A mobile edge computing task scheduling method based on a quantum heuristic algorithm

Abstract

The application discloses a kind of based on quantum heuristic algorithm's mobile edge computing task scheduling method, belong to network communication technical field. Including the following steps: construction scene;System architecture;System parameter update;Problem construction;Solution method.The application is used in the task scheduling optimization in mobile edge computing (MEC) environment, by integrating the concept of quantum mechanics into double deep Q network (DDQN), successfully improves the adaptability and robustness of task scheduling.QI-DRL algorithm is demonstrated in theory The advancement of, also by extensive experimental verification with real-world dataset Its actual application effectiveness.The application also provides a new perspective and technical path for handling resource management and task scheduling problems in MEC environment, so that in the case of high demand change and resource limitation, the system can more effectively manage and schedule tasks, significantly optimize the operation efficiency and response speed in MEC environment.

Description

Technical Field

[0001] This invention relates to the field of network communication technology, and in particular to a mobile edge computing task scheduling method based on a quantum heuristic algorithm. Background Technology

[0002] In modern computing environments, with the surge in data volume and the diversification of application demands, traditional cloud computing centers can no longer fully meet the needs of latency-sensitive or geographically widespread application scenarios. Edge computing, as an emerging computing paradigm, greatly alleviates these problems by processing and making decisions at the edge where data is generated. Especially driven by the application of mobile edge computing, intelligent task scheduling, deep reinforcement learning, and quantum heuristic algorithms, edge computing is rapidly developing, providing more efficient, reliable, and responsive services. The four background technologies mainly involved in this invention include the following aspects:

[0003] Edge computing and mobile edge computing: Edge computing solves the problems of high data transmission latency and expensive bandwidth by processing data close to the data source. In this architecture, data does not need to be transmitted to a remote cloud center but is processed on local nodes, enabling real-time data analysis and decision-making. Key advantages of edge computing include low latency, high bandwidth utilization, and enhanced privacy protection. Mobile edge computing (MEC) is an extension of edge computing, specifically designed for mobile users and wireless network environments. MEC provides computing and storage resources at traditional mobile network base stations, enabling mobile devices to execute complex applications without relying on a core cloud. This model is particularly suitable for connected vehicles, mobile video analytics, location-based services, and the Internet of Things (IoT).

[0004] Task Scheduling Strategy: In mobile edge computing environments, effective task scheduling is central to resource management. An efficient scheduling strategy not only improves resource utilization but also ensures application performance and user satisfaction. The complexity of task scheduling lies in the need to consider multiple factors, including computing resources, network conditions, task priorities, and data dependencies. Modern task scheduling strategies increasingly employ intelligent algorithms to address the dynamic and uncertain edge computing environment. For example, prediction-based scheduling algorithms can predict future loads based on historical data, thus making scheduling decisions in advance. Furthermore, multi-objective optimization strategies are designed to simultaneously meet multiple Quality of Service (QoS) requirements, such as minimizing latency and maximizing energy efficiency.

[0005] Deep Reinforcement Learning (DRL) algorithms are advanced techniques that combine deep learning with reinforcement learning, particularly well-suited for solving decision-making problems in highly uncertain and complex environments. With the help of deep learning, DRL can handle high-dimensional state spaces, learning the mapping from environment states to action values ​​through neural networks. This is especially important in edge computing, where edge nodes often need to make rapid decisions amidst dynamically changing network conditions and fluctuating user demands. DRL trains its policy through interaction with the environment, employing a reward mechanism where the environment rewards or penalizes each decision made by the algorithm. Through long-term training, DRL can optimize its policy to maximize cumulative rewards. For example, DRL can be applied to dynamic resource allocation problems, where the algorithm learns how to allocate limited computational and network resources based on the current system state to maximize overall system performance.

