CN-122019139-A - Lightweight privacy computing task scheduling method and device integrating edge-cloud cooperation

CN122019139ACN 122019139 ACN122019139 ACN 122019139ACN-122019139-A

Abstract

The application discloses a lightweight privacy computing task scheduling method and device integrating edge-cloud cooperation, and relates to the technical field of privacy computing. Dividing a main task into a plurality of computing sub-tasks with security protocol identifications according to a strategy, processing the sub-tasks by a scheduling node in a trusted environment at an edge side by utilizing a lightweight engine to generate a protected local computing result, aggregating each edge result by a verification and multiparty collaboration mechanism at a cloud end, finally decoding to obtain a target privacy computing result, and dynamically optimizing the scheduling strategy by a reinforcement learning model based on performance indexes and utility evaluation executed at the time. By the method, the self-adaptive task scheduling with both calculation efficiency and data privacy safety is realized in the edge environment with limited resources, and the overall efficiency and reliability of the collaborative calculation system are improved.

Inventors

JIANG YUAN
WANG CHANGQING
WU XIAOYAN
Qiao Qichao
REN JUNQI
GAO BEIBEI

Assignees

中国工业互联网研究院(工业和信息化部密码应用研究中心)

Dates

Publication Date: 20260512
Application Date: 20251230

Claims (10)

1. The lightweight privacy computing task scheduling method integrating the edge-cloud cooperation is characterized by comprising the following steps of: obtaining task calculation amount and data privacy level according to a privacy calculation task request initiated by the edge terminal equipment; Based on the task calculated amount and the data privacy level, obtaining a cooperative scheduling strategy of the edge terminal equipment and the cloud server; dividing the privacy calculation task into a plurality of security calculation sub-tasks according to the cooperative scheduling strategy; Based on the security calculation sub-task, dispatching an edge node to perform local privacy calculation to obtain a local privacy calculation result; And aggregating the local privacy calculation result based on a cooperative mechanism of the cloud server to obtain a target privacy calculation result, and optimizing the cooperative scheduling strategy according to the target privacy calculation result to obtain a target cooperative scheduling strategy.
2. The method of claim 1, wherein the step of obtaining the co-scheduling policy of the edge terminal device and the cloud server based on the task computation amount and the data privacy level comprises: according to the task calculated amount, predicting available calculation resources of the edge terminal equipment and expected loads of the cloud server; based on the data privacy level, determining a security constraint condition required by data transmission between the edge and the cloud; generating an initial collaborative scheduling policy according to the available computing resources, the expected load and the security constraint condition; And carrying out evaluation and calibration on the initial cooperative scheduling strategy based on the historical scheduling performance data to obtain the cooperative scheduling strategy of the edge terminal equipment and the cloud server.
3. The method of claim 1, wherein the step of dividing the privacy computing task into a plurality of security computing sub-tasks according to the co-scheduling policy comprises: determining a cooperative mode in the cooperative scheduling strategy, and analyzing the dependency relationship between the data flow and the control flow in the privacy computing task according to the cooperative mode; determining a calculation boundary and a data desensitization rule of each subtask based on the dependency relationship and the data privacy level; Reconstructing an original task into a plurality of computing units with independent execution capacity according to the computing boundary and the data desensitization rule; and adding a lightweight security protocol identifier to each computing unit to obtain a security computing subtask.
4. The method of claim 1, wherein the step of scheduling the edge node for local privacy calculations based on the security calculation sub-tasks to obtain local privacy calculation results comprises: Determining a security protocol identifier of the security computation sub-task based on the security computation sub-task, and distributing a corresponding local trusted execution environment for the edge node according to the security protocol identifier; loading and initializing a lightweight privacy calculation engine corresponding to the security calculation sub-task based on the local trusted execution environment; scheduling the lightweight privacy calculation engine to process the raw data associated with the subtasks in the local trusted execution environment to generate protected intermediate state data; And carrying out integrity check and signature on the intermediate state data, and outputting the intermediate state data as a local privacy calculation result.
5. The method of claim 4, wherein the step of scheduling the lightweight privacy calculation engine to process raw data associated with a subtask within the local trusted execution environment, generating protected intermediate state data comprises: Scheduling the lightweight privacy calculation engine, carrying out lightweight homomorphic encryption processing on the original data in the local trusted execution environment based on a data desensitization rule of a subtask and the lightweight privacy calculation engine to obtain processed data; Performing slicing operation on the processed data to obtain sliced data; coding according to the fragment data to obtain a preliminary result; noise is added to the preliminary results, generating protected intermediate state data.
6. The method of claim 1, wherein the step of aggregating the local privacy calculations to obtain the target privacy calculations comprises: performing data verification on local privacy results from a plurality of edge nodes to obtain data verification results; Based on a collaboration mechanism of a cloud server, initiating multiparty collaborative calculation for the data verification result to obtain an edge calculation result fed back by each edge node; And aggregating the edge calculation result to obtain aggregated data, and decoding the aggregated data to obtain a target privacy calculation result.
7. The method of claim 6, wherein the step of initiating multiparty collaborative calculations based on the data validation results to obtain edge calculation results for each of the edge nodes feedback comprises: Determining a security protocol identification of the local privacy calculation result, and selecting a corresponding cloud aggregation calculation graph according to the security protocol identification, wherein the cloud aggregation calculation graph reflects the data transmission relation between the edge nodes; partitioning is carried out based on the cloud aggregation calculation graph to obtain a plurality of calculation partition graphs; Encrypting the data verification result according to the map identification of the calculated map to obtain an encrypted data verification result; And distributing the encryption data verification result according to the calculation partition map, and sending multi-party collaborative calculation to obtain an edge calculation result fed back by each edge node.
8. The method of claim 1, wherein the step of optimizing the co-scheduling policy based on the target privacy calculation result to obtain a target co-scheduling policy comprises: Generating a performance evaluation index according to the calculation time consumption, the resource consumption and the result quality of each edge node in the task scheduling process; calculating a utility difference value between actual utility and expected utility of the target privacy calculation result, and determining a strategy effectiveness evaluation result of the cooperative scheduling strategy according to the utility difference value; inputting the performance evaluation index and the strategy effectiveness evaluation result into a reinforcement learning model to generate strategy adjustment parameters; And optimizing the cooperative scheduling strategy based on the strategy adjustment parameters to obtain a target cooperative scheduling strategy.
9. The method of claim 8, wherein the step of inputting the performance evaluation index and the policy effectiveness evaluation result to a reinforcement learning model, generating a policy adjustment parameter comprises: Determining a network state, an edge resource state and task characteristics in the task scheduling, and coding the network state, the edge resource state and the task characteristics to obtain a state vector; Taking the performance evaluation index and the strategy effectiveness evaluation result as reward signals; Fitting a cost function based on the state vector and the bonus signal; and inputting the state vector, the reward signal and the cost function into a reinforcement learning model to generate strategy adjustment parameters.
10. The utility model provides a merge edge-high in clouds cooperative lightweight privacy calculation task scheduling device which characterized in that, merge edge-high in clouds cooperative lightweight privacy calculation task scheduling device includes: the task analysis module is used for obtaining task calculation amount and data privacy level according to the privacy calculation task request initiated by the edge terminal equipment; The policy generation module is used for obtaining a cooperative scheduling policy of the edge terminal equipment and the cloud server based on the task calculated amount and the data privacy level; The task dividing module is used for dividing the privacy calculation task into a plurality of security calculation sub-tasks according to the cooperative scheduling strategy; The edge computing module is used for scheduling edge nodes to perform local privacy computation based on the security computing sub-tasks to obtain a local privacy computing result; The cloud aggregation module is used for aggregating the local privacy calculation result based on a cooperative mechanism of the cloud server to obtain a target privacy calculation result, and optimizing the cooperative scheduling policy according to the target privacy calculation result to obtain a target cooperative scheduling policy.

