Edge calculation method and system for new energy electric field task unloading

Abstract

The invention discloses an edge calculation method and system for new energy electric field task unloading, and the method comprises the steps: building a new energy electric field mobile edge calculation unloading model through combining the parameters of equipment and a base station and the system communication bandwidth; based on a new energy electric field mobile edge computing unloading model, dividing unloading priorities of data collected by various sensing devices; considering system unloading time delay and data task processing time delay constraints, and constructing a time delay optimization problem of load balancing; long-term energy consumption constraints of MEC in a Lyapunov decoupling time delay optimization problem are adopted, and the time delay optimization problem is simplified into an optimization problem without long-term constraints by constructing a virtual energy deficiency queue; and an AC algorithm is adopted to perform joint optimization of task unloading and resource allocation on the optimization problem without long-term constraint, and maximization of the service quality of the new energy electric field data acquisition system is realized. According to the method, data with relatively high time delay requirements in the new energy electric field can be processed in real time, and optimal acquisition and processing of intensive data are realized.

Description

Technical Field

[0001] The present invention relates to the field of mobile edge computing technology, and specifically to an edge computing method and system for offloading new energy electric field tasks. Background Art

[0002] With the surge in the number of mechanical equipment in new energy power plants, especially the dense deployment of wind turbines in wind farms, the amount of data carried by the entire system has shown an explosive growth trend, and the complexity and diversity of the data have also increased significantly. Faced with the real-time requirements of such a huge amount of data, new energy power plants must rely on advanced data processing technology and low-latency communication capabilities to ensure rapid data collection, processing and analysis, thereby supporting intelligent decision-making and rapid response.

[0003] However, the particularity of the new energy electric field also brings many challenges to its data processing. First, the wide distribution and frequent mobility of sensor devices make the data collection network face the risk of dynamic changes and instability, which increases the complexity and difficulty of data transmission. Secondly, the MEC environment is affected by many unpredictable factors such as geographical location, climate conditions, and power supply. Its stability and reliability have become key factors restricting data processing efficiency. With the diversification and personalized development of user service needs, the service migration process must not only consider the compatibility at the technical level, but also weigh the migration cost to ensure the continuity and stability of the service. It is particularly important that in the application scenario of the new energy electric field, which has extremely high environmental protection requirements, how to effectively reduce the energy consumption of MEC equipment and achieve green computing has become an important indicator for measuring the sustainability of the system. Therefore, while promoting digital transformation, the new energy electric field also needs to actively explore new technologies and methods for energy saving and consumption reduction, such as adopting efficient energy management systems, optimizing the working mode of MEC equipment, and promoting the use of renewable energy power supply, so as to balance the relationship between data processing capabilities and energy consumption and promote the sustainable development of the new energy electric field. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides an edge computing method and system for offloading new energy electric field tasks to solve the problems of insufficient stability of edge computing and dynamic networks, weak real-time data processing capabilities, and difficulty in controlling business migration costs.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides an edge computing method for offloading tasks in a new energy electric field, comprising: constructing a mobile edge computing offloading model for a new energy electric field in combination with parameters of equipment and base stations and system communication bandwidth; dividing the offloading priority of data collected by various sensor devices in the new energy electric field based on the mobile edge computing offloading model of the new energy electric field; considering the unloading delay of the new energy electric field system and the data task processing delay constraints, constructing a load balancing delay optimization problem; using Lyapunov to decouple the long-term energy consumption constraints of MEC in the delay optimization problem, and simplifying the delay optimization problem to an optimization problem without long-term constraints by constructing a virtual energy deficit queue; using an AC algorithm to jointly optimize task offloading and resource allocation for the optimization problem without long-term constraints, so as to maximize the service quality of the new energy electric field data acquisition system.

[0008] As a preferred solution of the edge computing method for offloading tasks in a new energy electric field described in the present invention, wherein: the construction of a new energy electric field mobile edge computing offloading model includes:

[0009] Assume that N = {1, 2, ..., n} represents the set of sensor devices, M = {1, 2, ..., m} represents the set of edge servers, each sensor device generates a task that needs to be calculated in each time slot, part of which is executed locally on the sensor device and part is offloaded to the nearby edge server, f n The local computing resources of SE n are limited, and the edge server has a certain computing resource f m and bandwidth resource B m , when SE n is within the communication coverage of the edge server, the edge server can offload services to SE n;

[0010] The collected data types are divided into three types: high, medium, and low. k = {1, 2, 3}. Assume that the user has an unloading task in time slot t. in, represents the computational resources required by SE n in time slot t, A n (t) represents the amount of data generated by SE n in time slot t∈T, Represents task k t Maximum latency requirement;

[0011] The running time of the whole system is divided into multiple time slots T = {T1, T2, ..., T n}, all SE n data transmission adopts time division multiple access technology, defining is the unloading ratio coefficient, where b n (t)∈[0,1], SE n performs the calculation of the task locally for a period of The period for unloading the remaining tasks to the edge server is Definition P n (t) = {p n (t),n∈N,t∈T}, where p n (t) is the transmission power from SE n to the edge server.

