53results about How to "Improve training stability" patented technology

The invention discloses a dynamic carpooling scheduling method for unmanned electric taxis. The method comprises the following steps: firstly, acquiring data and cleaning the data; mapping the data in each hexagonal grid; establishing a travel demand feature matrix and a spatial dependency matrix; constructing a space-time diagram convolutional neural network based on an attention mechanism, and obtaining a travel demand prediction result; setting constraint conditions matched with the car orders in the dynamic car-pooling scene; based on the travel demand prediction result and the vehicle order matching constraint condition, constructing a vehicle order matching decision model based on distributed deep reinforcement learning; and designing a Double DQN solving algorithm, carrying out real-time order matching and idle vehicle scheduling based on a result, notifying the matched vehicle to complete a corresponding order, and notifying the unmatched idle vehicle to carry out regional scheduling or charging energy supplementation. According to the method, in dynamic carpooling of the multi-passenger bicycle, the endurance process constraint of the electric automobile and the charging energy complementation requirement of the bicycle are considered, the actual operation requirement is better met, and the quality is improved.

The invention belongs to the technical field of medical data processing, and discloses a temporal-mandibular joint disease automatic diagnosis method and system based on artificial intelligence, and the method comprises the steps: collecting and preprocessing data; expanding the patient data, dynamically adjusting the structure and parameters of a data generator, and generating the patient data; performing feature extraction on the patient data, and automatically adjusting an evolution rule of a feature extraction network to obtain patient data features; performing dimensionality reduction on the data features to obtain low-dimensional patient data features; inputting the low-dimensional patient data features into a classifier, and dynamically adjusting the configuration and weight of hidden layer nodes of the classifier according to the low-dimensional patient data features to obtain a classification result; and defining a loss function, and training the model. Parameters can be automatically adjusted in the data feature extraction process, the data dimension reduction process is optimized by adopting quantum state mapping and an adaptive adjustment mechanism, the classifier can dynamically adjust the number of hidden layer nodes and an activation function, and the accuracy and robustness of a diagnosis result are improved.

The invention discloses a weak supervision defogging method based on uncertainty driving, and the method employs a weak supervision defogging network of a U-Net architecture, and the weak supervision defogging network comprises a regional dynamic attention module RDAM, a lower sampling layer, a multi-scale feature fusion module MSFFM, and an upper sampling layer. The RDAM is used for processing an input image or the output of a previous layer of RDAM, constructing a dynamic convolutional layer and obtaining a regional dynamic attention feature map through the dynamic convolutional layer; the down-sampling layer is used for down-sampling the output of the RDAM and reducing the spatial resolution; the MSFFM is used for processing input features through a window multi-head self-attention mechanism so as to capture a long-distance dependency relationship between pixels and performing multi-scale feature extraction and fusion; the upper sampling layer is used for recovering the feature map output by the MSFFM to the resolution ratio of the feature map output by the MSFFM of the upper layer and performing feature splicing with the feature map output by the MSFFM of the upper layer; and enabling the up-sampled feature map to sequentially pass through a 3 * 3 convolutional layer, a ReLU activation function and a 1 * 1 convolutional layer, and finally outputting a defogged image.

The invention discloses a quadruped robot gait generation method based on an improved SAC algorithm. The quadruped robot gait generation method comprises the steps that current state data and motion data of a quadruped robot are obtained; obtaining new state data based on the action data; setting a multi-dimensional reward function, and forming tuple data with the current state data, the action data and the new state data; inputting the tuple data into an intelligent agent formed by an actor network and a multi-reviewer network to obtain a dynamic weight and an action value; obtaining a mixed strategy gradient based on the dynamic weight and the action value; based on the mixed strategy gradient and the loss function, training the actor network and the multi-head reviewer network to obtain a trained actor network and a trained multi-head reviewer network; and obtaining current state data of the quadruped robot, and inputting the current state data into the trained actor network and the trained multi-head commentator network to obtain optimized gait data. And the gait generation efficiency and the gait robustness of the quadruped robot are improved.

