A child health care data storage protection method and system based on federated learning

Abstract

The present application relates to the technical field of data security access, in particular to a child health care data storage protection method and system based on federated learning, comprising obtaining a distributed attribute authorization data set, the attribute authorization data set including an attribute definition data set, a policy management data set and a key distribution data set; receiving an authorization instruction through the policy management data set, and generating an access control policy according to the authorization instruction; when a federated learning platform initiates a data access request, creating an isolated cloud computing sandbox environment for a federated learning task according to the access control policy; dynamically calculating and using a differential privacy parameter matched with a use purpose, and applying the parameter to child health care data within the authorized range; executing the federated learning task in the cloud computing sandbox environment, monitoring and limiting data processing operations; providing a permission dynamic adjustment mechanism, when modifying the authorized range and / or revoking the authorization, updating the access control policy to take effect immediately, and pushing a permission change instruction to all computing nodes.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data security access, in particular to a child health data storage protection method and system based on federated learning. BACKGROUND

[0002] With the in-depth application of information technology in the field of medical health, the electronic storage and analysis of child health data has become an important support for improving the level of child health services and promoting related medical research. Child health data is protected by platform data access, such as using keys or access control rules, which is the core technology direction of current data security protection. In the prior art, mechanisms such as access control list, role-based access control, and more flexible attribute-based access control are generally used, combined with encryption technology, to protect data within a single institution or trust domain. These mechanisms can ensure compliance use of data within the authorized range to some extent.

[0003] However, in the field of child health care, the multi-source nature of data and the demand for collaborative sharing are increasingly prominent. For example, multiple medical institutions, research institutions and public health departments may need to jointly use child health data distributed everywhere to conduct large-scale epidemiological research, rare disease analysis or public health policy making. Traditional centralized data sharing mode is difficult to implement due to the risk of privacy leakage and compliance barriers. Federated learning, as an emerging distributed machine learning technology, allows each participant to jointly train the model without exposing local raw data, providing a new way to solve the data island problem and privacy protection dilemma. Although federated learning replaces direct sharing of raw data with exchange of model parameters, thus protecting data privacy at the macro level, the combination of existing access control platforms and federated learning frameworks still has technical problems in specific child health data application scenarios.

[0004] One of the significant technical problems is the complexity of cross-domain identity authentication and dynamic authorization. How to establish a trusted cross-domain identity mutual recognition mechanism while ensuring the autonomy of each domain, and according to the changing authorization wishes of child data subjects, the changes in rights brought by the growth of children, and the dynamic needs of specific research projects for data range. Another core technical problem is the difficulty of implementing the data minimization principle and informed consent in the federated environment.

[0005] Therefore, a child health data storage protection method and system based on federated learning are proposed. SUMMARY

[0006] The purpose of the present application is to provide a child health care data storage protection method and system based on federated learning, by obtaining a distributed attribute authorization dataset, the attribute authorization dataset includes attribute definition dataset, policy management dataset and key distribution dataset; receive authorization instructions through the policy management dataset, generate access control policies according to the authorization instructions; when the federated learning platform initiates a data access request, create an isolated cloud computing sandbox environment for the federated learning task according to the access control policy; dynamically calculate and use the differential privacy parameters matched with the purpose, and apply the parameters to the child health care data within the authorized range; execute the federated learning task in the cloud computing sandbox environment, monitor and limit the data processing operation; provide a dynamic permission adjustment mechanism, when modifying the authorized range and / or revoking the authorization, update the access control policy immediately, and push the permission change instruction to all computing nodes.

[0007] To achieve the above purpose, the present application provides the following technical solutions:

[0008] A child health care data storage protection method based on federated learning, comprising:

[0009] Obtain a distributed attribute authorization dataset, the attribute authorization dataset includes attribute definition dataset, policy management dataset and key distribution dataset;

[0010] Receive authorization instructions through the policy management dataset, generate access control policies according to the authorization instructions;

[0011] When the federated learning platform initiates a data access request, create an isolated cloud computing sandbox environment for the federated learning task according to the access control policy; dynamically calculate and use the differential privacy parameters matched with the purpose, and apply the parameters to the child health care data within the authorized range; execute the federated learning task in the cloud computing sandbox environment, monitor and limit the data processing operation;

[0012] Provide a dynamic permission adjustment mechanism, when modifying the authorized range and / or revoking the authorization, update the access control policy, and push the permission change instruction to all computing nodes, forcibly terminate the unfinished computing tasks beyond the authorized range.

