CN115033359A - Internet of things agent multi-task scheduling method and system based on time delay control - Google Patents

Abstract

The invention provides a time delay control-based Internet of things agent multi-task scheduling method and a time delay control-based Internet of things agent multi-task scheduling system, wherein the time delay control-based Internet of things agent multi-task scheduling method comprises the following steps: acquiring information of a plurality of power businesses reaching a resource pool of an Internet of things agent; inputting the information into a pre-constructed optimization problem model of time delay controlled multi-task deployment and resource allocation to solve to obtain a deployment mode of the power service and a resource allocation mode of an internet of things agent; scheduling the power business according to the deployment mode of the power business and the resource allocation mode of the Internet of things agent; the optimization problem model of the time delay control multitask deployment and resource allocation is constructed based on the aim of minimizing the total time delay of a plurality of power services; by solving the optimization problem model of the multi-task deployment and the resource allocation of the time delay control, the invention completes the cooperative deployment of a plurality of power tasks and realizes the high-efficiency allocation of the resources facing the time delay control.

Description

A multi-task scheduling method and system for IoT agent based on delay control

technical field

The invention belongs to the technical field of the Internet of Things in electric power, and in particular relates to a multi-task scheduling method and system of the Internet of Things agent based on time delay control.

Background technique

With the continuous deepening of the construction of the power Internet of Things, the scale of the power system continues to expand, the number of intelligent terminals increases rapidly, and the power business also presents a trend of diversity and real-time. With the increasingly widespread application of Internet of Things technology, new sensor technology and artificial intelligence technology, the operation of power grids is shifting to machine intelligence, perception intelligence and computational intelligence, resulting in massive amounts of heterogeneous data. The traditional cloud computing architecture cannot efficiently meet all the business needs of the smart grid. The edge computing model increases the ability of task execution and data caching and analysis on network devices, and migrates some or all computing tasks of the original cloud computing model to network edge devices, which can reduce the computing load of the cloud computing center and ease the network. Bandwidth pressure, improve data processing efficiency, and become a new power business data processing solution.

However, at this stage, there are still many problems in the power Internet of Things: 1. There are many types of power terminal equipment, complex and diverse physical interfaces, and the underlying connection protocols are very different, which brings long business application development cycles, difficult expansion, and dependence on terminal manufacturers. At the same time, it also restricts the interconnection and data sharing between different business systems; 2. The information collected by all business terminals of the power Internet of Things will be sent to the background business system for processing. This centralized processing mode is causing While the performance of some delay-sensitive applications is degraded, it will also bring huge processing pressure to the backbone communication network and background business systems, and the delay performance of real-time services cannot be guaranteed.

Invention No. CN112988285B "task offloading method and device, electronic device and storage medium" relates to the technical field of task offloading. The task offloading method is applied to an electronic device, the electronic device is connected in communication with a task offloading system, the task offloading system includes a second device and at least one first device, and the task offloading method includes: first, acquiring a task to be processed of at least one first device; secondly , input the task to be processed into a preset task offloading model to obtain a task offloading strategy; then, send the task offloading strategy to at least one first device, so that the at least one first device offloads the target task to the second device based on the task offloading strategy device, and the second device performs execution processing on the target task. Through the above method, the efficiency of task offloading can be improved.

Invention No. CN113553165A "A Game Theory-Based Mobile Edge Computing Task Offloading and Resource Scheduling Method". The invention discloses a mobile edge computing task offloading and resource scheduling method based on game theory. The method aims to jointly minimize the energy consumption of the mobile edge computing server and the user delay, and models the user task offloading and resource scheduling problem as For a specific optimization problem, a computing system based on multi-user task offloading with different task offload priorities is constructed, and a migration model is established by offloading multi-user tasks to an edge base station. In a specific optimization problem, based on game theory to solve the transmission rate and cost coefficient as constraints, the task offloading method is designed with the ultimate goal of minimizing server energy consumption. This method can effectively balance the interests of users and systems, and provide guarantees for implementing task offloading in mobile edge computing systems.

Invention No. CN113590232A, "A Relay Edge Network Task Offloading Method Based on Digital Twin" discloses a digital twin-based relay edge network task offloading method, including building a relay edge network task offloading strategy model; updating digital The state of the corresponding parts in the twin environment; the parameters of the digital twin are transferred to the simulated task offloading system for iterative training to obtain the optimal task offloading strategy model; the optimal task offloading strategy model is transferred to the simulated manual control interface for backup ;Transfer the current digital twin parameter training model and the optimal task offloading strategy model to the digital twin environment cache, and forward it to each relay node by the real edge server, and the relay node forwards it to the user terminal that communicates with it. ; User terminals and relay nodes perform corresponding task offloading according to the optimal task offloading strategy model. The present invention can reduce the trial and error cost of the actual 5G edge computing technology in the process of landing, and improve the landing efficiency.