The invention relates to a boiler flue gas temperature data correction method and system based on hydrodynamic analysis, and belongs to the technical field of boiler flue gas temperature correction. The method comprises the following steps: acquiring historical operation parameters of a boiler, modeling through hydrodynamic analysis to obtain a flue gas temperature field turbulence model, and solving the flue gas temperature field turbulence model through CFD software to obtain a flue gas temperature simulation value; according to the flue gas temperature simulation value and actually measured flue gas temperature data under the corresponding simulation condition, a boiler flue gas temperature data set is constructed through data preprocessing; according to the boiler flue gas temperature data set, a boiler flue gas temperature correction model is obtained through improved DNN network training and testing; boiler flue gas temperature data are detected in real time, and the boiler flue gas correction temperature is obtained through the boiler flue gas temperature correction model, so that boiler flue gas temperature measurement errors caused by factors such as cold surface radiation are reduced, rapid and accurate boiler flue gas correction temperature prediction is achieved, and combustion process control in boiler operation is optimized.

The invention provides a comparative learning training method and system for recalling military documents according to user questions. The method comprises the following steps: preprocessing document data, extracting document keywords, and generating document features; extracting query keywords from the input question to generate query features; based on the input problem, constructing a positive sample pair and a negative sample pair, the positive sample pair comprising the input problem and a corresponding positive sample, and the negative sample pair comprising the input problem and a corresponding negative sample; respectively processing the query features and the document features by using a double-tower model to generate query feature vectors and document feature vectors; calculating the similarity between the query feature vector and the document feature vector, and distributing similarity labels for the positive samples and the negative samples; carrying out soft mining and hard mining processing on the negative sample; the batch size of contrast learning is increased by using a momentum contrast mechanism and a negative sample queue. The processing efficiency and the recall accuracy can be improved.

The application claims a small sample bearing fault diagnosis method based on Mamba and double granularity cross-domain alignment, which is used to solve the problem of low diagnosis accuracy caused by large distribution difference of bearing monitoring data across working conditions and scarcity of target working condition samples. The method constructs a double branch space-channel feature extraction network CA-Mamba, adopts an improved Mamba module and channel attention to enhance key response, and fuses deep features through a selective bridging module. Then, a domain discriminator and a group discriminator are constructed to realize cross-domain feature confusion and class boundary constraint; relying on the confidence weighted pseudo label PSMMD and dynamic memory enhanced contrast learning DMECL, the target domain feature is constrained from the class-instance double granularity, so that the target domain sample is aligned to the homologous source domain prototype, and finally the covariance measurement is used to complete the target domain classification. The application can extract domain-invariant and high-discriminative features under small sample conditions, and improve the diagnosis accuracy under cross-domain conditions.

This invention relates to a semi-supervised segmentation method for thyroid nodules based on morphological progressive pseudo-supervised refinement. Addressing issues such as blurred boundaries, background interference, and low pseudo-label quality in semi-supervised learning within ultrasound images, this invention proposes a plug-and-play semi-supervised segmentation module. This module progressively filters blurred boundary responses, recovers reliable foreground regions, and introduces morphologically-aware geometric constraints to systematically refine noisy pseudo-supervised signals. The semi-supervised segmentation module constructed by this method includes: a boundary and uncertainty curriculum module to suppress blurred boundary responses; a dynamic region expansion module to recover weak but reliable foreground regions under spatial constraints; and a gradient-driven geometric refinement module to further standardize boundary quality and structural consistency through topology-aware feedback and differentiable geometric refinement. Through the collaboration of these three modules, the semi-supervised segmentation method gradually transforms noisy pseudo-labels into more reliable and structurally sound supervised signals.

This invention relates to an active learning-based data-free black-box attack method and system based on multidimensional value assessment, belonging to the field of artificial intelligence security technology. This method constructs a pre-emptive "sample screening funnel," utilizing a local substitution model to perform multidimensional assessments of sample boundary approximation, information uncertainty, and geometric diversity before sending images to a commercial cloud API. Only high-value samples are selected for querying, thereby achieving low-cost, high-efficiency model theft and adversarial attacks. This invention ensures the diversity and training stability of data-free generated samples, significantly improves the transfer success rate of adversarial examples, and achieves "low-cost, low-risk" economical attacks. It has strong versatility and can be seamlessly integrated into various existing data-free attack frameworks, facilitating deployment and implementation in practical security assessment systems.