[0013] Preferably, the attribute definition dataset is used to define the attribute set related to the child health care data, and the attribute set at least includes data type attribute, use purpose attribute, timeliness attribute and sensitivity attribute;

[0014] The access control policy includes the data range, use purpose, valid time and minimization requirement allowed to be accessed by the federated learning task.

[0015] Preferably, the step of creating an isolated cloud computing sandbox environment for the federated learning task specifically comprises: generating a temporary session key based on the key distribution dataset and assigning the temporary session key to the authorized federated learning task; building a cloud computing sandbox environment in a physically isolated virtual container, which has an independent file system, network namespace and process space; deploying a resource monitoring agent in the virtual container to monitor the occupation of computing resources in real time and set an upper limit on resource usage; assigning a unique identifier to the cloud computing sandbox environment and associating it with the corresponding access control policy.

[0016] Preferably, the step of dynamically calculating and using the differential privacy parameter matched with the use purpose specifically comprises:

[0017] Extracting the sensitivity attribute value related to the current authorized data from the attribute definition dataset; querying the preset differential privacy parameter mapping table based on the data type attribute and the use purpose attribute to obtain the initial privacy budget and the sensitivity;

[0018] According to the sensitivity attribute value of the child health care data, the initial privacy budget is dynamically adjusted; a privacy parameter decay mechanism is set based on the timeliness attribute to dynamically adjust the differential privacy parameter; differential privacy parameters are set based on different stages of the federated learning task, and different noise addition mechanisms are used for model training and aggregation stages;

[0019] Tracking the privacy budget consumption in real time, and automatically increasing the privacy protection strength and terminating data access when the cumulative consumption exceeds the preset threshold.

[0020] Preferably, the step of executing the federated learning task in the cloud computing sandbox environment specifically comprises:

[0021] Deploying a data access audit module to record all data operation behaviors, including data reading time, range, processing method and result output; implementing an instruction-level monitoring mechanism to verify the authority of each execution instruction of the federated learning algorithm and blocking unauthorized data processing operations; setting a data outflow control point to ensure that the processing result only contains statistical features and / or model parameters within the authorized range, and prohibiting the outflow of raw data;

[0022] Verifying the calculation process of the participant through zero-knowledge proof technology to conform to the access control policy; establishing an abnormal behavior detection mechanism to automatically suspend the calculation task and issue an alarm to the security management module when potential unauthorized data access is detected.

[0023] Preferably, the step of providing a dynamic adjustment mechanism for authority specifically comprises:

[0024] Building an authority change event listener to receive the authorization instructions issued by the guardian through the strategy management dataset in real time; performing digital signature verification on the authorization instructions;

[0025] The permission change instruction is converted into a structured access control policy update instruction, including permission type, change range, effective time and priority; based on a distributed consistency protocol, the permission change is ensured to be synchronized and effective on all computing nodes;

[0026] The permission re-evaluation is performed on the executing federated learning task, the computing task exceeding the new authorized range is identified; the task interruption signal is sent to the affected cloud computing sandbox environment, triggering the security cleaning process, including memory data erasure, temporary file deletion and computing state saving.

[0027] A child health care data storage protection system based on federated learning, comprising:

[0028] A data acquisition module is configured to acquire a distributed attribute authorization dataset, wherein the attribute authorization dataset comprises an attribute definition dataset, a policy management dataset and a key distribution dataset;

[0029] An access policy control module is configured to receive an authorization instruction through the policy management dataset and generate an access control policy according to the authorization instruction;

[0030] An access judgment module is configured to create an isolated cloud computing sandbox environment for a federated learning task according to an access control policy when a federated learning platform initiates a data access request; dynamically calculate and use a differential privacy parameter matched with a use purpose, and apply the parameter to child health care data within an authorized range; and execute a federated learning task in the cloud computing sandbox environment, monitor and limit data processing operations;

[0031] An access restriction module is configured to provide a permission dynamic adjustment mechanism, update an access control policy when modifying an authorized range and / or revoking an authorization, and push a permission change instruction to all computing nodes to forcibly terminate an unfinished computing task exceeding the authorized range.

