Lightweight privacy computing task scheduling method and device based on fusion of edge-cloud collaboration

Abstract

This application discloses a lightweight privacy computing task scheduling method and apparatus integrating edge-cloud collaboration, relating to the field of privacy computing technology. The method includes: parsing task requests to obtain their computational load and data privacy level, thereby generating an initial edge-cloud collaborative scheduling strategy; dividing the main task into multiple computational subtasks with security protocol identifiers according to the strategy; on the edge side, the scheduling node processes the subtasks in a trusted environment using a lightweight engine to generate protected local computation results; on the cloud side, the results from each edge are aggregated through verification and multi-party collaboration mechanisms, and finally decoded to obtain the target privacy computing result. Based on the performance indicators and utility evaluation of this execution, the scheduling strategy is dynamically optimized using a reinforcement learning model. Through this approach, adaptive task scheduling that balances computational efficiency and data privacy security is achieved in resource-constrained edge environments, improving the overall performance and reliability of the collaborative computing system.

Description

Technical Field

[0001] This application relates to the field of privacy computing and edge computing collaboration technology, and in particular to a lightweight privacy computing task scheduling method and apparatus that integrates edge-cloud collaboration. Background Technology

[0002] Existing methods often focus on traditional performance metrics such as resource utilization and computational latency when allocating computing tasks, failing to deeply integrate data privacy level as a dynamic and critical core decision factor. This results in a lack of responsiveness to the inherent security needs of tasks: tasks with high privacy requirements may be assigned to edge nodes with ambiguous security boundaries, increasing the risk of leakage; while pursuing absolute security by transmitting all data to the cloud for processing can cause network congestion and computational latency, severely impairing efficiency.

[0003] Secondly, there is an inherent contradiction between the limited resources at the edge and complex privacy-preserving computing technologies. Edge terminals typically have limited computing power, storage, and battery life, while robust privacy protection technologies (such as traditional homomorphic encryption or secure multi-party computation) often incur huge computational and communication overheads, making direct deployment difficult. This resource bottleneck makes achieving efficient and lightweight privacy-preserving computing at the edge exceptionally challenging. The lack of a collaborative mechanism that can dynamically adapt lightweight security protocols and computing resource allocation based on task security requirements and the real-time status of edge nodes ultimately forces the system to make an either-or compromise between "security" and "efficiency."

[0004] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main purpose of this application is to provide a lightweight privacy computing task scheduling method and device that integrates edge-cloud collaboration, aiming to solve the technical problem that it is difficult to balance privacy protection and task scheduling efficiency in edge computing scenarios in the prior art.

[0006] To achieve the above objectives, this application provides a lightweight privacy computing task scheduling method that integrates edge-cloud collaboration, the method comprising: Based on the privacy computing task request initiated by the edge terminal device, the task computing workload and data privacy level are obtained; Based on the task computation volume and the data privacy level, a collaborative scheduling strategy between the edge terminal device and the cloud server is obtained; According to the collaborative scheduling strategy, the privacy computing task is divided into multiple secure computing subtasks; Based on the aforementioned secure computing subtask, edge nodes are scheduled to perform local privacy computing to obtain local privacy computing results; The local privacy computation results are aggregated based on the collaborative mechanism of the cloud server to obtain the target privacy computation result. The collaborative scheduling strategy is then optimized based on the target privacy computation result to obtain the target collaborative scheduling strategy.

[0007] In one embodiment, the step of obtaining the collaborative scheduling strategy between the edge terminal device and the cloud server based on the task computation volume and the data privacy level includes: Based on the computational workload of the task, predict the available computing resources of the edge terminal devices and the expected load of the cloud server; Based on the aforementioned data privacy level, determine the security constraints required for data transmission between the edge and the cloud; An initial cooperative scheduling strategy is generated based on the available computing resources, the expected load, and the security constraints. The initial collaborative scheduling strategy is evaluated and calibrated based on historical scheduling performance data to obtain the collaborative scheduling strategy between the edge terminal device and the cloud server.

[0008] In one embodiment, the step of dividing the privacy computing task into multiple security computing subtasks according to the cooperative scheduling strategy includes: Determine the collaboration mode in the collaborative scheduling strategy, and parse the dependency relationship between the data flow and control flow inside the privacy computing task based on the collaboration mode; Based on the aforementioned dependencies and the aforementioned data privacy level, the computational boundaries and data anonymization rules for each subtask are determined. Based on the computational boundary and the data anonymization rules, the original task is reconstructed into multiple computational units with independent execution capabilities; A lightweight security protocol identifier is attached to each of the computing units to obtain a secure computing subtask.

[0009] In one embodiment, the step of scheduling edge nodes to perform local privacy computation based on the secure computation subtask and obtaining the local privacy computation result includes: Based on the secure computing subtask, a security protocol identifier for the secure computing subtask is determined, and a corresponding local trusted execution environment is allocated to the edge node according to the security protocol identifier; Based on the local trusted execution environment, load and initialize the lightweight privacy computing engine corresponding to the secure computing subtask; The lightweight privacy computing engine is scheduled to process the raw data associated with subtasks within the local trusted execution environment, generating protected intermediate state data. The intermediate state data is verified for integrity and signed, and the result is output as the local privacy calculation result.

[0010] In one embodiment, the step of scheduling the lightweight privacy computing engine to process the raw data associated with the subtasks within the local trusted execution environment and generate protected intermediate state data includes: The lightweight privacy computing engine is scheduled, and based on the data desensitization rules of the subtask and the lightweight privacy computing engine, the original data is processed using lightweight homomorphic encryption within the local trusted execution environment to obtain the processed data; The processed data is then segmented to obtain segmented data; The fragmented data is encoded to obtain preliminary results; Noise is added to the preliminary results to generate protected intermediate state data.

[0011] In one embodiment, the step of aggregating the local privacy computation results using the cloud server-based collaborative mechanism to obtain the target privacy computation result includes: Data verification is performed on the local privacy results from multiple edge nodes to obtain data verification results; Based on the collaborative mechanism of the cloud server, multi-party collaborative computation is initiated on the data verification results to obtain the edge computing results fed back by each edge node; The edge computing results are aggregated to obtain aggregated data, and the aggregated data is decoded to obtain the target privacy computing results.

[0012] In one embodiment, the step of initiating multi-party collaborative computation based on the data verification result to obtain the edge computing result fed back by each edge node includes: Determine the security protocol identifier of the local privacy computation result, and select the corresponding cloud aggregation computation graph based on the security protocol identifier. The cloud aggregation computation graph reflects the data transmission relationship between the edge nodes. Based on the cloud-based aggregated computing graph, multiple computing sub-graphs are obtained by partitioning the graph. The data verification results are encrypted according to the partition identifier of the computational partition graph to obtain encrypted data verification results; The encrypted data verification results are distributed according to the computational graph, and multi-party collaborative computation is initiated to obtain the edge computation results fed back by each edge node.