The present application belongs to the technical field of Internet of Vehicles and communication security, and discloses a federated learning training method based on vehicle state perception and applicable to dynamic Internet of Vehicles. The method collects vehicle running state information of vehicle clients, preliminarily screens candidate vehicle clients, calculates comprehensive scores of the screened vehicle clients, adopts a client selection strategy combining selection of the top-ranked clients with random supplementation to determine a set of vehicle clients participating in the current round of federated learning, configures local training parameters for the selected vehicle clients according to vehicle speed and computing capacity, performs local training on local data sets to obtain model updates, performs weighted aggregation on the model updates of the effective participating vehicle clients at the server end, introduces a momentum mechanism to smooth the model update process, and obtains new global model parameters. This method effectively improves the stability, robustness and training efficiency of federated learning in a high-dynamic Internet of Vehicles environment.

The application discloses a face forgery detection method and system based on federal incremental learning, comprising the following steps: constructing a federal incremental learning framework, each client is provided with a local face forgery detection model, and a global server is provided with a global face forgery detection model; the local face forgery detection model is subjected to cross-entropy training in a basic training stage, generates a forgery substitute sample in an incremental training stage, and is used as training input together with a real face sample and a forged face sample; after training, the model parameters are uploaded to the global server for aggregation; the global face forgery detection model is updated based on a difference perception aggregation strategy; an adversarial perturbation is trained based on the global face forgery detection model converged in the present stage and the real face sample, and a perturbation pool is refreshed until a preset termination condition is reached; and a to-be-tested face image is identified as real or fake based on the trained global face forgery detection model. The application can relieve catastrophic forgetting and improve cross-task generalization ability and robustness.

The invention discloses a gastrointestinal tract image classification method based on an improved Adam optimization algorithm, and belongs to the technical field of medical image processing. The method comprises the following steps: a data preprocessing module for acquiring gastrointestinal tract image data for preprocessing, constructing a data set, and randomly dividing the preprocessed gastrointestinal tract image into a training set and a test set; the feature extraction module is used for extracting key features from the preprocessed image; the image classification module is used for classifying the images subjected to feature extraction by adopting an improved Adam algorithm (AdaWAGC algorithm); the result output module is used for displaying an image classification result; and the evaluation module is used for evaluating the classification result. By adopting the method to classify the gastrointestinal tract image, the accuracy and efficiency of gastrointestinal tract image recognition can be improved.

The application provides a large language model individualization alignment method and device based on difficulty perception learning, and belongs to the technical field of model training. The method comprises the following steps: obtaining multiple groups of training samples; each group of training samples comprises a user question, a user portrait, and a target reply corresponding to the user question under the user portrait; calculating the perplexity of the target reply in each group of training samples; dividing the multiple groups of training samples into high-perplexity training samples and low-perplexity training samples according to a preset perplexity threshold and the perplexity of the target reply in each group of training samples; performing first training on a preset large language model based on the low-perplexity training samples and a likelihood loss function, performing second training on the preset large language model after the first training based on the high-perplexity training samples and reinforcement learning, and obtaining a target large language model, so that the target large language model completes individualization alignment. The application can improve the individualization alignment effect of the model.

This invention provides a federated learning method based on prototype alignment and knowledge distillation, belonging to the field of distributed machine learning technology. It includes: a server initializing global model parameters and a global prototype set; each client performing local model training based on its own private dataset, extracting sample feature representations, and constructing local prototypes for each category of local samples; each client performing hierarchical prototype alignment during local training; each client calculating distribution offset based on the degree of offset between the local feature distribution and the global feature distribution; after completing local training, each client uploading the updated local model parameters and local prototypes to the server; the server aggregating the local model parameters uploaded by each client and aggregating the local prototypes uploaded by each client by category; and the server proceeding to the next round of federated training. This method can improve the training stability, classification accuracy, and communication efficiency of federated learning models in heterogeneous scenarios.

This invention relates to the field of computer vision, specifically to a passive domain adaptive target detection method and electronic device based on pseudo-label confidence feedback, comprising the following steps: S1, acquiring a pre-trained source domain model and unlabeled target domain data; S2, constructing a mean-teacher self-training architecture; S3, generating pseudo-labels and calculating confidence feedback signals; S4, implementing an adaptive dynamic mean-teacher update strategy. This invention, by introducing a pseudo-label confidence feedback mechanism, enables the model to have adaptive adjustment capabilities, effectively suppressing the adverse effects of low-quality pseudo-labels on model training; it can dynamically reduce the decay rate in high-quality pseudo-label scenarios, improving cross-domain transfer efficiency and detection accuracy; this invention effectively alleviates the "confirmation bias" problem in passive domain adaptive tasks, significantly improving the model's training stability, noise resistance, and cross-domain target detection performance in the target domain, possessing good application value and promising prospects for promotion.

