Edge computing intelligent resource management system and method based on service insurance mechanism

Abstract

The invention belongs to the technical field of computer and network communication, and particularly relates to an edge computing intelligent resource management system and method based on a service insurance mechanism, and the system comprises a request receiving module which receives a user computing task request; the system state sensing module is used for collecting current system state information of the edge computing node; the service insurance market module is used for obtaining service insurance contract options and an optimal execution plan based on the calculation task request and the current system state information; the task scheduling and execution module provides the service insurance contract options and the service quality default probability of the corresponding optimal execution plan for the user to select; and the edge computing node executes the optimal execution plan corresponding to the optimal service insurance contract selected by the user. According to the method, the uncertainty of edge computing service performance is converted into quantifiable economic risk cost, pure pursuit of technical indexes is converted into an intelligent decision normal form oriented to economic optimization, and service quality guarantee and system economic benefits are effectively balanced.

Description

Technical Field

[0001] This invention relates to the field of computer and network communication technology, and specifically to an edge computing intelligent resource management system and method based on a service insurance mechanism. Background Technology

[0002] With the widespread deployment of the Internet of Things (IoT), Artificial Intelligence (AI), Augmented Reality (AR), and 5G communication technologies, edge computing, as a distributed computing paradigm that pushes computing power and data storage to the network edge closer to the data source, is playing an increasingly crucial role in meeting the low latency, high bandwidth, and data privacy requirements of various emerging applications. Typical application scenarios include real-time decision-making for autonomous vehicles, intelligent monitoring of the Industrial Internet of Things (IIoT), and real-time analysis of high-definition video.

[0003] However, edge computing environments are inherently highly dynamic and uncertain. The computing resources (such as CPU / GPU load), storage resources, and network connectivity (such as bandwidth, latency, and packet loss rate) of edge computing nodes can fluctuate dramatically due to sudden traffic surges, user movement, or changes in the physical environment. This inherent resource instability presents significant and inherent business risks for service providers (ESPs) in fulfilling their Service Level Agreements (SLAs) to users, such as a defined latency cap or processing reliability.

[0004] Therefore, there is an urgent need to build a comprehensive service quality insurance market that can adapt to the edge computing environment. The aim is to dynamically price and underwrite different levels of service quality assurance based on users' risk preferences, so as to more intelligently balance the relationship between service quality assurance, user demand matching and economic benefits under conditions of limited resources and environmental uncertainty. Summary of the Invention

[0005] Based on this, the present invention proposes an edge computing intelligent resource management system and method based on a service insurance mechanism to overcome the shortcomings of existing edge computing environments in handling service quality uncertainty.

[0006] This invention transforms the performance uncertainty of edge computing services (such as real-time video analytics and AI inference) under dynamic network and load environments into a priced and tradable insurance product through an innovative mechanism. It also establishes an intelligent closed loop for edge computing node resource allocation and task acceptance decisions, enabling edge computing resource management to shift from being driven by purely technical indicators (such as changes in response latency) to proactive management and economically optimal decision-making based on risk and cost. This achieves personalized customization of service guarantees for different users and maximizes the overall long-term economic benefits of the system.

[0007] In a first aspect, embodiments of this application provide an edge computing intelligent resource management system based on a service insurance mechanism, including: The request receiving module is used to receive computing task requests from different users; The system status awareness module is used to collect and aggregate the current system status information of the edge computing nodes managed by the system in real time and periodically. The service insurance market module is used to obtain service insurance contract options and their corresponding optimal execution plans based on computing task requests and the current system status information of edge computing nodes; The task scheduling and execution module is used to provide the user with the service insurance contract options, along with the service quality default probability of the optimal execution plan corresponding to each service insurance contract option, for the user to choose from. When the user selects the optimal service insurance contract from the service insurance contract options, the module obtains the optimal execution plan corresponding to the optimal service insurance contract from the service insurance market module and sends the optimal execution plan to the designated edge computing node for execution.

[0008] In one possible implementation, the service insurance contract options include a service price, a service level commitment, and a penalty value for breach of contract.

[0009] In one possible implementation, the service insurance market module includes: An insurance product generator is used to dynamically provide at least one service insurance contract option for each received computation task request; The risk actuarial engine is used to generate a candidate execution plan group based on each service insurance contract option and the current system status information of the edge computing node; predict the service quality default probability of each candidate execution plan in the candidate execution plan group; and calculate the expected compensation cost corresponding to each candidate execution plan based on the service quality default probability and the default penalty value of the service insurance contract option. The underwriting decision and risk control module is used to generate the total expected underwriting cost based on basic operating costs and expected claims costs, and to determine an optimal execution plan and service price for each service insurance contract option by minimizing the total expected underwriting cost.

[0010] In one possible implementation, the formula for calculating the expected compensation cost is: in, Indicates the expected compensation cost; Indicates the first Tasks for each user; Indicates the first One service insurance contract option; Indicates the first in the candidate implementation plan group One candidate implementation plan; Indicates the probability of service quality breach; Indicates time slot System status information of edge computing nodes; This represents the penalty value for breach of contract.

