Task age aware on-demand scheduling optimization method in digital twin-driven vehicle-mounted computing power network

Abstract

The invention provides an on-demand scheduling optimization method for task age awareness in a digital twinborn driven vehicle-connected computing power network. The method comprises the following steps: constructing a digital twinborn model of a physical network; determining the task age of the current task according to the generation time difference between the current time and the calculation task which returns the calculation result recently, the maximum task age demand, the task age at the previous moment and whether the directly associated calculation power pool accepts the request of the calculation task or not; constructing a scheduling optimization model with an optimization target of minimizing task ages of all tasks and taking a service decision, a calculation association decision, a power allocation variable and a calculation resource allocation variable as optimization variables; and designing a self-adaptive scheduling optimization algorithm, solving the scheduling optimization model, and obtaining a task scheduling scheme according to a solving result. By accurately sensing the age of the task, diversified vehicle-mounted application requirements can be met, and the flexibility and accuracy of task scheduling are improved.

Description

Technical Field

[0001] The present invention relates to the field of vehicle scheduling technology, and in particular to an on-demand scheduling optimization method based on task age awareness in a vehicle-to-vehicle computing network driven by digital twins. Background Art

[0002] Existing research on task timeliness has mostly focused on optimization in static environments, failing to consider variations in information urgency. This makes it unsuitable for vehicle-connected computing power networks (VCPNs) and lacks novel joint metrics for task timeliness and urgency. Due to the limited resources in VCPNs, the need to ensure the timeliness and urgency of task transmission, and the time-varying channels caused by mobility, a key challenge is to simultaneously meet the dynamic requirements of heterogeneous information freshness and task urgency across different vehicles in dynamic, time-varying environments with limited resources. Furthermore, in task-oriented VCPNs, most studies fail to consider the maximum permissible age of information. Establishing and maintaining strict AoI (Age of Information) constraints is difficult under dynamic, heterogeneous network conditions and resource constraints. Summary of the Invention

[0003] The purpose of the present invention is to overcome the shortcomings of the existing technology and provide an on-demand scheduling optimization method for task age awareness in a vehicle-connected computing network driven by digital twins.

[0004] In order to achieve the above-mentioned purpose, the present invention adopts the following technical solutions.

[0005] In a first aspect, the present invention provides a method for on-demand scheduling optimization based on task age awareness in a vehicle-connected computing network driven by a digital twin, comprising:

[0006] Build digital twin models of physical networks;

[0007] The task age of the current task is determined based on the time difference between the current time and the most recently returned computing task, the maximum task age requirement, the task age at the previous moment, and whether the directly associated computing power pool accepts the computing task request;

[0008] Construct a scheduling optimization model with the optimization goal of minimizing the task age of all tasks and service decision, computing association decision, power allocation variable, and computing resource allocation variable as optimization variables;

[0009] An adaptive scheduling optimization algorithm is designed to solve the scheduling optimization model, and a task scheduling solution is obtained according to the solution result.

[0010] Furthermore, the task age of the current task is determined based on the time difference between the current time and the computing task that most recently returned the computing result, the maximum task age requirement, the task age at the previous moment, and whether the directly associated computing power pool accepts the computing task request, including:

[0011] If the directly associated computing power pool accepts the request for the computing task, the task age of the current task is the time difference between the time when the task calculation result is returned and the time when it is generated.

[0012] If the directly associated computing power pool does not accept the request for the computing task, the task age of the current task is the maximum task age requirement. and The minimum value among them, is the task age of the previous computation task, and a and b are adjustable parameters.

[0013] Furthermore, the scheduling optimization model is:

[0014]

[0015] Where X is the service decision, Φ is the computation association decision, P is the power allocation variable, and F is the computation resource allocation variable. Is the mission age, is the association decision between the task requester u and the directly associated computing power pool n and the indirectly associated computing power pool n', x u,n It is the task service decision between the task requester u and the directly associated computing power pool n, is a collection of computing capacity providers, is the set of task requesters, is the computing power allocated by the computing power provider to the task requester u at time t, F n' is the maximum available computing power of the computing power provider at time t, is the uplink power of task requester u at time t, is the maximum transmission power of the task requester u, is the energy calculation of the task initiator u, is the maximum energy consumption requirement, is the task execution time, is the maximum delay requirement.

