1622 results about "Global model" patented technology

The invention discloses a power distribution room anomaly detection system based on cloud side-end cooperation, and belongs to the technical field of intelligent power grids. In order to solve the problems of high network bandwidth pressure, insufficient edge computing capability, low anomaly detection accuracy, difficulty in multi-source data fusion and the like caused by the adoption of an end-cloud direct connection architecture in an existing power distribution room monitoring system, the system comprises: a data acquisition layer configured with various heterogeneous sensors to acquire operating parameters and environmental data in real time; the edge storage and calculation layer carries out local real-time processing, anomaly detection, model training and visual display, an anomaly detection module of the edge storage and calculation layer carries out research and judgment on real-time data to generate early warning information, and a prediction and detection linkage module monitors an anomaly probability trend and adjusts a sampling frequency; the edge gateway realizes protocol conversion and data forwarding; and the cloud decision-making layer aggregates multi-edge node data, optimizes a global model through federal learning, and issues and updates a local model. The system is used for improving the accuracy, real-time performance and reliability of anomaly detection of the power distribution room, reducing the operation and maintenance cost and realizing intelligent operation and maintenance.

The invention discloses a personalized federated learning optimization method and system, and mainly solves the problem of poor performance of an existing personalized federated learning model. The method comprises the following steps: establishing a communication link between a client and a server; each client receives a current global sharing model parameter broadcasted by the server, loads the current global sharing model parameter to a local model, and introduces a total loss function of a dynamic alignment strength definition model; training and optimizing local model parameters, and updating shared parameters by using the parameters; the client side calculates a self-adaptive aggregation weight based on the parameter updating quantity norm, the data volume weight and the synchronous frequency weight of the client side, and uploads the self-adaptive aggregation weight and the updated shared parameters to the server; and the server receives the parameter and weight information uploaded by the server, executes global model aggregation to obtain an updated global model, and outputs the global model reaching accuracy convergence or a preset training round on the verification set. According to the method, local personalization and global consistency can be balanced, the robustness and efficiency of global aggregation are improved, and the method can be used for processing scenes of high data isomerism and dynamic change of client participation states.

Disclosed in the present invention are a large language model federated fine-tuning method and apparatus based on gradient compression. The method comprises the following steps: constructing, on the basis of a gradient tensor generated during fine tuning of a large language model, a raw data set having a time series relationship, performing inference by means of an autoencoder to obtain a reconstructed gradient data set, and constructing a reconstruction loss function to optimize the autoencoder; and initializing a base model of the large language model as a global model at a server end, the server end updating the global model to a client, using a pre-trained encoder to obtain a compressed gradient at the client, and at the server end, using a pre-trained decoder to decode and aggregate the compressed gradient, and then updating the global model. The present invention can improve the fine-tuning efficiency of the large language model and reduce computing resource requirements while ensuring data privacy protection, and is suitable for application scenarios such as communication optimization improvement and privacy protection enhancement in the process of scientific computing-oriented large model fine-tuning and training.

An efficient heterogeneous federated learning method based on hybrid distillation includes: initializing, by a server, global model parameters, and setting a preset total number of training rounds and a number of clients participating in each of the training rounds; loading local datasets in the clients respectively, performing random transformations on the local datasets to generate client distillation data for the clients, sampling multiple sub-networks from an original network of each client, training each sub-network on the client distillation data to obtain updated local model parameters of each client, and uploading the updated local model parameters to the server; and receiving, by the server, the updated local model parameters, performing, by the server, server distillation based on the updated local model parameters and a preset auxiliary dataset to obtain updated global model parameters and an updated global model, and sending, by the server, the updated global model to the clients.

The invention relates to a landslide grading early warning method based on multi-modal data change characteristics. The method comprises the following steps: establishing a mountain digital twinborn body; the method comprises the following steps: deploying a multi-node sensor network in a target area, configuring an edge computing unit, collecting geological data in real time, and screening the geological data based on mountain digital twin to form effective local data; constructing a lightweight multi-modal neural network model at each node, dynamically searching hyper-parameters by using an ant colony optimization algorithm, and generating an encryption model weight update quantity packet; the central server dynamically calculates node weights according to disaster feature vectors output by the digital twins, generates a global model through weighted aggregation, and directionally distributes and updates the global model; real-time monitoring data and a model prediction result are fused, millimeter-level disaster evolution simulation is executed through a variable step size physical engine, an advanced early warning signal is triggered when a deduced prediction risk exceeds a threshold value, and the edge model adaptability, federal aggregation precision and early warning advancement are remarkably improved.