[0013] In one embodiment, the step of optimizing the cooperative scheduling strategy based on the target privacy calculation result to obtain the target cooperative scheduling strategy includes: Based on the computation time, resource consumption, and result quality of each edge node during this task scheduling process, performance evaluation indicators are generated. Calculate the utility difference between the actual utility and the expected utility of the target privacy calculation result, and determine the policy effectiveness evaluation result of the collaborative scheduling strategy based on the utility difference; The performance evaluation metrics and the policy effectiveness evaluation results are input into the reinforcement learning model to generate policy adjustment parameters. The cooperative scheduling strategy is optimized based on the strategy adjustment parameters to obtain the target cooperative scheduling strategy.

[0014] In one embodiment, the step of inputting the performance evaluation metric and the policy effectiveness evaluation result into the reinforcement learning model to generate policy adjustment parameters includes: Determine the network state, edge resource state, and task characteristics in this task scheduling, and encode the network state, edge resource state, and task characteristics to obtain a state vector; The performance evaluation metrics and the effectiveness evaluation results of the strategy will be used as reward signals. A value function is fitted based on the state vector and the reward signal; The state vector, the reward signal, and the value function are input into the reinforcement learning model to generate policy adjustment parameters.

[0015] Furthermore, to achieve the above objectives, this application also proposes a lightweight privacy computing task scheduling device that integrates edge-cloud collaboration. The lightweight privacy computing task scheduling device that integrates edge-cloud collaboration includes: The task parsing module is used to obtain the task computation amount and data privacy level based on the privacy computing task request initiated by the edge terminal device; The strategy generation module is used to obtain the collaborative scheduling strategy between the edge terminal device and the cloud server based on the task computation volume and the data privacy level. The task partitioning module is used to divide the privacy computing task into multiple security computing subtasks according to the collaborative scheduling strategy. The edge computing module is used to schedule edge nodes to perform local privacy computing based on the secure computing subtask, and obtain local privacy computing results; The cloud aggregation module is used to aggregate the local privacy computation results based on the collaborative mechanism of the cloud server to obtain the target privacy computation result, and optimize the collaborative scheduling strategy based on the target privacy computation result to obtain the target collaborative scheduling strategy.

[0016] Furthermore, to achieve the above objectives, this application also proposes a lightweight privacy computing task scheduling device that integrates edge-cloud collaboration. The lightweight privacy computing task scheduling device that integrates edge-cloud collaboration includes: a memory, a processor, and a computer program stored on the memory and executable on the processor. The computer program is configured to implement the steps of the lightweight privacy computing task scheduling method that integrates edge-cloud collaboration as described above.

[0017] In addition, to achieve the above objectives, the present invention also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the steps of the lightweight privacy computing task scheduling method for converged edge-cloud collaboration as described above.

[0018] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the lightweight privacy computing task scheduling method for converged edge-cloud collaboration as described above.

[0019] This application provides a lightweight privacy-preserving computing task scheduling method that integrates edge-cloud collaboration. It parses task requests to obtain their computational load and data privacy level, generating an initial edge-cloud collaborative scheduling strategy. Based on the strategy, the main task is divided into multiple computational subtasks with security protocol identifiers. At the edge, scheduling nodes process the subtasks in a trusted environment using a lightweight engine, generating protected local computation results. In the cloud, the results from each edge are aggregated through verification and multi-party collaboration mechanisms, ultimately decoding to obtain the target privacy-preserving computation result. Based on the performance metrics and utility evaluation of this execution, the scheduling strategy is dynamically optimized using a reinforcement learning model. This approach achieves adaptive task scheduling that balances computational efficiency and data privacy security in resource-constrained edge environments, improving the overall performance and reliability of the collaborative computing system. Attached Figure Description

[0020] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0021] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a flowchart illustrating an embodiment of the lightweight privacy computing task scheduling method integrating edge-cloud collaboration proposed in this application. Figure 2 This is a dependency graph of an embodiment of the lightweight privacy computing task scheduling method integrating edge-cloud collaboration in this application; Figure 3 This is a schematic diagram of the module structure of a lightweight privacy computing task scheduling device that integrates edge-cloud collaboration, as described in an embodiment of this application. Figure 4 This is a schematic diagram of the device structure of the hardware operating environment involved in the lightweight privacy computing task scheduling method that integrates edge-cloud collaboration in the embodiments of this application.

[0023] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0024] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0025] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0026] The main solution of this application embodiment is: to obtain the task computation amount and data privacy level based on the privacy computing task request initiated by the edge terminal device; Based on the task computation volume and the data privacy level, a collaborative scheduling strategy between the edge terminal device and the cloud server is obtained; According to the collaborative scheduling strategy, the privacy computing task is divided into multiple secure computing subtasks; Based on the aforementioned secure computing subtask, edge nodes are scheduled to perform local privacy computing to obtain local privacy computing results; The local privacy computation results are aggregated based on the collaborative mechanism of the cloud server to obtain the target privacy computation result. The collaborative scheduling strategy is then optimized based on the target privacy computation result to obtain the target collaborative scheduling strategy.

[0027] Currently, existing methods often focus on traditional performance metrics such as resource utilization and computational latency when allocating computing tasks, failing to deeply integrate data privacy level as a dynamic and critical core decision factor. This results in a lack of responsiveness to the inherent security needs of tasks: tasks with high privacy requirements may be assigned to edge nodes with ambiguous security boundaries, increasing the risk of leakage; while pursuing absolute security by transmitting all data to the cloud for processing can cause network congestion and computational latency, severely impairing efficiency.

[0028] Secondly, there is an inherent contradiction between the limited resources at the edge and complex privacy-preserving computing technologies. Edge terminals typically have limited computing power, storage, and battery life, while robust privacy protection technologies (such as traditional homomorphic encryption or secure multi-party computation) often incur huge computational and communication overheads, making direct deployment difficult. This resource bottleneck makes achieving efficient and lightweight privacy-preserving computing at the edge exceptionally challenging. The lack of a collaborative mechanism that can dynamically adapt lightweight security protocols and computing resource allocation based on task security requirements and the real-time status of edge nodes ultimately forces the system to make an either-or compromise between "security" and "efficiency."

[0029] This application provides a solution that parses task requests to obtain their computational load and data privacy level, thereby generating an initial edge-cloud collaborative scheduling strategy. Based on this strategy, the main task is divided into multiple computational subtasks with security protocol identifiers. At the edge, scheduling nodes process the subtasks in a trusted environment using a lightweight engine, generating protected local computation results. In the cloud, the results from each edge are aggregated through verification and multi-party collaboration mechanisms, and finally, the target privacy-preserving computation result is decoded. Based on the performance metrics and utility evaluation of this execution, the scheduling strategy is dynamically optimized using a reinforcement learning model. Through this approach, adaptive task scheduling that balances computational efficiency and data privacy security is achieved in resource-constrained edge environments, improving the overall performance and reliability of the collaborative computing system.

[0030] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or an electronic device capable of performing the above functions, a lightweight privacy computing task scheduling device integrating edge-cloud collaboration, etc. This embodiment does not specifically limit it. The following uses a lightweight privacy computing task scheduling device integrating edge-cloud collaboration as an example to describe this embodiment and the following embodiments.

[0031] All actions involving the acquisition of signals, information, or data in this application are carried out in accordance with the relevant data protection laws and policies of the country where the application is located, and with the authorization of the owner of the relevant device.

[0032] This application provides a lightweight privacy computing task scheduling method that integrates edge-cloud collaboration, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the lightweight privacy computing task scheduling method integrating edge-cloud collaboration in this application.