Technical solution 1 firstly obtains the task to be processed of at least one first device; secondly, input the task to be processed into a preset task offloading model to obtain a task offloading strategy; then, send the task offloading strategy to at least one first device, so that the At least one first device offloads the target task to the second device based on the task offloading policy, and the second device executes the target task. Although this scheme improves the efficiency of task offloading, it only considers task offloading in a single-task collaboration scenario.

Technical solution 2 This method aims to jointly minimize the energy consumption of the mobile edge computing server and the user delay, and models the user task offloading and resource scheduling problem as a specific optimization problem. This solution does not consider the data routing between multiple tasks. Optimization.

Technical solution 3: The current digital twin parameter training model and the optimal task offloading strategy model are transmitted to the digital twin environment cache, and are forwarded by the real edge server to each relay node, and then the relay node is forwarded to the communication with it. User terminal; the user terminal and the relay node perform corresponding task offloading according to the optimal task offloading strategy model. Like the scheme based on scheme 1, this scheme also only considers tasks in a single-task collaborative scenario, and lacks universality for specific scenarios.

SUMMARY OF THE INVENTION

In order to overcome the above-mentioned deficiencies of the prior art, the present invention proposes a multi-task scheduling method for IoT agents based on delay control, including:

Obtain the information that multiple power services arrive at the resource pool of the IoT agent;

Inputting the information into a pre-built time-delay controlled multi-task deployment and resource allocation optimization problem model to solve, to obtain the deployment mode of the power service and the resource allocation mode of the IoT agent;

According to the deployment method of the power service and the resource allocation method of the IoT agent, the power service is scheduled;

Wherein, the multi-task deployment and resource allocation optimization problem model of delay control is constructed based on the goal of minimizing the total delay of multiple power services.

Preferably, the construction of the optimization problem model of the delay-controlled multi-task deployment and resource allocation includes:

The objective function is constructed with the goal of minimizing the total delay in the deployment of multiple power services in the resource pool of each IoT agent and the resource allocation mode;

Build constraints based on resource constraints of IoT agents;

Build an optimization problem model of delay-controlled multi-task deployment and resource allocation based on the objective function and constraints;

Wherein, the resources include computing resources and bandwidth resources; the information includes at least one or more of the following: the sequence of containers in the resource pool required by the power service request, the priority of the power service, and the data of the power service The time interval distribution of packets leaving the resource pool corresponds to the parameters of the Poisson process, the processing speed of the unit computing resource of the resource pool to the power service, and the transmission speed of the unit bandwidth resource of the resource pool to the power service.

Preferably, the calculation formula of the objective function is as follows:

In the formula, F represents the total delay of the deployment and resource allocation of multiple power services in the resource pool of each IoT agent, K _S represents the number of power services in the same resource pool in the same time period, ω _s represents the power service The priority of s, D _s represents the total delay of the power service s, and S represents the set of all power services;

The calculation formula of the total delay D _s of the power service s is as follows:

表示电力业务s在离开资源池z后的链路时延；In the formula, |N _s | represents the number of containers in the resource pool that process the power service s, D _s,z represents the average delay of the power service s at the resource pool z,

represents the link delay of the power service s after leaving the resource pool z;

The calculation formula of the average delay D _s,z of the power service s at the resource pool z is as follows:

In the formula, c _s,1 represents the processing speed of the power service s after resource allocation, c _s,2 represents the transmission speed of the power service s after resource allocation, and P _s,z represents the power service s in the resource pool. The non-empty probability in the processing queue of z, λ _s represents the parameters of the Poisson process corresponding to the time interval distribution of the data packets of the power service s leaving the resource pool;

的计算式如下：The link delay of the power service s after leaving the resource pool z

The calculation formula is as follows:

In the formula, n _z represents the size of the data to be transmitted in the resource pool z.

Preferably, the calculation formula of the resource constraint is as follows:

表示电力业务在资源池中可获得的资源量集合，

表示电力业务在资源池z中的容器使用所有计算资源时的处理速度，

表示电力业务在资源池z中的容器使用所有带宽资源时的传输速度。In the formula, _KB represents the total number of power services deployed on the resource pool, κ _z,i represents the proportion of computing resources allocated by resource pool z to power service i, and ν _z,i represents the bandwidth resources allocated by resource pool z to power service i , σ represents the maximum amount of resources available to the power business in the resource pool,

Represents the set of resources available for the power business in the resource pool,

represents the processing speed of the power business when the containers in the resource pool z use all computing resources,

Indicates the transmission speed of the power service when all the bandwidth resources are used by the containers in the resource pool z.