[0006] Quantum-inspired algorithms draw inspiration from the concepts of quantum computing. Although they do not require quantum computer hardware, these algorithms are designed considering the superposition and entanglement properties of qubits, making them highly efficient in handling certain types of problems, such as combinatorial optimization and search problems. Quantum heuristics are often used to solve problems that are computationally too demanding for classical algorithms. In edge computing scenarios, quantum heuristics can be used to optimize task scheduling by finding the optimal solution among all potential scheduling schemes, thereby reducing computational latency and energy consumption. These algorithms are typically implemented using quantum simulations, simulating quantum behavior on classical computers. For example, the quantum annealing algorithm solves optimization problems by simulating the quantum annealing process. Specifically, it explores the possible solution space by simulating the annealing process of quantum particles, gradually cooling to the lowest energy state to find a globally optimal or approximate solution. Summary of the Invention

[0007] The purpose of this invention is to provide a mobile edge computing task scheduling method based on quantum heuristic algorithms. In a mobile edge computing environment, a quantum heuristic deep reinforcement learning (QI-DRL) algorithm is used to solve the computational task scheduling optimization problem. By introducing concepts from quantum mechanics into a dual deep Q-network (DDQN), and comprehensively considering the uncertainty and dynamic changes of the system service state, a more adaptive and robust task scheduling method is proposed. Ultimately, this significantly optimizes task scheduling in high-demand scenarios, reducing task drop rate and latency.

[0008] To achieve the above objectives, this invention provides a mobile edge computing task scheduling method based on a quantum heuristic algorithm, comprising the following steps:

[0009] S1. Construct a scenario where mobile terminals are represented by a set I = {1, ..., I} and are located within the service range of multiple MEC servers, which are represented by a set M = {1, ..., M}. Task requests are initiated by the mobile terminals, and each request contains one or more tasks. These tasks are identified by index j and are processed by the MEC server or the terminal itself. The execution order of tasks is first-in, first-out.

[0010] S2, System Architecture: In each cycle j, the system generates an offload decision (r) in response to a request from terminal i. j ,e j ); where r j Indicates the execution location: r j =0 indicates that the task is executed locally on the terminal, r j =m indicates that the task is offloaded to the MEC server; m∈M; e j This indicates the number of energy units allocated to the task;

[0011] S3. System parameter update: Based on the task processing system model in step S2, simplify the total task execution delay t. j And describe the updates of the energy queue and task queue for each cycle j;

[0012] S4. Problem Construction: Define the service status and comprehensive observations within each period j, and optimize the problem using the optimal scheduling strategy ζ(·).

[0013] S5. Solution method: The solution method is obtained through multi-period service mixed state representation, service mixed state uncertainty measurement, and quantum heuristic deep reinforcement learning.

[0014] Preferably, in step S2, according to e j The system determines the different values ​​of e based on... j Different values ​​of generate corresponding processing tasks:

[0015] If e j =0, the task will not be executed, but will be stored in the task queue waiting for subsequent energy. The task will continue if e j Exceeding the number of energy units in the current energy queue or exceeding The task request will be discarded.

[0016] Preferably, in step S2, considering the computational efficiency and power consumption of executing the task locally on the mobile terminal, the time required to execute the task on the terminal itself is evaluated, denoted as... in This is the power consumption of the terminal when it executes a task within period j.

[0017] The power consumption calculation formula is: Where ν i α and α represent CPU frequency and average activity factor, respectively, while k0 and k1 represent dynamic and static power consumption coefficients.

[0018] Preferably, in step S2, when the system selects to offload the task to the MEC server, a connection is established between the terminal and the server. If the connection is different from the one used in the previous cycle, then it is established via s j This is used to identify the connection status between the terminal and the MEC server at the start of period j, and to set the corresponding handover delay h based on changes in the connection status. j The task takes t seconds to process on the MEC server. e .

[0019] Preferably, in step S3, the total task execution delay t j The simplified method is shown below:

[0020] If the task is executed locally, then

[0021] If the task is offloaded to the MEC server, then

[0022] If the decision is made to discard the task, then t j =0;

[0023] In each period j, the update description of the energy queue is as follows:

[0024]

[0025] in This represents the number of energy units accumulated by terminal i during period j;

[0026] The update description of the task queue is as follows:

[0027]

[0028] in This represents the number of tasks generated in period j.

[0029] Preferably, in step S4, the specific steps for problem construction are as follows:

[0030] Define the service state of each period j as N. j Specifically, it is expressed as:

[0031]

[0032] Where Ω represents the state space of the service state, defined as follows:

[0033]

[0034] The comprehensive observation of each period j is defined as γ. j Specifically, it is expressed as:

[0035] γ j =N j ×R j ∈Γ

[0036] Where Γ is composed of Ω and a positive real triplet (R) + ) 3 The set that constitutes;

[0037] Utility function U(γ) j ,ζ(γ j The system performance for each cycle j is evaluated using the following mathematical expression:

[0038]