[0033] In this embodiment, the lightweight privacy computing task scheduling method integrating edge-cloud collaboration includes steps S10~S50: Step S10: Based on the privacy computing task request initiated by the edge terminal device, obtain the task computing load and data privacy level.

[0034] It should be noted that task computation volume refers to a quantitative assessment of the computing resources required to complete the privacy-preserving computation task, including CPU cycles, memory usage, storage space, and network bandwidth, which determines the scale and complexity of the task execution. Data privacy level is a security level that is pre-defined or dynamically assigned based on data sensitivity (such as personal identification information, medical records, geographical location, etc.), used to define the strength of protection and de-identification rules that data must follow during transmission and processing.

[0035] In its implementation, upon receiving a task request, the module parses the request metadata, estimates the computational load using built-in resource evaluation algorithms (such as empirical models based on task type and data size), and queries the data tags or privacy policy library associated with the task to accurately obtain its privacy level, providing a quantitative basis for subsequent resource scheduling and security policy formulation.

[0036] Step S20: Based on the task computation volume and the data privacy level, obtain the collaborative scheduling strategy between the edge terminal device and the cloud server.

[0037] In the implementation, the computational load and data privacy level are taken as key inputs, and combined with real-time network conditions (such as bandwidth and latency) and resource load (such as CPU / memory availability at edge nodes and in the cloud), the calculation is performed through a pre-defined decision model (e.g., based on a rule engine or a lightweight machine learning model). Ultimately, the model outputs a specific strategy, such as: for tasks with high privacy requirements but low computational load, prioritizing scheduling to resource-sufficient edge nodes and applying local differential privacy processing; for tasks with massive computational load and extremely high privacy requirements, a federated learning model of edge preprocessing + secure cloud aggregation might be adopted.

[0038] In one feasible implementation, the step of obtaining the collaborative scheduling strategy between the edge terminal device and the cloud server based on the task computation volume and the data privacy level includes: Based on the computational workload of the task, predict the available computing resources of the edge terminal devices and the expected load of the cloud server; Based on the aforementioned data privacy level, determine the security constraints required for data transmission between the edge and the cloud; An initial cooperative scheduling strategy is generated based on the available computing resources, the expected load, and the security constraints. The initial collaborative scheduling strategy is evaluated and calibrated based on historical scheduling performance data to obtain the collaborative scheduling strategy between the edge terminal device and the cloud server.

[0039] It should be noted that available computing resources refer to the quantified value of the remaining computing power (CPU, memory, etc.) of the edge terminal device at the time of scheduling decision, which can be immediately allocated to process new tasks. Expected load refers to the total amount of tasks that the cloud server will need to handle in the future, predicted based on historical patterns or queue status. Security constraints are technical security requirements that must be followed during data transmission, directly derived from data privacy levels. For example, they may specify that data transmission must use encryption algorithms of a certain strength (such as AES-256), specific secure transmission protocols (such as TLS 1.3), or limit the maximum latency and number of hops of data packets to reduce the risk of theft or tampering. Historical scheduling performance data refers to the actual operating performance indicators recorded when the system executed similar collaborative scheduling strategies in the past. These typically include the actual task completion time, energy consumption, data transmission time, and privacy protection effectiveness verification results, used to quantitatively evaluate the true effectiveness of the strategy.

[0040] In practice, this process is executed by a central scheduler or by agents distributed at the edge. First, the scheduler uses lightweight agents deployed on edge devices to collect real-time information such as CPU utilization and available memory, and then uses predictive algorithms such as simple moving averages or exponential smoothing to estimate available computing resources. Simultaneously, by interacting with cloud monitoring services to obtain their current task queue length and processing rate, the expected load can be predicted. Secondly, the system maintains a privacy level-security constraint mapping table, based on the privacy level of the task. Automatically retrieve the corresponding security constraints For example, the required encryption algorithm and the maximum permissible network latency. These parameters are then fed into a weighted or rule-based decision function. Generate an initial cooperative scheduling policy that prioritizes efficiency or safety, such as "if..." and If local processing is allowed, the data is routed to the edge; otherwise, it is encrypted and uploaded to the cloud.

[0041] To improve the accuracy of the strategy, the system introduces a feedback calibration process. The scheduler compares the "initial cooperative scheduling strategy" with the "historical scheduling performance data" (such as historical task completion times) recorded in the database. Energy consumption A comparative analysis is conducted. By calculating the deviation between the expected value of the strategy and the historical actual value, the weights or thresholds in the decision function are fine-tuned using regression models or reinforcement learning algorithms. This results in a final, evaluated, and calibrated collaborative scheduling strategy that better reflects the actual network environment and resource status. This process enables the system to continuously learn and optimize, dynamically balancing privacy and efficiency.

[0042] Step S30: According to the collaborative scheduling strategy, the privacy computing task is divided into multiple secure computing subtasks.

[0043] It should be noted that secure computing subtasks refer to logical units that can be executed independently or collaboratively, formed by breaking down the original privacy-preserving computing task according to specific security principles and parallelization requirements. Each subtask not only contains a portion of computational logic and data fragments, but more importantly, it is given clear security boundaries and data processing rules to ensure that the exposure or failure of a single subtask during subsequent distributed execution will not lead to the overall leakage of the original sensitive data.

[0044] Understandably, the system parses the decisions regarding computation location (edge / cloud) and security requirements (such as encryption methods) in the collaborative scheduling strategy and segments the task based on its logical dependencies (such as a directed acyclic graph (DAG) description). For example, for a federated learning task, the engine might divide it into: a local model training subtask executed on an edge device (using differential privacy techniques to add noise), and a model security aggregation subtask executed in the cloud (using a secure multi-party computation protocol). The decomposition process ensures that each subtask meets the security constraints of its target execution environment and specifies encrypted channels and formats for necessary data exchange between subtasks, ultimately generating a task graph containing multiple secure computation subtasks and their dependencies.

[0045] In one feasible implementation, the step of dividing the privacy computing task into multiple secure computing subtasks according to the cooperative scheduling strategy includes: Determine the collaboration mode in the collaborative scheduling strategy, and parse the dependency relationship between the data flow and control flow inside the privacy computing task based on the collaboration mode; Based on the aforementioned dependencies and the aforementioned data privacy level, the computational boundaries and data anonymization rules for each subtask are determined. Based on the computational boundary and the data anonymization rules, the original task is reconstructed into multiple computational units with independent execution capabilities; A lightweight security protocol identifier is attached to each of the computing units to obtain a secure computing subtask.

[0046] In the specific implementation, the cooperative scheduling strategy is first read to identify the cooperative mode (e.g., determined to be federated learning mode). Based on this mode, the code or workflow description file (e.g., a DAG) of the privacy-preserving computation task is analyzed to parse the data flow dependencies (e.g., node B needs the output of node A) and control flow dependencies (e.g., node C only executes after node B's success signal is triggered), and formally represented as a dependency graph G(V, E), where vertex V represents the basic computation operation and edge E represents the dependency relationship. The generated dependency graph is as follows: Figure 2 As shown.