[0012] As a preferred solution of the edge computing method for offloading tasks in a new energy electric field described in the present invention, the offloading priority of the data collected by various sensor devices in the new energy electric field includes:

[0013] Data is unloaded in sequence according to the division of data types. When SE n unloads part of the task to the nearby edge server, the signal-to-noise ratio from SE n to MEC m in time slot t is defined as:

[0014]

[0015] Among them, p n is the transmission power of SE n, is the channel gain from SE n to MEC m, σ 2 is the noise density;

[0016] The transmission rate of MEC m and SE n in time slot t is calculated as:

[0017]

[0018] Where W is the communication bandwidth;

[0019] The transmission delay of the offloaded task from SE n to MEC m at time slot t is calculated as:

[0020]

[0021] The transmission energy consumption is

[0022] The offload task is processed simultaneously on the local device and the edge server. When the k-th type of data of SE n is calculated locally, the local calculation delay in time slot t is expressed as:

[0023]

[0024] When the k-th type of data of SE n is offloaded to MEC m, the MEC calculation delay in time slot t is:

[0025]

[0026] Using edge servers to offload tasks generates additional long-term MEC computing energy consumption:

[0027]

[0028] in, represents the energy budget of MEC in time slot t;

[0029] Through the direct cellular link between each sensor device, the task is offloaded to a MEC in time slot t. In the next time slot, SE n roams to the communication coverage of another edge node and selects MEC m as the edge server. The service migration cost of SE n can be expressed as:

[0030]

[0031] Where γ represents the cost of migrating a task from one edge server to another edge server. represents the decision result of service migration. When the edge server of time slot t is different from the edge server of time slot t-1, otherwise

[0032] As a preferred solution of the edge computing method for offloading new energy electric field tasks described in the present invention, the delay optimization problem of constructing load balancing includes:

[0033] The local computation of data and task offloading are performed in parallel, and the execution time of the k-th data of SE n is The optimization target is determined by the task delay and migration cost of the device, which can be expressed as:

[0034]

[0035] Among them, Q n (t) represents the penalty factor of QoS of SE n at time slot t, and the weight parameter β represents the QoS preference factor of the user at time slot t;

[0036] Define the optimization variables of the system in time slot t as P n (t)=[p1(t),p2(t),...,p n (t)],B n (t)=[b1(t),b2(t),...,b n (t)],F n (t)=[f1(t),f2(t),...,f n (t)], the optimization function is formally expressed as:

[0037]

[0038] stC1:p n (t)≤p max

[0039] C2:0≤b n (t)≤1

[0040]

[0041] Among them, C1 is the device transmission power constraint, C2 is the load ratio coefficient constraint, C3 is the device resource constraint, C4 is the device delay requirement, and C5 is the long-term energy consumption constraint.

[0042] As a preferred solution of the edge computing method for offloading renewable energy electric field tasks described in the present invention, the delay optimization problem is simplified to an optimization problem without long-term constraints, including:

[0043] Design a dynamic virtual energy loss queue and define the virtual energy loss queue of MEC m in time slot t as In the initial time slot, the virtual energy loss queue is set to 0, that is, D n (0) = 0, the evolution of the virtual energy loss queue of a slot is expressed as:

[0044]

[0045] in It is the deviation between energy expenditure and energy budget;

[0046] Define the Lyapunov function to describe the sum of the squares of all virtual energy loss queue backlogs at the tth time slot:

[0047]

[0048] Where Θ(t)={D1(t),D2(t),...,D n (t)}, Δ(Θ(t)) is introduced as the conditional Lyapunov drift to ensure the stability of the system:

[0049] Δ(Θ(t))=Ε[L(Θ(t+1))-L(Θ(t))|Θ(t)].

[0050] Control Δ(Θ(t)) to reduce queue backlog and stabilize the system. The drift penalty algorithm is used to obtain the Lyapunov drift penalty function Δ V (Θ(t)):

[0051] Δ V (Θ(t))=Δ(Θ(t))+VΕ[D n (t)]

[0052] Wherein, V represents a non-negative weight coefficient; the upper bound of the Lyapunov drift penalty function is expressed as:

[0053]

[0054] in,

[0055] The delay optimization problem is simplified to an optimization problem without long-term constraints by using Lyapunov queue optimization technology:

[0056]

[0057] stC1-C4

[0058] Among them, X t = {P n (t),B n (t),F n (t} represents the task offloading and resource allocation strategy of all SE n at time slot t, represents the feasible task offloading and resource allocation strategy at time slot t, and the weight of the positive control parameter V is Q n (t) Achieve an ideal balance between user QoS and MEC energy consumption.

[0059] As a preferred solution of the edge computing method for task offloading of new energy electric fields described in the present invention, the joint optimization of task offloading and resource allocation for optimization problems without long-term constraints includes:

[0060] The Actor-Critic based DRL algorithm is used to process the optimization problem without long-term constraints, following the changes of Markov property computing resources and the changes of the described wireless environment, wherein the three key elements of the learning environment system are the state set S, the action set A and the reward function set R;

[0061] The state of MDP is the communication state of the available edge servers around SE and the energy consumption when SE offloads tasks to the edge servers. At each decision time t, the state of the MEC system is expressed as:

[0062]

[0063] in, is the signal-to-noise ratio from SE n to MEC, is the transmission energy consumption offloaded from SE n to MEC;

[0064] According to the current system state s(t), the mobile unloading module takes action a(t), which is expressed as:

[0065] a(t)={p n (t),b n (t),f n (t)}

[0066] According to the current system state s(t) and the selected action a(t), the reward obtained by SE at time slot t is expressed as:

[0067]

[0068] The entire state space constituted by the sensing device is S(t) = [s(t), t∈T], the entire action space constituted by the device is A(t) = [a(t), t∈T], and the entire reward space constituted by the device is R(t) = [r(t), t∈T].

[0069] As a preferred solution of the edge computing method for offloading new energy electric field tasks described in the present invention, it also includes:

[0070] The Critic network is responsible for estimating the value of the state value function under the current state s(t) and updating its parameters according to the reward value r(t) given by the environment. The Actor network plays the role of a decision maker and is responsible for generating corresponding action decisions based on the current state s(t).