The present application relates to the technical field of intelligent manufacturing and intelligent scheduling, and particularly relates to a flexible processing unit dynamic regulation method, system and model based on multiple agents. The method comprises the following steps: S1: establishing an event-driven Markov decision process model; S2: constructing corresponding task selection agents and workpiece selection agents to form a hierarchical multi-agent deep reinforcement learning architecture; S3: designing the system state features, action space and reward function of the task selection agents and the workpiece selection agents; S4: jointly training the task selection agents and the workpiece selection agents based on a multi-agent proximal policy optimization training algorithm to obtain optimal policy parameters of the task selection agents and the workpiece selection agents; and S5: the task selection agents and the workpiece selection agents output task actions and workpiece actions respectively based on the optimal policy parameters obtained through training. The present application can improve the adaptability of the system to dynamic disturbances and the real-time decision-making capability.

The invention provides a large-model multi-language enhancement training method capable of continuously expanding experts, which comprises the following steps: step 1, extracting an original FFN module in a large-scale language model LLM to be trained and segmenting the original FFN module into a plurality of new fine-grained FFN modules as fine-grained experts; 2, setting a fine-grained router, and selecting a preset number of newly-added fine-grained experts according to a preset rule to obtain the output of the fine-grained router; 3, setting a parent router, controlling the weight between the output of the original FFN module and the output of the fine-grained router, and calculating the final output; 4, freezing an original FFN module, and independently training the fine-grained expert; step 5, freezing an original FFN module, adding the newly added fine-grained expert, and training the newly added fine-grained expert; step 6, freezing all other parameters, and only training a parent router; and 7, continuously repeating the step 4 to the step 6 to realize large-model multi-language enhancement training of sustainable extension experts.

The application provides a bone tumor fine-grained classification model training, classification method and device, constructs a multi-modal positive sample containing global / local positive lateral X-ray image, lesion attribute and patient information, combines text semantic similarity to remove false negative part to construct high-quality negative sample; global / local image features are extracted through a double-image encoder, the lesion attribute keywords are converted into "entity-translation-existence" triplets relying on a medical knowledge base, and global / local semantic features of patient basic information and lesion attributes are extracted by using a text encoder; based on contrast learning, the global image and global semantic are constructed InfoNCE contrast loss, the model is trained based on the local mutual information loss calculated based on the DV variation formula and the classification loss, the global text feature alignment and the local text feature alignment are realized, the medical terminology semantics are deeply combined, the training stability is improved, the bone tumor subtype classification accuracy and robustness are significantly improved, and reliable support is provided for clinical accurate diagnosis.

The application provides a kind of based on feature enhancement and composite model's ultra-short-term wind power prediction method, belongs to ultra-short-term wind power prediction technical field, this method includes: by calculating the mutual information of multiple meteorological characteristics and wind power output and normalizing, screening to obtain key features, and constructing wind power prediction fitness function, by minimizing wind power prediction fitness function, obtain enhanced input features;By introducing one-dimensional convolutional neural network and bidirectional long short gate recurrent unit, construct the residual network of bidirectional long short gate recurrent unit and convolutional neural network fusion, and introduce space-time attention mechanism, construct to obtain composite residual network ultra-short-term wind power prediction model;And use multiple meteorological characteristics and enhanced input features to predict to obtain ultra-short-term wind power prediction result;The application solves the problem that the influence of meteorological multi-scale fluctuation on output cannot be accurately grasped, prediction is disconnected, disturbance is not considered and prediction accuracy continues to decline with step increase.

The application discloses a Mongolian-Chinese multi-expert neural machine translation method and system based on decision layer relative strategy optimization, and the method comprises the following steps: acquiring Mongolian-Chinese bilingual parallel corpus and preprocessing to obtain Mongolian-Chinese translation training data; training a basic translation model based on the Mongolian-Chinese translation training data to obtain a Mongolian-Chinese task adaptation model; performing sensitivity analysis on the Mongolian-Chinese task adaptation model to obtain hierarchical sensitivity results of each network layer; according to the hierarchical sensitivity results and the complexity of an input sentence, different network layers of the Mongolian-Chinese task adaptation model are allocated with different quantization precisions to obtain a light-weight translation model; a course type training sequence is constructed, and an optimized Mongolian-Chinese translation model is obtained based on a composite reward function; a Mongolian sentence to be translated is input into the optimized Mongolian-Chinese translation model to generate a corresponding Chinese initial translation; and a final Chinese translation is determined according to the Mongolian sentence to be translated and the Chinese initial translation.