[0016] Furthermore, after constructing a scheduling optimization model whose optimization goal is to minimize the task age of all tasks and whose optimization variables are service decision, computing association decision, power allocation variable, and computing resource allocation variable, the method further includes:

[0017] Given the power allocation variable P and the computing resource allocation variable F, the scheduling optimization model is optimized into the first sub-model:

[0018]

[0019] The problem corresponding to the power allocation variable P and the computing resource allocation variable F is transformed into a joint power and computing power allocation sub-problem to obtain the second sub-model:

[0020]

[0021] Where, is the current task execution time, is the uplink delay of the current task, is the transmission delay of the current task from computing pool n to computing pool n', is the computation time of the task in DT, is the set of scheduled users;

[0022] Accordingly, solving the scheduling optimization model and obtaining a task scheduling solution according to the solution result includes:

[0023] The first sub-model and the second sub-model are solved respectively, and the first solution result and the second solution result are obtained accordingly. The first solution result includes the calculation association decision coefficient Φ and the service decision X, and the second solution result includes the power allocation variable P and the computing resource allocation variable F.

[0024] Furthermore, solving the first sub-model includes:

[0025] A DRL-based algorithm is designed to solve the first sub-model to obtain a first solution result.

[0026] Furthermore, solving the scheduling optimization model and obtaining a task scheduling solution according to the solution result includes:

[0027] The scheduling optimization model is solved using a multi-head DDQN network, and a task scheduling solution is obtained based on the solution results.

[0028] Furthermore, before optimizing the uplink transmission time in the second sub-model to obtain the second sub-model, the method further includes:

[0029] According to the transmission power of the current task requester Noise power σ 2 , channel gain and the transfer task size Associated variables Calculate the uplink delay of TR-CPP

[0030] According to the transfer rate between the directly associated computing power pool CPP n and the indirectly associated computing power pool CPP n' And the associated variables with the indirect computing power pool CPP n' Calculate the transmission time of CPP n and CPP n'

[0031] In a second aspect, the present application also provides a task age-aware on-demand scheduling optimization device in a vehicle-connected computing network driven by a digital twin, comprising:

[0032] Digital twin mapping module, used to build digital twin models of physical networks;

[0033] The task age calculation module is used to determine the task age of the current task based on the time difference between the current time and the most recently returned calculation result, the maximum task age requirement, the task age at the previous moment, and whether the directly associated computing power pool accepts the calculation task request;

[0034] A model building module is used to build a scheduling optimization model with the optimization goal of minimizing the task age of all tasks and with service decision, computing association decision, power allocation variable, and computing resource allocation variable as optimization variables;

[0035] The solution module is used to design an adaptive scheduling optimization algorithm, solve the scheduling optimization model, and obtain a task scheduling solution based on the solution result.

[0036] In a third aspect, the present invention further provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned method when executing the program.

[0037] In a fourth aspect, the present invention further provides a computer program product, comprising a computer program, which implements the above method when executed by a processor.

[0038] Beneficial effects of the present invention: The on-demand scheduling optimization method based on task age awareness in a vehicle-connected computing power network driven by a digital twin provided by the present invention determines a task scheduling scheme by solving a scheduling model, wherein the optimization objective of the scheduling model is to minimize the task age of all tasks, and the constraint variables are determined by the service decision X, the computing association decision Φ, the power allocation variable P, and the computing resource allocation variable F. The task age can reflect the time sensitivity and urgency of the computing task, so that the final scheduling scheme can achieve the goal of on-demand resource scheduling and task allocation, adaptively adjust the service association and resource allocation decisions, effectively match the dynamic unevenness of resource distribution with the differentiated requirements of tasks, reduce task age, and improve the real-time performance of the overall task. The on-demand scheduling optimization method based on task age awareness in a vehicle-connected computing power network driven by a digital twin provided by the present invention can adapt to diverse vehicle application requirements and improve the flexibility and accuracy of task scheduling through accurate perception of task age.

[0039] Additional aspects and advantages of the present invention will be set forth in part in the following description, will be obvious from the following description, or may be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0041] Figure 1 A flowchart of an on-demand scheduling optimization method based on task age awareness in a vehicle-connected computing network driven by a digital twin according to an embodiment of the present invention;

[0042] Figure 2 A schematic diagram of the structure of the vehicle-mounted computing power network provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0043] The embodiments of the present invention are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention and are not to be construed as limiting the present invention.