Preferably, the information is input into a pre-built time-delay controlled multi-task deployment and resource allocation optimization problem model for solving, and the deployment mode of the power service and the resource allocation mode of the IoT agent are obtained, including:

The information is input into a pre-built optimization problem model of delay-controlled multi-task deployment and resource allocation, and an improved ant colony algorithm based on dynamic pheromone volatility coefficient is used to optimize the delay-controlled multi-task deployment and resource allocation. The optimization problem model is solved to obtain the deployment mode of each container in the resource pool of the IoT agent for multiple power tasks;

Based on the deployment methods of multiple power tasks in the resource pool of the IoT agent, the near-end strategy optimization algorithm is used to solve the optimization problem model of the multi-task deployment and resource allocation for delay control, and the IoT agent is obtained. The resource allocation method of the resource pool to each container.

Preferably, the improved ant colony algorithm based on the dynamic pheromone volatility coefficient is used to solve the optimization problem model of the multi-task deployment and resource allocation of delay control, so as to obtain the resource pool of the IoT agent for multiple power tasks The deployment method of each container in , including:

Use the computing resources in the resource pool of the IoT agent to initialize the pheromone of the ant colony algorithm;

Calculate the disappearance of the pheromone trajectory of each resource pool based on the dynamic pheromone volatilization coefficient;

Calculate the probability of selecting each resource pool for task deployment based on the disappearance of the pheromone trajectory, and select the next resource pool to deploy the power task based on the probability;

Based on the disappearance of the pheromone trajectory and the selected resource pool for the next deployment power task, the ant colony is used for iterative calculation, and the most frequent planning in the ant colony path planning is obtained as multiple power tasks in each container in the resource pool of the IoT agent deployment method.

Preferably, the calculation formula of the pheromone of the initialization ant colony algorithm is as follows:

In the formula, τ _0,j represents the initialized pheromone corresponding to resource pool j, r' _m,j represents the available memory of resource pool j, subscript m represents memory, r _m,j represents the total memory of resource pool j, ψ _m represents the total memory of all resource pools, r' _p,j represents the available CPU of resource pool j, subscript p represents CPU, r _p,j represents the total CPU of resource pool j, ψ _p represents the total CPU of all resource pools.

Preferably, the calculation formula for the disappearance of the pheromone track is as follows:

In the formula, τ(t,j) represents the disappearance of the pheromone trajectory corresponding to resource pool j at time t, τ _0,j represents the initialized pheromone corresponding to resource pool j, Δ represents the change of pheromone, and P(k) represents the first pheromone. The resource pool that can be selected in k iterations, ρ represents the dynamic pheromone volatility coefficient;

The calculation formula of the dynamic pheromone volatilization coefficient ρ is as follows:

Based on the same inventive concept, the present application also provides an IoT agent multi-task scheduling system based on delay control, including: a data acquisition module, a solution module and a scheduling module;

The data acquisition module is used to acquire the information that multiple power services reach the resource pool of the IoT agent;

The solving module is used for inputting the information into a pre-built time-delay controlled multi-task deployment and resource allocation optimization problem model for solving, to obtain the deployment mode of the power service and the resource allocation mode of the IoT agent;

The scheduling module is used to schedule the power service according to the deployment mode of the power service and the resource allocation mode of the IoT agent;

Wherein, the multi-task deployment and resource allocation optimization problem model of delay control is constructed based on the goal of minimizing the total delay of multiple power services.

Preferably, the solving module is further configured to construct the objective function with the goal of minimizing the total delay in the deployment of multiple power services in the resource pool of each IoT agent and the resource allocation mode;

Build constraints based on resource constraints of IoT agents;

Build an optimization problem model of delay-controlled multi-task deployment and resource allocation based on the objective function and constraints;

Wherein, the resources include computing resources and bandwidth resources; the information includes at least one or more of the following: the sequence of containers in the resource pool required by the power service request, the priority of the power service, and the data of the power service The time interval distribution of packets leaving the resource pool corresponds to the parameters of the Poisson process, the processing speed of the unit computing resource of the resource pool to the power service, and the transmission speed of the unit bandwidth resource of the resource pool to the power service.