The invention belongs to the field of energy consumption optimization, and particularly relates to a two-stage electrolytic copper foil electrodeposition energy consumption optimization method based on federal learning. The method comprises the steps that each client side collects multi-dimensional technological parameters in the electrolytic copper foil electro-deposition process, and technological parameter vectors are constructed; constructing a federated optimization framework, training a local energy consumption predictor by each client in a local autonomous optimization stage, and calling an intelligent optimization algorithm for optimization; in the federated collaborative optimization stage, each client updates a local energy consumption predictor based on the structure parameters of the global energy consumption predictor generated by the server, and calls an intelligent optimization algorithm to perform collaborative optimization; and based on the latest local energy consumption predictor, carrying out backtracking evaluation on the optimal process parameter vector set, and screening out the process parameter vector which has the lowest predicted energy consumption and is finally applied to production control. According to the method, flexible support for various intelligent optimizers is realized, and generalization, stability and reliability of the model in a real industrial scene are improved.

The present application belongs to the technical field of nondestructive testing of building materials, and particularly relates to a rebound nondestructive testing method and system for the compressive strength of rapidly hardened cement-based materials. The technical solution is as follows: sample data of rapidly hardened cement-based materials is collected to construct a sample data set; the sample data set is subjected to outlier rejection, normalization processing and feature weight mapping to form a multi-dimensional input feature vector; based on the multi-dimensional input feature vector, a compressive strength prediction model is constructed, and the model parameters are optimized through k-fold cross-validation; the rebound value, curing age and mix proportion parameters of the rapidly hardened cement-based material to be tested are input into the trained compressive strength prediction model to output the corresponding compressive strength detection value. The present application can comprehensively consider sample data, establish a nonlinear mapping relationship between multi-dimensional features and compressive strength, thereby realizing rapid, continuous and high-precision evaluation of the compressive strength of rapidly hardened cement-based materials without damaging the materials or structure.

This invention provides a method and apparatus for predicting three-dimensional voxel damage based on component consistency constraints. The method first acquires structural information, constructs a three-dimensional voxel mesh and a mapping relationship between components and voxels, and integrates structural properties and load effects to construct a multi-channel input tensor. This tensor is then input into a specially trained three-dimensional segmentation network model to obtain initial voxel damage probabilities. The core of this invention lies in introducing cross-scale engineering consistency constraints: based on the mapping relationship, voxel predictions are aggregated into component-level damage representations, and a consistency correction is calculated according to a reasonable range defined by the component's preset damage level. This allows for the coordinated adjustment of the initial probabilities of all voxels within the component, ultimately outputting a three-dimensional damage distribution map that conforms to component-level engineering rationality. This method effectively solves three major engineering problems: class imbalance in voxel-level prediction, overall component inconsistency, and overfitting to noise labels.

The invention relates to the technical field of intelligent clothing design, in particular to an intelligent clothing building method and system based on deep learning. Comprising a line draft scanning module, a partition module, a block integration module, a block adjustment module, an image generation module, a stereoscopic model import module, a wearing state generation module and a display terminal, the partition module is connected with the line draft scanning module, the block integration module is connected with the partition module, the block adjustment module is connected with the partition module, and the image generation module is connected with the display terminal. The image generation module is connected with the block integration module, the stereoscopic model import module is connected with the image generation module, the wearing state production module is connected with the image generation module, and the display terminal is connected with the image generation module; the design modification of the designated area can be completed according to the actual thought of a creator, meanwhile, rapid updating and presentation are carried out according to the corresponding modification content, the situation of repeated modification caused by large difference between a design scheme and actual display is avoided, and actual operation is more convenient.

The invention discloses a medical image classification method based on curriculum-type enhancement scheduling and hybrid enhancement. The method comprises the following steps: S100, constructing a curriculum-type data enhancement mechanism; s200, performing a staged hybrid enhancement strategy; s300, carrying out multi-stage model training and optimization; and S400, carrying out model interpretability output generation. According to the method, the stability and generalization performance of the model in the multi-stage training process are improved, the interpretability of a model output result is enhanced, and reliable technical support is provided for a clinical auxiliary diagnosis system.

