427 results about "Recovery mechanism" patented technology

A system and methods for implementing adversarial-robust compression and reconstruction using a vector quantized variational autoencoder (VQ-VAE) with secure latent space management. The system provides comprehensive protection against adversarial attacks through multi-channel threat detection, adaptive defensive parameters, and coordinated response mechanisms. Input data is continuously monitored for potential threats, and defensive parameters are dynamically adjusted based on detected threat levels. The system implements bounded constraints and hierarchical projections to maintain latent space security while preserving compression efficiency. Multi-stage reconstruction with progressive validation ensures reliable data recovery even under adversarial conditions. The system coordinates defensive responses across all compression and reconstruction processes, implementing various recovery mechanisms when security violations are detected. This approach enables robust compression and reconstruction of time-series data while maintaining protection against various forms of adversarial manipulation.

The invention relates to a method and a system for storing and backing up electric power data of a data-in-data station based on use popularity, and belongs to the technical field of data-in-data stations, and the method comprises the following steps: classifying electric power business data into hot data, temperature data and cold data based on preset access frequency, time window and data volume index; according to a classification result and a business importance degree, storing hot data in an SSD cloud disk or a memory database, storing temperature data in a distributed column database, and compressing cold data and then archiving the cold data to an object storage or a tape library; dynamically configuring differentiated backup strategies for different types of data; automatic scheduling of storage and backup is realized through a cloud platform component, and hot switching of master and slave databases, OSS archived data recovery by a DataWorks tool and a remote multi-copy recovery mechanism are supported; and migrating data in advance in combination with a historical access prediction popularity trend, and triggering cold data compression and archiving when storage resources exceed 80%. The method is suitable for power industry scenes, the storage cost is optimized, and the data calling efficiency is improved.

The invention belongs to the technical field of deep learning model training, and particularly relates to a large model automatic multi-dimensional parallelization training system based on a heterogeneous GPU. The system comprises: (1) a 3D parallel composer which automatically generates a combination of data parallelism, tensor parallelism and pipeline parallelism to realize load balancing and efficient utilization of resources; (2) a parallel executor, in which a manager is placed, and communication overhead is reduced by optimizing physical mapping of a GPU (Graphics Processing Unit); the model divider is used for optimizing division of a model layer according to the performance of the GPUs and realizing load balance among different GPUs; the communication scheduler is used for realizing gradient synchronization among the asymmetric data parallel groups; and (3) a dynamic adapter which adopts a parameter-level check point recovery mechanism to quickly adjust and continue training when GPU resources are dynamically changed, so as to ensure the stability and continuity of the training process. Experimental results show that the method can effectively improve the training efficiency of the large model in the heterogeneous GPU environment.

The invention discloses an audit monitoring method and system based on multi-agent dynamic permission grading. The method comprises the following steps: constructing a trusted security basis; generating a dynamic token based on the authority control strategy, wherein the load of the dynamic token comprises an authority identifier of an atomic operation unit defined by the authority control strategy; micro-permission life cycle control is implemented, and when it is detected that any permission recovery triggering condition is met, a corresponding permission recovery operation is triggered; performing full-process auditing and anomaly detection, collecting full-link logs of the system, and performing four-layer collaborative detection; hierarchical response is implemented through a multi-level alarm and intervention mechanism, and optimization is carried out based on audit data. According to the method, a dynamic token, micro-permission atomization and dynamic recovery mechanism is introduced, the permission granularity is controlled in a minimum operation unit, and automatic real-time recovery of the permission is achieved. Through a four-layer cooperative detection model and in combination with a weighted voting decision-making mechanism, known threats, behavior abnormalities, complex attack chains and unknown threats are efficiently dealt with.

The invention relates to the technical field of software testing, and discloses an automatic test script generation method and system, a storage medium and electronic device.The method comprises the steps that test case information and tested environment information are received, and structured case representation is generated; enhancing the structured use case representation based on a preset knowledge base; generating a plurality of positioning candidates for the target control in the testing step, and establishing a priority relationship among the positioning candidates; generating an executable automatic test script; running the script, and when execution failure is detected, automatically adjusting a positioning strategy, a waiting strategy or a business process based on a failure type, and executing again; and feeding back the successfully executed script information to the knowledge base, and updating the content and priority of the knowledge base. Through organic combination of knowledge base enhancement, multi-candidate positioning and a self-healing mechanism, the technical problems that an existing automatic test script is high in maintenance cost, poor in stability and lack of domain knowledge are solved.