[0011] In one possible implementation, the formula for calculating the total expected underwriting cost is: in, This represents the total expected underwriting cost; Indicates the first Tasks for each user; Indicates the first One service insurance contract option; Indicates the first in the candidate implementation plan group One candidate implementation plan; Indicates basic operating costs, This indicates the expected cost of compensation.

[0012] In one possible implementation, the system further includes a strategy learning and optimization module, which collects the execution results of the optimal execution plan and learns and updates the decision model of the service insurance market module based on the execution results.

[0013] In one possible implementation, the policy learning and optimization module is a reinforcement learning agent that aims to maximize the long-term cumulative return of the system, with the objective function being: st in, This represents the overall decision-making strategy of a reinforcement learning agent. Represents the mathematical expectation operator; t Indicates the sequence number of the time slot. t =0 indicates the initial moment or the start of the first decision cycle; T This represents the total time slots of the optimization process; T -1 indicates the previous time slot in the total time slot; Indicates the discount factor; Indicates the first Tasks for each user; Indicates the first One service insurance contract option; Indicates the first The task of the user and the first The task is obtained by combining service insurance contract options; Indicates time slot t The collection of all best-service insurance contracts successfully signed within the country; Indicates time slot t The following service prices; This represents the optimal execution plan; This represents the total expected underwriting cost of executing the optimal execution plan; Indicates time slot t All edge computing nodes currently in use k The task set; Represents an edge computing node; Indicates the optimal execution plan Under the guidance of edge computing nodes Computational tasks The amount of computing resources consumed; Represents edge computing nodes k Total resource capacity; This represents the set of resources for all edge computing nodes; Indicates the execution of the optimal execution plan. Certainty accuracy; Indicates the minimum accuracy required to make a commitment; Indicates time slot t+ System status information of edge computing nodes at time 1; Represents the system state transition function. Indicates time slot System status information of edge computing nodes at that time. This indicates that the reinforcement learning agent is in the time slot. t Internally, it refers to the systematic actions taken based on the overall decision-making strategy. Indicates time slot t Random events that occur within the system and cannot be fully controlled.

[0014] Secondly, embodiments of this application provide an edge computing intelligent resource management method based on a service insurance mechanism. The method is based on the edge computing intelligent resource management system described in the first aspect and includes: Step S1, in any time slot t It obtains one or more computing task requests from the user and the current system status information of the edge computing nodes; Step S2: Based on the computing task request and the current system status information of the edge computing node, formulate at least one service insurance contract option; Step S3: Perform an actuarial assessment of the risks and costs for each service insurance contract option; Step S4: Based on the actuarial results of each service insurance contract option, determine an optimal execution plan for each service insurance contract option, and dynamically formulate or adjust the service price of the corresponding service insurance contract option based on the total expected underwriting cost of the optimal execution plan. Step S5: Provide the user with the at least one service insurance contract option, along with the service quality default probability of the optimal execution plan corresponding to each service insurance contract option, for the user to choose from. Step S6: If the user receives at least one service insurance contract option and selects one as the optimal service insurance contract, proceed to S7; if the user does not receive at least one service insurance contract option, the task ends. Step S7: After receiving the optimal service insurance contract selected by the user, the system sends the optimal execution plan corresponding to the optimal service insurance contract to the edge computing node for execution. Step S8: Collect execution results and update the decision model based on the execution results.

[0015] In one possible implementation, step S3 includes: Step S31: Generate a candidate execution plan group based on each service insurance contract option and the current system status information of the edge computing node; Step S32: Predict the service quality default probability of any candidate execution plan in the candidate execution plan group; Step S33: Calculate the expected compensation cost corresponding to the candidate execution plan based on the service quality default probability and the default penalty value of the service insurance contract option; Step S34: Combine the expected claims cost with the basic operating costs required to execute the candidate execution plan to obtain the total expected underwriting cost required to execute the candidate execution plan; Step S35: Repeat steps S32-S34 to obtain the total expected underwriting cost for all candidate execution plans in the candidate execution plan group.

[0016] In one possible implementation, the optimal execution plan is the candidate execution plan from the group of candidate execution plans that minimizes the total expected underwriting cost, as shown in the following objective formula: in, This represents the optimal execution plan; Indicates the first Tasks for each user; Indicates the first One service insurance contract option; Indicates the first in the candidate implementation plan group One candidate implementation plan; Indicates the candidate implementation plan group; Indicates basic operating costs; Indicates the probability of service quality breach; Indicates time slot System status information of edge computing nodes; This represents the penalty value for breach of contract.

[0017] Compared with the prior art, the technical solution disclosed in this invention has the following significant technical effects: This invention constructs a complete service insurance market and achieves business model innovation. For the first time, this invention establishes a complete insurance mechanism in the field of edge computing, encompassing customizable insurance products, dynamic premiums, actuarial science, and intelligent risk control.