[0039] Where Δj represents the number of periods considering the average utility, starting from period j-Δj and continuing until period j is included, the function u(γ) k ,ζ(γ k ))Measured within a single period k, based on comprehensive observations of γ k and the strategy adopted ζ(γ) k The immediate utility of )

[0040] To maximize long-term cumulative utility by determining the optimal scheduling policy ζ(·), the optimization problem can be formalized as follows:

[0041]

[0042] Preferably, in step S5, the specific steps for representing the multi-cycle service hybrid state are as follows:

[0043] Each service state N j Represented as a right vector |ψ j >, normalize it to a pure state |φ j >, through |ψ j >and its norm|ψ j | Divide:

[0044]

[0045] Ensure |φ j The norm is 1, making it a quantum-like pure state;

[0046] Capturing the impact of recent service states on the current state, the mixture state density matrix ρ of period j' includes the normalized pure states |φ| of the most recent k periods. j >:

[0047]

[0048] Where p j Represents the relationship between each state |φ j The relevant probabilities reflect the likelihood and impact of each past state on the current state;

[0049] The summation from j'-k+1 to j' balances the need for timeliness with avoiding overfitting to older, less relevant data. The total sum of the probabilities of the above states is...

[0050] Preferably, in step S5, the specific steps for measuring the uncertainty of the service mixture are as follows:

[0051] ρ j' Decompose it into its eigenbases:

[0052]

[0053] Where p α Represents the relationship between each eigenvector |χ α >The relevant probability, and |χ α > Form an orthogonal basis for the service states, with eigenvectors {|χ α >} represents a potential measurement operator;

[0054] After obtaining the measurement operator, the Born rule is used to measure the probability of each state occurring in the mobile edge computing network, which is determined by |χ α > represents the probability p of the state α It is given by the following formula:

[0055] p α =Tr(|χ α ><χ α |ρ j' )

[0056] The method selects the maximum probability p α Status | χ α As a network observation:

[0057]

[0058] Mixed state ρ j' Collapse into the most likely state This state This serves as a quantification and precise representation of the current service status.

[0059] Preferably, in step S5, the specific steps of quantum-inspired deep reinforcement learning are as follows:

[0060] Execute action (r) j' ,e j' And observe the resulting new service state N. j'+1 and its system utility U j' Then, N is processed using a quantum heuristic framework. j'+1 This is integrated with the service states of the past k periods to form a quantum mixed state:

[0061]

[0062] For ρ j'+1 Perform quantum-like measurements to determine the most likely state of service. This state is determined by the mixed service state ρ j'+1 Highest similarity measurement operator express;

[0063] After performing quantum-like measurement, and the new request status R j'+1 Extended to its higher-dimensional representation in space Γ:

[0064]

[0065] The integrated observations form the updated observations for the next cycle:

[0066]

[0067] Updated observations of γ j'+1 This information is provided as input to the DDQN and enters the main Q network for subsequent action selection.

[0068]

[0069] Where θ represents the adjustable parameters of the main Q-network.

[0070] The parameters of the main Q-network are continuously updated to minimize the discrepancy between the predicted results and observed reality. The updated target value y... j'+1 The following recalculation combines immediate system utility with expected future benefits:

[0071]

[0072]

[0073] Therefore, the mobile edge computing task scheduling method based on a quantum heuristic algorithm using the above structure, as described in this invention, has the following beneficial effects:

[0074] (1) This invention proposes an innovative quantum-inspired deep reinforcement learning (QI-DRL) algorithm for task scheduling optimization in mobile edge computing (MEC) environments. By integrating quantum mechanical concepts into a dual deep Q-network (DDQN), this invention successfully improves the adaptability and robustness of task scheduling. Specifically, the method utilizes quantum-like mixed states to comprehensively represent the system's service state, combining past and present information to reflect the system's dynamic characteristics. Furthermore, through quantum measurement techniques, the method can accurately extract observation information, thereby improving the decision-making process. These innovations not only enhance the method's performance under uncertain and dynamic conditions but also validate its superiority over traditional scheduling algorithms and standard DRL methods in experimental evaluations, particularly in terms of significant improvements in task drop rate and latency.

[0075] (2) This invention not only demonstrates the theoretical advancement of the QI-DRL algorithm, but also verifies its effectiveness in practical applications through extensive experiments with real-world datasets. Furthermore, this invention provides a novel perspective and technical approach for addressing resource management and task scheduling issues in MEC environments, enabling the system to manage and schedule tasks more effectively under conditions of high demand variability and limited resources, significantly optimizing operational efficiency and response speed in MEC environments.