The invention relates to the technical field of data center high availability, in particular to a cross-data center fault isolation and switching method and system in a multi-tenant environment. The resource mapping table is constructed by taking the tenants as the minimum control units, the affected tenants are accurately identified, logic isolation is executed, the problem of waste of full-tenant service migration resources caused by node-level or cluster-level switching in the prior art is avoided, and the fault influence range is minimized. The optimal data center is dynamically selected by combining the multi-factor objective function with the tenant SLA level, the defects that an existing scheduling strategy is opaque and tenant priorities cannot be distinguished can be overcome, and it is ensured that key tenant services are preferentially recovered. And finally, through combination of a gray takeover mechanism and health feedback confirmation and multi-dimensional recovery verification after migration, the condition that an existing recovery mechanism is extensive can be changed, the fault processing precision and the resource scheduling efficiency of the data center in a multi-tenant environment are remarkably enhanced, and the service continuity is guaranteed.

The invention discloses a test method, device and equipment and a computer readable storage medium, which are applied to the technical field of computers, and comprise the following steps: obtaining multi-source data of a distributed storage cluster, the multi-source data comprising hardware state data, service associated operation data and business dependence data; constructing a relation dependency graph among the hardware nodes, the service nodes and the service nodes based on the multi-source data; adjusting the chaos test script according to the real-time state of the distributed storage cluster and the dynamic adjustment strategy to obtain a target chaos test script; and executing the target chaos test script, determining a fault range and a recovery mechanism based on the relation dependency graph, and outputting a test report. According to the method, comprehensive and accurate presentation of the dependency relationship of each node is realized, and the chaos test script can adapt to the dynamic change of the cluster, so that the test scene is more practical, the invalid test is avoided, the effectiveness, reliability and efficiency of the cluster chaos test are remarkably improved, and the high availability and service continuity of the cluster are powerfully guaranteed.

The invention relates to the technical field of network security awareness, in particular to a security awareness method and system for a GPU cluster, and the method comprises the steps: collecting multi-source security situation data, and providing basic information for subsequent security analysis; by constructing and updating a cluster security situation map in real time, information timeliness is updated and ensured in real time, and powerful support is provided for risk assessment; a dynamic risk assessment model is utilized to calculate vertex and edge scores, high-risk vertexes and key risk conduction paths are identified, and a basis is provided for resource isolation and scheduling; the targeted instruction is generated according to the security risk score and the preset security policy library, different security risks are effectively handled, and safe and stable operation of the cluster is guaranteed; by converting the instruction into a specific control command and distributing and executing the specific control command, effective implementation of security measures is ensured, and the cluster security protection capability is improved; by setting a finite-state machine monitoring instruction execution state, a recovery mechanism is automatically triggered, and continuous and safe operation of the cluster is guaranteed.

The invention relates to the technical field of unmanned ships, and discloses a novel wave-light complementary power generation-based multifunctional unmanned operation and maintenance ship, which comprises a ship body, an environment monitoring mechanism, an energy recovery mechanism, a full-rotation electric propulsion device, an unmanned aerial vehicle landing platform, an ROV automatic laying and recovery mechanism and an AUV automatic laying and recovery mechanism, the energy recovery mechanism comprises two clean energy power generation devices including wave energy power generation devices and solar power generation devices, the wave energy power generation devices are installed on the two sides of the ship body, the solar power generation devices are installed on a deck of the ship body, and clean energy is stored and redistributed through a storage battery in the cabin. An energy recovery mechanism of the device comprises wave energy power generation devices installed on the two sides of the ship body and a solar power generation device installed on the deck, comprehensive utilization of wave energy and solar energy can be achieved, the utilization rate of clean energy is increased, and the requirement for long endurance of the unmanned operation and maintenance ship is met.

The invention relates to the technical field of audio control, and discloses a distributed audio group mute and recovery method and system, a storage medium and a device, and the method comprises the steps: determining a master device in a distributed audio network, and enabling the master device to find and register a plurality of slave devices to establish an audio group; receiving a group mute or recovery instruction sent by a user through the master device, and performing validity verification on the instruction; the master device allocates a uniform timestamp to the verified instruction, and sends a synchronous control command to all slave devices in the group based on the timestamp; the slave equipment executes mute or recovery operation at the moment corresponding to the timestamp, synchronous response of the equipment in the group is realized, all the equipment is ensured to accurately and synchronously execute the mute / recovery operation through unified management of the master equipment and the adoption of global timestamp synchronization and adaptive delay prediction technologies, and an intelligent state monitoring and exception recovery mechanism is provided, so that the service life of the slave equipment is prolonged. Finally, efficient and reliable multi-room audio group control with excellent user experience is realized.