[0032] Compared with the prior art, the present application has the following advantages:

[0033] 1、The present application constructs three datasets of attribute definition, policy management and key distribution, and automatically generates an isolated cloud computing sandbox environment according to an access control policy when a federated learning task is initiated. The virtual container has an independent file system, network namespace and process space, internally deploys a resource monitoring agent and an instruction-level permission verification mechanism, can track each data reading, processing and output operation in real time, and immediately intercepts unauthorized instructions. At the same time, by assigning a unique identifier to each sandbox and associating it with an access policy, strict isolation between different tasks is ensured, avoiding the risks of "over-authorization" and "permission boundary crossing" under traditional centralized or role-based control, significantly improving the isolation degree and security level of data storage and computing process.

[0034] 2、For the high sensitivity of children's health data, the application acquires the initial privacy budget and sensitivity value from the preset differential privacy parameter mapping table before the federated learning task runs according to the attribute definition of the sensitivity attribute and the use purpose attribute in the data set; combined with the timeliness attribute, a decay mechanism is dynamically introduced, and the noise intensity is automatically adjusted as the data use period moves; different noise addition mechanisms are used for the two stages of model training and parameter aggregation, respectively, to ensure the model efficiency while maximizing privacy protection.

[0035] 3、The application integrates a permission change event listener, which can receive authorization change or revocation instructions issued by guardians through a policy management data set in real time, and ensures that the instruction source is legal through digital signature verification. Under the guarantee of a distributed consistency protocol, the updated access control policy is immediately issued to all computing nodes, and the permission of the federated learning task being executed is re-evaluated to identify computing processes that exceed the new authorized range. For unfinished and unauthorized tasks, send an interrupt signal to trigger a security cleaning process. At the same time, combined with the zero-knowledge proof and abnormal behavior detection module, the computing process of each party is verified and monitored to ensure that any subsequent access strictly follows the latest permission configuration, realizing a closed-loop management from authorization change to forced execution. BRIEF DESCRIPTION OF DRAWINGS

[0036] Figure 1 A flowchart of a child health care data storage protection method based on federated learning provided by the application;

[0037] Figure 2 A structural diagram of a child health care data storage protection system based on federated learning provided by the application;

[0038] Figure 3 A structural diagram of data storage protection provided by an embodiment of the application. DETAILED DESCRIPTION

[0039] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the application.

[0040] The application provides a child health care data storage protection method based on federated learning, which is applied to a child health care data storage protection system based on federated learning. The flowcharts of the specific method and system are referred to Figure 1 and Figure 2 .

[0041] Embodiment one

[0042] Referring to Figure 1 The application provides a child health care data storage protection method based on federated learning, and the technical solution is as follows: a distributed attribute authorization dataset is obtained, the attribute authorization dataset includes an attribute definition dataset, a policy management dataset, and a key distribution dataset;

[0043] An authorization instruction is received through the policy management dataset, and an access control policy is generated according to the authorization instruction;

[0044] When a federated learning platform initiates a data access request, a segregated cloud computing sandbox environment is created for a federated learning task according to the access control policy; a differential privacy parameter matched with a use purpose is dynamically calculated and applied to child health care data within the authorized range; the federated learning task is executed in the cloud computing sandbox environment, and data processing operations are monitored and limited;

[0045] A dynamic permission adjustment mechanism is provided, when the authorized range is modified and / or the authorization is revoked, the access control policy is updated and takes effect immediately, and permission change instructions are pushed to all computing nodes to forcibly terminate incomplete computing tasks that exceed the new authorized range.

[0046] Specifically, the attribute definition dataset is used to define an attribute set related to child health care data, and the attribute set at least includes a data type attribute, a use purpose attribute, a timeliness attribute, and a sensitivity attribute;

[0047] The access control policy includes a data range allowed to be accessed by a federated learning task, a use purpose, a valid time, and a minimization requirement.