[0044] It will be understood by those skilled in the art that, unless expressly stated otherwise, the singular forms "a", "an" and "the" used herein may also include the plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be intermediate elements. In addition, "connected" or "coupled" as used herein may include wireless connections or couplings. The term "and / or" used herein includes any unit and all combinations of one or more associated listed items.

[0045] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those skilled in the art in the art to which the present invention pertains. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with their meanings in the context of the prior art and, unless defined as such herein, will not be interpreted in an idealized or overly formal sense.

[0046] To address the above issues, embodiments of the present invention provide a method for on-demand scheduling optimization based on task age awareness in a vehicle-connected computing network driven by digital twins. Before describing the method in detail, the following definitions are provided for the terms used in the present invention:

[0047] (1) Computing power network

[0048] The development of the Internet of Vehicles (IoV) and the explosive growth of artificial intelligence (AI) applications have spurred a variety of compute-intensive and latency-sensitive in-vehicle applications. This surge has also led to increased demand for computing resources and a growing need for data processing and analysis at the network edge, posing unprecedented challenges to current computing architectures. Computing networks, as a new architecture, integrate with the IoV to form a connected computing network that delivers real-time, immersive experiences. The introduction of VCPNs (connected computing power networks) heralds a new era in intelligent transportation. By integrating heterogeneous resources from vehicles, edge servers, and cloud infrastructure, they manage and schedule computing tasks in a coordinated manner, enabling real-time applications such as autonomous driving and traffic management. Other studies have discussed the research background, key technologies, and key application scenarios of computing networks from the perspective of computing and network convergence. Furthermore, some research has combined computing networks with distributed machine learning and proposed a resource efficiency metric that comprehensively considers training performance and system cost. By establishing a network within the AI computing power pool, "Net-in-AI" has been proposed for ubiquitous AI.

[0049] In addition to the aforementioned application features, CPN also offers significant advantages from an implementation and deployment perspective. The computing network provides efficient real-time data processing and decision-making capabilities for the Internet of Vehicles (IoV). By integrating distributed computing resources across vehicles, edge nodes, and infrastructure, it improves computing and communication efficiency.

[0050] (3) Digital Twin (DT) Technology

[0051] With the rapid development of technologies such as autonomous driving, intelligent transportation, and in-vehicle computing, the Internet of Vehicles (IoV) faces the challenges of high-frequency real-time data processing and complex task scheduling. Against this backdrop, digital twin technology enables comprehensive monitoring and optimized management of the entire IoV environment by creating real-time virtual models of physical objects (such as vehicles, roadside units, and network nodes). Digital twins not only reflect the vehicle's current location, speed, status, and sensor data, but also integrate and analyze network conditions, computing resources, and traffic flow information in real time, supporting dynamic decision-making and efficient scheduling.

[0052] Furthermore, the introduction of digital twin technology enables system simulation testing in a virtual environment, evaluating task execution and resource allocation strategies in various scenarios and proactively identifying potential risks and bottlenecks. This capability is crucial for ensuring the reliability and safety of autonomous driving and intelligent traffic management systems. Therefore, digital twins not only enhance the intelligence of connected vehicles but also provide strong support for optimizing resource utilization, improving system performance, and enhancing safety, becoming a key technology for realizing future intelligent transportation.

[0053] (4) Age of Information (AoI) optimization

[0054] The emergence of 6G and edge intelligence has significantly increased the demand for timely and reliable information updates and relevant data availability to support applications such as autonomous driving. In such applications, decisions based on outdated information can lead to conflicting actions, resulting in inefficiencies and compromising vehicle safety and performance. Tasks in VCPNs often rely on fresh information to ensure accurate and timely decision-making, making information age crucial. Some techniques formulate the problem as minimizing the AoI and energy consumption in drone-assisted massive IoT scenarios, where AoI is defined as the time from the first data collection to the return to the warehouse. Other techniques analyze mixed traffic flows involving two users, considering both deadline constraints and AoI considerations, aiming to identify effective scheduling strategies that minimize the average AoI. Other studies assume that each message is observable by multiple SNs and optimize the state AoI from the perspective of multi-source information perception. Other studies consider the interplay between transmission and offloading strategies in edge networks and design joint scheduling and offloading strategies based on AoI to minimize AoI. Despite these efforts, increasing communication latency and the ever-growing amount of data required for AI tasks pose challenges for computing nodes to collect and analyze operational data for online optimization. As a result, achieving dynamic and fast responses in complex, heterogeneous vehicle-connected computing networks remains difficult.