The invention discloses an edge service task rescheduling method based on HG-Trans, which comprises the following steps: S1, modeling a dynamic task scheduling problem into a Markov decision process, and converting a scheduling state into a heterogeneous graph structure; s2, inputting the heterogeneous graph structure into an HG-Trans network to obtain an estimated state value and probability distribution of all actions; and S3, inputting the estimated state value and the probability distribution of all actions into a decision network, and outputting a task scheduling strategy. And an edge fault scene task rescheduling strategy based on a heterogeneous graph neural network is utilized to comprehensively improve the task scheduling performance of an edge computing system in a fault scene. The problems that an existing scheduling method depends on local information or simple rules, lacks an efficient fault recovery mechanism and cannot quickly adjust a task execution path after a fault occurs are solved, and global optimal allocation of resources is achieved while scheduling delay is reduced.

The invention relates to a delivery recovery control method and device during delivery fault of a delivery robot and computer equipment, and the method comprises the steps: obtaining fault sensing information collected by a plurality of types of sensors, the fault sensing information comprising mechanical state information, power system operation information, navigation information and communication state information; when it is determined that a distribution fault exists according to the fault sensing information, fault classification is determined based on the fault sensing information; based on the fault sensing information, fault classification, a fault classification decision mechanism, a dynamic emergency response mechanism and a preset intelligent recovery protocol, distribution recovery control is carried out; the dynamic emergency response mechanism comprises the step of opening an anti-falling air bag of the conveying robot to support the goods under the condition that the goods falling risk is identified according to the mechanical state information. Through fault detection, cargo anti-falling emergency processing and a recovery mechanism, smooth completion of a distribution task can be ensured.

The invention discloses an account private key security storage and recovery system for a block chain development container, and relates to the technical field of block chains. The system comprises a private key format processing and verification module, an access control and authentication module, a private key encryption storage module and a security recovery and monitoring module. Through the above modules, the system can significantly improve the security, reliability and usability of private key management in a block chain development container through technical means such as multi-level encrypted storage, refined access control, an effective recovery mechanism and real-time monitoring on the basis of ensuring the security of the private key; and a more efficient and safer guarantee is provided for development and operation of the block chain application.

The invention discloses a BMC firmware starting method and device, equipment, a medium and a program product, and relates to the technical field of server management.The BMC firmware starting method comprises the steps that when a main mirror image file is damaged, BMC firmware is restarted on the basis of a standby mirror image file, if starting fails, the BMC firmware is restarted on the basis of a latest mirror image file of a remote server, and if starting fails, the BMC firmware is restarted on the basis of the latest mirror image file of the remote server; if the BMC firmware is started abnormally, the BMC firmware is restarted based on the local minimum mirror image file, so that the problem that a double-mirror image backup mechanism still faces system paralysis can be solved, a multi-stage mirror image file recovery mechanism is constructed through the multiple pre-deployed mirror image files, when the BMC firmware is started abnormally, the system is recovered step by step according to the multi-stage mirror image file recovery mechanism, and the system recovery efficiency is improved. The technical effects of ensuring the minimum availability of the system and improving the reliability and safety of the system are achieved.

The invention relates to the field of time capsule interaction, and discloses a time capsule interaction method and system, and the method comprises the steps: carrying out the logic unit segmentation of a to-be-sealed content, and building an encryption baseline which continuously evolves along with a time factor; under the support of an encryption baseline, performing fragmentation processing on the evolved and encrypted content, and triggering copy reconstruction through a witness mechanism when a node failure is detected; on the basis of a content storage link, continuously executing periodic health check in a storage period, and generating a state index in combination with a node online rate, key evolution integrity and user interaction history; when the key has an unrecoverable risk in the state index, triggering a multi-role collaborative recovery mechanism, and requiring a user side to submit a standby identity certificate; and after missing key reconstruction is completed, when preset opening time arrives, decryption operation is executed based on the key set which is finally evolved and completed and the recovered fragments, and an opening response with an interaction attribute is generated. The method has the advantage of improving the trust of the user.