[0048] In this embodiment, the attributes of child health care data are finely defined, covering dimensions such as data type, use purpose, timeliness, and sensitivity, and these attributes are mapped to the access control policy of the federated learning task, so that the system can explicitly determine the accessible data range, use purpose, valid time limit, and minimization processing requirement at the beginning of task initiation, and fundamentally guarantee the legality and compliance of data use and the minimization principle. Not only does it realize the accurate classification and hierarchical management of various child health care information, dynamically adapt to the privacy protection needs of different research or service scenarios, but also effectively curbs the risk of data abuse and unauthorized access through the enforcement of valid time and minimization requirements, providing solid technical support for the safe sharing and long-term controllable use of child health care data in the federated learning process.

[0049] Further, the step of creating an isolated cloud computing sandbox environment for the federated learning task specifically comprises: generating a temporary session key based on the key distribution dataset and assigning the temporary session key to the authorized federated learning task; building a cloud computing sandbox environment in a physically isolated virtual container, which has an independent file system, network namespace and process space; deploying a resource monitoring agent in the virtual container to monitor the occupation of computing resources in real time and set an upper limit on resource usage; assigning a unique identifier to the cloud computing sandbox environment and associating it with the corresponding access control policy.

[0050] In the present embodiment, by generating and assigning a temporary session key to each federated learning task based on the key distribution dataset, strong isolation and dynamic authorization of data access are achieved, effectively reducing the risk of long-term key leakage; by building a cloud computing sandbox environment in a physically isolated virtual container and deploying a resource monitoring agent, not only is the safe operation of each task in an independent file system, network namespace and process space guaranteed, but real-time monitoring and limitation of computing resource consumption are also achieved to prevent resource abuse or cross-border access; by assigning a unique identifier to each sandbox environment and corresponding to its access control policy, not only is the traceability and auditability of the system enhanced, but also the task that does not comply with the new policy can be quickly locked and terminated when the authorization changes, comprehensively improving the security and controllability of child health data in the whole process of federated learning.

[0051] Further, the step of dynamically calculating and using a differentially private parameter matching the purpose specifically comprises:

[0052] Extracting the sensitivity attribute value related to the currently authorized data from the attribute definition dataset; based on the data type attribute and the purpose of use attribute, querying the preset differentially private parameter mapping table to obtain the initial privacy budget and sensitivity;

[0053] Based on the sensitivity attribute value of the child health data, the initial privacy budget is dynamically adjusted; based on the timeliness attribute, a privacy parameter decay mechanism is set to dynamically adjust the differentially private parameter; based on the different stages of the federated learning task, different privacy parameters are set, and different noise addition mechanisms are adopted for the model training and aggregation stages;

[0054] Tracking the consumption of the privacy budget in real time, and when the cumulative consumption exceeds the preset threshold, automatically increasing the privacy protection strength and terminating the data access.

[0055] In this embodiment, the sensitivity attribute is dynamically extracted, and the initial privacy budget is obtained from the preset mapping table in combination with the data type and the use purpose. During the operation, the budget is automatically adjusted according to the sensitivity attribute of the child health data, and the privacy parameter decay mechanism is introduced according to the timeliness attribute, so that the real-time dynamic adjustment of the differential privacy strength is realized. At the same time, different noise addition strategies are adopted for the model training and aggregation stages of the federated learning, which takes into account the data utility and strengthens the privacy protection. The system also tracks the privacy budget consumption in real time, and when the cumulative consumption reaches the threshold, it can automatically increase the privacy strength or even terminate the access, ensuring that the minimum exposure and dynamic defense principles are always followed in the whole process, greatly improving the security, compliance and usability of child health data on the federated learning platform.

[0056] Further, the step of performing a federated learning task in the cloud computing sandbox environment specifically comprises:

[0057] A data access audit module is deployed to record all data operation behaviors, including data reading time, range, processing method and result output; a command-level monitoring mechanism is implemented to verify the authority of each execution instruction of the federated learning algorithm, and unauthorized data processing operations are blocked; a data outflow control point is set to ensure that the processing result only contains authorized statistical features and / or model parameters, and raw data outflow is prohibited;

[0058] The computing process of the participant is verified by zero-knowledge proof technology to comply with the access control policy; an abnormal behavior detection mechanism is established, and when potential unauthorized access to data is detected, the computing task is automatically suspended and an alarm is sent to the security management module.