[0018] This invention enables risk internalization decision-making based on economic rationality, thereby improving decision-making quality. Innovatively, this invention internalizes expected compensation costs as a real-time, quantifiable economic cost item into the decision-making model for resource allocation and task acceptance. This ensures that all system decisions are based on the economic accounting of risk-adjusted total costs, rather than purely technical indicators. This fundamentally guarantees the economic rationality and optimality of resource allocation, effectively avoiding unreasonable economic costs incurred to achieve technical targets.

[0019] This invention improves the accuracy of risk prediction and the system's adaptability through collaborative learning. The multi-task learning evaluation network employed in this invention enables the prediction of risk probability and the assessment of long-term value to share underlying features and mutually reinforce each other, thereby improving the accuracy of risk assessment. The entire system aims to maximize risk-adjusted long-term returns and can intelligently adapt to changes in market demand and resource status, constructing a robust and efficient edge computing intelligent service method. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a schematic diagram illustrating the application of the edge computing intelligent resource management system and method based on service insurance mechanism provided in the embodiments of the present invention in the low-altitude economic scenario; Figure 2 A schematic diagram of the overall architecture of the edge computing intelligent resource management system based on a service insurance mechanism provided in an embodiment of the present invention; Figure 3A flowchart of an edge computing intelligent resource management method based on a service insurance mechanism provided in an embodiment of the present invention; Figure 4 A comparative diagram of the cumulative revenue of service providers under different strategies provided in the embodiments of this application; Figure 5 A schematic diagram comparing the SLA default rates of high-priority tasks under different system loads, provided for embodiments of this application. Figure 6 This diagram illustrates the comparison of average computing resource utilization under different strategies provided in the embodiments of this application. Detailed Implementation

[0022] To better understand the technical solution of this application, the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0023] It should be understood that the described embodiments are merely some, not all, of the embodiments in this application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without inventive effort are within the scope of protection of this application.

[0024] The terminology used in the embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. The singular forms “a,” “the,” and “the” used in the embodiments of this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise.

[0025] It should be understood that the term "and / or" used in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0026] The edge computing intelligent resource management system and method based on a service insurance mechanism provided in this invention can be applied to scenarios with requirements for service quality reliability and real-time performance, such as augmented reality, low-altitude economy, vehicle-to-everything (V2X) communication, and autonomous driving assistance. This embodiment uses a simple low-altitude economy scenario as an example. (See attached image.) Figure 1 This is a schematic diagram illustrating the application of the edge computing intelligent resource management system and method based on a service insurance mechanism provided in this invention in a low-altitude economic scenario. Figure 1As shown, in a complex urban environment, a medical drone performing an emergency medical supply transport mission and a regular logistics drone performing routine delivery are flying simultaneously. On the ground, multiple 5G base stations and edge computing nodes (MECs) provided by edge service providers are deployed. In this scenario, the Quality of Service (QoS) requirements of the drones performing different tasks are drastically different. For edge service providers, the core challenge lies in how to provide ultimate reliability assurance for high-value medical drones while also cost-effectively serving regular logistics drones with limited computing and network resources, thereby optimizing the overall system's economic benefits. To address this challenge, this embodiment provides an edge computing intelligent resource management system and method based on a service insurance mechanism. Its core lies in building an intelligent computing service insurance market for users (drone operators in low-altitude economic scenarios) and edge service providers. This intelligent computing service insurance market, by introducing risk pricing and dynamic underwriting mechanisms, meets the differentiated reliability needs of different users while optimizing the overall system's economic benefits.

[0027] See Figure 2 This is a schematic diagram of the overall architecture of an edge computing intelligent resource management system based on a service insurance mechanism provided in an embodiment of the present invention. The system can be deployed on a cloud server or a 5G core network controller to provide users with highly reliable, low-latency intelligent computing services. Figure 2 As shown, the system includes: The request receiving module 110 is the communication interface between the system and the user, used to receive computing task requests from different users. In this embodiment... Figure 1 In the low-altitude economic scenario shown, the system may simultaneously receive visual obstacle avoidance calculation requests from a medical drone and a similar request from a conventional logistics drone. In this scenario, the mission value and safety risks of the medical drone are extremely high; therefore, drone operators will consider user risk preference parameters. Setting it to an extremely high value close to 1 (e.g., 0.95), in contrast, for conventional logistics drones, drone operators might set the user risk preference parameter to... Set it to a low value (e.g., 0.5).

[0028] The system state awareness module 120 acts as the system's "sensor," collecting and aggregating the current system state information of the edge computing nodes (MECs) managed by the system in real time and periodically. This system state information is high-dimensional, including: the current load percentage, memory usage, network interface throughput and latency of each edge computing node's computing units (such as CPUs and GPUs); the task processing queue length of each edge computing node; the expected total computational load of all tasks in the queue; the detailed status of queues with different priorities; and the global network state, such as the historical average round-trip time (RTT) between user devices and each edge computing node. The system state awareness module 120 integrates the collected current system state information into a structured state vector or tensor for system decision-making.