The invention relates to the technical field of federated learning, in particular to a dynamic federated mutual learning method and system for balancing individuation and generalization, and the method specifically comprises the following steps: each client carries out the preprocessing of data to be processed of a model, and carries out the strong enhancement and weak enhancement processing; inputting the data subjected to strong enhancement processing into a shared model, inputting the data subjected to weak enhancement processing into a private model, and performing iterative training on the two models; related parameters of the shared model after each round of iterative training and a difference item between two model parameters are uploaded to a federation server; the federated server adopts a multi-dimensional adaptive aggregation strategy to obtain an updated global model, and returns the updated global model to each client to replace the shared model in the next round of training; and finally generating a generalization result and a personalized result. According to the method, the private-shared model architecture is constructed, and dynamic federated mutual learning is carried out in combination with the federated server, so that balance and collaborative improvement of individuation and generalization performance can be realized.

The invention discloses an energy storage battery health state evaluation method and system based on federated learning, and relates to the technical field of energy storage batteries. The method comprises the following steps: training a local health state evaluation model based on a federal loss function containing physical prior constraints at each client device, and enhancing sparse working condition data by adopting a local generation model to generate local model update; in the central server, performing value-guided heterogeneous aggregation processing, performing weighted aggregation on local model update submitted by the client, and generating a global health state assessment model; in the central server, executing digital twinborn consistency calibration, and performing post-aggregation fine tuning on the global model by using a reference signal generated by a cloud digital twinborn body; and communicating between the client equipment and the central server by adopting an event triggering and gradient sparse quantization compression mechanism, and deploying an adaptive differential privacy policy based on gradient inversion auditing. According to the method, the problems of high data privacy leakage risk, low model precision under heterogeneous data and high communication overhead in the prior art are solved.

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 dynamic portrait construction method and system fusing a large model and user behavior data, belongs to the technical field of artificial intelligence and big data analysis, and aims to solve the problems of insufficient real-time performance, difficulty in multi-source data integration and high privacy risk in the traditional technology. The method comprises the following steps: collecting basic attributes, behavior sequences and unstructured data by burying points, and processing through an Apache Flink session window and a dynamic watermark; multi-modal feature extraction (discrete feature embedding, bidirectional LSTM coding behavior sequence and BERT coding text) is carried out, and joint embedding is generated through cross-modal contrast learning; generating three types of labels, namely a static label (rule engine), a dynamic label (1.3 B parameter quantity Nano-vLLM) and a predictive label (XGBoost), and dynamically adjusting weights; and realizing global model updating through federated learning and differential privacy. The system comprises a data acquisition layer, a feature extraction layer and a label generation updating layer. The real-time performance and accuracy of the portrait are improved, the privacy of the user is protected, and the commercial value in e-commerce, finance and other scenes is remarkable.

Disclosed in the present invention are a model training method and apparatus based on federated learning, and a device and a storage medium. The method comprises: training a local model on the basis of power data, so as to generate a fault analysis model used for fault analysis of a power device; and uploading local model parameters to a cloud server, such that the cloud server aggregates the local model parameters, then updates a corresponding global model, and issues updated global model parameters to edge nodes, wherein when aggregating the local model parameters, the cloud server marks the edge node corresponding to an abnormal local model parameter as a suspected abnormal node, monitors the marked suspected abnormal node, and removes, when it is detected that the suspected abnormal node is abnormal, the local model parameters uploaded by the suspected abnormal node. By means of the present invention, it can be ensured that the performance of a fault analysis model obtained by training a local model is not degraded.

The invention discloses a non-independent identically distributed data asynchronous federal learning method based on an improved aggregation algorithm. The method comprises the steps that a server initializes a global model and issues the global model to all clients; and the client performs local training on the received global model by using local data, and uploads the model and model parameters to the server after training is completed. Then, the server adjusts a model lag degree based on a client data volume proportion, calculates model difference consistency, client historical contribution stability, old degree penalty of the client model and cosine similarity of the client model and the global model based on parameters of the client model and the current global model, and generates an asynchronous federal aggregation factor accordingly; and updating the global model parameters to generate a new global model. And finally, testing the global model by the server, and judging whether the learning process is stopped or not. According to the method, fair and effective model aggregation can be realized, and the model convergence stability and the final model detection precision are improved.