[0076] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0077] Figure 1 This is a schematic diagram illustrating the uncertainty of mobile edge computing in a mobile edge computing task scheduling method based on a quantum heuristic algorithm according to the present invention.

[0078] Figure 2 This is a schematic diagram of the system flow within a single cycle of a mobile edge computing task scheduling method based on a quantum heuristic algorithm according to the present invention.

[0079] Figure 3 This is a schematic diagram of the quantum-like feature extraction process of a mobile edge computing task scheduling method based on a quantum heuristic algorithm according to the present invention. Detailed Implementation

[0080] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0081] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0082] This invention proposes an innovative task scheduling method aimed at optimizing task processing performance in mobile edge computing scenarios. The method intelligently offloads tasks between the local machine and the MEC server through a dynamic decision-making process, while considering device energy efficiency and task queue management. Each task request is evaluated and processed based on data transfer volume, required memory, and the number of CPU cores. This scheduling method utilizes quantum heuristics and deep reinforcement learning techniques, constructing a quantum representation of hybrid service states to process and optimize service decisions for each task cycle. The method's design not only improves task processing efficiency but also effectively manages energy consumption and system resources, thereby achieving continuous performance optimization in dynamic and uncertain network environments.

[0083] Example

[0084] like Figure 1-3 As shown, this invention provides a mobile edge computing task scheduling method based on a quantum heuristic algorithm, comprising the following steps:

[0085] S1. Construct a scenario where mobile terminals are represented by a set I = {1, ..., I} and are located within the service range of multiple MEC servers, which are represented by a set M = {1, ..., M}. Task requests are initiated by the mobile terminals, and each request contains one or more tasks. These tasks are identified by index j and are processed by the MEC server or the terminal itself. The execution order of tasks is first-in, first-out.

[0086] For each request R j Using a triple (D) j B j C j ) indicates that, among which, D j B represents the amount of data transfer required for the unloading task. j Represents the memory required to process a task, Cj This represents the required number of CPU cores. Additionally, each terminal i has a task queue. Maximum length is Used for temporary storage tasks. Additionally, considering the battery and power limitations of the terminal devices, each terminal i is also configured with an energy queue. Its maximum length is Used to store energy units.

[0087] S2, System Architecture: In each cycle j, the system generates an offload decision (r) in response to a request from terminal i. j ,e j ); where r j Indicates the execution location: r j =0 indicates that the task is executed locally on the terminal, r j =m indicates that the task is offloaded to the MEC server; m∈M; e j This indicates the number of energy units allocated to the task;

[0088] S3. System parameter update: Based on the task processing system model in step S2, simplify the total task execution delay t. j And describe the updates of the energy queue and task queue for each cycle j;

[0089] S4. Problem Construction: Define the service status and comprehensive observations within each period j, and optimize the problem using the optimal scheduling strategy ζ(·).

[0090] S5. Solution method: The solution method is obtained through multi-period service mixed state representation, service mixed state uncertainty measurement, and quantum heuristic deep reinforcement learning.

[0091] In step S2, according to e j The system determines the different values ​​of e based on... j Different values ​​of generate corresponding processing tasks:

[0092] If e j =0, the task will not be executed, but will be stored in the task queue waiting for subsequent energy. The task will continue if e j Exceeding the number of energy units in the current energy queue or exceeding The task request will be discarded.

[0093] In step S2, considering the computational efficiency and power consumption of executing the task locally on the mobile terminal, the time required to execute the task on the terminal itself is evaluated, denoted as... in This is the power consumption of the terminal when it executes a task within period j.

[0094] The power consumption calculation formula is: Where ν i α and α represent CPU frequency and average activity factor, respectively, while k0 and k1 represent dynamic and static power consumption coefficients.

[0095] In step S2, when the system chooses to offload the task to the MEC server, a connection needs to be established between the terminal and the server. If this connection is different from the one used in the previous cycle, a network switch is required, which may cause additional latency. j This is used to identify the connection status between the terminal and the MEC server at the start of period j, and to set the corresponding handover delay h based on changes in the connection status. j The task takes t seconds to process on the MEC server. e .