[0029] The service insurance market module 130 is the core decision-making hub of the system. It is designed to simulate an automated insurance underwriting and pricing department. It takes the computing task requests received by the request receiving module 110 and the current system status information of the edge computing nodes collected by the system status awareness module 120 as decision inputs, and outputs service insurance contract options and their corresponding optimal execution plans.

[0030] Furthermore, the service insurance market module 130 includes an insurance product generator, a risk actuarial engine, and an underwriting decision and risk control module. Specifically: The insurance product generator dynamically provides at least one service insurance contract option for each received computing task request. In a preferred embodiment, the insurance product generator can access a database storing multiple insurance templates, each defining a basic coverage framework. For example, insurance template A defines a high-level Service Level Agreement (SLA) and a high penalty for breach of contract, while insurance template B defines a lower-level SLA and a lower penalty for breach of contract. The insurance product generator works by dynamically generating a competitive service price (i.e., a "premium") for these insurance templates based on current system state information and user risk preference parameters, thereby combining them into the final service insurance contract option offered to the user. For example, when a computing task request for a medical drone is received, based on its extremely high user risk preference parameters, the insurance product generator might create a "life-protection level" contract draft, which promises extremely low latency (e.g., <30ms) and extremely high obstacle avoidance accuracy (e.g., >99.999%), and sets a high penalty for breach of contract. For computational task requests from conventional logistics drones, the insurance product generator may generate a draft "standard express delivery-grade" contract with relatively low service level agreements (SLAs) and penalties for breach of contract. At the same time, the insurance product generator will dynamically generate a competitive service price, i.e., a "premium," for each of these draft contracts based on the current system status. The generator will then combine these draft contracts with their corresponding service prices to create service insurance contract options for users.

[0031] Preferably, the service insurance contract options include a service price (i.e., "premium"), a service level commitment (i.e., "coverage"), a penalty value for service default (i.e., "insurance amount"), and an optional service priority.

[0032] The risk actuarial engine, a key innovation distinguishing this invention from existing technologies, generates a candidate execution plan group based on each service insurance contract option and the current system state information of the edge computing node; predicts the service quality default probability of each candidate execution plan in the candidate execution plan group; and calculates the expected compensation cost corresponding to each candidate execution plan based on the service quality default probability and the default penalty value of the service insurance contract option, so as to transform the "probability of occurrence" into an economic cost, which is the mathematical expectation of future potential compensation.

[0033] In other words, before the service insurance contract option is formally offered to the user (i.e., the drone operator), the risk actuarial engine first performs a precise quantitative assessment of the insured risk. It needs to predict the probability of service quality default (i.e., the "probability of failure") for each candidate execution plan in the candidate execution plan group (e.g., allocating the video stream from the medical drone to the MEC node A with the lowest current load and reserving a dedicated 5G network slice for it) under the current system state information of the edge computing node. In this scenario, "failure" means service quality default, for example, the actual end-to-end visual processing latency exceeds the 30ms upper limit promised in the contract.

[0034] Furthermore, default events can be used. This indicates a breach of service quality; the breach event. Defined as ,in, Indicates the first Tasks for each user; Indicates the first One service insurance contract option; Indicates the execution of candidate execution plans. Deterministic accuracy at the time; Indicates the minimum accuracy required to make a commitment; Indicates time slot t System status information of edge computing nodes; Indicates time slot t Edge computing nodes execute candidate execution plans The expected end-to-end latency is a random variable that depends on the current system state information of the edge computing nodes; Indicates the maximum delay of the commitment.

[0035] Furthermore, each candidate execution plan in the candidate execution plan group defines a series of operations such as DNN partitioning, early exit, node allocation, resource allocation, and priority. For example, candidate execution plans... Assign a high priority to the task on the medical drone, lock its visual obstacle avoidance task onto the high-performance MEC node with the lowest current load, and use full DNN inference without premature exit; candidate execution plan. Tasks from conventional logistics drones are rescheduled to edge computing nodes that have high-performance GPUs, even though the workload is slightly higher.

[0036] The working principle of generating a candidate execution plan group based on each service insurance contract option and the current system status information of the edge computing node is as follows: After receiving the current system status information and the service level commitment (SLA) required by the service insurance contract option as input, the risk actuarial engine combines decision dimensions such as node selection, resource allocation, and inference strategy (e.g., whether to exit early, which split point to choose) to generate a set of candidate plans.

[0037] Furthermore, the service quality default probability can be predicted using a trained machine learning model (e.g., a deep neural network). The machine learning model takes current system state information (such as node load and network latency) and candidate execution plans (such as resource allocation and task priority) as input features. Its output layer is activated by a sigmoid function, directly generating a scalar value between 0 and 1, which is the predicted service quality default probability. The accuracy of the machine learning model is ensured through supervised learning on historical system data. The training objective is to minimize the cross-entropy loss between the predicted service quality default probability and the actual default result.