The invention discloses an application method and device of a battery model management iteration system based on federated learning and block chain verification, and relates to the field of block chain energy storage management, and the method comprises the steps: deploying an intelligent contract on a block chain to publish a federated learning task, and initializing a global model; each participant uses a private data local training model, encrypts and uploads the private data local training model to a central coordination server for security aggregation; submitting the aggregated global model to a block chain verification layer, and automatically evaluating the performance and recording a result by the smart contract; the verified model is deployed to an actual scene for battery state prediction; collecting actual operation data of the battery through an oracle machine feedback network, and feeding back the data to a chain; and the intelligent contract incentive engine automatically re-evaluates the performance of the model according to feedback data, distributes rewards, and iteratively optimizes a loop. According to the method, the problem that data islands exist in key parameter prediction models such as the state of health and the remaining service life of the battery is solved, and the credibility and the sustainable evolution ability of the prediction models are enhanced.

The invention discloses a multi-mode federated learning method and system, computer equipment and a readable storage medium, and belongs to the technical field of federated learning. The multi-modal federated learning method comprises the following steps: on each client node, mapping local data of various modals into a plurality of vectors in a unified semantic space, determining an incidence matrix of the data of the various modals, and fusing the plurality of vectors according to the incidence matrix to obtain a local semantic vector; training a local model by using the local semantic vector to obtain local model parameters, and uploading the local model parameters to a server; on the server, identifying the difference degree between the data distribution condition of each client node and the global data distribution condition, and determining the node weight vector of each client node; and performing weighted aggregation on the corresponding local model parameters by using the node weight vector of each client node to generate global model parameters for next federated learning. Therefore, the performance of the training model can be improved.

The invention discloses a federated learning backdoor attack defense method based on layer perception detection, which relates to the technical field of artificial intelligence security, and comprises the steps of client training and uploading, malicious client identification, backdoor key layer detection and deletion, robust aggregation updating and adversarial training for enhancing robustness. According to the method, the model parameters uploaded by the clients are subjected to clustering analysis, and the maximum benign cluster is identified by adopting an unsupervised clustering method, so that instability caused by single threshold judgment is avoided, and the benign client and the malicious client can be distinguished; and meanwhile, layer perception detection and elimination are used in the scheme, so that the backdoor introduced by a malicious client can be effectively identified and eliminated on the premise of ensuring the global model precision, and the influence of backdoor attack on the model is inhibited.

The invention provides a neighbor projection type gradient coordination compression method and device and a federated learning system, and relates to the technical field of federated learning. According to the method, the uploading gradient of the client side is obtained, direction normalization is carried out, direction embedding updating of the client side is carried out through index moving average, then the similarity between the client sides is calculated, a neighbor set is generated, a similar graph is constructed, and then conflict detection and weight setting are executed. And performing weighted orthogonal projection in the frozen neighbor direction of each client to correct the gradient, and finally performing weighted aggregation on the gradient and updating global model parameters. According to the method, gradient conflicts are efficiently detected and corrected in a local range by constructing the client similar graph and combining a neighbor projection mechanism, the problems of high complexity and excessive information reduction caused by global processing are avoided, the convergence speed, stability and precision of a global model are improved under the condition that additional calculation and communication overhead of the client is not increased, and the user experience is improved. The method is suitable for large-scale non-independent identically distributed data scenes.

The invention relates to a cross-network collaborative security alarm noise reduction method based on security federal learning, and belongs to the field of network security. Comprising the steps that a server prepares and distributes a large model injected with personalized and global parameter modules; the client performs fine adjustment based on local alarm data and only uploads gradient update of the global parameter module; the server dynamically divides a network layer into a low-conflict layer and a high-conflict layer by calculating the gradient conflict degree of each client on each network layer of the global module; the server only aggregates the low-conflict layers to update the global model and distributes the update back to the client, while the client retains its locally trained low-conflict layer. To-be-detected alarms are input into the client large model to be analyzed and processed, and accurate alarms and disposal suggestions are fed back to the data set for continuous optimization after the result is audited by experts. According to the method, client drift is relieved, and the noise reduction accuracy and individuation effect of the model on heterogeneous data are remarkably improved while data privacy is protected.

