained layered model, and outputting a point-by-point prediction sequence; carrying out smooth filtering on the point-by-point prediction sequence and merging layer boundary candidate points; and alternately determining the layer top and the layer bottom according to a depth sequence, dividing continuous stratum sections, and outputting the top boundary depth and the bottom boundary depth of each stratum section. According to the invention, high automation and intelligentization of the logging layering process are realized, and the dependence on artificial experience is greatly reduced.

The invention provides a time-space layered arrangement method and system for multi-user interactive light show, and a storage medium. The method comprises the following steps: acquiring user customized data submitted after a plurality of user terminals enter a system information input page by scanning a two-dimensional code, wherein the user customized data comprises a selected light pattern mode and uploaded material information; calculating a scheduling weight for each piece of user customized data in real time, and mapping the user customized data into a space-time occupation block with a preset duration and a space size according to the scheduling weight; respectively distributing and arranging the plurality of space-time occupation blocks to a main show layer, a group show layer or a background layer of a space-time layering model by utilizing a scheduling optimization algorithm to obtain an arranged space-time layering model; and performing multi-layer fusion rendering on the arranged space-time layering model, generating light show customized content, and sending the light show customized content to a projector to be projected on a light show medium. It can be ensured that the light show picture is clear in primary and secondary, clear and distinguishable.

The invention discloses a transient electromagnetic wave logging response calculation method and device, electronic equipment and a storage medium. The method comprises the following steps: determining a target function from at least two reference functions associated with a target cased well radial stratification model, wherein the at least two reference functions comprise a first function and a second function; the signal receiving time of a receiving coil in the target cased well is determined, and a three-component transmitting coil and a three-component receiving coil are placed in a borehole of the target cased well; substituting the time of receiving the signal into the target function to obtain a target result, wherein the target result is a solving result used for indicating a derivative term of time domain magnetic field intensity to time in a frequency-time transformation equation of the target function; and determining the induced electromotive force of the receiving coil position at the borehole middle position of the target cased well according to the target result. According to the technical scheme, high-precision and efficient calculation of transient electromagnetic wave logging response can be achieved.

The application discloses a kind of financial risk control evaluation method and system based on multi-source data fusion, comprising: access internal, third party and low-credibility supplementary data and clean up;With matching rate, completeness, freshness, fluctuation coefficient quantification accuracy, integrity, timeliness, stability, combined with type coefficient λ, correlation degree μ, sensitivity σ to calculate the score of credibility with subjective and objective coupling algorithm, according to layer distribution fusion weight and extract low-quality data risk signal;Support visual self-defined index and hierarchical model;Splice feature calculation score and map risk level, realize pre-event / middle / post-event dynamic update;Matching credit policy, when the credibility change ≥5 points or high priority risk, dynamically adjust weight.The application includes data governance, index modeling, full-cycle scoring, intelligent decision-making, traceability audit module, to improve evaluation accuracy and compliance.

The invention provides an automatic flow identification and conversion method based on an unstructured document, and relates to the technical field of enterprise flow digital transformation, and the method comprises the steps: collecting the unstructured document, carrying out the automatic identification, and generating an image document and a text document; based on a hierarchical AI model strategy, extracting image features of the image document by adopting a convolutional neural network model, and extracting text features of the text document by adopting a natural language processing model; the image features and the text features are mapped and fused into element semantic features through a multi-modal information fusion algorithm, verification and complementation are carried out in combination with an element recognition method, and fusion feature data are generated; and on the basis of the conversion rule, converting the fused feature data into the structured process data, and executing dual verification and error repair to generate the standardized flow chart, so that efficient, accurate and automatic conversion from the unstructured document to the standardized flow chart can be realized, the enterprise process digitization efficiency is remarkably improved, and the labor cost is reduced.