[0047] Next, considering data privacy levels, the dependency graph is securely partitioned. Based on the principle of minimizing privacy protection, computational boundaries are defined for each potential partition point in the graph (i.e., sub-task boundaries), ensuring that sensitive data is confined to a specific domain (e.g., raw user data is processed only within edge devices). Simultaneously, for each data item that needs to be transmitted across boundaries... Customized data anonymization rules ,in This refers to the privacy level of the data item. For example, the rule might be... (i.e., using public key encryption), and (That is, generalization processing). Based on these boundaries and rules, the planner reconstructs the original task into a set of loosely coupled computational units with independent execution capabilities.

[0048] Finally, a configuration file is generated for each computing unit, with a lightweight security protocol identifier appended. This identifier is a machine-readable instruction that tells the execution environment how to handle the unit. For example, for a computing unit that needs to run on a resource-constrained edge device, its identifier might specify the use of a lightweight differential privacy algorithm, with parameters. This final computational unit, wrapped in secure metadata, becomes a secure computational subtask that can be distributed to the appropriate edge or cloud nodes for execution.

[0049] Step S40: Based on the secure computing subtask, schedule edge nodes to perform local privacy computing to obtain local privacy computing results.

[0050] It should be noted that local privacy computation results refer to the output generated after scheduling one or more edge nodes to perform computations on their assigned security computation subtasks.

[0051] Understandably, upon receiving the secure computation subtask and its accompanying lightweight security protocol identifier, a controlled secure execution environment (such as a Trusted Execution Environment (TEE) or a resource-isolated container) is first created locally. Then, the corresponding cryptographic library or privacy algorithm module (such as a differential privacy library for adding Laplace noise) is loaded according to the protocol identifier. Finally, the computational logic of the subtask is run in the isolated environment. The entire process ensures that the original sensitive data is not leaked, and the final output is a secure result that has been encrypted, perturbed, or transformed. This result is then returned to the scheduler or the next stage task node through a secure channel.

[0052] In one feasible implementation, the step of scheduling edge nodes to perform local privacy computation based on the secure computation subtask and obtaining the local privacy computation result includes: Based on the secure computing subtask, a security protocol identifier for the secure computing subtask is determined, and a corresponding local trusted execution environment is allocated to the edge node according to the security protocol identifier; Based on the local trusted execution environment, load and initialize the lightweight privacy computing engine corresponding to the secure computing subtask; The lightweight privacy computing engine is scheduled to process the raw data associated with subtasks within the local trusted execution environment, generating protected intermediate state data. The intermediate state data is verified for integrity and signed, and the result is output as the local privacy calculation result.

[0053] In the implementation, a centralized edge resource scheduler collaborates with a TEE management daemon deployed on each edge node. The scheduler first parses the security protocol identifier of the secure computation subtask, for example, an identifier like "TEE-SGX: LightHE". Based on this, the scheduler selects an edge node with SGX hardware capabilities and available resources, and sends a command to its TEE management daemon requesting the allocation of a local trusted execution environment configured to perform Lightweight Homomorphic Encryption (LightHE) computation. This command includes the code hash of the subtask, used for code integrity measurement upon TEE startup.

[0054] Upon receiving the instruction, the edge node's TEE management daemon (running in a normal rich execution environment) invokes processor-specific instructions (such as Intel's ECREATE) to create a secure enclave as a TEE instance. Subsequently, it securely loads and initializes the core computing module (signed and verified) corresponding to LightHE from the Lightweight Privacy Computing Engine. The initialization process includes generating key materials required for cryptographic algorithms within the TEE or importing them from a secure location.

[0055] After the computing engine is initialized, the scheduler associates the raw data with the subtasks. For example, a set of user data records in a local database. The data is transmitted to the TEE via a secure, encrypted channel. The engine processes this data in the TEE's isolated memory, performing operations such as homomorphic encryption to generate protected intermediate state data, i.e., the ciphertext result. Before outputting, to ensure the results are not tampered with during subsequent transmission and aggregation, the TEE uses its built-in attestation key to calculate a digital signature for the intermediate state data. and the ciphertext set Along with the signature, it is output as a local privacy computation result to the outside of the TEE and collected by the scheduler.

[0056] In one feasible implementation, the step of scheduling the lightweight privacy computing engine to process the raw data associated with the subtasks within the local trusted execution environment and generate protected intermediate state data includes: The lightweight privacy computing engine is scheduled, and based on the data desensitization rules of the subtask and the lightweight privacy computing engine, the original data is processed using lightweight homomorphic encryption within the local trusted execution environment to obtain the processed data; The processed data is then segmented to obtain segmented data; The fragmented data is encoded to obtain preliminary results; Noise is added to the preliminary results to generate protected intermediate state data.

[0057] In its implementation, the engine first reads the data anonymization rules from the subtask configuration. If the rule specifies encryption requirements (such as "encryption: LightHE-CKKS"), the engine calls the built-in lightweight CKKS homomorphic encryption library. The engine uses a pre-defined or dynamically negotiated public key to encrypt the original data vector. Encryption is performed to obtain the processed data, i.e., the ciphertext vector. The advantage of the CKKS scheme is that it can directly approximate calculations on real or complex numbers, making it very suitable for scenarios such as machine learning.

[0058] Next, to further enhance security and prevent the TEE itself from being compromised, the engine will process the ciphertext vector. Perform sharding operations. It may employ a secret sharing scheme, for example, by... Divided into Data in segments Among them, at least Each fragment ( Only then can the original ciphertext be reconstructed. This process is completed within the TEE, and the fragment itself can be in ciphertext form.

[0059] The engine then encodes the fragmented data according to a predefined protocol. For example, to efficiently add noise that satisfies differential privacy later, the engine might encode each fragment... Mapped to a polynomial The coefficients, or by performing specific vector transformations, can be used to obtain preliminary results. .

[0060] Finally, the engine performs the noise-adding step. It does so based on the differential privacy budget. and global sensitivity From the Laplace distribution Mid-sampling generates noise vector Then, the noise was combined with the preliminary results: This final result This refers to protected intermediate state data. It incorporates the confidentiality of homomorphic encryption, distributes risks through secret sharding, and integrates the strict mathematical guarantees of differential privacy, forming a multi-layered, defense-in-depth privacy protection effect.

[0061] Step S50: Aggregate the local privacy calculation results based on the collaborative mechanism of the cloud server to obtain the target privacy calculation result, and optimize the collaborative scheduling strategy based on the target privacy calculation result to obtain the target collaborative scheduling strategy.

[0062] It should be noted that the target privacy computation result refers to the final computation result obtained by the cloud server through the fusion of local privacy computation results from multiple edge nodes using secure aggregation protocols (such as secure multi-party computation, federated learning aggregation algorithms, etc.). This result meets the requirements of the global computation objective while inheriting the privacy protection characteristics of each local result, ensuring that no original data information of any individual edge node or all details of the local computation result are leaked during the aggregation process.

[0063] In practical implementation, the cloud server acts as an aggregator and coordinator. First, it collects the signed or encrypted local privacy computation results uploaded by each edge node through a secure channel (such as TLS). Then, the server runs a pre-built secure aggregation algorithm (such as calculating a weighted average in federated learning). Alternatively, homomorphic ciphertext can be decrypted / merged in secure multi-party computation to remove potential malicious data and obtain the target privacy computation result. Finally, based on the performance indicators (such as computation latency, resource consumption, and result accuracy) and analysis results of this aggregation task, reinforcement learning or heuristic algorithms are used to dynamically update the strategy parameters such as task segmentation and resource allocation to generate a more efficient and secure target collaborative scheduling strategy for use in the next task.