The invention discloses a machine room group control multi-objective optimization digital twin platform and an optimization method, and relates to the technical field of machine room group control. Constructing a multi-objective optimization model, wherein the multi-objective optimization model comprises a three-dimensional objective function and constraint conditions; the three-dimensional objective functions comprise an energy consumption minimization function, a temperature and humidity control precision maximization function and an equipment life loss minimization function. Compared with a traditional static twinborn model, the method has the advantages that the deviation is greatly reduced, and a precise virtual-real mapping basis is provided for optimization decision making. According to the whale optimization algorithm, the global search capability and the NSGA-IIPareto solution screening capability are fused, the convergence speed of the algorithm is improved through fuzzy membership degree fitness calculation and crowding degree screening, when deviation exceeds a threshold value, optimization can be shortened to 15 seconds, compared with an existing algorithm, the response speed is greatly improved, and the real-time control requirement under the dynamic load of a machine room is met.

The invention discloses a federal learning-based cross-device brain grain model training system. The system comprises a terminal device layer, an edge coordination layer and a central server layer, the terminal equipment layer is responsible for collecting original electroencephalogram data and executing local lightweight training; the edge coordination layer is used for aggregating model updates of a plurality of terminal devices in the jurisdiction to generate a regionalized sub-model; according to the central server layer, a central server performs secondary optimization on each regional model to generate a global unified model; the encrypted global model and strategy configuration are pushed to all edge coordination layers; and the edge coordination layer pushes the optimized model and strategy to terminal equipment. Through a three-layer federated architecture and multi-dimensional optimization, the system efficiency is remarkably improved: the terminal equipment only uploads an encrypted model increment, and source data protection is realized in combination with differential privacy and homomorphic encryption; federal transfer learning shortens the cold start time of new equipment.

The invention discloses a federal learning security training method and system based on differential privacy, belongs to the technical field of artificial intelligence, and aims to solve the technical problem of how to realize data privacy protection and model precision balance in cross-mechanism model training. Comprising the following steps: constructing a distributed structure comprising a plurality of clients and a server; each client constructs a noise variance calculation model by taking the gradient feature, the privacy budget coefficient, the data sensitivity level coefficient and the training stage coefficient as calculation parameters after each round of local training is finished, and Gaussian noise is calculated according to the noise variance; verifying the noise variance of each round, and adjusting the calculation parameters of the noise variance based on the comparison result of the accumulated budget consumption and the global privacy budget; dynamically allocating the data sensitivity level of the local training data and the budget allocation proportion of the client; and the server performs secure aggregation through a secret sharing algorithm, and distributes the updated global model parameters to each client.

The invention relates to the technical field of internet big data and information security, and designs a gearbox fault diagnosis method based on multi-source agent federal field generalization by combining the interaction capability of wide-area data and a local diagnosis agent. According to the method, a dual regularization mechanism is introduced in the training process of each local diagnosis agent, and the problem of negative migration is solved by actively learning domain-invariant fault features, so that the cross-working-condition and cross-equipment generalization diagnosis performance of a final model is remarkably enhanced on the premise of protecting data privacy; meanwhile, a central server is replaced by a decentralized coordination network realized by a block chain technology, and a committee consensus protocol is executed to ensure that the generation process of the global model is open, transparent and non-tampering, so that the single-point fault risk is fundamentally eliminated, the security and robustness guarantee is provided for the whole coordination process, and the method has the advantages of being high in practicability and the like. And the long-term applicability and the fault diagnosis accuracy and reliability of the method in a dynamically changing industrial environment are ensured.