[0038] Furthermore, the formula for calculating the expected compensation cost is as follows: in, Indicates the first Tasks for each user; Indicates the first One service insurance contract option; Indicates the expected compensation cost; Indicates the first in the candidate implementation plan group One candidate implementation plan; Indicates the probability of service quality breach; Indicates time slot System status information of edge computing nodes; This represents the penalty value for breach of contract.

[0039] The underwriting decision and risk control module is the "final decision-maker" of the system. It is used to generate the total expected underwriting cost based on the basic operating cost and expected claims cost, and determine an optimal execution plan and service price for each service insurance contract option by minimizing the total expected underwriting cost.

[0040] The formula for calculating total expected underwriting cost is: in, This represents the total expected underwriting cost; Indicates basic operating costs; This indicates the expected cost of compensation.

[0041] Basic operating costs The calculation formula is: in, This represents the price per unit of computing resources, referring to the price at an edge computing node. k The price required to perform one standard unit of calculation; This indicates that the candidate execution plan is executed on the edge node. The total computational workload required, i.e., the edge computing workload; Indicates the calculation cost; Indicates the unit price of network transmission; Indicates the amount of intermediate data transmitted; This indicates network costs.

[0042] The optimal execution plan is the candidate execution plan that minimizes the total expected underwriting cost from the candidate execution plan group. In other words, it involves solving the following optimization problem: in, This represents the optimal execution plan; Indicates the first Tasks for each user; Indicates the first One service insurance contract option; Indicates the first in the candidate implementation plan group One candidate implementation plan; Indicates the candidate implementation plan group; Indicates basic operating costs; Indicates the probability of service quality breach; Indicates time slot System status information of edge computing nodes; This represents the penalty value for breach of contract.

[0043] The task scheduling and execution module 140 is used to execute the signed "service insurance contract". It provides the user with service insurance contract options, along with the service quality default probability of the optimal execution plan corresponding to each service insurance contract option, for the user to choose from. When the user selects the optimal service insurance contract from the service insurance contract options, the module obtains the optimal execution plan corresponding to the optimal service insurance contract from the service insurance market module and sends the optimal execution plan to the designated edge computing node for execution.

[0044] The Strategy Learning and Optimization Module 150 is an optional module for implementing adaptive optimization of the system. It collects the execution results of the optimal execution plan (e.g., actual profit, whether there is a default) and learns and updates the decision model of the service insurance market module based on the execution results.

[0045] Preferably, the strategy learning and optimization module 150 is a reinforcement learning agent that continuously updates the decision logic (e.g., the weights of one or more neural networks) of the decision model within the service insurance market module, so that the system's service price pricing strategy and underwriting risk (e.g., the probability of service quality default) risk management strategy can be continuously improved over time and with changes in the environment, so as to maximize the long-term cumulative benefits of the system.

[0046] The reinforcement learning agent aims to maximize the long-term cumulative return of the system. The objective function is: st in, It represents the overall decision-making strategy of the reinforcement learning agent and is the optimization variable of the objective function; This represents the mathematical expectation operator, used to handle uncertainties in system operation. These uncertainties mainly stem from unpredictable external events (such as the random arrival of new tasks) and random fluctuations in environmental conditions (such as network latency jitter). Therefore, the optimization objective of this invention is not to optimize the instantaneous gain of a specific possibility, but to find an overall decision-making strategy that is optimal in the statistical average sense. This ensures the robustness and long-term optimality of the decision-making process. t Indicates the sequence number of the time slot. t =0 indicates the initial moment or the start of the first decision cycle; T This represents the total time slots of the optimization process; T -1 indicates the previous time slot in the total time slot; Indicates the discount factor; Indicates the first Tasks for each user; Indicates the first One service insurance contract option; Indicates the first The task of the user and the first The task is obtained by combining service insurance contract options; Indicates time slot t The collection of all best-service insurance contracts successfully signed within the country; Indicates time slot t The following service prices; This represents the optimal execution plan; This represents the total expected underwriting cost of executing the optimal execution plan; Indicates time slot t All edge computing nodes currently in use k The task set; Represents an edge computing node; Indicates the optimal execution plan Under the guidance of edge computing nodes Computational tasks The amount of computing resources consumed; Represents edge computing nodes k Total resource capacity; This represents the set of resources for all edge computing nodes; Indicates the execution of the optimal execution plan. Certainty accuracy; Indicates the minimum accuracy required to make a commitment; Indicates time slot t+ System status information of edge computing nodes at time 1; This represents the system state transition function, which uses time slots. t System status information of edge computing nodes System Actions and random events Input, output time slot t+ System status information of edge computing nodes at time 1 ; Indicates time slot System status information of edge computing nodes; This indicates that the reinforcement learning agent is in the time slot. t Internally, it refers to the system actions taken based on the overall decision-making strategy (e.g., generating service prices for new optimal execution plans). Indicates time slot t Random events that occur within the system that cannot be fully controlled (e.g., fluctuations in network links). Constraint (1) represents resource capacity constraints, constraint (2) represents service level commitment constraints, and constraint (3) is a state transition constraint.