[0055] (5) Multi-task scheduling and resource optimization

[0056] In the connected vehicle (IoV) environment, multi-task scheduling and resource optimization are key factors in ensuring efficient system operation. With the widespread adoption of intelligent transportation, autonomous driving, and in-vehicle applications, the vast amounts of data generated by vehicles and infrastructure need to be processed in real time in dynamic and complex environments. This creates a high demand for computing resources and also requires systems to flexibly schedule multiple tasks. The introduction of mobile edge computing (MEC) offers new possibilities for multi-task scheduling in IoV. Some research aims to minimize total weighted energy consumption by jointly optimizing multi-task partitioning, computing, and communication resource allocation. Other research addresses computational offloading by considering task dependencies to improve information freshness. Other research explores effective data compression schemes to reduce wireless device energy consumption and improve computational performance. A fine-grained task offloading method for green computing networks based on extensive attention has been proposed. By focusing on key information, it reduces task latency and energy consumption. Another study investigates an intelligent collaborative task offloading and service caching mechanism for multi-cell MEC networks, aiming to maximize the system utility based on quality of service. Furthermore, a multi-agent reinforcement learning algorithm based on graph attention is proposed to learn optimal task offloading and service caching strategies.

[0057] To facilitate understanding of the embodiments of the present invention, several specific embodiments will be further explained below with reference to the accompanying drawings, and each embodiment does not constitute a limitation on the embodiments of the present invention.

[0058] Example 1

[0059] like Figure 1 As shown in FIG, a method for on-demand scheduling optimization based on task age awareness in a vehicle-connected computing network driven by a digital twin includes the following steps:

[0060] S101, build a digital twin model of the vehicle-connected computing network.

[0061] S102, based on the above digital twin model architecture, generates a time difference between the current time and the calculation task that most recently returned the calculation result Maximum mission age requirement And the task age at the last moment And whether the directly related computing power pool CPP (Computing Power Pool) accepts the request for computing tasks to determine the task age of the current task

[0062] Specifically, if the directly associated computing power pool accepts the request for computing tasks, the task age of the current task is the time difference between the time when the task calculation result is returned and the time when it is generated.

[0063] If the directly associated computing power pool does not accept the request for the computing task, the task age of the current task is the maximum task age requirement. and The minimum value among them, is the task age of the previous computation task, a and b are adjustable parameters, where It increases exponentially and is calculated based on the urgency of the task.

[0064] That is to say, Among them, a and b are adjustable parameters, and the calculation task When , it means that the service CPP n directly associated with the computing task accepts the computing request of the task; if The directly associated service CPP cannot accept the task computing request at time t.

[0065] S103 , constructing a scheduling optimization model with the optimization goal of minimizing the average task age of all tasks and with the service decision X, the computing association decision φ, the power allocation variable P, and the computing resource allocation variable F as optimization variables.

[0066] Specifically, the scheduling optimization model is:

[0067]

[0068] Where X is the service decision, Φ is the computation association decision, P is the power allocation variable, and F is the computation resource allocation variable. Is the mission age, is the association decision between the task requester u and the directly associated computing power pool n and the indirectly associated computing power pool n', x u,n It is the task service decision between the task requester u and the directly associated computing power pool n, is a collection of computing capacity providers, is the set of task requesters, is the computing power allocated by the computing power provider to the task requester u at time t, F n' is the maximum available computing power of the computing power provider at time t, is the uplink power of task requester u at time t, is the maximum transmission power of the task requester u, is the energy calculation of the task initiator u, is the maximum energy consumption requirement, is the task execution time, is the maximum delay requirement.

[0069] S104, designing an adaptive scheduling optimization algorithm, solving the scheduling optimization model, and obtaining a task scheduling solution based on the solution result.

[0070] In some embodiments of the present invention, task-oriented optimization problems are difficult to solve directly due to the non-convexity of the objective problem and constraints, as well as the mobility of vehicles, the time-varying nature of the wireless environment, and the dynamic nature of computing power. The optimal solution to problem P can be observed by solving the following two subproblems: given P and F, the original problem P is transformed into a joint service decision and TR-CPP association decision problem. Once the service decision and TR-CPP association decision are fixed, the original problem P is transformed into a joint power optimization and computational power allocation subproblem.