[0071] Actor policy network: from a random policy π(a(t)|s(t); θ t ), expressed as:

[0072] a(t)~π(a(t)|s(t);θ t )

[0073] Among them, π(a(t)|s(t); θ t )=Pr(s(t+1)|s(t),a(t)) indicates the random strategy π Mapping environmental states to probabilities of taking actions;

[0074] The measurement strategy J(θ) is set as the statistical average of the rewards, indicating that the Critic network is trained based on temporal difference (TD) learning. The TD error predicts the new state and evaluates the future quality. It is calculated by the state value function of two adjacent time slots and the current reward, expressed as:

[0075]

[0076] in, is the steady-state distribution under π, r(t) is the cumulative reward, Pr(s′(t),r(t)|s(t),a(t)) represents the probability of transitioning from state s(t) and action a(t) to state s′(t) and obtaining reward r(t);

[0077] Using TD error δ t Calculate the gradient of the objective function online:

[0078]

[0079] Among them, the Actor strategy network can be updated by the gradient ascent method, expressed as:

[0080]

[0081] Among them, α θ ∈(0,1) is the learning rate of the Actor strategy network;

[0082] Critic value network: Critic uses the state value function V(s(t),ω t ) is used to measure the return of state s(t). In each time slot t, when the agent strategy network makes the agent action a(t) based on the current state s(t), the evaluator value network obtains the reward r(t) and the next time slot state information s(t+1) from the environment. The evaluator can calculate the TD error of time slot t, which is expressed as:

[0083] δ t =r(t)+ηV(S(t+1);ω t )-V(S(t);ω t )

[0084] Among them, η∈(0,1) is the discount factor, and the least squares time difference method is used, that is, minimizing the loss (δ t ) 2 , update the critical value network:

[0085] ω(t+1)=ω(t)+α ω δ t ▽ θ V(S(t);ω t )

[0086] Among them, α ω ∈(0,1) is the learning rate of the evaluation value network.

[0087] In a second aspect, the present invention provides an edge computing system for offloading tasks of a new energy electric field, comprising:

[0088] A model building unit, used to build a new energy electric field mobile edge computing offloading model by combining the parameters of the device and the base station and the system communication bandwidth;

[0089] A priority division unit, used to divide the unloading priorities of the data collected by various sensor devices of the new energy electric field based on the mobile edge computing unloading model of the new energy electric field;

[0090] An optimization problem construction unit is used to consider the unloading delay of the new energy electric field system and the delay constraints of data task processing, construct a delay optimization problem of load balancing, use Lyapunov decoupling to solve the long-term energy consumption constraints of MEC in the delay optimization problem, and simplify the delay optimization problem into an optimization problem without long-term constraints by constructing a virtual energy deficit queue;

[0091] The joint optimization unit is used to use the AC algorithm to jointly optimize task offloading and resource allocation for optimization problems without long-term constraints, so as to maximize the service quality of the new energy electric field data acquisition system.

[0092] In a third aspect, the present invention provides a computing device, comprising:

[0093] Memory and processor;

[0094] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the method are implemented.

[0095] In a fourth aspect, the present invention provides a computer-readable storage medium storing computer-executable instructions, which implement the steps of the method when executed by a processor.

[0096] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention constructs a mobile edge computing unloading model for a new energy electric field by combining various parameters of the equipment and base station and the communication bandwidth of the system; in order to promote the construction of new energy electric fields and realize the automation and informatization of data collection and processing, it is necessary to timely process and analyze the data-intensive and delay-sensitive computing tasks in the Internet of Things of new energy electric fields, prioritize the data collected by various sensor devices in the new energy electric field according to the importance and real-time requirements of the data, and transmit the data in real time in sequence; under the condition of meeting the processing delay of the data task of the new energy electric field system, construct a maximum QoS optimization problem to ensure that each node minimizes the overall efficiency of the system while maintaining a good response time; adopt Lyapunov optimization to establish a virtual energy deficit queue to decouple the long-term energy constraint problem of MEC in the QoS optimization problem; adopt the AC algorithm to jointly optimize the task unloading and resource allocation for the QoS optimization problem without long-term constraints, so as to maximize the QoS and optimize the overall performance of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0097] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0098] Figure 1 This is an overall flow chart of an edge computing method for offloading new energy electric field tasks according to an embodiment of the present invention;

[0099] Figure 2 An edge computing model diagram of an edge computing method for offloading tasks of a new energy electric field according to an embodiment of the present invention;

[0100] Figure 3 A schematic diagram of the architecture of an AC algorithm of an edge computing method for offloading tasks in a new energy electric field according to an embodiment of the present invention;

[0101] Figure 4 A schematic diagram of the convergence performance of the AC algorithm of the edge computing method for offloading new energy electric field tasks according to an embodiment of the present invention at different learning rates;

[0102] Figure 5 This is a comparison diagram of average delays under different numbers of users and numbers of MECs for an edge computing method for offloading new energy electric field tasks according to an embodiment of the present invention;

[0103] Figure 6 A schematic diagram of average energy loss of MEC under different schemes of the edge computing method for offloading renewable energy electric field tasks described in an embodiment of the present invention. DETAILED DESCRIPTION

[0104] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0105] Example 1

[0106] Reference Figure 1 to Figure 3 As an embodiment of the present invention, an edge computing method for offloading tasks in a new energy electric field is provided. Figure 2As shown in Figure 1, the application includes N sensor devices and M base stations, which builds a timely unloading system for intensive data new energy electric field tasks, and ultimately achieves the goal of maximizing the QoS of each sensor device. In actual applications, such as Figure 1 As shown, an edge computing method for data-intensive new energy electric field task offloading specifically includes the following steps:

[0107] S100: Combine the parameters of the equipment and base station as well as the system communication bandwidth to build a new energy electric field mobile edge computing offloading model;

[0108] In the embodiment of the present application, N={1,2,...,n} is set to represent a set of sensor devices, M={1,2,...,m} represents a set of edge servers, each sensor device generates a task that needs to be calculated in each time slot, part of the task is executed locally on the sensor device, and part is offloaded to a nearby edge server, f n The local computing resources of SE n are limited, and the edge server has a certain computing resource f m and bandwidth resource B m , when SE n is within the communication coverage of the edge server, the edge server can offload services to SE n;