[0096] In step S3, the total task execution delay t j The simplified method is shown below:

[0097] If the task is executed locally, then

[0098] If the task is offloaded to the MEC server, then

[0099] If the decision is made to discard the task, then t j =0;

[0100] In each period j, the update description of the energy queue is as follows:

[0101]

[0102] in This represents the number of energy units accumulated by terminal i during period j;

[0103] The update description of the task queue is as follows:

[0104]

[0105] in This represents the number of tasks generated in period j.

[0106] In step S4, the specific steps for problem construction are as follows:

[0107] First, define the service state of each period j as N. j Specifically, it is expressed as:

[0108]

[0109] Where Ω represents the state space of the service state, defined as follows:

[0110]

[0111] Next, the comprehensive observation for each period j is defined as γ. j The format is:

[0112] γ j =N j ×R j ∈Γ

[0113] Where Γ is composed of Ω and a positive real triplet (R) + ) 3 A set of components.

[0114] To optimize task scheduling in a MEC environment, a framework aimed at maximizing utility was designed. The utility function U(γ) j ,ζ(γ j This is used to evaluate the system performance for each cycle j, covering task processing efficiency, energy conservation, and queue management, thus comprehensively reflecting the system's effectiveness. Mathematically, it is expressed as:

[0115]

[0116] Where Δj represents the number of periods considering the utility average, starting from period j-Δj and continuing until period j is included. The function u(γ) k ,ζ(γ k ))Measured within a single period k, based on comprehensive observations of γ k and the strategy adopted ζ(γ) k The immediate utility of the decision is considered. This formulation not only reflects the direct benefits of decisions made within each cycle but also evaluates the cumulative performance over the Δj-cycle window, ensuring a balance between short-term actions and their long-term impact on system utility.

[0117] Therefore, by determining the optimal scheduling policy ζ(·) to maximize the long-term cumulative utility, the problem can be formalized as solving the following optimization problem:

[0118]

[0119] In step S5, when dealing with the inherent uncertainties in the MEC environment, the method innovatively applies quantum probability theory by using the density matrix ρ to represent different service states over multiple periods as a quantum mixed state. This approach is particularly valuable in MECs because it not only captures the probabilistic properties of service states at a single moment but also depicts their evolution and interactions over time.

[0120] The specific steps for representing a hybrid state in a multi-cycle service are as follows:

[0121] Each service state N j Represented as a right vector |ψj >. Because |ψ j It is usually not a pure state (because it is not a unit vector), it needs to be normalized to a pure state |φ. j >, specifically through |ψ j >and its norm|ψ j |(i.e.|ψ) j Divide by the square root of its inner product:

[0122]

[0123] This normalization ensures that |φ j The norm is 1, making it a quantum-like pure state.

[0124] To effectively capture the impact of recent service states on the current state, the mixed-state density matrix ρ of period j' includes the normalized pure states |φ| of the most recent k periods. j >:

[0125]

[0126] Where p j Represents the relationship between each state |φ j The relevant probabilities reflect the likelihood and impact of each past state on the current state. The summation from j'-k+1 to j' balances the need for timeliness with avoiding overfitting to older, less relevant data. This method ensures that the sum of the total probabilities of these states is... It provides a coherent and comprehensive quantum-like representation of the current service environment.

[0127] In step S5, the proposed quantum-like uncertainty measurement technique is used to measure the mixed-state density matrix ρ j' Extract the state that best matches the current system state.

[0128] The specific steps for measuring service mixture uncertainty are as follows:

[0129] To obtain the mixture density matrix ρ j' A reasonable set of measurement operators, first ρ j' Decompose it into its eigenbase: ρ j' Decompose it into its eigenbases:

[0130]

[0131] Where p α Represents the relationship between each eigenvector |χ α >The relevant probability, and |χ α > Form an orthogonal basis for the service states, with eigenvectors {|χ α >} represents a potential measurement operator;

[0132] The completeness of eigenvectors is derived from their orthogonality and normality from ρ. j' The Hermitian properties of it have been confirmed:

[0133] Orthogonality and normality: for each eigenvector |χ α >Orthogonal to other vectors and normalized, satisfying χ α |χ β >=δ αβ , where δ αβ It is the Kronecker product.

[0134] Completeness relation: The sum of the inner products of all eigenvectors equals the identity operator on the Hilbert space, establishing its completeness: ∑|χ α ><χ α | = I, where I is the identity operator.

[0135] The above properties confirm that any state |ψ> can be represented as these eigenvectors |χ. α The linear combination of > verifies the applicability of these eigenvectors as measurement operators.