[0071] Specifically, after constructing a scheduling optimization model whose optimization objective is to minimize the average task age of all tasks and whose optimization variables are service decision X, computing association decision Φ, power allocation variable P, and computing resource allocation variable F, the method further includes:

[0072] Given the power allocation variable P and the computing resource allocation variable F, the scheduling optimization model is optimized into the first sub-model:

[0073]

[0074] Among them, once the service decision and TR-CPP association are fixed, the power problem corresponding to the power allocation variable P can be transformed into a joint power and computing power allocation sub-problem to obtain the second sub-model:

[0075]

[0076] Where, is the current task execution time, is the uplink delay of the current task, is the transmission delay of the current task from computing pool n to computing pool n', is the computation time of the task in DT, The set of scheduled users.

[0077] Accordingly, solving the scheduling optimization model and obtaining a task scheduling solution according to the solution result includes:

[0078] The first sub-model and the second sub-model are solved respectively to obtain a first solution result and a second solution result respectively. The first solution result includes the calculation association decision φ and the service decision X, and the second solution result includes the power allocation variable P and the calculation resource allocation variable F.

[0079] In some embodiments of the present invention, solving the first sub-model includes:

[0080] A DRL-based algorithm is designed to solve the first sub-model to obtain a first solution result.

[0081] Specifically, the complexity of problem P1 is handled by the DRL algorithm, which has excellent adaptability to dynamic environments. In order to minimize the AoT of the system, a Markov decision process (MDP) associated with the joint service decision and TR-CPP is first constructed. Generally, the MDP consists of four tuples: in The system state, the actions that the agent can take in each state, the reward, and the next state. The digital twins of the TRs nodes and CPP computing power nodes in the mapping layer represent the relevant states. DT not only monitors the operating status of the entire network in real time but also records the historical operating status of the computing power network. At each time period, the agent obtains state from the environment and takes actions related to service decisions and TR-CPP association decisions. It then receives rewards based on its actions and transitions from the current state to the next state.

[0082] In order to minimize the AoT of the entire network, we first construct the Markov decision process of problem P1 as follows:

[0083] State: The state observed by the agent consists of several parts: attribute information of the task generated by the task requester, the age of the task at the last moment, the location of the task initiator and the computing power pool, and the transmission rate between computing power pools.

[0084] Action: Service decision X, calculate associated decision φ.

[0085] Reward: The TR algorithm, with the assistance of the DT algorithm, adaptively migrates tasks to the most appropriate CPP on demand to reduce AoT. In the proposed network environment, if the TR performs an action that reduces AoT, it will immediately receive a positive reward. Therefore, in order to efficiently approximate the optimal policy for multi-task scheduling and service association that meets multiple constraints after each training set, while minimizing AoT, the problem of how to minimize AoT is transformed into the problem of how to maximize the agent's reward. Therefore, the reward function is designed as

[0086] To handle multidimensional discrete action spaces, a Multi-head Double Deep Q-Network (MDDQN) with multiple output layers and Prioritized Experience Replay (PER) was designed to learn optimal policies and improve training efficiency. This architecture learns the optimal policy by sharing state features extracted from the first few layers of the neural network, reducing the number of parameters and achieving low computational complexity. The designed DRL network primarily consists of a main network, a target network, PER, and a loss function. The target network shares the same architecture as the main network and is responsible for generating target values for training the main network, enhancing the stability of the learning process.

[0087] Because the state is large, we construct an approximate representation of the action-value function, assuming that the Q-function corresponding to the state and action is recorded as Q(s,a). The value of Q(s,a) can be approximated by the neural network parameters in MDDQN, expressed as Q(s,a)≈Q(s(t),a(t);θ), where θ is the parameter of the main neural network.

[0088] For the target network, the input of the target network maintains the next state, but the action is not selected based on the maximum Q value. Instead, the corresponding action is obtained through the main network Therefore, the target network value in MDDQN is expressed as:

[0089]

[0090] Where γ represents the cumulative discount factor. Subsequently, the state-action pairs are stored in PER with random priority and importance sampling. The network parameters θ are then updated using gradient descent until convergence.

[0091] In addition, the optimized second sub-model can be solved by the relevant KKT (Karush-Kuhn-Tucker) conditions.