The invention discloses a method for heterogeneous scheduling of computing power under a large model training reasoning framework, relates to the technical field of computing power scheduling, and solves the technical problem that task interruption or resource waste is easily caused by incomplete task migration, splitting and recovery mechanisms when computing power is insufficient or equipment is abnormal. According to the method, parameters such as the load rate, the residual computing power and the temperature of the CPU, the GPU and the FPGA are dynamically collected, the calculation amount, the duration and the delay requirement of the task are combined, a quantitative model is established, the resource utilization rate is increased, a temperature prediction model is established in stages, a dynamic threshold value is set in combination with the equipment type, the environment temperature and the health degree, temperature early warning and migration in the task execution process are achieved, and the task execution efficiency is improved. The frequency reduction risk caused by overheating of equipment is reduced, the task interruption loss is ensured to be reduced by more than 50% through task splitting, intelligent migration and pause recovery mechanisms aiming at insufficient computing power or equipment abnormity, and the stability of the system during peak flow or hardware failure is improved.

The utility model relates to the technical field of cooling tower heat recovery, in particular to a cooling tower heat recovery mechanism capable of rapidly dissipating heat. Comprising a recycling box, a box cover is arranged on the upper side of the recycling box, two supporting columns are arranged in the recycling box, a recycling pipe is arranged in the recycling box and penetrates through the interiors of the supporting columns, a partition plate is fixedly connected to the interior of the recycling box, an air inlet pipe is fixedly connected to one side of the recycling box, and an air outlet pipe is fixedly connected to the other side of the recycling box. According to the device, the recycling box is sleeved with the recycling pipe, water is injected into the recycling pipe, heat in the recycling box and water in the recycling pipe are subjected to heat exchange, heat in water vapor in the recycling box can be rapidly absorbed, the heat recycling efficiency is improved, the motor is started to drive the rotating shaft to rotate, and the heat in the recycling box is recycled. And therefore, the stirring blades are driven to rotate, the flow speed of gas in the recycling box is increased through the stirring blades, heat in the recycling box is fully absorbed by the recycling pipe, and heat waste is reduced.

The invention discloses a pre-aggregation storage table creation method, which relates to the technical field of data processing, and comprises the following steps: S1, dynamic rule generation: firstly, through a big data analysis system, carrying out deep analysis on a large amount of historical data in an original table, comprehensively mastering data type classification, and analyzing daily average data volume and data time window characteristics; the method comprises the following steps of S1, performing multi-level aggregation window division, S3, performing real-time feedback optimization, S4, performing distributed cooperative processing, S5, performing a self-adaptive compression mechanism, S6, performing a fault-tolerant recovery mechanism, S7, performing data quality monitoring, S8, performing metadata management and S9, performing external system butt joint. Through a dynamic rule generation real-time feedback optimization step and a self-adaptive compression mechanism step, the effects of data processing accuracy, high efficiency and storage cost reduction are achieved, and through a multi-stage aggregation window division distributed cooperative processing step and a fault-tolerant recovery mechanism step, the reliability of data processing is improved. And the effects of enhancing the complex data processing capability of the system, improving the reliability of the system and guaranteeing the service continuity are achieved.

The invention discloses a triplex nested logic-based park energy storage and load cooperative scheduling method, system and device and a medium, and belongs to the technical field of intelligent power grid and energy storage operation control, and the method comprises the steps: obtaining park operation data, and generating first scheduling prediction information; generating an executable power candidate scheme based on the prediction information and the real-time operation data; action frequency constraint processing is executed, and an action recovery mechanism is established and verified; cooling window control is implemented, and the execution time, the power change amplitude and the power change sequence of actions are limited; and then comprehensive judgment is carried out, a final scheduling instruction is generated, and scheduling parameters are adaptively corrected according to execution feedback. Through triple nested logic, the number of times of start and stop, direction switching and power climbing can be limited in real time under the conditions of high-proportion new energy access and rapid load fluctuation, a cooling window is actively planned, the action sequence is optimized, and thermal instability and voltage impact caused by frequent charging and discharging or continuous high-power operation are avoided.

The invention discloses an intelligent driving cooperative control terminal based on a visual language action model, which comprises an instruction optimization processing unit, an execution management analysis unit and a dynamic adaptive cooperative processing unit, relates to the technical field of intelligent cooperative control, and solves the problem of how to realize strategy dynamic switching based on real-time scene classification. In order to solve the technical problem of improving the self-adaptive capability of the system to complex traffic scenes, according to the invention, through a dynamic self-adaptive cooperative processing unit, coherent actions are disassembled and then conflict analysis is carried out, the grading standards of emergency actions and non-emergency actions are determined, and a pause-recovery mechanism is adopted for the non-emergency actions, so that the self-adaptive capability of the system to the complex traffic scenes is improved. The method comprises the following steps: solving conflicts for emergency actions through parameter adjustment or equivalent security substitution, improving action execution coherence and security, realizing scene fine-grained classification based on a multi-modal fusion model, triggering strategy switching through scene confidence, and starting corresponding strategies for a composite scene after confidence of all sub-scenes reaches the standard.