[0059] In this embodiment, by deploying data access audit and command-level monitoring in the sandbox environment, each data reading and processing operation can be recorded and verified in real time, and unauthorized instruction execution is strictly prevented; through the data outflow control point, only authorized statistical features or model parameters are allowed to be output, and raw data leakage is prevented; combined with the zero-knowledge proof technology, the compliance of the participant's computing process is verified to ensure that each node strictly follows the access control policy; and when potential unauthorized behavior is detected, the task is automatically suspended and an alarm is sent, realizing full-process and multi-level dynamic protection and traceable audit, greatly improving the data security and trustworthiness in the federated learning process.

[0060] Further, the step of providing a dynamic permission adjustment mechanism specifically comprises:

[0061] A permission change event listener is constructed to receive the authorization instructions issued by the guardian through the strategy management dataset in real time; the authorization instructions are digitally signed and verified;

[0062] The permission change instruction is converted into a structured access control policy update instruction, including permission type, change range, effective time and priority; based on a distributed consistency protocol, the permission change is ensured to be synchronized and effective on all computing nodes;

[0063] The ongoing federated learning task is re-evaluated for permission, and the computing task that exceeds the new authorized range is identified; a task interruption signal is sent to the affected cloud computing sandbox environment, triggering a security cleaning process, including memory data erasure, temporary file deletion and computing state saving.

[0064] In this embodiment, by constructing a permission change event listener and digitally signing the authorization change instruction issued by the guardian, the system can capture and confirm the legality of the change source in real time; after converting the change instruction into a structured policy update, the distributed consistency protocol is used to ensure that each computing node is synchronized and effective, so that the policy remains consistent in any network or node failure; after updating, the running federated learning task is re-evaluated for permission, and once the task exceeds the new authorized range, an interruption signal is immediately sent to the corresponding sandbox environment and a security cleaning process (including memory erasure, temporary file deletion and computing state saving) is started, realizing the immediate blocking and subsequent audit of unauthorized computing, and fully enhancing the real-time controllability of data access and the safety and reliability of the system.

[0065] This invention achieves efficient, secure, and controllable sharing of children's health care data throughout the federated learning process through multiple means, including refined attribute definition, policy generation, sandbox isolation, key distribution, differential privacy, and dynamic permission adjustment. First, refined modeling is performed on attributes such as data type, purpose of use, timeliness, and sensitivity. Based on this, control policies including access scope, purpose, validity period, and minimization requirements are generated to ensure that data calls begin and end in compliance. Temporary session keys generated based on the key distribution dataset and independent virtual container sandboxes not only prevent long-term key leakage and resource overruns between multiple tasks, but also monitor and limit resource usage within the container in real time. Dynamic differential privacy parameter calculation and budget decay mechanisms apply differentiated noise at different stages of model training and aggregation, ensuring both data utility and enhanced privacy protection. When the budget is exhausted, the intensity can be automatically increased or access terminated. Data access auditing, instruction-level monitoring, and data outflow control deployed within the sandbox, combined with zero-knowledge proofs and anomaly detection, block all unauthorized operations and ensure that only statistical features or model parameters are output. Based on a distributed consensus protocol and digital signature verification permission change monitoring and push mechanism, after the guardian adjusts the authorization, it can be instantly synchronized to each node, interrupting out-of-scope tasks in real time and performing memory and file cleanup. The above design not only significantly reduces the risks of chain breakage and key leakage, but also significantly improves the system's security, reliability, compliance, and operational efficiency through end-to-end traceability and the principle of minimum exposure, laying a solid foundation for the long-term trusted sharing of sensitive children's health care data.

[0066] Example 2

[0067] This invention provides a child healthcare data storage and protection method based on federated learning, applied to a child healthcare data storage and protection system based on federated learning, as detailed below. Figure 2 This example demonstrates a specific application scenario for the secure sharing of children's growth and development data. In this scenario, multiple hospitals, community health centers, and educational institutions need to jointly analyze children's physical development data, vaccination records, nutritional status assessments, and learning ability test results. The specific process is detailed below. Figure 3 The specific technical solution is as follows: a data acquisition module is used to acquire a distributed attribute authorization dataset, which includes an attribute definition dataset, a policy management dataset, and a key distribution dataset;

[0068] The access policy control module is used to receive authorization instructions through the policy management dataset and generate access control policies based on the authorization instructions.