[0092] In some embodiments of the present invention, before optimizing the uplink transmission time in the second sub-model to obtain the second sub-model, the method further includes:

[0093] According to the transmission power of the current task requester Noise power σ 2 , channel gain and the transfer task size Associated variables Calculate the uplink delay of TR-CPP

[0094] Furthermore, according to the transmission rate between the directly associated computing power pool CPP n and the indirectly associated computing power pool CPP n' And the associated variables with the indirect computing power pool CPP n' Calculate the transmission time of CPP n and CPP n'.

[0095] More specifically, when TR generates a computing task to CPPs, the total time required for the computing task consists of three parts: the communication time required for task migration, the computing time required to calculate the power pool, and the communication time required to return the results. The calculation of each time is described below.

[0096] In addition to executing computational tasks on the directly associated CPP n, tasks can also be transferred to other CPP n' for efficient execution via CPP n. Once a task is generated by TR, the execution request and computational task will be transferred to the directly associated CPP n.

[0097] 1) Migrate to TR n

[0098] The available uplink rate of TR u in time slot t is:

[0099]

[0100] in is the transmission power, σ 2 is the noise power. represents the channel gain, d u,n is the distance from TR u to CPPu, and α is the path loss exponent. Then the uplink delay for TR u to send the task is:

[0101]

[0102] The energy generated by TR u is calculated as:

[0103]

[0104] 2) TR u migrate to other CPP n'

[0105] Since the computing resources provided by the directly associated CPP n are insufficient, the computing tasks will be migrated to other CPP n′. Execution. For any TR u, the goal of computing association is to select the target CPP to perform the computing task. Define association <TR u ,CP Pn '> means If the computation task of TR u is computed by CPP n'. Otherwise, Note that if but R n,n' Expressed as the transmission rate between CPP n and CPP n', the transmission delay from CPP n to CPP n' is

[0106] By deploying an intelligent scheduling algorithm in the DT network, distributed tasks are assigned to the most appropriate computing nodes to reduce AoI. The task execution delay of TR u on CPP n is expressed as

[0107] In some embodiments of the present invention, a digital twin model of a physical network is constructed as follows:

[0108] Considering the DT-driven vehicle computing network scenario, a base station equipped with a server can be represented as a computing power pool (CPP), which can provide computing and communication resource services according to user needs. and They are represented as a set of computing capacity providers (CPPs) and task requesters (TRs), where task requesters Directly associated with computing power provider CPP n. Figure 2The proposed three-layer architecture is described, comprising the user layer, the computing resource pool layer, and the digital twin layer. The user layer consists of vehicles acting as computing power task requesters (TRs), initiating pending computing tasks and aiming to complete them using the computing resources associated with the CPP. The computing power pool layer is composed of distributed computing power nodes of various types. This layer primarily utilizes deployed optimization algorithms to ensure timely task response and meet the diverse needs of TRs. In the digital twin layer, different types of DTs are considered, namely twin computing power nodes and twin TRs. These twin networks are maintained on edge servers to reflect the key states of the entire DVCPN in digital space and execute learning algorithms to support intelligent resource collaborative orchestration. Taking the TR-to-CPP communication as an example, the DT space not only reflects the time-varying computing power information of TRs and CPPs but also establishes a dynamic, multi-dimensional, and multi-timescale virtual model of the communication environment between them. Based on this, complex decision-making in the DT space can reduce interaction latency between physical entities.

[0109] The present invention provides an on-demand scheduling optimization method for task age in a digital twin-driven vehicle-connected computing network. This method, through a digital twin-driven vehicle-connected computing network architecture, achieves real-time awareness of the vehicle-connected computing network system status to optimize task allocation and scheduling, ensuring that compute-intensive and latency-sensitive tasks are processed in the shortest possible time. The method also captures the freshness and urgency of tasks through the Age of Task (AoT), thereby supporting efficient scheduling decisions. Furthermore, to address the conflicting challenges of highly heterogeneous resource distribution, dynamically changing network conditions, and diverse task requirements, the present invention proposes an adaptive task age-aware scheduling algorithm to minimize the sum of the AoTs of multiple tasks. This algorithm adaptively adjusts service association and resource allocation decisions based on changes in task requirements and the system environment, effectively matching the dynamic heterogeneity of resource distribution with the differentiated requirements of tasks. Specifically, an improved reinforcement learning algorithm is designed to determine the optimal service association strategy that maximizes benefits, and an optimization algorithm based on convex optimization is used to derive the optimal resource allocation. This allows for flexible response to dynamic task requirements and resource conditions, improving the flexibility and stability of the system. This invention aims to solve the core problems of task scheduling and resource optimization in the Internet of Vehicles through the realization of the above goals, improve the overall performance and reliability of the system, and provide a solid technical foundation for the development of intelligent transportation and autonomous driving applications.