[0047] The reinforcement learning agent employs a multi-task learning evaluation network. This network takes the system state information vector and candidate execution plan vector as input vectors. After shared feature extraction, a bifurcated output layer outputs two features: one is a long-term value signal used to guide policy learning, and the other is used by the service insurance market module to calculate the service quality default probability. The loss function of the evaluation network... Correspondingly, the time difference error loss in value prediction Cross-entropy loss for probability prediction Together constitute, that is, ,in, This represents the hyperparameters that balance the two tasks. This design achieves the inherent unity and collaborative learning between risk assessment and value judgment.

[0048] Building upon the above embodiments, this embodiment further provides an edge computing intelligent resource management method based on a service insurance mechanism. See also... Figure 3 This is a flowchart of an edge computing intelligent resource management method based on a service insurance mechanism provided in an embodiment of the present invention. Figure 3 As shown, the method includes: Step S1, in any time slot t It obtains one or more computing task requests from the user and the current system status information of the edge computing nodes.

[0049] Furthermore, the computing task request is high-dimensional, including: computing task request sequence number, computing model required by the computing task, size of input data, bandwidth requirements, user's local computing power, and user risk preference parameters.

[0050] Step S2: Based on the computing task request and the current system status information of the edge computing node, formulate at least one service insurance contract option.

[0051] Furthermore, the service insurance contract options include a service price (i.e., "premium"), a service level commitment (i.e., "coverage"), a penalty value for service default (i.e., "insurance amount"), and an optional service priority.

[0052] Furthermore, the service level commitment includes a promised minimum accuracy rate. and the maximum delay promised .

[0053] Step S3 involves performing an actuarial assessment of the risks and costs for each service insurance contract option. This specifically includes: Step S31: Generate a candidate execution plan group based on each service insurance contract option and the current system status information of the edge computing node.

[0054] Step S32: Predict the service quality default probability of any candidate execution plan in the candidate execution plan group.

[0055] Step S33: Based on the service quality default probability and the default penalty value of the service insurance contract option, calculate the expected compensation cost corresponding to the candidate execution plan. That is, transform the "probability of loss" into an economic cost, which is the mathematical expectation of future potential compensation.

[0056] Step S34: Combine the expected payout cost with the basic operating costs required to execute the candidate execution plan to obtain the total expected underwriting cost required to execute the candidate execution plan.

[0057] Total expected underwriting cost The calculation formula is: in, Indicates the first Tasks for each user; Indicates the first One service insurance contract option; This represents the total expected underwriting cost; Indicates basic operating costs; This indicates the expected cost of compensation.

[0058] Basic operating costs The calculation formula is: in, This represents the price per unit of computing resources, referring to the price at an edge computing node. k The price required to perform one standard unit of calculation; This indicates that the candidate execution plan is executed on the edge node. The total computational workload required, i.e., the edge computing workload; Indicates the calculation cost; Indicates the unit price of network transmission; Indicates the amount of intermediate data transmitted; This indicates network costs.

[0059] Step S35: Repeat steps S32-S34 to obtain the total expected underwriting cost for all candidate execution plans in the candidate execution plan group.

[0060] Step S4: Based on the actuarial results of each service insurance contract option, determine an optimal execution plan for each service insurance contract option, and dynamically formulate or adjust the service price of the corresponding service insurance contract option based on the total expected underwriting cost of the optimal execution plan to ensure the expected profit margin.

[0061] It is important to note that the service price of the corresponding service insurance contract option, which is dynamically determined or adjusted based on the total expected underwriting cost of the optimal execution plan, is based on a cost-plus pricing logic. Specifically, the system presets a profit margin. α The total expected underwriting cost of executing the optimal execution plan. Based on profit margin and the total expected underwriting cost of executing the optimal execution plan. The product is the profit, and the time slot is obtained. t The following service prices The calculation formula is: Furthermore, the optimal execution plan is the candidate execution plan that minimizes the total expected underwriting cost from the candidate execution plan group. That is, it involves solving the following optimization problem: in, This represents the optimal execution plan; Indicates the first Tasks for each user; Indicates the first One service insurance contract option; Indicates the first in the candidate implementation plan group One candidate implementation plan; Indicates the candidate implementation plan group; Indicates basic operating costs; Indicates the probability of service quality breach; Indicates time slot System status information of edge computing nodes; This represents the penalty value for breach of contract.

[0062] Step S5: Provide the user with the at least one service insurance contract option, along with the service quality default probability of the optimal execution plan corresponding to each service insurance contract option, for the user to choose from.