[0064] In one feasible implementation, the step of aggregating the local privacy computation results using the cloud server-based collaborative mechanism to obtain the target privacy computation result includes: Data verification is performed on the local privacy results from multiple edge nodes to obtain data verification results; Based on the collaborative mechanism of the cloud server, multi-party collaborative computation is initiated on the data verification results to obtain the edge computing results fed back by each edge node; The edge computing results are aggregated to obtain aggregated data, and the aggregated data is decoded to obtain the target privacy computing results.

[0065] In the specific implementation, the cloud server first acts as the verifier. It receives the results of its local privacy computation uploaded by the edge node i. and TEE attestation signature The data packet. The server uses the pre-configured TEEattestation public key of each node to verify the signature validity, that is, to check the equation. Check if it's valid and verify the data format. Those that pass verification... The set is denoted as the data validation result. (Assume that m nodes pass the verification).

[0066] Next, the server initiates multi-party collaborative computation based on a pre-defined collaborative mechanism (such as federated averaging or the Secure Aggregation protocol). For example, in a federated learning scenario, the server will use the current global model parameters... The broadcast is sent to all nodes providing VVV. Each node uses its raw data to compute model updates within its local TEE. It then uses a secure aggregation client library to mask or encrypt the results, generating edge computing output. And return it to the server.

[0067] Then, the server processes all received... Perform aggregation. For simple additive homomorphic encryption, the aggregation operation is equivalent to calculating... .this It's aggregated data; after decryption, it should equal... .

[0068] Finally, the server performs the decoding process. If homomorphic encryption is used, the server uses its private key. Decrypting Aggregated Data: If secret sharing is used, the server needs to collect a sufficient number of shares before performing reconstruction algorithms such as Lagrange interpolation. This is the target privacy computation result of this round of federated learning (i.e., global model update), which can be used to update the global model. The entire process ensures that local data or model updates on a single node are not leaked to the server or other nodes.

[0069] In one feasible implementation, the step of initiating multi-party collaborative computation based on the data verification result to obtain the edge computation result fed back by each edge node includes: Determine the security protocol identifier of the local privacy computation result, and select the corresponding cloud aggregation computation graph based on the security protocol identifier. The cloud aggregation computation graph reflects the data transmission relationship between the edge nodes. Based on the cloud-based aggregated computing graph, multiple computing sub-graphs are obtained by partitioning the graph. The data verification results are encrypted according to the partition identifier of the computational partition graph to obtain encrypted data verification results; The encrypted data verification results are distributed according to the computational graph, and multi-party collaborative computation is initiated to obtain the edge computation results fed back by each edge node.

[0070] In the specific implementation, the cloud server first parses the `protocol_id` field carried in each verified local privacy computation result as a security protocol identifier. Assuming the protocol is identified as the "Shamir Secret Sharing (t, n) Threshold Scheme," the server then retrieves and loads the cloud-based aggregated computation graph that matches this protocol from its policy library. The graph may define a hierarchical aggregation logic, for example, requiring edge nodes to perform partial calculations between each other first, then designated aggregation nodes to perform intermediate aggregation, and finally send the results to the cloud.

[0071] Next, the server, based on the natural stages of the computation process or to optimize communication overhead, will... By partitioning the data, multiple computational subgraphs are obtained. .For example, It may include all edge nodes and their first-round interaction relationships. This includes the specified region aggregation nodes. Each subgraph It is assigned a unique sub-image identifier .

[0072] Then, the server performs encrypted distribution preparation. For each computational subgraph... and the nodes that need to receive data The server will use Temporary session public key Alternatively, based on the secret sharing protocol, the original data verification result (such as the secret share of node k) can be verified. This allows for encryption or re-sharing. Specifically, if point-to-point encryption is required, then generate... If (t, n) sharing is required within a subgraph, a new polynomial share is generated for each data item. All these encrypted or encoded data for a specific subgraph and node together constitute the set of encrypted data verification results.

[0073] Finally, the server distributes the encrypted data verification results based on the computational graph. It will belong to the graph. Data packets (containing encrypted data and execution instructions) are sent in batches. Within the relevant edge nodes, multi-party collaborative computation is initiated. After receiving the data, the node decrypts it within its local TEE and executes the computation specified by the subgraph (e.g., decrypting using its own private key). get Then calculate the intermediate function. Finally, the edge computing results generated locally (which may be new secret shares or partial aggregation results) are returned to the server or the next-hop node, thereby advancing the completion of the entire multi-party computation.

[0074] In one feasible implementation, the step of optimizing the cooperative scheduling strategy based on the target privacy calculation result to obtain the target cooperative scheduling strategy includes: Based on the computation time, resource consumption, and result quality of each edge node during this task scheduling process, performance evaluation indicators are generated. Calculate the utility difference between the actual utility and the expected utility of the target privacy calculation result, and determine the policy effectiveness evaluation result of the collaborative scheduling strategy based on the utility difference; The performance evaluation metrics and the policy effectiveness evaluation results are input into the reinforcement learning model to generate policy adjustment parameters. The cooperative scheduling strategy is optimized based on the strategy adjustment parameters to obtain the target cooperative scheduling strategy.

[0075] In its implementation, after completing a round of privacy-preserving computation, the cloud server first initiates an evaluation process. This process involves extracting data from the logs of each edge node. computation time Resource consumption (Can be normalized to a single comprehensive score) and result quality (e.g., the norm or accuracy of the local model update). Then, a performance evaluation metric is generated through a weighted function. ,For example: ,in These are weighting coefficients, reflecting the system's emphasis on different dimensions.

[0076] Meanwhile, server computational utility difference Assuming expected utility It is an improvement in model accuracy based on historical data prediction, but the actual utility... This is the new accuracy measured on the test set using the results of target privacy computation (such as the aggregated global model). The utility difference can then be expressed as... A larger positive A positive value indicates that the strategy is more effective than expected, while a negative value indicates the opposite. Based on... The size and sign of the policy, combined with business tolerance, allow the server to determine the effectiveness assessment result. For example, it can be quantized as a scalar between -1 (completely invalid) and 1 (very valid).

[0077] Finally, the server will evaluate performance metrics. and strategy effectiveness evaluation results As state features, these are input into a reinforcement learning model (such as the DDPG or PPO algorithm). The model takes the policy configuration for the current round as the action and evaluates the reward function. (in The model learns from hyperparameters. Through inference, it generates policy adjustment parameters, such as "increasing the weight of tasks assigned to nodes with high computational power by 0.1" or "extending the aggregation wait timeout by 5 seconds." Based on these parameters, the server optimizes the cooperative scheduling policy, updates the policy configuration file, and thus obtains a better-performing target cooperative scheduling policy that can be used for the next round of computation.