[0109] In the embodiment of the present application, in a new energy electric field with intensive tasks, the data types collected are divided into three types: high, medium, and low, k = {1, 2, 3};

[0110] It should be noted that high-level data are key data, including wind speed, wind direction, temperature, humidity and other data that affect the performance of wind turbines, which need to be processed in a timely manner; medium-level data are operating status data, such as wind turbine operating status, power output and other data, which need to be processed in real time, but can tolerate a slightly longer data processing delay; low-level data are environmental monitoring data, such as noise and vibration data in the surrounding environment, which have lower real-time requirements, but still need to be collected and processed regularly;

[0111] Assume that the user has an unloading task at time slot t in, A represents the computational resources required by SE n in time slot t, n (t) represents the amount of data generated by SE n in time slot t∈T, Represents task k t Maximum latency requirement;

[0112] The running time of the whole system is divided into multiple time slots T = {T1, T2, ..., T n}, the data transmission of all SE n adopts time division multiple access technology;

[0113] definition is the unloading ratio coefficient, where b n (t)∈[0,1], SEn performs the calculation of the task locally for a period of The period for unloading the remaining tasks to the edge server is Definition P n (t) = {p n (t),n∈N,t∈T}, where p n (t) is the transmission power from SE n to the edge server.

[0114] S200: Based on the mobile edge computing unloading model of the new energy electric field, the data collected by various sensor devices of the new energy electric field are divided into unloading priorities, and real-time data transmission is performed in sequence;

[0115] In the embodiment of the present application, data is unloaded in sequence according to the classification of data types. When SE n unloads part of the task to the nearby edge server, the signal-to-noise ratio from SE n to MEC m in time slot t is defined as:

[0116]

[0117] Among them, p n is the transmission power of SE n, is the channel gain from SE n to MEC m, σ 2 is the noise density;

[0118] The transmission rate of MEC m and SE n in time slot t is calculated as:

[0119]

[0120] Where W is the communication bandwidth;

[0121] The transmission delay of the offloaded task from SE n to MEC m at time slot t is calculated as:

[0122]

[0123] The transmission energy consumption is

[0124] In the embodiment of the present application, the offload task is processed simultaneously on the local device and the edge server. When the k-th type of data of SE n is calculated locally, the local calculation delay in time slot t is expressed as:

[0125]

[0126] The corresponding energy consumption is

[0127] When the k-th type of data of SE n is offloaded to MEC m, the MEC calculation delay in time slot t is:

[0128]

[0129] The corresponding energy consumption is The local energy consumption is

[0130] In the embodiment of the present application, enabling MEC task offloading is beneficial to improving the QoS of end users, but it will generate additional MEC computing energy consumption, and using edge servers to offload tasks will generate additional long-term MEC computing energy consumption:

[0131]

[0132] in, represents the energy budget of MEC in time slot t;

[0133] In the embodiment of the present application, since SE n may roam in multiple areas supported by different edge nodes, dynamic service migration should be considered to maintain effective task co-offloading. For example, SE n offloads its tasks to an MEC in time slot t through the direct cellular link between them. In the next time slot, SE n roams to the communication coverage provided by another edge node. Since the user is far away from the current edge server, the ongoing service should be offloaded to another appropriate edge server. At this time, the transmission network distance is greatly shortened, but additional migration overhead will be generated. The co-offloading decision that follows the movement of SE n is conducive to smaller transmission delays, but also generates additional migration overhead. Therefore, it is very important to study the impact of dynamic service migration on task collaborative offloading. Use γ to represent the cost of migrating a task from one edge server to another edge server. Then, MEC m is selected as the edge server, and the service migration cost of SE n can be expressed as:

[0134]

[0135] Where γ represents the cost of migrating a task from one edge server to another edge server. represents the decision result of service migration. When the edge server of time slot t is different from the edge server of time slot t-1, otherwise

[0136] S300: Considering the unloading delay of the new energy power plant system and the data task processing delay constraints, construct a load balancing delay optimization problem;

[0137] In the embodiment of the present application, local computation of data and task offloading are performed in parallel, and the execution time of the k-th type of data of SE n is

[0138] In order to reduce network distance and transmission delay, frequent service migration is inevitable. Therefore, the optimization target is determined by the task delay and migration cost of the device. The goal is to maximize the QoS of SE n, which is equivalent to minimizing the penalty factor Q n (t), define the following optimization objective, expressed as:

[0139]

[0140] Among them, Q n (t) represents the penalty factor of QoS of SE n in time slot t, involving task delay and migration cost, Q n The increase of (t) will lead to the decrease of device QoS. The weight parameter β represents the QoS preference factor of the user at time slot t. The larger β is, the more important the task delay is than the migration cost.

[0141] In the embodiment of the present application, the optimization variables of the system in time slot t are defined as P n (t)=[p1(t),p2(t),...,p n (t)],B n (t)=[b1(t),b2(t),...,b n (t)],F n (t)=[f1(t),f2(t),...,f n (t)], the optimization function is formally expressed as:

[0142]

[0143] stC1:p n (t)≤p max

[0144] C2:0≤b n (t)≤1

[0145]

[0146] Among them, C1 is the device transmission power constraint, C2 is the load ratio coefficient constraint, C3 is the device resource constraint, C4 is the device delay requirement, and C5 is the long-term energy consumption constraint.

[0147] S400: Lyapunov decouples the long-term energy consumption constraints of MEC in the delay optimization problem, and simplifies the delay optimization problem into an optimization problem without long-term constraints by building a virtual energy deficit queue.