[0063] In step S6, if the user receives at least one service insurance contract option and selects one as the optimal service insurance contract, proceed to step S7; if the user does not receive at least one service insurance contract option, the task ends.

[0064] Preferably, the user bases their expected utility function. Choose the best service insurance contract from at least one service insurance contract option.

[0065] Furthermore, the expected utility function User risk preference parameter The function.

[0066] Step S7: After receiving the optimal service insurance contract selected by the user, the system sends the optimal execution plan corresponding to the optimal service insurance contract to the edge computing node for execution.

[0067] Step S8: Collect execution results and update the decision model based on the execution results.

[0068] This invention constructs a complete service insurance market, achieving business model innovation. For the first time in the field of edge computing, this invention establishes a complete insurance mechanism encompassing customizable insurance products, dynamic premiums, actuarial science, and intelligent risk control. This invention achieves risk internalization decision-making based on economic rationality, improving decision-making quality. This invention innovatively internalizes expected compensation costs as a real-time, quantifiable economic cost item into the decision-making model for resource allocation and task acceptance. This ensures that all system decisions are based on the economic accounting of risk-adjusted total costs, rather than purely technical indicators, fundamentally guaranteeing the economic rationality and optimality of resource allocation and effectively avoiding unreasonable economic costs incurred to achieve technical targets. This invention improves the accuracy of risk prediction and the system's adaptability through collaborative learning. The multi-task learning evaluation network employed in this invention enables the prediction of risk probability and the judgment of long-term value to share underlying features and mutually promote each other, improving the accuracy of risk assessment. The entire system aims to maximize risk-adjusted long-term returns and can intelligently adapt to changes in market demand and resource status, constructing a robust and efficient edge computing intelligent service method.

[0069] Compared to the greedy and static QoS strategies commonly used in existing technologies, the technical solution of this invention proposes an intelligent decision-making strategy based on a service insurance mechanism. The greedy strategy's decision-making is based solely on the instantaneous physical state of the edge computing nodes, such as always selecting the node with the lowest current computing load for offloading; it is a heuristic algorithm with local optima. The static QoS strategy reserves a portion of high-performance edge computing nodes as a dedicated resource pool for high-priority tasks, allocating resources through this hard rule. The edge computing intelligent resource management system and method based on a service insurance mechanism provided in this application have significant advantages in terms of service provider cumulative revenue, SLA default rate of high-priority tasks under different system loads, and average computing resource utilization. See also Figure 4 This is a diagram illustrating the comparison of cumulative revenue for service providers under different strategies provided in this application's embodiments. See also... Figure 5 This is a schematic diagram comparing the SLA default rates of high-priority tasks under different system loads, provided in an embodiment of this application. See also... Figure 6 This is a schematic diagram comparing the average computing resource utilization under different strategies provided in the embodiments of this application.

[0070] The above description is merely a specific embodiment of the present invention. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this invention should be included within the protection scope of this invention. The protection scope of this invention should be determined by the scope of the claims.

Claims

1. An edge computing intelligent resource management system based on a service insurance mechanism, characterized in that, The system comprises: a request receiving module configured to receive computing task requests from different users; a system state sensing module configured to collect and aggregate current system state information of the edge computing nodes managed by the system in real time or periodically; a service insurance market module configured to obtain service insurance contract options and their corresponding optimal execution plans based on the computing task requests and the current system state information of the edge computing nodes; a task scheduling and execution module configured to provide the service insurance contract options to the users together with the service quality breach probability of the optimal execution plan corresponding to each service insurance contract option for the users to select; when the users select an optimal service insurance contract from the service insurance contract options, the optimal execution plan corresponding to the optimal service insurance contract is obtained from the service insurance market module, and the optimal execution plan is issued to the designated edge computing node for execution. 2.The edge computing intelligent resource management system based on service insurance mechanism of claim 1, wherein, The service insurance contract options comprise a service price, a service level commitment and a breach penalty value. 3.The edge computing intelligent resource management system based on service insurance mechanism of claim 2, wherein, The service insurance market module comprises: an insurance product generator configured to dynamically provide at least one service insurance contract option for each received computing task request; a risk actuarial engine configured to generate a candidate execution plan set corresponding to each service insurance contract option and the current system state information of the edge computing nodes; predict the service quality breach probability of each candidate execution plan in the candidate execution plan set; and calculate an expected loss cost corresponding to each candidate execution plan based on the service quality breach probability and the breach penalty value of the service insurance contract option; an underwriting decision and risk control module configured to generate a total expected underwriting cost based on the basic operating cost and the expected loss cost, and determine an optimal execution plan and a service price for each service insurance contract option by minimizing the total expected underwriting cost. 4.The edge computing intelligent resource management system based on service insurance mechanism of claim 3, wherein, The calculation formula of the expected loss cost is: ， wherein, represents the expected cost of a claim; represents the task of the th user; represents the th service insurance contract option; represents the th candidate execution plan in the candidate execution plan group; represents the probability of service quality breach; represents the system state information of the edge computing node at the time slot ; and represents the breach penalty value. 5.The edge computing intelligent resource management system based on service insurance mechanism of claim 3, wherein, The calculation formula of the total expected underwriting cost is: ， wherein, represents the total expected underwriting cost; represents the task of the user; represents the service insurance contract option of the user; represents the candidate execution plan in the candidate execution plan group; represents the base operating cost; represents the expected claim cost.