[0078] In one feasible implementation, the step of inputting the performance evaluation metric and the policy effectiveness evaluation result into the reinforcement learning model to generate policy adjustment parameters includes: Determine the network state, edge resource state, and task characteristics in this task scheduling, and encode the network state, edge resource state, and task characteristics to obtain a state vector; The performance evaluation metrics and the effectiveness evaluation results of the strategy will be used as reward signals. A value function is fitted based on the state vector and the reward signal; The state vector, the reward signal, and the value function are input into the reinforcement learning model to generate policy adjustment parameters.

[0079] In its implementation, the system initiates a reinforcement learning process at the end of each scheduling cycle. First, it retrieves a snapshot of the environment at the start of the cycle to determine the network state at that time (e.g., link latency). ,bandwidth ), edge resource status (e.g., CPU idle rate of each node) Memory availability ) and task characteristics (e.g., total amount of task data) , estimated number of units Next, this heterogeneous information is standardized and encoded to obtain a fixed-dimensional state vector. .For example, It can be a concatenated vector:

[0080] ,in It represents the number of nodes.

[0081] Then, the system combines the performance evaluation index P and the strategy effectiveness evaluation result E obtained in this round according to a predetermined rule as a reward signal. A simple linear combination is: The weight and This is set by the system administrator based on business priorities. Quantified in the previous state The old strategy (action) will be adopted. The immediate benefits obtained.

[0082] Next, the reinforcement learning agent needs to fit a value function to evaluate the long-term value of the state. Assuming the Q-learning algorithm family is used, the agent will maintain a Q-network. (Parameters are) Approximate action value function. It uses transition samples stored in the experience replay buffer. The network parameters are updated by minimizing the temporal difference error. The loss function is typically: ,in It is a discount factor. These are the parameters of the target network.

[0083] Finally, in the policy improvement phase, the agent will change the current state vector. The reward signal obtained and the well-fitted value function The input is fed into the reinforcement learning model. The model, based on the current value function, applies a policy (such as...) (Greedy strategy or direct output of deterministic actions) Generates the strategy adjustment parameters to be used in the next cycle. (For example, adjusting the task assignment weight vector). This process enables the scheduling strategy to adapt to the dynamic environment and continuously evolve towards maximizing long-term cumulative rewards.

[0084] This embodiment provides a lightweight privacy-preserving computing task scheduling method that integrates edge-cloud collaboration. It parses task requests to obtain their computational load and data privacy level, generating an initial edge-cloud collaborative scheduling strategy. Based on the strategy, the main task is divided into multiple computational subtasks with security protocol identifiers. At the edge, the scheduling node processes the subtasks in a trusted environment using a lightweight engine, generating protected local computation results. In the cloud, the results from each edge are aggregated through verification and multi-party collaboration mechanisms, and finally, the target privacy-preserving computation result is decoded. Based on the performance metrics and utility evaluation of this execution, the scheduling strategy is dynamically optimized using a reinforcement learning model. This approach achieves adaptive task scheduling that balances computational efficiency and data privacy security in resource-constrained edge environments, improving the overall performance and reliability of the collaborative computing system.

[0085] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the lightweight privacy computing task scheduling method of edge-cloud collaboration in this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0086] This application also provides a lightweight privacy computing task scheduling device that integrates edge-cloud collaboration. Please refer to [reference needed]. Figure 3 The lightweight privacy computing task scheduling device that integrates edge-cloud collaboration includes: The task parsing module 10 is used to obtain the task computation amount and data privacy level based on the privacy computing task request initiated by the edge terminal device; The strategy generation module 20 is used to obtain a collaborative scheduling strategy between the edge terminal device and the cloud server based on the task computation volume and the data privacy level. The task partitioning module 30 is used to divide the privacy computing task into multiple security computing subtasks according to the collaborative scheduling strategy. Edge computing module 40 is used to schedule edge nodes to perform local privacy computing based on the secure computing subtask, and obtain local privacy computing results; The cloud aggregation module 50 is used to aggregate the local privacy calculation results based on the collaborative mechanism of the cloud server to obtain the target privacy calculation result, and optimize the collaborative scheduling strategy based on the target privacy calculation result to obtain the target collaborative scheduling strategy.

[0087] In one feasible implementation, the strategy generation module 20 is further configured to predict the available computing resources of the edge terminal device and the expected load of the cloud server based on the task computation volume. Based on the aforementioned data privacy level, determine the security constraints required for data transmission between the edge and the cloud; An initial cooperative scheduling strategy is generated based on the available computing resources, the expected load, and the security constraints. The initial collaborative scheduling strategy is evaluated and calibrated based on historical scheduling performance data to obtain the collaborative scheduling strategy between the edge terminal device and the cloud server.

[0088] In one feasible implementation, the task partitioning module 30 is further configured to determine the cooperation mode in the cooperative scheduling strategy, and parse the dependency relationship between the data flow and control flow within the privacy computing task according to the cooperation mode; Based on the aforementioned dependencies and the aforementioned data privacy level, the computational boundaries and data anonymization rules for each subtask are determined. Based on the computational boundary and the data anonymization rules, the original task is reconstructed into multiple computational units with independent execution capabilities; A lightweight security protocol identifier is attached to each of the computing units to obtain a secure computing subtask.

[0089] In one feasible implementation, the edge computing module 40 is further configured to determine the security protocol identifier of the security computing subtask based on the security computing subtask, and allocate a corresponding local trusted execution environment to the edge node according to the security protocol identifier; Based on the local trusted execution environment, load and initialize the lightweight privacy computing engine corresponding to the secure computing subtask; The lightweight privacy computing engine is scheduled to process the raw data associated with subtasks within the local trusted execution environment, generating protected intermediate state data. The intermediate state data is verified for integrity and signed, and the result is output as the local privacy calculation result.

[0090] In one feasible implementation, the edge computing module 40 is further configured to schedule the lightweight privacy computing engine, and based on the data desensitization rules of the subtask and the lightweight privacy computing engine, perform lightweight homomorphic encryption processing on the original data within the local trusted execution environment to obtain the processed data; The processed data is then segmented to obtain segmented data; The fragmented data is encoded to obtain preliminary results; Noise is added to the preliminary results to generate protected intermediate state data.

[0091] In one feasible implementation, the cloud aggregation module 50 is further configured to perform data verification on local privacy results from multiple edge nodes to obtain data verification results; Based on the collaborative mechanism of the cloud server, multi-party collaborative computation is initiated on the data verification results to obtain the edge computing results fed back by each edge node; The edge computing results are aggregated to obtain aggregated data, and the aggregated data is decoded to obtain the target privacy computing results.

[0092] In one feasible implementation, the cloud aggregation module 50 is further configured to determine the security protocol identifier of the local privacy computation result, and select the corresponding cloud aggregation computation graph according to the security protocol identifier, wherein the cloud aggregation computation graph reflects the data transmission relationship between the edge nodes; Based on the cloud-based aggregated computing graph, multiple computing sub-graphs are obtained by partitioning the graph. The data verification results are encrypted according to the partition identifier of the computational partition graph to obtain encrypted data verification results; The encrypted data verification results are distributed according to the computational graph, and multi-party collaborative computation is initiated to obtain the edge computation results fed back by each edge node.