[0148] In the embodiment of the present application, a dynamic virtual energy loss queue is designed, and the virtual energy loss queue of MEC m in time slot t is defined as In the initial time slot, the virtual energy loss queue is set to 0, that is, D n (0) = 0, the evolution of the virtual energy loss queue of a slot is expressed as:

[0149]

[0150] in It is the deviation between energy expenditure and energy budget;

[0151] In the embodiment of the present application, a Lyapunov function is defined to describe the sum of squares of all virtual energy loss queue backlogs in the tth time slot:

[0152]

[0153] Where Θ(t)={D1(t),D2(t),...,D n (t)}, Δ(Θ(t)) is introduced as the conditional Lyapunov drift to ensure the stability of the system:

[0154] Δ(Θ(t))=Ε[L(Θ(t+1))-L(Θ(t))|Θ(t)]

[0155] In the embodiment of the present application, Δ(Θ(t)) is controlled to reduce the queue backlog and stabilize the system. The drift penalty algorithm is used to obtain the Lyapunov drift penalty function Δ V (Θ(t)):

[0156] Δ V (Θ(t))=Δ(Θ(t))+VΕ[D n (t)]

[0157] Among them, V represents the non-negative weight coefficient; the upper bound of the Lyapunov drift penalty function is expressed as:

[0158]

[0159] in,

[0160] In the embodiment of the present application, the delay optimization problem is simplified to an optimization problem without long-term constraints by using Lyapunov queue optimization technology:

[0161]

[0162] stC1-C4

[0163] Among them, X t = {Pn (t),B n (t),F n (t} represents the task offloading and resource allocation strategy of all SE n at time slot t, represents the feasible task offloading and resource allocation strategy at time slot t, and the weight of the positive control parameter V is Q n (t) Achieve an ideal balance between user QoS and MEC energy consumption.

[0164] S500: The AC algorithm is used to jointly optimize task offloading and resource allocation for optimization problems without long-term constraints, so as to maximize the service quality of the new energy electric field data acquisition system. Figure 3 As shown;

[0165] In the embodiment of the present application, due to dynamic requests and network topology, it is difficult to predict the state of the future environment. Therefore, it is necessary to design an online decision-making mobile offloading solution to make the best decision without considering future environmental information. The goal of reinforcement learning is to obtain online decision-making capabilities through continuous interactive learning with the environment, and to make the best decision sequence to optimize the long-term cumulative rewards in the future. First, a discrete-time MDP is used to describe the dynamic MEC scenario, and then an Actor-Critic-based DRL algorithm is used to deal with optimization problems without long-term constraints. It is generally believed that the changes in available computing resources and the changes in the described wireless environment follow the Markov property. The three key elements of the learning environment system are the state set S, the action set A, and the reward function set R;

[0166] In the embodiment of the present application, the state of the MDP is the communication state of the available edge servers around the SE, and the energy consumption when the SE offloads tasks to the edge servers. At each decision time t, the state of the MEC system is expressed as:

[0167]

[0168] in, is the signal-to-noise ratio from SE n to MEC, is the transmission energy consumption offloaded from SE n to MEC;

[0169] The entire state space composed of the sensing devices is S(t) = [s(t), t∈T];

[0170] In the embodiment of the present application, according to the current system state s(t), the mobile unloading module performs an action a(t), which is expressed as:

[0171] a(t)={p n (t),b n (t),f n (t)}

[0172] The entire action space formed by the device is A(t) = [a(t), t∈T];

[0173] In the embodiment of the present application, according to the current system state s(t) and the selected action a(t), the reward obtained in the time slot tSE is expressed as:

[0174]

[0175] The entire reward space composed of devices is R(t) = [r(t), t∈T]. Since the reward function measures the unloading delay and migration cost consumption, obtaining a series of operations that maximize the cumulative reward can achieve the goal;

[0176] In the embodiment of the present application, the Actor-Critic algorithm is used as the framework of the DRL algorithm. The Actor-Critic algorithm is based on an idea similar to adversarial training of neural networks. The Actor neural network is trained to generate decision actions, and the Critic neural network is trained to estimate the quality of the current system state. Specifically, the Critic network is responsible for estimating the value of the state value function under the current state s(t) and updating its parameters according to the reward value r(t) given by the environment. The Actor network plays the role of a decision maker and is responsible for generating corresponding action decisions according to the current state s(t). The Actor network adjusts its parameters according to the judgment value given by the Critic network to increase or decrease the output probability of the selected action. In the Actor-Critic algorithm, the learning agent consists of two independent entities, the Actor and the Critic. The training process of these two entities will be discussed in detail below:

[0177] Actor policy network: from a random policy π(a(t)|s(t); θ t ), expressed as:

[0178] a(t)~π(a(t)|s(t);θ t )

[0179] Among them, π(a(t)|s(t); θ t )=Pr(s(t+1)|s(t),a(t)) indicates the random strategy π Mapping environmental states to probabilities of taking actions;

[0180] The measurement strategy J(θ) is set as the statistical average of the rewards, indicating that the Critic network is trained based on temporal difference (TD) learning. The TD error predicts the new state and evaluates the future quality. It is calculated by the state value function of two adjacent time slots and the current reward, expressed as:

[0181]

[0182] in, is the steady-state distribution under π, r(t) is the cumulative reward, Pr(s′(t),r(t)|s(t),a(t)) represents the probability of transitioning from state s(t) and action a(t) to state s′(t) and obtaining reward r(t);

[0183] Temporal difference TD error δ based on policy gradient method t Calculate the gradient of the objective function online:

[0184]

[0185] Among them, the Actor strategy network can be updated by the gradient ascent method, expressed as:

[0186]

[0187] Among them, α θ ∈(0,1) is the learning rate of the Actor strategy network;

[0188] Critic Value Network: The goal of the Critic agent is to estimate the expected cumulative reward value corresponding to the environment state s(t). The critic agent relies on a neural network to approximate the state value function V π (s(t)), that is, linear combination to fit the state value function corresponding to the environmental state;

[0189] Critic uses the state value function V(s(t),ω t ) is used to measure the return of state s(t). In each time slot t, when the agent strategy network makes the agent action a(t) based on the current state s(t), the evaluator value network obtains the reward r(t) and the next time slot state information s(t+1) from the environment. The evaluator can calculate the TD error of time slot t, which is expressed as:

[0190] δ t =r(t)+ηV(S(t+1);ω t )-V(S(t);ω t )

[0191] Among them, η∈(0,1) is the discount factor, and the least squares time difference method is used, that is, minimizing the loss (δ t ) 2 , update the critical value network:

[0192]

[0193] Among them, α ω ∈(0,1) is the learning rate of the evaluation value network.