Provided is a catching device for surface rescue based on an aircraft platform, and a surface rescue method. The catching device for surface rescue based on the aircraft platform includes a throwing mechanism, a flying mechanism, a recovery mechanism, an automatic control module, and a catching net. Both the throwing mechanism and the recovery mechanism are disposed on the flying mechanism, and the automatic control module automatically controls the throwing mechanism and the recovery mechanism. After the flying mechanism flies above a drowning person, the throwing mechanism is configured to throw the catching net towards the drowning person. After the catching net nets the drowning person, the recovery mechanism can drive a bottom of the catching net to close and lift the catching net upwards to approach the flying mechanism, and the flying mechanism is configured to drag the drowning person to a safe location.

The invention belongs to the technical field of gas filling, and particularly relates to an automatic filling device for medical high-purity gas production, which comprises a three-axis movable frame, a mounting part, a filling mechanism, a protection mechanism, a recovery mechanism, a filtering mechanism and a capturing mechanism, the mounting part comprises a mounting box and a water tank mounted at the bottom of the mounting box, and the mounting box is controlled to move through a three-axis movable frame; the filling mechanism comprises a connecting nozzle and a conveying pipe connected with an external air source, and the connecting nozzle is communicated with an inner cavity of the conveying pipe; the protection mechanism comprises a nozzle and a control pump, and the control pump pumps liquid in the water tank and discharges the liquid to the joint of the connecting nozzle and the gas tank through the nozzle. The recycling mechanism comprises a collecting hopper and a connecting pipe, and the collecting hopper is located under the connecting nozzle; through cooperation of the structure, water flow continuously absorbs low-temperature heat, frosting blockage caused by sudden temperature drop inside and outside the mouthpiece can be prevented, static electricity can be conducted away through the electric conduction characteristic of the water film, and the hidden danger of burning explosion of combustible gas is reduced.

The invention discloses a basic input and output system protection method, device and equipment and a readable medium, which are applied to the technical field of computers, and comprise the following steps: monitoring a power-on pin and return information of a basic input and output system, and determining a starting process state of the basic input and output system; if the state of the starting process is abnormal, performing a read-write test on a storage medium of the basic input and output system, and determining an abnormal reason; if the exception reason is firmware data exception, recovering the basic input and output system through the storage medium of the double basic input and output system and the substrate management controller. According to the invention, various states of the basic input and output system in the starting process can be identified; and in the case of exception, automatically determining the reason for starting exception of the basic input / output system through a read-write test, and realizing an automatic recovery mechanism when firmware data exception occurs in the basic input / output system. And the protection capability and the abnormity self-repairing capability of the basic input and output system are integrally improved.

The invention discloses a computer multi-cloud data security backup method and system based on scene security, and the method comprises the steps: carrying out the risk assessment based on scene parameters, and formulating a security backup strategy; the method comprises the following steps: processing cloud data through adaptive fragmentation encryption, and dividing the cloud data into core data and non-core data; based on the relevance between the fragmented data, adding an identifier; performing evaluation based on the security index data of the multiple cloud storage nodes of each cloud storage platform to obtain the cloud storage platform meeting the storage requirement; through intelligent routing distribution, the encrypted data fragments are distributed and stored in the heterogeneous cloud storage platform, and it is ensured that the fragments stored in any single cloud platform are insufficient to reconstruct complete data; and constructing a multi-factor data recovery mechanism based on a data exception condition, and implementing dynamic migration on the fragments of the high-risk region. The method has the advantages that the flexibility and recovery capability of data backup are effectively improved through adaptive fragmentation encryption and intelligent routing distribution, and potential security threats are avoided.

In one embodiment, an apparatus includes: first and second cores to execute instructions, and an interface circuit coupled to the first and second cores. In a lockstep mode in which the first and second cores are to execute redundantly, the interface circuit is to identify a miscompare between the first core and the second core that is due to a corrected error in one of the first core or the second core, and indicate the miscompare as a recoverable error. Other embodiments are described and claimed.