The invention relates to the technical field of model transfer learning, and discloses a model adaptive optimization method based on transfer learning. The method comprises the steps that source domain model structure parameters and target domain task initial data distribution are obtained, the feature mapping relation of all levels of a source domain model is extracted, and a cross-domain feature migration reference topological framework is generated; dividing a migratable feature layer and a to-be-reconstructed feature layer according to a target domain data distribution difference, and dynamically adjusting a migration priority in combination with a sample distribution density; freezing and unfreezing the transferable feature layer layer by layer based on the priority, synchronously constructing a local feature reconstructor, and optimizing domain offset through iterative feature alignment; collecting a feature reconstruction error and a migration feature retention degree in each iteration, and calculating a dynamic balance coefficient to adjust a freezing proportion and reconstruction intensity; and fusing the two types of features through a global model integrator, and generating mixed feature representation to drive end-to-end training of a target domain task.

The invention discloses a multi-tenant adaptive cooperative defense method and system in a hybrid cloud scene, and the method comprises the following steps: obtaining a remote credible proof of hardware, verifying the credibility of a tenant agent and the integrity of a code based on the remote credible proof, granting a mark, and detecting cloud information to generate a machine readable portrait containing key parameters; collecting local multi-mode log data coding embedding vectors of tenants, and after privacy processing, calculating that the abnormal confidence exceeds a threshold value by a local model to trigger current limiting or blocking; and constructing a hierarchical federated architecture containing local nodes of tenants, a regional cloud and a global control plane, encrypting aggregation parameters to generate a cross-tenant attack recognition global model and issuing the cross-tenant attack recognition global model to each tenant on the premise that log data is not out of a domain. According to the invention, integrity verification is carried out on the tenant side security agent through the trusted access mechanism, and the tenant portrait is constructed based on the trusted mark, so that the real and verifiable cooperative defense capability among a plurality of tenants is realized, and the cooperative process has dependency.

The embodiment of the invention provides a federated learning contribution evaluation method and device, and the method comprises the steps: carrying out the grouping of a plurality of edge computing devices, and obtaining a plurality of sub-federated learning sets; and for the target federated learning sub-set, aggregating model update information corresponding to each edge computing device in the target federated learning sub-set, and determining a collaborative contribution value of the target federated learning sub-set based on the performance index of the updated global model on the common test set. Through a first linear programming solver, according to the collaborative contribution values of the multiple federated learning sub-sets, obtaining the maximum loss value corresponding to all the federated learning sub-sets and optimizing the maximum loss value to obtain the minimized maximum loss value, and through a second linear programming solver, obtaining the maximum loss value corresponding to all the federated learning sub-sets; and according to the maximum loss value after all the sub federated learning sets are minimized and the reference contribution values corresponding to the plurality of edge computing devices, target contribution vectors corresponding to the plurality of edge computing devices are determined, and the contribution degree of each edge computing device in the training process is accurately quantified.

The invention relates to the technical field of federated learning security, and discloses a federated learning poisoning defense method based on time-frequency spectrogram and comparative learning, which comprises the following steps: receiving model update uploaded by each client, grouping and vectorizing parameters according to model layers, and generating a time-frequency spectrogram by applying short-time Fourier transform to parameter vectors of each layer; based on the time-frequency spectrogram, constructing a positive sample pair through data enhancement, carrying out difficult negative sample mining, and training an encoder by using a contrast loss function to extract an embedded vector with high discriminant power; and performing unsupervised clustering on the embedded vector by using a DBSCAN clustering algorithm, judging the maximum cluster as a benign client, performing final judgment in combination with historical malicious records, and only aggregating model parameters of the benign client to update a global model. According to the invention, high-precision detection of attack features can be realized, and a more universal, more efficient and more practical federal learning poisoning attack defense method is realized.

The invention discloses a federated learning training method and system supporting heterogeneous data, belongs to the technical field of distributed machine learning, and is used for solving the technical problem that an existing federated learning technology cannot solve three major problems of data privacy, data heterogeneity and hierarchical communication bottleneck at the same time and lacks an integrated federated learning training framework. The method comprises the following steps that: a working node acquires a local model update quantity and performs quantitative compression based on local data, a historical update record and a latest global model to obtain a compressed update package and uploads the compressed update package to an affiliated edge node; the edge node aggregates the compressed update packet sent by each working node to obtain a regional aggregation update packet; carrying out second quantization compression on the regional aggregation update packet to obtain a final uploading packet, and uploading the final uploading packet to a central server to which the final uploading packet belongs; the central server globally aggregates the final upload packet sent by each edge node to obtain a global model update packet; and optimizing the global model according to the global model updating package to obtain an optimized model.