[0069] An access judgment module is configured to create an isolated cloud computing sandbox environment for a federated learning task according to an access control policy when a federated learning platform initiates a data access request; dynamically calculate a differential privacy parameter matched with a use purpose, and apply the parameter to child healthcare data within an authorized range; and execute the federated learning task in the cloud computing sandbox environment, monitor and limit data processing operations.

[0070] An access restriction module is configured to provide a permission dynamic adjustment mechanism, update the access control policy when modifying an authorized range and / or revoking authorization, and push permission change instructions to all computing nodes to forcibly terminate an unfinished computing task that exceeds the authorized range.

[0071] The system first acquires a distributed attribute authorization dataset through a data acquisition module. For the child growth and development scenario, the attribute definition dataset includes the following key attributes: a data type attribute, a use purpose attribute, a timeliness attribute, and a sensitivity attribute; the data type attribute includes height and weight data, immunization records, nutrition intake evaluation, cognitive development evaluation, and the like; the use purpose attribute includes growth trend research, nutrition intervention effect evaluation, regional development difference analysis, and the like; the timeliness attribute includes data available duration and research cycle limitation; and the sensitivity attribute is used to divide general development data and high-sensitivity data such as special disease or psychological evaluation.

[0072] The strategy management dataset is configured to receive and manage authorization instructions from a child guardian. In actual application, a joint research project of multiple hospitals, community health centers, and educational institutions needs to obtain explicit authorization of a child guardian on a specific data use range and purpose. The guardian may allow the use of height and weight data of his / her child for regional growth curve research, but limit the use range of nutrition status and learning ability evaluation. The system receives these fine-grained authorization instructions through the strategy management dataset, and generates an access control policy including a data range, a use purpose, a valid time, and a minimization requirement according to the authorization instructions. According to the sensitivity attribute in the attribute definition dataset, the sensitivity of child healthcare data is graded in combination with objective dimensions such as data type, use purpose, and timeliness, and is quantitatively evaluated through a preset differential privacy parameter mapping table and a dynamic adjustment mechanism.

[0073] When the federated learning platform initiates a data access request, the access judgment module first verifies the identity of the request subject, and confirms that the request subject has a legal research qualification; and then generates a temporary session key for the specific research task based on the key distribution dataset. In this scenario, the system allocates a unique and time-limited temporary session key for each authorized federated learning task (such as a “regional child growth and development curve research” project), instead of a traditional long-term static key mode, which significantly reduces the risk of key leakage.

[0074] The access judgment module then builds a cloud computing sandbox environment in a physically isolated virtual container. This virtual container has an independent file system (containing only authorized child development data), a network namespace (limiting communication with external systems), and a process space (preventing cross-process data leakage). For child physical development data analysis tasks for different medical institutions, the system creates multiple isolated containers, ensuring that institution A cannot access institution B's raw data. At the same time, the system deploys a resource monitoring agent inside the container to monitor real-time computing resource usage, set CPU, memory, and network bandwidth usage limits, and prevent resource abuse or denial-of-service attacks. Each sandbox environment is assigned a unique identifier and associated with a corresponding access control policy, facilitating subsequent audit tracking.

[0075] Due to the special sensitivity of children's health data, the access judgment module dynamically calculates and uses differential privacy parameters that match the purpose before performing federated learning tasks. In specific implementation, the system extracts the sensitivity attribute value of the current authorized data from the attribute definition data set; based on the data type (development indicators) and the purpose of use (regional difference analysis), it queries the pre-set differential privacy parameter mapping table to obtain the initial privacy budget and sensitivity benchmark value. For high-sensitivity data (such as psychological development abnormality assessment), the system will dynamically reduce the privacy budget and increase the protection strength; based on the timeliness attribute, such as the guardian's authorization period, a privacy parameter decay mechanism is set to automatically increase the noise level over time. Different noise addition mechanisms are used in the model training and parameter aggregation stages of federated learning, such as Gaussian noise in the training stage and Laplace noise in the aggregation stage, to ensure model convergence efficiency while improving privacy protection effect. The system also tracks privacy budget consumption in real time, and when the cumulative consumption of certain feature data, such as cognitive ability data for a specific age group, exceeds the pre-set threshold, it automatically increases the privacy protection strength of this type of data or completely terminates its access rights.