6. The edge computing intelligent resource management system based on service insurance mechanism according to claim 1, characterized in that, The system further comprises a strategy learning and optimization module configured to collect the execution results of the optimal execution plans, and learn and update the decision model of the service insurance market module based on the execution results. 7.The edge computing intelligent resource management system based on service insurance mechanism of claim 6, wherein, The strategy learning and optimization module is a reinforcement learning intelligent agent, and the reinforcement learning intelligent agent aims to maximize the long-term cumulative revenue of the system, and the objective function is: ， s.t. ， ， ， in, This represents the overall decision-making strategy of a reinforcement learning agent. Represents the mathematical expectation operator; t Indicates the sequence number of the time slot. t =0 indicates the initial moment or the start of the first decision cycle; T This represents the total time slots of the optimization process; T -1 indicates the previous time slot in the total time slot; Indicates the discount factor; Indicates the first Tasks for each user; Indicates the first One service insurance contract option; Indicates the first The task of the user and the first The task is obtained by combining service insurance contract options; Indicates time slot t The collection of all best-service insurance contracts successfully signed within the country; Indicates time slot t The following service prices; This represents the optimal execution plan; This represents the total expected underwriting cost of executing the optimal execution plan; Indicates time slot t All edge computing nodes currently in use k The task set; Represents an edge computing node; Indicates the optimal execution plan Under the guidance of edge computing nodes Computational tasks The amount of computing resources consumed; Represents edge computing nodes k Total resource capacity; This represents the set of resources for all edge computing nodes; Indicates the execution of the optimal execution plan. Certainty accuracy; Indicates the minimum accuracy required to make a commitment; Indicates time slot t+ System status information of edge computing nodes at time 1; Represents the system state transition function. Indicates time slot System status information of edge computing nodes at that time. This indicates that the reinforcement learning agent is in the time slot. t Internally, it refers to the systematic actions taken based on the overall decision-making strategy. Indicates time slot t Random events that occur within the system and cannot be fully controlled.

8. An edge computing intelligent resource management method based on a service insurance mechanism, characterized in that, The method is based on the edge computing intelligent resource management system of any one of claims 1-7, and comprises: Step S1, at any time slot t obtaining one or more computing task requests from a user and current system status information of the edge computing node; Step S2, formulating at least one service insurance contract option based on the computing task requests and the current system state information of the edge computing nodes; Step S3, performing risk and cost actuarial assessment on each service insurance contract option; Step S4, determining an optimal execution plan for each service insurance contract option based on the actuarial results of each service insurance contract option, and dynamically formulating or adjusting the service price of the corresponding service insurance contract option based on the total expected underwriting cost of the optimal execution plan; Step S5, providing the at least one service insurance contract option to the users together with the service quality breach probability of the optimal execution plan corresponding to each service insurance contract option for the users to select; Step S6, if the user receives at least one service insurance contract option and selects one service insurance contract option as the optimal service insurance contract from the at least one service insurance contract option, proceed to S7; if the user does not receive at least one service insurance contract option, the task ends; Step S7, after the system receives the optimal service insurance contract selected by the user, the optimal service insurance contract corresponding to the optimal execution plan is issued to the edge computing node for execution; Step S8, collect the execution result, and update the decision model according to the execution result. 9.The edge computing intelligent resource management method based on service insurance mechanism according to claim 8, characterized in that, Step S3 includes: Step S31, generate a candidate execution plan group according to each service insurance contract option and the current system state information of the edge computing node; Step S32, predict the service quality default probability of any candidate execution plan in the candidate execution plan group; Step S33, based on the service quality default probability and the default penalty value of the service insurance contract option, calculate the expected compensation cost corresponding to the candidate execution plan; Step S34, combine the expected compensation cost with the basic operation cost required for executing the candidate execution plan to obtain the total expected insurance cost required for executing the candidate execution plan; Step S35, repeat steps S32-S34 to obtain the total expected insurance cost corresponding to all candidate execution plans in the candidate execution plan group. 10.The edge computing intelligent resource management method based on service insurance mechanism according to claim 9, characterized in that, The optimal execution plan is the candidate execution plan in the candidate execution plan group that minimizes the total expected insurance cost, and the target formula is as follows: ， wherein, represents an optimal execution plan; represents a task of a th user; represents a service insurance contract option of a th user; represents a th candidate execution plan in a candidate execution plan group; represents a candidate execution plan group; represents a base operating cost; represents a service quality breach probability; represents system state information of an edge computing node at a time slot ; and represents a breach penalty value.

Images