[0194] Therefore, it can be seen from the above that the present invention constructs a mobile edge computing unloading model for a new energy electric field by combining various parameters of the equipment and base station and the communication bandwidth of the system; in order to promote the construction of new energy electric fields and realize the automation and informatization of data collection and processing, it is necessary to timely process and analyze the data-intensive and delay-sensitive computing tasks in the new energy electric field Internet of Things, and prioritize the data collected by various sensor devices in the new energy electric field according to the importance and real-time requirements of the data, and perform real-time data transmission in sequence; under the condition of meeting the data task processing delay of the new energy electric field system, construct a maximum QoS optimization problem to ensure that each node minimizes the overall efficiency of the system while maintaining a good response time; use Lyapunov optimization to establish a virtual energy deficit queue, decouple the long-term energy constraint problem of MEC in the QoS optimization problem; use the AC algorithm to perform joint optimization of task unloading and resource allocation for the QoS optimization problem without long-term constraints. To maximize QoS and optimize the overall performance of the system.

[0195] The above is a schematic scheme of an edge computing method for offloading tasks of a new energy electric field in this embodiment. It should be noted that the technical solution of the edge computing system for offloading tasks of a new energy electric field and the technical solution of the edge computing method for offloading tasks of a new energy electric field belong to the same concept. The details of the technical solution of the edge computing system for offloading tasks of a new energy electric field in this embodiment that are not described in detail can all be referred to the description of the technical solution of the edge computing method for offloading tasks of a new energy electric field.

[0196] This embodiment also provides an edge computing system for offloading new energy electric field tasks, including:

[0197] A model building unit, used to build a new energy electric field mobile edge computing offloading model by combining the parameters of the device and the base station and the system communication bandwidth;

[0198] A priority division unit, used to divide the unloading priority of the data collected by various sensor devices of the new energy electric field based on the mobile edge computing unloading model of the new energy electric field;

[0199] The optimization problem construction unit is used to consider the unloading delay of the new energy electric field system and the delay constraints of data task processing, construct the delay optimization problem of load balancing, use Lyapunov decoupling to consider the long-term energy consumption constraints of MEC in the delay optimization problem, and simplify the delay optimization problem into an optimization problem without long-term constraints by constructing a virtual energy deficit queue;

[0200] The joint optimization unit is used to use the AC algorithm to jointly optimize task offloading and resource allocation for optimization problems without long-term constraints, so as to maximize the service quality of the new energy electric field data acquisition system.

[0201] This embodiment also provides a computing device, including:

[0202] Memory and processor;

[0203] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the method proposed in the above embodiment.

[0204] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method proposed in the above embodiment is implemented.

[0205] The storage medium proposed in this embodiment and the method proposed in the above embodiment belong to the same inventive concept. The technical details not fully described in this embodiment can be referred to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0206] Through the above description of the implementation methods, the technicians in the relevant field can clearly understand that the present invention can be implemented by means of software and necessary general hardware, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ReadOnly, Memory, ROM), random access memory (RandomAccess Memory, RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform the methods of various embodiments of the present invention.

[0207] Example 2

[0208] Reference Figures 4 to 6 Based on the previous embodiment, this embodiment provides an application example of an edge computing method and system for offloading new energy electric field tasks, in order to verify and illustrate the technical effects adopted in this method.

[0209] Based on the simulation scenario, this embodiment obtained the following experimental data, such as Figure 4 As shown in the figure, the convergence effect of the AC algorithm applied to the task offloading of dense data renewable energy electric field is demonstrated, where M = 10, each round contains 200 offloading decision explorations, and the vertical axis represents the average reward obtained from the 200 explorations. A large reward means that the task processing delay and migration cost are relatively small, which means that better offloading services can be provided to users in an energy-saving way. Figure 4It can be seen that when the algorithm learning rate is 0.0001 and 0.001, as the training continues, the reward function tends to converge. This shows that the trained model is suitable for the scenario of the present invention and has good convergence performance.

[0210] like Figure 5 As shown in Figure 1, the average delay of the proposed scheme under different numbers of users N and edge servers M is compared. The average delay refers to the ratio of the total delay of the offloaded task to the number of tasks in the entire journey, which measures the average satisfaction of the user with the task offloading service throughout the journey. From a longitudinal perspective, when the number of users remains unchanged, the average delay of the proposed scheme decreases with the increase of the number of MECs. When the number of selectable edge servers is 3 (M=3), its average delay is about 1.2 times that of M=5, and even 1.5 times that of M=10. This is because the proposed DRL scheme makes the optimal decision considering the future state of the MEC system, which is not only the current optimal decision, but also the long-term optimal decision. As the number of users increases, the available resources of a single user will be limited, resulting in a higher task failure rate and making the average delay close to the task request delay. In addition, with the redundant computing resources of more edge servers, the average delay will intuitively decrease.

[0211] like Figure 6 The figure shows the average energy loss of MEC under different schemes. Scheme 1 does not consider the long-term MEC energy constraint scheme; Scheme 2 is a scheme without resource allocation, that is, the edge server only provides offloading services to SEs within its communication coverage, without considering the resource allocation of computing resources. When the energy budget remains unchanged, the time average energy loss of the proposed scheme is optimal. Figure 6 As shown, the proposed scheme can achieve optimal end-user QoS and energy deficit when the MEC energy budget is large.