[0110] Example 2

[0111] Based on Example 1, this Example 2 provides a device for on-demand scheduling optimization based on task age awareness in a vehicle-connected computing network driven by a digital twin. The device for on-demand scheduling optimization based on task age awareness in a vehicle-connected computing network driven by a digital twin corresponds to the above-mentioned method for on-demand scheduling optimization based on task age awareness in a vehicle-connected computing network driven by a digital twin, and specifically includes:

[0112] Digital twin mapping module, used to build digital twin models of physical networks.

[0113] The task age calculation module is used to determine the task age of the current task based on the time difference between the current time and the computing task that most recently returned the calculation result, the maximum task age requirement, the task age at the previous moment, and whether the directly associated computing power pool accepts the computing task request.

[0114] The model building module is used to build a scheduling optimization model with the optimization goal of minimizing the task age of all tasks and with service decision, computing association decision, power allocation variable, and computing resource allocation variable as optimization variables.

[0115] The solution module is used to design an adaptive scheduling optimization algorithm, solve the scheduling optimization model, and obtain a task scheduling solution based on the solution result.

[0116] For specific details, please refer to the description of the on-demand scheduling optimization method for task age awareness in the vehicle-connected computing network driven by digital twins, which will not be repeated here.

[0117] Example 3

[0118] Embodiment 3 of the present invention provides an electronic device including a memory and a processor, wherein the processor and the memory communicate with each other, the memory stores program instructions executable by the processor, and the processor invokes the program instructions to execute a method for on-demand scheduling optimization based on task age awareness in a vehicle-connected computing network driven by a digital twin. The method includes the following process steps:

[0119] Build digital twin models of physical networks;

[0120] The task age of the current task is determined based on the time difference between the current time and the most recently returned computing task, the maximum task age requirement, the task age at the previous moment, and whether the directly associated computing power pool accepts the computing task request;

[0121] Construct a scheduling optimization model with the optimization goal of minimizing the task age of all tasks and service decision, computing association decision, power allocation variable, and computing resource allocation variable as optimization variables;

[0122] An adaptive scheduling optimization algorithm is designed to solve the scheduling optimization model, and a task scheduling solution is obtained according to the solution result.

[0123] Example 4

[0124] Embodiment 4 of the present invention provides a computer program product, comprising a computer program. When executed by a processor, the computer program implements a method for on-demand scheduling optimization based on task age awareness in a vehicle-connected computing network driven by a digital twin. The method comprises the following steps:

[0125] Build digital twin models of physical networks;

[0126] The task age of the current task is determined based on the time difference between the current time and the most recently returned computing task, the maximum task age requirement, the task age at the previous moment, and whether the directly associated computing power pool accepts the computing task request;

[0127] Construct a scheduling optimization model with the optimization goal of minimizing the task age of all tasks and service decision, computing association decision, power allocation variable, and computing resource allocation variable as optimization variables;

[0128] An adaptive scheduling optimization algorithm is designed to solve the scheduling optimization model, and a task scheduling solution is obtained according to the solution result.

[0129] Those skilled in the art will appreciate that the accompanying drawings are merely schematic diagrams of an embodiment, and the modules or processes in the accompanying drawings are not necessarily required to implement the present invention.

[0130] Each embodiment in this specification is described in a progressive manner. The same or similar parts between the embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments. In particular, for method or system embodiments, since they are basically similar to method embodiments, the description is relatively simple. For relevant parts, refer to the partial description of the method embodiment. The method and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Ordinary technicians in this field can understand and implement it without expending creative work.