The invention discloses a service application unified data exchange mechanism and dynamic scheduling fault tolerance method and system, and relates to the technical field of service scheduling. The service application unified data exchange mechanism and dynamic scheduling fault tolerance method comprises a starting service scheduling module, a task management module, a configuration center module, a communication management module, a state monitoring module and an exception management module. The task management module receives the task configuration file, analyzes a service dependency relationship in the task configuration file, generates an execution node of the task, and promotes execution of the task in a data stream mode; and the task management module distributes a binary executable file required by the task according to the analyzed task configuration. According to the method, the problems that a current task scheduling system generally lacks a real-time monitoring and response mechanism for a service state, and service abnormity is difficult to process in time, so that the system stability is reduced; and in addition, the system fault-tolerant capability is insufficient, a rapid migration and recovery mechanism is lacked when a service fault occurs, and the system availability is low.

The invention discloses a fault self-check recovery method and system based on a cloud deployment system, and the method comprises the steps: carrying out the multi-dimensional deep detection of the operation states of an operation system layer, a database layer and an application layer of a cloud platform through constructing an all-dimensional real-time monitoring system, and automatically triggering an adaptive recovery mechanism once an abnormality is found, and continuous and stable operation of the system is ensured. The matching tool integrates core modules of state acquisition, anomaly analysis, recovery execution, log recording and the like, collaboratively realizes real-time acquisition of operation data, accurate anomaly identification through an intelligent algorithm, rapid matching and implementation of a recovery scheme, and records logs in the whole process for troubleshooting and system optimization. According to the tool, server nodes, database states, service accessibility and webpage availability are checked in sequence, emergency processing is started immediately when abnormity is found, the stability and reliability of a cloud platform are remarkably improved, and the manual intervention cost is reduced.

The invention discloses a decentralized domain event publishing and subscribing method and system. The invention relates to the technical field of decentralized services. Based on a Gossip protocol and a PAFD algorithm, service instance cluster nodes are kept in state synchronization; an event storage structure for holding event data and consumption status is initialized. And after the service instance receives the business operation request, generating a field event and publishing the field event to a local event storage. According to the invention, the publishing and subscribing service is built in the micro-service instance cluster, so that close integration of business logic and message service is realized, and extra framework service deployment resources are not needed. Meanwhile, the D-S evidence theory not only provides a method for processing uncertain information, but also provides intelligent support for the decision-making process. In fault detection, the system can automatically make decisions according to fused information, such as isolating fault nodes, triggering a recovery mechanism and the like, so that the decision efficiency and reliability are improved.

The invention discloses a federated learning privacy protection and attack defense method, which relates to the technical field of artificial intelligence and information privacy security crossing, and comprises the following steps: a client performs training based on local data and generates update, dynamically allocates privacy budget by evaluating the contribution degree of each network layer to model performance, and finally performs privacy protection and attack defense. The method comprises the following steps: reducing noise at a key layer to retain the utility of a model, enhancing disturbance at a non-key layer to improve privacy protection, uploading an update after noise addition to a server, predicting an update to be submitted by a client based on a historical training process by the server, calculating a deviation value between an actual update and a predicted update, and constructing a behavior evaluation value by combining multiple rounds of deviations. The method comprises the following steps: identifying and isolating abnormal clients by utilizing clustering analysis, dividing normal clients into fixed groups for intra-group aggregation, periodically executing global aggregation to update a model, starting an intra-group recovery mechanism when abnormality is detected, and performing local retraining and reconstruction aggregation output by the normal clients in the same group, thereby realizing safe and efficient federated learning.

The invention relates to the technical field of cloud storage data, in particular to a dynamic hierarchical self-healing method and system for cloud storage data, and the system comprises a data collection and access module, a data preprocessing and feature engineering module, an intelligent prediction and decision module, a dynamic hierarchical storage execution module, and a fault self-healing and load balancing module. According to the method, the health state of the storage system is continuously monitored through the Kafka real-time data stream, the data access trend is predicted in combination with the LSTM model, and potential faults are warned in advance; the node state is judged through indexes such as TCP timeout and heartbeat loss rate, redistribution or storage hierarchy adjustment of affected data is triggered, fault non-inductive repair is achieved through a self-healing mechanism, and service continuity is guaranteed; through deep linkage of hierarchical storage and a fault recovery mechanism, the storage hierarchy of the data is actively adjusted according to the change of the data integrity in the recovery process, it is ensured that the optimal storage scheduling can be obtained while the data is recovered, and the self-healing capacity of the system is improved.