[0212] Therefore, it can be seen from the above embodiments that the present invention constructs a mobile edge computing unloading model for a new energy electric field by combining various parameters of the equipment and base station and the communication bandwidth of the system; in order to promote the construction of new energy electric fields and realize the automation and informatization of data collection and processing, it is necessary to timely process and analyze the data-intensive and delay-sensitive computing tasks in the new energy electric field Internet of Things, and prioritize the data collected by various sensor devices in the new energy electric field according to the importance and real-time requirements of the data, and perform real-time data transmission in sequence; under the condition of meeting the data task processing delay of the new energy electric field system, construct a maximum QoS optimization problem to ensure that each node minimizes the overall efficiency of the system while maintaining a good response time; use Lyapunov optimization to establish a virtual energy deficit queue, decouple the long-term energy constraint problem of MEC in the QoS optimization problem; use the AC algorithm to perform joint optimization of task unloading and resource allocation for the QoS optimization problem without long-term constraints. To maximize QoS and optimize the overall performance of the system.

[0213] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An edge computing method for offloading new energy electric field tasks, characterized in that: include: Combining the parameters of equipment and base stations as well as the system communication bandwidth, a new energy electric field mobile edge computing offloading model is constructed; Based on the mobile edge computing unloading model of the new energy electric field, the data collected by various sensor devices in the new energy electric field are divided into unloading priorities, and real-time data transmission is performed in sequence; Considering the unloading delay of the new energy power plant system and the delay constraints of data task processing, a delay optimization problem of load balancing is constructed; Lyapunov decoupling is used to solve the long-term energy consumption constraint of MEC in the delay optimization problem, and a virtual energy deficit queue is constructed to simplify the delay optimization problem into an optimization problem without long-term constraints. The AC algorithm is used to jointly optimize task offloading and resource allocation for the optimization problem without long-term constraints, so as to maximize the service quality of the new energy electric field data acquisition system.

2. The edge computing method for offloading new energy electric field tasks according to claim 1, characterized in that: The construction of the new energy electric field mobile edge computing offloading model includes: Assume that N = {1, 2, ..., n} represents the set of sensor devices, M = {1, 2, ..., m} represents the set of edge servers, each sensor device generates a task that needs to be calculated in each time slot, part of which is executed locally on the sensor device and part is offloaded to the nearby edge server, f n The local computing resources of SE n are limited, and the edge server has a certain computing resource f m and bandwidth resource B m , when SE n is within the communication coverage of the edge server, the edge server can offload services to SE n; The collected data types are divided into three types: high, medium, and low. k = {1, 2, 3}. Assume that the user has an unloading task in time slot t. in, A represents the computational resources required by SE n in time slot t, n (t) represents the amount of data generated by SE n in time slot t∈T, Represents task k t Maximum latency requirement; The running time of the whole system is divided into multiple time slots T = {T1, T2, ..., T n }, all SE n data transmission adopts time division multiple access technology, defining is the unloading ratio coefficient, where b n (t)∈[0,1], SE n performs the calculation of the task locally for a period of The period for unloading the remaining tasks to the edge server is Define P n (t) = {p n (t),n∈N,t∈T}, where p n (t) is the transmission power from SE n to the edge server.

3. The edge computing method for offloading new energy electric field tasks according to claim 2, characterized in that: The unloading priorities of the data collected by various sensor devices in the new energy electric field include: Data is unloaded in sequence according to the division of data types. When SE n unloads part of the task to the nearby edge server, the signal-to-noise ratio from SE n to MECm in time slot t is defined as: Among them, p n is the transmission power of SE n, is the channel gain from SE n to MEC m, σ 2 is the noise density; The transmission rate of MECm and SE n in time slot t is calculated as: Where W is the communication bandwidth; The transmission delay of the offloaded task from SE n to MEC m in time slot t is calculated as: The transmission energy consumption is The offload task is processed simultaneously on the local device and the edge server. When the k-th type of data of SE n is calculated locally, the local calculation delay in time slot t is expressed as: When the k-th type of data of SE n is unloaded to MECm, the MEC calculation delay in time slot t is: Using edge servers to offload tasks generates additional long-term MEC computing energy consumption: in, represents the energy budget of MEC in time slot t; Through the direct cellular link between each sensor device, the task is offloaded to a MEC in time slot t. In the next time slot, SE n roams to the communication coverage of another edge node and selects MECm as the edge server. The service migration cost of SE n can be expressed as: Where γ represents the cost of migrating a task from one edge server to another edge server. represents the decision result of service migration. When the edge server of time slot t is different from the edge server of time slot t-1, otherwise 4. The edge computing method for offloading new energy electric field tasks according to claim 3, characterized in that: The delay optimization problem of building load balancing includes: The local computation of data and task offloading are performed in parallel, and the execution time of the k-th data of SE n is The optimization target is determined by the task delay and migration cost of the device, which can be expressed as: Among them, Q n (t) represents the penalty factor of QoS of SE n at time slot t, and the weight parameter β represents the QoS preference factor of the user at time slot t; Define the optimization variables of the system in time slot t as P n (t)=[p1(t),p2(t),...,p n (t)],B n (t)=[b1(t),b2(t),...,b n (t)],F n (t)=[f1(t),f2(t),...,f n (t)], the optimization function is formally expressed as: s.t.C1:p n (t)≤p max C2:0≤b n (t)≤1 C3:0≤f n ≤f n max C4:T n (t)≤T n max Among them, C1 is the device transmission power constraint, C2 is the load ratio coefficient constraint, C3 is the device resource constraint, C4 is the device delay requirement, and C5 is the long-term energy consumption constraint.