[0136] After obtaining the measurement operator, the Born rule is used to measure the probability of each state occurring in the mobile edge computing network. The measurement reveals that the state is determined by |χ... α > represents the probability p of the state α It is given by the following formula:

[0137] p α =Tr(|χ α ><χ α |ρ j' )

[0138] This step translates the theoretical framework into actionable data, determining which states are most important in any given period.

[0139] To effectively utilize this quantum-like measurement in subsequent DRL environments, the method selects the maximum probability p α Status | χ α As a network observation:

[0140]

[0141] This method effectively mixes ρ j' Collapse into the most likely state This state This serves as a quantified and precise representation of the current conditions of the service status, reflecting both past influences and current observations.

[0142] In step S5, the specific steps of quantum-inspired deep reinforcement learning are as follows:

[0143] Execute action (r) j' ,e j' And observe the resulting new service state N. j'+1 and its system utility U j' Then, N is processed using a quantum heuristic framework. j'+1 This is integrated with the service states of the past k periods to form a quantum mixed state:

[0144]

[0145] For ρ j'+1 Perform quantum-like measurements to determine the most likely state of service. This state is determined by the mixed service state ρ j'+1 Highest similarity measurement operator express;

[0146] After performing quantum-like measurement, and the new request status R j'+1 Extended to its higher-dimensional representation in space Γ:

[0147]

[0148] The integrated observations form the updated observations for the next cycle:

[0149]

[0150] Updated observations of γ j'+1 This information is provided as input to the DDQN and enters the main Q network for subsequent action selection.

[0151]

[0152] Where θ represents the adjustable parameters of the main Q-network.

[0153] The parameters of the main Q-network are continuously updated to minimize the discrepancy between the predicted results and observed reality. The updated target value y... j'+1 The following recalculation combines immediate system utility with expected future benefits:

[0154]

[0155] L(θ)=E[(y j'+1 -Q(γ j'+1 ,(r j'+1 ,e j'+1 );θ)) 2 ].

[0156] Therefore, this invention adopts the aforementioned quantum-heuristic algorithm-based task scheduling method for mobile edge computing and proposes an innovative quantum-heuristic deep reinforcement learning (QI-DRL) algorithm for task scheduling optimization in mobile edge computing (MEC) environments. This invention successfully improves the adaptability and robustness of task scheduling by integrating quantum mechanics concepts into a dual deep Q-network (DDQN). It demonstrates the theoretical advancement of the QI-DRL algorithm and verifies its practical effectiveness through extensive experiments with real-world datasets. This invention also provides a novel perspective and technical approach for addressing resource management and task scheduling problems in MEC environments, enabling the system to manage and schedule tasks more effectively under conditions of high demand variability and resource constraints, significantly optimizing operational efficiency and response speed in MEC environments.