The invention discloses an integrated federated learning optimization method based on clustering weight sampling, and the method specifically comprises the following steps: a federated learning system comprises a plurality of clients and a server, and the server calculates the similarity between the clients through model updating information uploaded by the clients, clustering the clients by adopting a dynamic clustering method according to the similarity; the server carries out secondary clustering according to a set sampling rule and judges whether a first-stage iteration threshold value is reached, all the clients obtain a latest global model and freeze a model feature recognition layer for fine tuning, the server collects parameters of all the clients after fine tuning, and then the parameters are clustered according to similarity and are subjected to secondary clustering according to the sampling rule; and combining into an enhanced global model through an ensemble learning strategy. The method can be widely applied to data privacy protection scenes in the fields of medical image analysis, financial risk control, intelligent transportation and the like, and a new technical solution is provided for efficient application of federal learning in a heterogeneous environment.

The invention relates to the technical field of network security, in particular to a defense method and system for federated learning backdoor attacks, a server initializes a federated learning global model, and identifies at least one key layer easy to implant a backdoor in the global model; distributing the current global model parameters to a plurality of clients, receiving model update information uploaded by each client, extracting a multi-dimensional gradient feature vector of each client based on gradient information of each client in a key layer, calculating a mahalanobis distance anomaly score of each client relative to the gradient distribution of the whole client, and calculating the mahalanobis distance anomaly score of each client according to the mahalanobis distance anomaly score; and screening and rejecting the clients which are judged to be abnormal, aggregating model updates uploaded by the clients which are judged to be benign, and generating a new global model. Malicious updating is effectively isolated while the performance of the main task is ensured, and the robustness and safety of a global model are improved.

The invention provides a model training method, device and system based on federated learning and safety protection, and relates to the technical field of artificial intelligence, and the method comprises the steps: carrying out the current training of a target model based on local training data, and obtaining a current model parameter; the local training data is obtained by performing quantum encryption on the original training data; performing homomorphic encryption on the current model parameter to obtain a current encrypted model parameter; uploading the current encryption model parameter to an aggregation node; and receiving global model parameters distributed by the aggregation nodes, and carrying out next training on the target model based on the global model parameters. According to the method, the device and the system provided by the invention, quantum encryption is adopted in an acquisition stage, and homomorphic encryption is adopted in an interaction stage, so that the privacy of an aggregation process is ensured, and a potential attacker cannot obtain model updating details of any single node; in the collaborative training process, it is ensured that data privacy is strictly protected, data are not attacked or tampered, and the efficiency of multi-party collaborative training is improved.

The invention discloses a privacy protection federated learning method and system based on Transform and knowledge distillation, and belongs to the technical field of artificial intelligence and network security, and the method comprises the steps: taking an attention mechanism of a Transform model as a core component of local feature extraction, so as to capture a data long-distance dependency relationship and improve the feature representation quality; a Paillier encryption protocol is introduced to realize homomorphic encryption transmission of model weights; the knowledge distillation technology is adopted at the central server side, the aggregation global model serves as a teacher model to extract soft knowledge, and the feedback client side compresses the model and optimizes the model; according to the invention, the Transform is used as a local feature extractor, the Paillier encryption protocol is combined, and the knowledge distillation technology is adopted after the central server is aggregated, so that the detection performance under Non-IID data is optimized, and both data security and detection efficiency are realized.

The invention relates to the technical field of audio forgery detection, in particular to an audio forgery detection method and system based on lightweight federated adversarial training, and the method comprises the steps: a client extracts audio features through a lightweight network, and updates a model through local adversarial training; the server aggregates the client parameters to generate a global model and issues the global model; and the client performs forgery detection by using the global model. According to the method, a federal learning framework is adopted, original audio is not out of the local, only compressed model parameters are uploaded, high-performance collaborative learning under privacy protection is realized, a data island is broken, and the generalization ability is improved; time domain, frequency domain and frequency spectrum residual multi-dimensional features are fused with cross-modal attention, and adversarial disturbance training is introduced locally, so that feature resolution and anti-interference robustness are enhanced; through combination of pruning, quantification and distillation lightweight processing, the communication and calculation overhead is remarkably reduced, and mobile and IoT edge equipment can be efficiently deployed.