[0131] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A task age-aware on-demand scheduling optimization method in a digital twin-driven vehicle-connected computing network, characterized by: include: Build digital twin models of physical networks; The task age of the current task is determined based on the time difference between the current time and the most recently returned computing task, the maximum task age requirement, the task age at the previous moment, and whether the directly associated computing power pool accepts the computing task request; Construct a scheduling optimization model with the optimization goal of minimizing the task age of all tasks and service decision, computing association decision, power allocation variable, and computing resource allocation variable as optimization variables; An adaptive scheduling optimization algorithm is designed to solve the scheduling optimization model, and a task scheduling solution is obtained according to the solution result.

2. The method according to claim 1, characterized in that The task age of the current task is determined based on the time difference between the current time and the most recently returned computing task, the maximum task age requirement, the task age at the previous moment, and whether the directly associated computing power pool accepts the computing task request, including: If the directly associated computing power pool accepts the request for the computing task, the task age of the current task is the time difference between the time when the task calculation result is returned and the time when it is generated. If the directly associated computing power pool does not accept the request for the computing task, the task age of the current task is the maximum task age requirement. and The minimum value among them, is the task age of the previous computation task, and a and b are adjustable parameters.

3. The method according to claim 1, characterized in that The scheduling optimization model is: Where X is the service decision, Φ is the computation association decision, P is the power allocation variable, and F is the computation resource allocation variable. Is the mission age, is the association decision between the task requester u and the directly associated computing power pool n and the indirectly associated computing power pool n', x u,n It is the task service decision between the task requester u and the directly associated computing power pool u, is a collection of computing capacity providers, is the set of task requesters, is the computing power allocated by the computing power provider to the task requester u at time t, F n' is the maximum available computing power of the computing power provider at time t, is the uplink power of task requester u at time t, is the maximum transmission power of the task requester u, is the energy calculation of the task initiator u, is the maximum energy consumption requirement, is the task execution time, is the maximum delay requirement.

4. The method according to claim 3, characterized in that After constructing a scheduling optimization model whose optimization goal is to minimize the task age of all tasks and whose optimization variables are service decision, computing association decision, power allocation variable, and computing resource allocation variable, the method further includes: Given the power allocation variable P and the computing resource allocation variable F, the scheduling optimization model is optimized into the first sub-model: The problem corresponding to the power allocation variable P and the computing resource allocation variable F is transformed into a joint power and computing power allocation sub-problem to obtain the second sub-model: Where, is the current task execution time, is the uplink delay of the current task, is the transmission delay of the current task from computing pool n to computing pool n', is the computation time of the task in DT, is the set of scheduled users; Accordingly, solving the scheduling optimization model and obtaining a task scheduling solution according to the solution result includes: The first sub-model and the second sub-model are solved respectively, and the first solution result and the second solution result are obtained accordingly. The first solution result includes the calculation association decision coefficient Φ and the service decision X, and the second solution result includes the power allocation variable P and the computing resource allocation variable F.

5. The method according to claim 4, characterized in that Solving the first sub-model includes: A DRL-based algorithm is designed to solve the first sub-model to obtain a first solution result.

6. The method according to claim 1, characterized in that Solving the scheduling optimization model and obtaining a task scheduling solution according to the solution result includes: The scheduling optimization model is solved using a multi-head DDQN network, and a task scheduling solution is obtained based on the solution results.

7. The method according to claim 4, characterized in that Before optimizing the uplink transmission time in the second sub-model to obtain the second sub-model, the method further includes: According to the transmission power of the current task requester Noise power σ 2 , channel gain and the transfer task size Associated variables Calculate the uplink delay of TR-CPP According to the transmission rate between the directly associated computing power pool CPPn and the indirectly associated computing power pool CPPn' And the associated variables with the indirect computing power pool CPPn' Calculate the transmission time of CPPn and CPPn' 8. A task age-aware on-demand scheduling optimization device in a digital twin-driven vehicle-connected computing network, characterized in that: include: Digital twin mapping module, used to build digital twin models of physical networks; The task age calculation module is used to determine the task age of the current task based on the time difference between the current time and the most recently returned calculation result, the maximum task age requirement, the task age at the previous moment, and whether the directly associated computing power pool accepts the calculation task request; A model building module is used to build a scheduling optimization model with the optimization goal of minimizing the task age of all tasks and with service decision, computing association decision, power allocation variable, and computing resource allocation variable as optimization variables; The solution module is used to design an adaptive scheduling optimization algorithm, solve the scheduling optimization model, and obtain a task scheduling solution based on the solution result.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the method according to any one of claims 1 to 7 is implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images