The invention relates to an electric power facility disaster damage monitoring method and an identification device, and belongs to the technical field of electric power facility disaster damage monitoring. The method comprises the steps of constructing a multi-dimensional feature portrait of a disaster damage identification device by collecting multi-source data, and dynamically scheduling and matching a pre-layering model for concurrent tasks based on the multi-dimensional feature portrait; performing decoupling analysis on the multi-dimensional feature portrait through the pre-layering model, and completing classification and positioning in parallel; data uploaded by all disaster damage identification devices are aggregated to train and generate a new model, a multi-dimensional feature weight is set based on a release target to calculate a comprehensive score, and a gray release target is screened. According to the method, dynamic scheduling of model resources is realized through the multi-dimensional feature portraits, and the utilization efficiency and response real-time performance of edge-side resources are improved; classification and positioning tasks are processed in parallel by adopting a decoupling architecture, so that the recognition precision is improved; a closed-loop self-evolution system is constructed by means of an intelligent gray release mechanism based on cloud edge cooperation, and the monitoring accuracy, predictability and continuous optimization capability are comprehensively improved.

The invention relates to the technical field of graphic rendering, and discloses a game frame rate optimization method and system, electronic equipment and a storage medium, and the method comprises the steps: collecting a rendered graphic buffer area and associated metadata, and obtaining a game frame rate based on the collected graphic buffer area and metadata; predicting image contents of subsequent frames by using a machine learning model library comprising a plurality of models with different complexities, and dynamically selecting an adaptive model according to frame loss duration and confidence scores output by the models; when frame loss is detected, generating a compensation frame according to the predicted image content; and displaying and outputting the compensation frame. Through multi-dimensional metadata collection, hierarchical model matrix design, a dynamic model selection strategy and game feature recognition and special model adaptation, the frame loss problem in game operation is solved, game fluency and user experience are improved, and meanwhile good cross-platform compatibility and system stability are achieved.

The invention discloses a method for determining a pull-torsion coupling effect of a semi-parallel steel wire rope. The method comprises the following steps: step 1, establishing a layered model of the semi-parallel steel wire rope; 2, defining the overall strain of the semi-parallel steel wire rope; step 3, establishing a deformation coordination equation of the semi-parallel steel wire rope and the internal steel wires thereof; 4, solving the force and moment of force of the spiral steel wire; 5, establishing an overall mechanical equation of the semi-parallel steel wire rope; and step 6, solving the tension-torsion coupling response of the cable under the specific boundary condition. The analytic model provided by the invention is clear in physical significance, can quickly and accurately predict the tension-torsion coupling behavior of the semi-parallel steel wire rope in the elastic deformation stage, and provides a reliable analysis tool for design, construction safety and service performance evaluation of structures such as bridge inhaul cables and suspenders.

The invention discloses a non-decoupling rapid calculation method for a seismic electromagnetic field caused by a piezomagnetic effect, and belongs to the technical field of geophysics, and the method comprises the steps: S1, converting a frequency space domain piezomagnetic effect three-dimensional seismoelectric coupling equation into a frequency wavenumber domain through introducing a surface harmonic coordinate system, and carrying out the joint solving of an elastic wave kinetic equation and an electromagnetic field Maxwell equation; s2, solving the three groups of coupled equations in a frequency wavenumber domain to obtain corresponding earthquake and electromagnetic wave fields; and S3, transforming the wave field from a frequency wavenumber domain to a time-space domain through a two-time independent domain transformation method. According to the calculation method, it is proposed that the piezomagnetic effect of ferromagnetic minerals in the earth crust can cause the electromagnetic phenomenon in the earthquake fracture process, a complete three-dimensional piezomagnetic effect seismoelectric coupling equation is given, displacement, stress, an electric field and a magnetic field in a layered model can be obtained at the same time through combined solving, and finally, when conversion is carried out back to a frequency-space domain, the electromagnetic phenomenon can be generated in the earthquake fracture process. Wave number integration only needs to be carried out once.

The invention relates to a federal learning personalized collaboration method based on a hierarchical model and privacy calculation, and belongs to a federal learning technology. According to the method, a local model is split into a presentation layer and a classification head for federal aggregation through a hierarchical cooperation method based on representation learning; the presentation layer deals with the problem of data heterogeneity through personalized collaborative learning feature representation capability; the classification head reviews old task knowledge through personalized collaboration to relieve the problem of disastrous forgetting; secondly, task distribution similarity is defined through JS divergence, and the server can calculate the task similarity between clients through a secure multi-party calculation method under the condition that real data distribution is not exposed; based on task similarity, collaborative aggregation weights are respectively optimized for a presentation layer and a classification head of each client to effectively transfer universal knowledge and personalized knowledge. According to the invention, while data privacy is protected, the personalized performance of different clients is improved, and the problem of two-dimensional data heterogeneity is avoided.