5. The edge computing method for offloading new energy electric field tasks according to claim 4, characterized in that: Simplifying the delay optimization problem into an optimization problem without long-term constraints includes: Design a dynamic virtual energy loss queue and define the virtual energy loss queue of MECm in time slot t as In the initial time slot, the virtual energy loss queue is set to 0, that is, D n (0) = 0, the evolution of the virtual energy loss queue of a slot is expressed as: in It is the deviation between energy expenditure and energy budget; Define the Lyapunov function to describe the sum of the squares of all virtual energy loss queue backlogs at the tth time slot: Where Θ(t)={D1(t),D2(t),...,D n (t)}, Δ(Θ(t)) is introduced as the conditional Lyapunov drift to ensure the stability of the system: Δ(Θ(t))=Ε[L(Θ(t+1))-L(Θ(t))|Θ(t)]. Control Δ(Θ(t)) to reduce queue backlog and stabilize the system. The drift penalty algorithm is used to obtain the Lyapunov drift penalty function Δ V (Θ(t)): D V (Θ(t))=Δ(Θ(t))+VE[D n (t)] Wherein, V represents a non-negative weight coefficient; the upper bound of the Lyapunov drift penalty function is expressed as: in, The delay optimization problem is simplified to an optimization problem without long-term constraints by using Lyapunov queue optimization technology: stC1-C4 Among them, X t = {P n (t),B n (t),F n (t} represents the task offloading and resource allocation strategy of all SE n at time slot t, represents the feasible task offloading and resource allocation strategy at time slot t, and the weight of the positive control parameter V is Q n (t) Achieve an ideal balance between user QoS and MEC energy consumption.

6. The edge computing method for offloading new energy electric field tasks according to claim 5, characterized in that: Joint optimization of task offloading and resource allocation for optimization problems without long-term constraints includes: The Actor-Critic based DRL algorithm is used to process the optimization problem without long-term constraints, following the changes of Markov property computing resources and the changes of the described wireless environment, wherein the three key elements of the learning environment system are the state set S, the action set A and the reward function set R; The state of MDP is the communication state of the available edge servers around SE and the energy consumption when SE offloads tasks to the edge servers. At each decision time t, the state of the MEC system is expressed as: in, is the signal-to-noise ratio from SE n to MEC, is the transmission energy consumption offloaded from SE n to MEC; According to the current system state s(t), the mobile unloading module takes action a(t), which is expressed as: a(t)={p n (t),b n (t),f n (t)} According to the current system state s(t) and the selected action a(t), the reward obtained by SE at time slot t is expressed as: The entire state space constituted by the sensing device is S(t) = [s(t), t∈T], the entire action space constituted by the device is A(t) = [a(t), t∈T], and the entire reward space constituted by the device is R(t) = [r(t), t∈T].

7. The edge computing method for offloading new energy electric field tasks according to claim 6, characterized in that: Also includes: The Critic network is responsible for estimating the value of the state value function under the current state s(t) and updating its parameters according to the reward value r(t) given by the environment. The Actor network plays the role of a decision maker and is responsible for generating corresponding action decisions based on the current state s(t). Actor policy network: from a random policy π(a(t)|s(t); θ t ), expressed as: a(t)~π(a(t)|s(t);θ t ) Among them, π(a(t)|s(t); θ t )=Pr(s(t+1)|s(t),a(t)) indicates the random strategy π Mapping environmental states to probabilities of taking actions; The measurement strategy J(θ) is set as the statistical average of the reward, indicating that the Critic network is trained based on temporal difference (TD) learning. The TD error predicts the new state and evaluates the future quality. It is calculated by the state value function of two adjacent time slots and the current reward, which is expressed as: in, is the steady-state distribution under π, r(t) is the cumulative reward, Pr(s′(t),r(t)|s(t),a(t)) represents the probability of transitioning from state s(t) and action a(t) to state s′(t) and obtaining reward r(t); Using TD error δ t Calculate the gradient of the objective function online: Among them, the Actor strategy network can be updated by the gradient ascent method, expressed as: Among them, α θ ∈(0,1) is the learning rate of the Actor strategy network; Critic value network: Critic uses the state value function V(s(t),ω t ) is used to measure the return of state s(t). In each time slot t, when the agent strategy network makes the agent action a(t) based on the current state s(t), the evaluator value network obtains the reward r(t) and the next time slot state information s(t+1) from the environment. The evaluator can calculate the TD error of time slot t, which is expressed as: δ t =r(t)+ηV(S(t+1);ω t )-V(S(t);ω t ) Among them, η∈(0,1) is the discount factor, and the least squares time difference method is used, that is, minimizing the loss (δ t ) 2 , update the critical value network: ω(t+1)=ω(t)+α ω d t ▽ θ V(S(t);ω t ) Among them, α ω ∈(0,1) is the learning rate of the evaluation value network.

8. A system using the edge computing method for offloading new energy electric field tasks as described in any one of claims 1 to 7, characterized in that: include: A model building unit, used to build a new energy electric field mobile edge computing offloading model by combining the parameters of the device and the base station and the system communication bandwidth; A priority division unit, used to divide the unloading priorities of the data collected by various sensor devices of the new energy electric field based on the mobile edge computing unloading model of the new energy electric field; An optimization problem construction unit is used to consider the unloading delay of the new energy electric field system and the delay constraints of data task processing, construct a delay optimization problem of load balancing, use Lyapunov decoupling to solve the long-term energy consumption constraints of MEC in the delay optimization problem, and simplify the delay optimization problem into an optimization problem without long-term constraints by constructing a virtual energy deficit queue; The joint optimization unit is used to use the AC algorithm to jointly optimize task offloading and resource allocation for optimization problems without long-term constraints, so as to maximize the service quality of the new energy electric field data acquisition system.

9. An electronic device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the method described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having computer program instructions stored thereon, wherein the computer-executable instructions, when executed by a processor, implement the steps of the method according to any one of claims 1 to 7.

Images