Description

Lightweight privacy computing task scheduling method and device integrating edge-cloud cooperation Technical Field The application relates to the technical field of privacy computation and edge computation cooperation, in particular to a lightweight privacy computation task scheduling method and device integrating edge-cloud cooperation. Background When the existing method distributes calculation tasks, the traditional performance indexes such as resource utilization rate, calculation delay and the like are often focused, but the data privacy level cannot be used as a dynamic and key core decision factor for deep integration. This results in a policy lacking response capability to the intrinsic security requirements of the task, i.e. tasks with high privacy requirements may be assigned to edge nodes with fuzzy security boundaries, increasing the risk of leakage, while passing all data to the cloud for absolute security processing, again causing network congestion and computational delays, severely compromising efficiency. Second, there is an inherent contradiction between limited resources on the edge side and complex privacy computing techniques. Edge terminals generally have limited computing power, storage and battery life, and powerful privacy protection technologies (such as traditional homomorphic encryption or secure multiparty computing) tend to have huge computing and communication overheads, which are difficult to directly deploy. This resource bottleneck makes it extremely difficult to implement efficient and lightweight privacy preserving computation on the edge side. The lack of a synergistic mechanism to dynamically adapt lightweight security protocols to computing resource allocation based on task security requirements and edge node real-time status ultimately results in a system that can only make a compromise between "security" and "efficiency". The foregoing is provided merely for the purpose of facilitating understanding of the technical solutions of the present invention and is not intended to represent an admission that the foregoing is prior art. Disclosure of Invention The application mainly aims to provide a lightweight privacy computing task scheduling method and device integrating edge-cloud cooperation, and aims to solve the technical problem that privacy protection and task scheduling efficiency are difficult to be compatible in an edge computing scene in the prior art. In order to achieve the above purpose, the application provides a lightweight privacy calculation task scheduling method integrating edge-cloud cooperation, which comprises the following steps: obtaining task calculation amount and data privacy level according to a privacy calculation task request initiated by the edge terminal equipment; Based on the task calculated amount and the data privacy level, obtaining a cooperative scheduling strategy of the edge terminal equipment and the cloud server; dividing the privacy calculation task into a plurality of security calculation sub-tasks according to the cooperative scheduling strategy; Based on the security calculation sub-task, dispatching an edge node to perform local privacy calculation to obtain a local privacy calculation result; And aggregating the local privacy calculation result based on a cooperative mechanism of the cloud server to obtain a target privacy calculation result, and optimizing the cooperative scheduling strategy according to the target privacy calculation result to obtain a target cooperative scheduling strategy. In an embodiment, the step of obtaining the cooperative scheduling policy of the edge terminal device and the cloud server based on the task computation amount and the data privacy level includes: according to the task calculated amount, predicting available calculation resources of the edge terminal equipment and expected loads of the cloud server; based on the data privacy level, determining a security constraint condition required by data transmission between the edge and the cloud; generating an initial collaborative scheduling policy according to the available computing resources, the expected load and the security constraint condition; And carrying out evaluation and calibration on the initial cooperative scheduling strategy based on the historical scheduling performance data to obtain the cooperative scheduling strategy of the edge terminal equipment and the cloud server. In an embodiment, the step of dividing the privacy computation task into a plurality of security computation sub-tasks according to the co-scheduling policy comprises: determining a cooperative mode in the cooperative scheduling strategy, and analyzing the dependency relationship between the data flow and the control flow in the privacy computing task according to the cooperative mode; determining a calculation boundary and a data desensitization rule of each subtask based on the dependency relationship and the data privacy level; Reconstructing an original tas