[0076] During the execution of federated learning tasks in the sandbox environment, the system deploys a data access audit module to record all data operation behaviors. For example, it records the reading time, query range, statistical processing method, and result output of researchers on 5-7 year-old children's nutrition status data; implements instruction-level monitoring mechanisms to verify the authority of each execution instruction of the federated learning algorithm, such as immediately blocking attempts to extract unauthorized child family background information; sets data outflow control points to ensure that output results only contain authorized statistical features such as height standard deviation, body weight distribution percentile, or model parameters, and strictly prohibit the outflow of raw sensitive data.

[0077] The system verifies the calculation process of each participating institution by zero-knowledge proof technology to comply with the access control policy, for example, proving that a hospital does not exceed the authorized query range and calculation logic when processing children's nutritional status data, and does not save intermediate results; at the same time, an abnormal behavior detection mechanism is established, when potential data unauthorized access is detected, such as attempting to associate unauthorized learning ability data with nutritional status, the calculation task is suspended and an alarm is sent to the security management module.

[0078] In view of the high frequency characteristics of the authorization change of children's health data, the access restriction module provides a powerful dynamic adjustment mechanism for the right. The system constructs a right change event listener and receives the authorization instructions issued by the guardian through the strategy management dataset in real time, such as a guardian deciding to revoke the use authorization of the learning ability evaluation data of his or her children; the system verifies the digital signature of the authorization instruction to ensure that the change comes from the authorized guardian rather than an attacker pretending to be the guardian. Then, the right change instruction is converted into a structured access control policy update instruction, including the right type including reading, analysis, export, etc., the change range including a specific data subset, the effective time including immediate or timed effectiveness, and the priority including emergency revocation or regular adjustment; based on the distributed consensus protocol (such as Raft or Paxos), the right change is ensured to be synchronized on all computing nodes to avoid security vulnerabilities caused by inconsistent strategies between nodes.

[0079] The system re-evaluates the rights of the ongoing federated learning tasks, such as the ongoing children's development pattern analysis, and identifies the calculation tasks that exceed the new authorization range; sends a task interruption signal to the affected cloud computing sandbox environment to trigger a security cleanup process, including memory data erasure (clearing unauthorized data loaded), temporary file deletion (removing intermediate calculation results), and calculation state saving (preserving the model training progress of the compliant part). This mechanism ensures that guardians can adjust or revoke the use authorization of their children's data at any time, truly realizing dynamic protection of data rights.

[0080] The application effect of the system in the embodiment can be verified by Table 1;

[0081] Table 1 Data protection effect table

[0082]

[0083] Through the implementation of the system in the child growth and development data security sharing scene, hospitals, community health centers and educational institutions can safely carry out cross-institutional research cooperation without sharing original data, effectively protect the safety of children's sensitive health data, and at the same time guarantee the complete control right of guardians over data use. Through dynamic differential privacy and precise authorization control, the system not only maximizes the utilization of data value, but also minimizes the risk of privacy leakage, providing a safe and reliable data foundation for children's health development research.

[0084] While embodiments of the application have been shown and described, it is to be understood that the application is not limited to the details of the embodiments described, since numerous changes, modifications, substitutions and variations can be made thereto without departing from the spirit and scope of the application as defined by the appended claims and their equivalents.