[0093] In one feasible implementation, the cloud aggregation module 50 is further configured to generate performance evaluation indicators based on the computation time, resource consumption, and result quality of each edge node during the current task scheduling process. Calculate the utility difference between the actual utility and the expected utility of the target privacy calculation result, and determine the policy effectiveness evaluation result of the collaborative scheduling strategy based on the utility difference; The performance evaluation metrics and the policy effectiveness evaluation results are input into the reinforcement learning model to generate policy adjustment parameters. The cooperative scheduling strategy is optimized based on the strategy adjustment parameters to obtain the target cooperative scheduling strategy.

[0094] In one feasible implementation, the cloud aggregation module 50 is further configured to determine the network status, edge resource status, and task characteristics in the current task scheduling, and to encode the network status, edge resource status, and task characteristics to obtain a state vector; The performance evaluation metrics and the effectiveness evaluation results of the strategy will be used as reward signals. A value function is fitted based on the state vector and the reward signal; The state vector, the reward signal, and the value function are input into the reinforcement learning model to generate policy adjustment parameters.

[0095] The lightweight privacy computing task scheduling device integrating edge-cloud collaboration provided in this application adopts the lightweight privacy computing task scheduling method integrating edge-cloud collaboration in the above embodiments, which can solve the technical problem of difficulty in balancing privacy protection and task scheduling efficiency in edge computing scenarios. Compared with the prior art, the beneficial effects of the lightweight privacy computing task scheduling device integrating edge-cloud collaboration provided in this application are the same as the beneficial effects of the lightweight privacy computing task scheduling method integrating edge-cloud collaboration provided in the above embodiments, and other technical features in the lightweight privacy computing task scheduling device integrating edge-cloud collaboration are the same as the features disclosed in the methods of the above embodiments, and will not be repeated here.

[0096] This application provides a lightweight privacy computing task scheduling device that integrates edge-cloud collaboration. The lightweight privacy computing task scheduling device that integrates edge-cloud collaboration includes: at least one processor; and a memory communicatively connected to at least one processor; wherein the memory stores instructions that can be executed by at least one processor, and the instructions are executed by at least one processor to enable at least one processor to execute the lightweight privacy computing task scheduling method that integrates edge-cloud collaboration in the above embodiment 1.

[0097] The following is for reference. Figure 4 This document illustrates a structural schematic of a lightweight privacy computing task scheduling device suitable for implementing the converged edge-cloud collaboration embodiments of this application. The lightweight privacy computing task scheduling device for converged edge-cloud collaboration in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 4 The lightweight privacy computing task scheduling device with converged edge-cloud collaboration shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0098] like Figure 4As shown, the lightweight privacy computing task scheduling device that integrates edge-cloud collaboration may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to programs stored in ROM (Read Only Memory) 1002 or programs loaded from storage device 1003 into RAM (Random Access Memory) 1004. RAM 1004 also stores various programs and data required for the operation of the lightweight privacy computing task scheduling device that integrates edge-cloud collaboration. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via bus 1005. Input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, LCDs (Liquid Crystal Displays), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard drives, etc.; and communication devices 1009. Communication device 1009 allows the converged edge-cloud collaborative lightweight privacy computing task scheduling device to wirelessly or wiredly communicate with other devices to exchange data. Although the figure shows a converged edge-cloud collaborative lightweight privacy computing task scheduling device with various systems, it should be understood that implementing or having all the systems shown is not required. More or fewer systems can be implemented alternatively.

[0099] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0100] The lightweight privacy computing task scheduling device integrating edge-cloud collaboration provided in this application adopts the lightweight privacy computing task scheduling method integrating edge-cloud collaboration in the above embodiments, and can solve the technical problem of lightweight privacy computing task scheduling integrating edge-cloud collaboration. Compared with the prior art, the beneficial effects of the lightweight privacy computing task scheduling device integrating edge-cloud collaboration provided in this application are the same as the beneficial effects of the lightweight privacy computing task scheduling method integrating edge-cloud collaboration provided in the above embodiments, and other technical features in the lightweight privacy computing task scheduling device integrating edge-cloud collaboration are the same as the features disclosed in the method of the previous embodiment, and will not be repeated here.

[0101] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0102] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0103] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, which are used to execute the lightweight privacy computing task scheduling method for fusion edge-cloud collaboration in the above embodiments.

[0104] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, RAM (Random Access Memory), ROM (Read Only Memory), EPROM (Erasable Programmable Read Only Memory or Flash Memory), optical fibers, CD-ROM (CD-Read Only Memory), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0105] The aforementioned computer-readable storage medium may be included in a lightweight privacy computing task scheduling device that integrates edge-cloud collaboration; or it may exist independently and not be assembled into a lightweight privacy computing task scheduling device that integrates edge-cloud collaboration.

[0106] The aforementioned computer-readable storage medium carries one or more programs. When the aforementioned one or more programs are executed by the lightweight privacy computing task scheduling device that integrates edge-cloud collaboration, the lightweight privacy computing task scheduling device that integrates edge-cloud collaboration obtains the task computing load and data privacy level based on the privacy computing task request initiated by the edge terminal device. Based on the task computation volume and the data privacy level, a collaborative scheduling strategy between the edge terminal device and the cloud server is obtained; According to the collaborative scheduling strategy, the privacy computing task is divided into multiple secure computing subtasks; Based on the aforementioned secure computing subtask, edge nodes are scheduled to perform local privacy computing to obtain local privacy computing results; The local privacy computation results are aggregated based on the collaborative mechanism of the cloud server to obtain the target privacy computation result. The collaborative scheduling strategy is then optimized based on the target privacy computation result to obtain the target collaborative scheduling strategy.

[0107] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including LAN (Local Area Network) or WAN (Wide Area Network)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0108] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0109] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0110] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described lightweight privacy computing task scheduling method for converged edge-cloud collaboration. This solves the technical problem of scheduling lightweight privacy computing tasks for converged edge-cloud collaboration. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the lightweight privacy computing task scheduling method for converged edge-cloud collaboration provided in the above embodiments, and will not be repeated here.

[0111] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the lightweight privacy computing task scheduling method for converged edge-cloud collaboration as described above.

[0112] The computer program product provided in this application can solve the technical problem of lightweight privacy computing task scheduling with converged edge-cloud collaboration. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the lightweight privacy computing task scheduling method with converged edge-cloud collaboration provided in the above embodiments, and will not be repeated here.