The application relates to a test phase adaptive learning method, device and equipment based on feature alignment, wherein the method comprises the following steps: obtaining target domain images and selecting a target domain image queue to be trained from the target domain images; obtaining image pseudo labels of the target domain image queue to be trained, calculating pseudo label probability advantages, screening target domain images meeting preset confidence and pseudo labels corresponding to the target domain images meeting preset confidence, obtaining a pseudo label loss based on a preset pseudo label loss function; and constructing target domain category prototype features, performing feature alignment training on the target domain category prototype features and preset source domain category prototype features, obtaining a feature alignment training loss of the source domain and the target domain, and obtaining a training target of a test phase model according to the feature alignment training loss and the pseudo label loss. Thus, the problem that a test phase model training work of related technologies is not suitable for actual requirements of an industrial scene is solved, and the classification prediction effect of the model is improved.

This invention discloses a method and system for constructing a dynamic loss function for large-scale models based on cognitive memory strength, belonging to the fields of large-scale model training optimization and cognitive dynamics. Existing large-scale model training generally employs a fixed cross-entropy loss function, using equal weights for all tokens in the supervision process. This fails to distinguish between key semantics, logical cores, and redundant padding information in the text, resulting in insufficient supervision of core semantics, overfitting of redundant information, gradient imbalance during long text training, and accumulation of cognitive biases in the model. Consequently, the model suffers from weak logical ability, poor memory of key information, and frequent hallucinations. This invention, based on an original cognitive memory dynamics mechanism, constructs a dynamically weighted loss function. It allocates training weights in real-time according to the four-dimensional spatiotemporal attributes and memory strength of each token, applying high weights to logical cores, key temporal information, and core semantics of the context, while applying low weights to redundant padding, invalid repetitions, and noise information. This invention breaks away from the traditional equal supervision mode of fixed loss, achieving a training process that aligns with the laws of human cognitive memory, matching the principle of minimum action in field theory, significantly improving the model's core semantic fitting ability, long text cognitive stability, and logical reasoning accuracy, thus reducing the model hallucination problem at its root.

The application discloses a kind of multi-modal sentiment analysis robust training methods for noise pollution and modal missing scene, including the following steps: constructing the multi-modal sentiment analysis dataset including text modality, audio modality and visual modality;Establish multi-modal sentiment analysis model, and identify dominant modality based on the performance representation of each modality during training process;Design dominant perception soft rollback modulation mechanism, continuously inhibit the reverse update of dominant modality to reduce the random gradient fluctuation introduced by noise;Introduce stability regularization constraint based on parameter variation, inhibit mutation update under noise and partial modal missing condition, improve the optimization stability of model;The soft rollback modulation mechanism and parameter stability regularization constraint are combined for model training, and a robust multi-modal sentiment analysis model is obtained without changing the original forward fusion process;Finally, the multi-modal sample to be predicted is input into the trained model, and the sentiment analysis result is output.The application can effectively alleviate the training bias and instability problem caused by modal dominance in multi-modal sentiment analysis, enhance the robustness and generalization ability of the model under noise interference, modal missing and complex scene, improve the multi-modal collaborative modeling effect and sentiment prediction performance.

The application discloses a chest radiograph report generation and lesion positioning method and system based on reinforcement learning, and belongs to the technical field of cross between artificial intelligence and medical image analysis. The method first uses a plurality of source chest radiograph images and query texts to supervise and fine-tune a basic model; then a GRPO reinforcement learning framework is used to optimize a strategy model, and an indicator function guided by token entropy is introduced into a target function, and finally, the optimized strategy model is used to realize end-to-end parallel generation of chest radiograph report texts and lesion boundary box coordinates. The method improves the quality of chest radiograph report generation while realizing the visualization and positioning of lesions, and the integrated output form is more in line with the actual clinical workflow.

The present application belongs to the technical field of urban computing and data mining, and in particular to a multi-view urban area embedding method based on spatial function consistency. The method comprises the following steps: S1: multi-view data collection and processing; S2: spatial function consistency intra-view representation learning; S3: intra-regional cross-view interaction; S4: inter-regional cross-view interaction; S5: double-feature attention fusion; and S6: downstream task application. The present application can effectively suppress information interference caused by noise edges and weakly related connections, and improve the accuracy and stability of regional local structure modeling. At the same time, the present application enhances the global semantic connectivity and cross-view information propagation capability of the regional representation. On this basis, through a double-layer attention fusion mechanism, adaptive integration of multi-view features and inter-regional associated features is realized, further improving the discriminability, robustness and generalization capability of the final regional embedding representation.