Claims

1. A method for protecting child healthcare data based on federated learning, characterized in that, include: Obtain the distributed attribute authorization dataset, which includes the attribute definition dataset, the policy management dataset, and the key distribution dataset; Receive authorization instructions through the policy management dataset and generate access control policies based on the authorization instructions; When the federated learning platform initiates a data access request, it creates an isolated cloud computing sandbox environment for the federated learning task according to the access control policy; it dynamically calculates differential privacy parameters that match the purpose of use, and applies the parameters to the child health data within the authorized scope. Perform federated learning tasks in a cloud computing sandbox environment, monitoring and restricting data processing operations; steps Specifically, this includes: generating temporary session keys based on the key distribution dataset and allocating the temporary session keys to authorized federated learning tasks; building a cloud computing sandbox environment in physically isolated virtual containers, with the virtual containers having independent file systems, network namespaces, and process spaces; deploying a resource monitoring agent in the virtual containers to monitor computing resource usage in real time and set resource usage limits; assigning a unique identifier to the cloud computing sandbox environment and associating it with the corresponding access control policy. Provides a dynamic permission adjustment mechanism. When the authorization scope is modified and / or authorization is revoked, the access control policy is updated and permission change instructions are pushed to all computing nodes to forcibly terminate unfinished computing tasks that exceed the authorization scope. The specific steps for providing a dynamic permission adjustment mechanism include: Build a permission change event listener to receive authorization instructions issued by guardians through the policy management dataset in real time; and perform digital signature verification on the authorization instructions. The permission change instructions are converted into structured access control policy update instructions, including permission type, scope of change, effective time and priority; based on the distributed consistency protocol, it is ensured that permission changes take effect synchronously on all computing nodes; The system re-evaluates permissions for ongoing federated learning tasks, identifies computational tasks that exceed the scope of the new authorization, and sends task interruption signals to affected cloud computing sandbox environments to trigger a security cleanup process, including memory data erasure, temporary file deletion, and computation state saving.

2. The method for protecting child healthcare data based on federated learning according to claim 1, characterized in that: The attribute definition dataset is used to define a set of attributes related to child health care data. The attribute set includes at least data type attributes, purpose of use attributes, timeliness attributes, and sensitivity attributes. The access control policy includes the scope of data that the federated learning task is allowed to access, the purpose of use, the effective time, and the minimum requirements.

3. The method for protecting child healthcare data based on federated learning according to claim 1, characterized in that: The specific steps involved in dynamically calculating differential privacy parameters that match the intended use include: Extract sensitivity attribute values ​​related to the currently authorized data from the attribute definition dataset; based on the data type attribute and the purpose of use attribute, query the preset differential privacy parameter mapping table to obtain the initial privacy budget and sensitivity; The initial privacy budget is dynamically adjusted based on the sensitivity attribute value of child health care data; a privacy parameter decay mechanism is set based on the timeliness attribute to dynamically adjust the differential privacy parameters; and differentiated privacy parameters are set based on different stages of the federated learning task, with different noise addition mechanisms applied to the model training and aggregation stages. It tracks privacy budget consumption in real time, and automatically increases privacy protection and terminates data access when the cumulative consumption exceeds a preset threshold.

4. The method for protecting child healthcare data based on federated learning according to claim 1, characterized in that: The specific steps for performing federated learning tasks in the cloud computing sandbox environment include: Deploy a data access audit module to record all data operation behaviors, including data reading time, scope, processing method, and result output; implement an instruction-level monitoring mechanism to verify the permissions of each execution instruction of the federated learning algorithm and block unauthorized data processing operations; set data outflow control points to ensure that the processing results only contain statistical features and / or model parameters within the authorized scope and prohibit the outflow of raw data. Zero-knowledge proof technology is used to verify that the participants' computation process complies with the access control policy; an abnormal behavior detection mechanism is established so that when potential unauthorized data access is detected, the computation task is automatically suspended and an alarm is sent to the security management module.

5. A child healthcare data storage and protection system based on federated learning, used to execute the child healthcare data storage and protection method based on federated learning as described in any one of claims 1-4, characterized in that, include: The data acquisition module is used to acquire a distributed attribute authorization dataset, which includes an attribute definition dataset, a policy management dataset, and a key distribution dataset. The access policy control module is used to receive authorization instructions through the policy management dataset and generate access control policies based on the authorization instructions. The access judgment module is used to create an isolated cloud computing sandbox environment for the federated learning task according to the access control policy when the federated learning platform initiates a data access request; dynamically calculate differential privacy parameters that match the purpose of use, and apply the parameters to the child health data within the authorized scope; Perform federated learning tasks in the cloud computing sandbox environment, and monitor and restrict data processing operations. The access restriction module provides a dynamic permission adjustment mechanism. When the authorization scope is modified and / or authorization is revoked, the access control policy is updated, and permission change instructions are pushed to all computing nodes to forcibly terminate unfinished computing tasks that exceed the authorization scope.

Images