[0113] The above are only some embodiments of this application and do not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the content of this application specification and drawings, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A method for scheduling lightweight privacy computing tasks in a fusion of edge-cloud collaboration, characterized in that, The lightweight privacy computing task scheduling method that integrates edge-cloud collaboration includes: Based on the privacy computing task request initiated by the edge terminal device, the task computing workload and data privacy level are obtained; Based on the task computation volume and the data privacy level, a collaborative scheduling strategy between the edge terminal device and the cloud server is obtained; According to the collaborative scheduling strategy, the privacy computing task is divided into multiple secure computing subtasks; Based on the aforementioned secure computing subtask, edge nodes are scheduled to perform local privacy computing to obtain local privacy computing results; The local privacy computation results are aggregated based on the collaborative mechanism of the cloud server to obtain the target privacy computation result. The collaborative scheduling strategy is then optimized based on the target privacy computation result to obtain the target collaborative scheduling strategy. The step of obtaining the collaborative scheduling strategy between the edge terminal device and the cloud server based on the task computation volume and the data privacy level includes: Based on the computational workload of the task, predict the available computing resources of the edge terminal devices and the expected load of the cloud server; Based on the aforementioned data privacy level, determine the security constraints required for data transmission between the edge and the cloud; An initial cooperative scheduling strategy is generated based on the available computing resources, the expected load, and the security constraints. The initial collaborative scheduling strategy is evaluated and calibrated based on historical scheduling performance data to obtain the collaborative scheduling strategy between the edge terminal device and the cloud server. The step of optimizing the cooperative scheduling strategy based on the target privacy calculation result to obtain the target cooperative scheduling strategy includes: Based on the computation time, resource consumption, and result quality of each edge node during this task scheduling process, performance evaluation indicators are generated. Calculate the utility difference between the actual utility and the expected utility of the target privacy calculation result, and determine the policy effectiveness evaluation result of the collaborative scheduling strategy based on the utility difference; The performance evaluation metrics and the policy effectiveness evaluation results are input into the reinforcement learning model to generate policy adjustment parameters. The cooperative scheduling strategy is optimized based on the strategy adjustment parameters to obtain the target cooperative scheduling strategy; The step of inputting the performance evaluation metric and the policy effectiveness evaluation result into the reinforcement learning model to generate policy adjustment parameters includes: Determine the network state, edge resource state, and task characteristics in this task scheduling, and encode the network state, edge resource state, and task characteristics to obtain a state vector; The performance evaluation metrics and the effectiveness evaluation results of the strategy will be used as reward signals. A value function is fitted based on the state vector and the reward signal; The state vector, the reward signal, and the value function are input into the reinforcement learning model to generate policy adjustment parameters.

2. The method of claim 1, wherein, The step of dividing the privacy computing task into multiple security computing subtasks according to the cooperative scheduling strategy includes: Determine the collaboration mode in the collaborative scheduling strategy, and parse the dependency relationship between the data flow and control flow inside the privacy computing task based on the collaboration mode; Based on the aforementioned dependencies and the aforementioned data privacy level, the computational boundaries and data anonymization rules for each subtask are determined. Based on the computational boundary and the data anonymization rules, the original task is reconstructed into multiple computational units with independent execution capabilities; A lightweight security protocol identifier is attached to each of the computing units to obtain a secure computing subtask.

3. The method of claim 1, wherein, The step of scheduling edge nodes to perform local privacy computation based on the secure computation subtask and obtaining the local privacy computation result includes: Based on the secure computing subtask, a security protocol identifier for the secure computing subtask is determined, and a corresponding local trusted execution environment is allocated to the edge node according to the security protocol identifier; Based on the local trusted execution environment, load and initialize the lightweight privacy computing engine corresponding to the secure computing subtask; The lightweight privacy computing engine is scheduled to process the raw data associated with subtasks within the local trusted execution environment, generating protected intermediate state data. The intermediate state data is verified for integrity and signed, and the result is output as the local privacy calculation result.

4. The method of claim 3, wherein, The step of scheduling the lightweight privacy computing engine to process the raw data associated with subtasks within the local trusted execution environment and generate protected intermediate state data includes: The lightweight privacy computing engine is scheduled, and based on the data desensitization rules of the subtask and the lightweight privacy computing engine, the original data is processed using lightweight homomorphic encryption within the local trusted execution environment to obtain the processed data; The processed data is then segmented to obtain segmented data; The fragmented data is encoded to obtain preliminary results; Noise is added to the preliminary results to generate protected intermediate state data.

5. The method of claim 1, wherein, The steps of aggregating the local privacy computation results using the cloud server-based collaborative mechanism to obtain the target privacy computation result include: Data verification is performed on the local privacy results from multiple edge nodes to obtain data verification results; Based on the collaborative mechanism of the cloud server, multi-party collaborative computation is initiated on the data verification results to obtain the edge computing results fed back by each edge node; The edge computing results are aggregated to obtain aggregated data, and the aggregated data is decoded to obtain the target privacy computing results.

6. The method of claim 5, wherein, The steps for initiating multi-party collaborative computation based on the data verification results to obtain the edge computation results fed back by each edge node include: Determine the security protocol identifier of the local privacy computation result, and select the corresponding cloud aggregation computation graph based on the security protocol identifier. The cloud aggregation computation graph reflects the data transmission relationship between the edge nodes. Based on the cloud-based aggregated computing graph, multiple computing sub-graphs are obtained by partitioning the graph. The data verification results are encrypted according to the partition identifier of the computational partition graph to obtain encrypted data verification results; The encrypted data verification results are distributed according to the computational graph, and multi-party collaborative computation is initiated to obtain the edge computation results fed back by each edge node.

7. A device for scheduling lightweight privacy computing tasks in a fusion edge-cloud collaboration, characterized in that, The lightweight privacy computing task scheduling device that integrates edge-cloud collaboration includes: The task parsing module is used to obtain the task computation amount and data privacy level based on the privacy computing task request initiated by the edge terminal device; The strategy generation module is used to obtain the collaborative scheduling strategy between the edge terminal device and the cloud server based on the task computation volume and the data privacy level. The task partitioning module is used to divide the privacy computing task into multiple security computing subtasks according to the collaborative scheduling strategy. The edge computing module is used to schedule edge nodes to perform local privacy computing based on the secure computing subtask, and obtain local privacy computing results; The cloud aggregation module is used to aggregate the local privacy computing results based on the collaborative mechanism of the cloud server to obtain the target privacy computing result, and optimize the collaborative scheduling strategy based on the target privacy computing result to obtain the target collaborative scheduling strategy. The step of obtaining the collaborative scheduling strategy between the edge terminal device and the cloud server based on the task computation volume and the data privacy level includes: Based on the computational workload of the task, predict the available computing resources of the edge terminal devices and the expected load of the cloud server; Based on the aforementioned data privacy level, determine the security constraints required for data transmission between the edge and the cloud; An initial cooperative scheduling strategy is generated based on the available computing resources, the expected load, and the security constraints. The initial collaborative scheduling strategy is evaluated and calibrated based on historical scheduling performance data to obtain the collaborative scheduling strategy between the edge terminal device and the cloud server. The step of optimizing the cooperative scheduling strategy based on the target privacy calculation result to obtain the target cooperative scheduling strategy includes: Based on the computation time, resource consumption, and result quality of each edge node during this task scheduling process, performance evaluation indicators are generated. Calculate the utility difference between the actual utility and the expected utility of the target privacy calculation result, and determine the policy effectiveness evaluation result of the collaborative scheduling strategy based on the utility difference; The performance evaluation metrics and the policy effectiveness evaluation results are input into the reinforcement learning model to generate policy adjustment parameters. The cooperative scheduling strategy is optimized based on the strategy adjustment parameters to obtain the target cooperative scheduling strategy; The step of inputting the performance evaluation metric and the policy effectiveness evaluation result into the reinforcement learning model to generate policy adjustment parameters includes: Determine the network state, edge resource state, and task characteristics in this task scheduling, and encode the network state, edge resource state, and task characteristics to obtain a state vector; The performance evaluation metrics and the effectiveness evaluation results of the strategy will be used as reward signals. A value function is fitted based on the state vector and the reward signal; The state vector, the reward signal, and the value function are input into the reinforcement learning model to generate policy adjustment parameters.

Images