[0157] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for mobile edge computing task scheduling based on quantum heuristic algorithm, characterized in that: Comprising the following steps: S1, constructing a scenario, in which mobile terminals are represented by a set I = {1, …, I} and located within the service range of multiple groups of MEC servers represented by a set M = {1, …, M}, task requests are initiated by mobile terminals, each request contains one or more tasks identified by index j, and the tasks are processed by MEC servers or the terminal itself, and the execution order of the tasks is first-in-first-out; S2, system architecture, at each period j, the system generates an offloading decision (r j ,e j ) for a request from terminal i; wherein r j represents the execution location: r j = 0 means executing the task locally on the terminal, r j = m means offloading the task to a MEC server m e M; e j represents the number of energy units allocated to the task; S3, system parameter update, based on the task processing system model in step S2, simplifying the total delay t of task execution j and describe the update of the energy queue and the task queue for each period j; S4, problem construction, defining the service state and comprehensive observation in each period j, and optimizing the problem through the best scheduling strategy ζ(·); S5, solution method, obtaining the solution method through multi-period service mixed state representation, service mixed state uncertainty measurement, and quantum heuristic deep reinforcement learning. 2.The method of claim 1, wherein: In step S2, the system generates a corresponding processing task according to the different values of e j In step S2, the system generates a corresponding processing task according to the different values of e j In step S2, the system generates a corresponding processing task according to the different values of e If e j = 0, the task will not execute, but will be saved in the task queue to wait for subsequent energy, if 0 The task will continue to execute, if e j exceeds the number of energy units in the current energy queue or exceeds The task request will be discarded. 3.The method of claim 1, wherein: In step S2, the time required for performing the task at the terminal itself is evaluated, denoted as wherein is the power consumption of the terminal when performing the task in period j; The power consumption calculation formula is wherein v i , a respectively represent the CPU frequency and the average activity factor, and k0 and k1 represent the dynamic and static power consumption coefficients. 4.The method of claim 1, wherein: In step S2, when the system selects to offload the task to the MEC server, a connection between the terminal and the server is established, and if the connection used in the previous period is different, the s j period j is identified, and the corresponding switching delay h is set according to the change in the connection state j； The processing time of the task on the MEC server is t e . 5.The method of claim 1, wherein: In step S3, the total delay t of the task execution j The simplified way is shown as follows: If the task is executed locally, then If the task is offloaded to the MEC server, then If it is decided to discard the task, then t j = 0; In each period j, the update of the energy queue is described as: wherein represents the number of energy units accumulated by terminal i in period j; The update of the task queue is described as: wherein denotes the number of tasks generated in cycle j. 6.The method of claim 1, wherein: In step S4, the specific steps of problem construction are as follows: Define the service state of each cycle j as N j , which is specifically represented as: Where Ω represents the state space of the service state, which is defined as follows: The combined observation of each period j is defined as γ j , which is specifically represented as: gamma j = N j x R j ∈ Γ where Γ is the set consisting of Ω and positive real number triplets (R + ) 3 consisting of Ω and positive real number triplets (R The utility function U(γ j , ζ(γ j )) is used to evaluate the system performance at each cycle j, mathematically expressed as: where Δj denotes the number of periods to consider for the average of utility, starting from period j - Δj and including period j, and the function u(γ k , ζ(γ k )) measures the instantaneous utility in a single period k based on the aggregate observation γ k and the strategy taken ζ(γ k ). By determining the best scheduling strategy ζ(·), the long-term cumulative utility is maximized, which is formalized as solving the following optimization problem:

7. The method of claim 1, wherein the method is based on a quantum heuristic algorithm. In step S5, the specific steps of multi-period service mixed state representation are as follows: Each service state N j Represented as a right vector |ψ j >, normalize it to a pure state |Φ j >, through |ψ j >and its norm|ψ j | Divide: Ensuring |φ j > norm 1, making it a quasi-quantum pure state; The mixed state density matrix p of period j' captures the influence of the recent service state on the current state, including the normalized pure state |φ j > where p j represents the probability associated with each state |φ j > reflects the likelihood and influence of each past state on the current state; The summation from j'-k+1 to j' balances the need for recency and the avoidance of overfitting to older, less relevant data, and the sum of the total probabilities of the above states is 8.The method of claim 1, wherein: In step S5, the specific steps of service mixed state uncertainty measurement are as follows: The ρ j' is decomposed into its eigenbases: where p α represents the probability associated with each eigenvector |χ α > and |χ α > forms an orthonormal basis for the service state, the eigenvectors {|χ α >} being the potential measurement operators; After obtaining the measurement operators, the probabilities of each state to occur in the mobile edge computing network are measured using the Born rule, in the state represented by |χ α > is given by: α is given by: p α = Tr(|χ α > χ α |ρ j' ) The law selects the maximum probability p α of the state |χ α > as the network observation: The mixed state p j' Collapse to the most probable state This state is then a quantitative and accurate representation of the current condition of the service state. 9.The method of claim 1, wherein: In step S5, the specific steps of quantum heuristic deep reinforcement learning are as follows: Execute action (r) j' ,e j' And observe the resulting new service state N. j'+1 and its system utility U j' Then, N is processed using a quantum heuristic framework. j'+1 This is integrated with the service states of the past k periods to form a quantum mixed state: On ρ j'+1 Performing a quasi-measurement to determine the most likely service state This state is represented by the measurement operator with the highest degree of similarity to the mixed service state ρ j'+1 ​​ After a quasi-quantum measurement, and a new request state R j'+1 is extended to its high-dimensional representation within the space Γ: Through integration, the next period update observation is formed: Updated observation γ j'+1 Provided as input to the DDQN, into the main Q network, for subsequent action selection: Where θ represents the adjustable parameters of the main Q network; The parameters of the main Q-network are continuously updated to minimize the difference between the predicted outcome and the observed reality; the target value y of the update j'+1 The recalculation, combining the immediate system utility and the expected future earnings, is as follows:

Images