Preferably, the calculation formula of the objective function constructed by the solving module is as follows:

In the formula, F represents the total delay of the deployment and resource allocation of multiple power services in the resource pool of each IoT agent, K _S represents the number of power services in the same resource pool in the same time period, ω _s represents the power service The priority of s, D _s represents the total delay of the power service s, and S represents the set of all power services;

The calculation formula of the total delay D _s of the power service s is as follows:

表示电力业务s在离开资源池z后的链路时延；In the formula, |N _s | represents the number of containers in the resource pool that process the power service s, D _s,z represents the average delay of the power service s at the resource pool z,

represents the link delay of the power service s after leaving the resource pool z;

The calculation formula of the average delay D _s,z of the power service s at the resource pool z is as follows:

In the formula, c _s,1 represents the processing speed of the power service s after resource allocation, c _s,2 represents the transmission speed of the power service s after resource allocation, and P _s,z represents the power service s in the resource pool. The non-empty probability in the processing queue of z, λ _s represents the parameters of the Poisson process corresponding to the time interval distribution of the data packets of the power service s leaving the resource pool;

的计算式如下：The link delay of the power service s after leaving the resource pool z

The calculation formula is as follows:

In the formula, n _z represents the size of the data to be transmitted in the resource pool z.

Preferably, the calculation formula of the resource constraint constructed by the solving module is as follows:

表示电力业务在资源池中可获得的资源量集合，

表示电力业务在资源池z中的容器使用所有计算资源时的处理速度，

表示电力业务在资源池z中的容器使用所有带宽资源时的传输速度。In the formula, _KB represents the total number of power services deployed on the resource pool, κ _z,i represents the proportion of computing resources allocated by resource pool z to power service i, and ν _z,i represents the bandwidth resources allocated by resource pool z to power service i , σ represents the maximum amount of resources available to the power business in the resource pool,

Represents the set of resources available for the power business in the resource pool,

represents the processing speed of the power business when the containers in the resource pool z use all computing resources,

Indicates the transmission speed of the power service when all the bandwidth resources are used by the containers in the resource pool z.

Preferably, the solving module is specifically used for:

The information is input into a pre-built optimization problem model of delay-controlled multi-task deployment and resource allocation, and an improved ant colony algorithm based on dynamic pheromone volatility coefficient is used to optimize the delay-controlled multi-task deployment and resource allocation. The optimization problem model is solved to obtain the deployment mode of each container in the resource pool of the IoT agent for multiple power tasks;

Based on the deployment methods of multiple power tasks in the resource pool of the IoT agent, the near-end strategy optimization algorithm is used to solve the optimization problem model of the multi-task deployment and resource allocation for delay control, and the IoT agent is obtained. The resource allocation method of the resource pool to each container.

Preferably, the solving module uses an improved ant colony algorithm based on dynamic pheromone volatilization coefficient to solve the optimization problem model of the multi-task deployment and resource allocation of time delay control, and obtains the results of multiple power tasks in the IoT agent. The deployment method of each container in the resource pool, including:

Use the computing resources in the resource pool of the IoT agent to initialize the pheromone of the ant colony algorithm;

Calculate the disappearance of the pheromone trajectory of each resource pool based on the dynamic pheromone volatilization coefficient;

Calculate the probability of selecting each resource pool for task deployment based on the disappearance of the pheromone trajectory, and select the next resource pool to deploy the power task based on the probability;

Based on the disappearance of the pheromone trajectory and the selected resource pool for the next deployment power task, the ant colony is used for iterative calculation, and the most frequent planning in the ant colony path planning is obtained as multiple power tasks in each container in the resource pool of the IoT agent deployment method.

Preferably, the calculation formula of the pheromone of the ant colony algorithm initialized by the solving module is as follows:

In the formula, τ _0,j represents the initialized pheromone corresponding to resource pool j, r' _m,j represents the available memory of resource pool j, subscript m represents memory, r _m,j represents the total memory of resource pool j, ψ _m represents the total memory of all resource pools, r' _p,j represents the available CPU of resource pool j, subscript p represents CPU, r _p,j represents the total CPU of resource pool j, ψ _p represents the total CPU of all resource pools.

Preferably, the calculation formula for calculating the disappearance of the pheromone trajectory by the solving module is as follows:

In the formula, τ(t,j) represents the disappearance of the pheromone trajectory corresponding to resource pool j at time t, τ _0,j represents the initialized pheromone corresponding to resource pool j, Δ represents the change of pheromone, and P(k) represents the first pheromone. The resource pool that can be selected in k iterations, ρ represents the dynamic pheromone volatility coefficient;

The calculation formula of the dynamic pheromone volatilization coefficient ρ is as follows:

The present invention also provides a computer device, comprising: one or more processors;

memory for storing one or more programs;

When the one or more programs are executed by the one or more processors, the aforementioned multitask scheduling method for IoT agents based on delay control is implemented.

The present invention also provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed, implements the aforementioned method for scheduling multiple tasks of an IoT agent based on delay control. Compared with the closest prior art, the present invention has the following beneficial effects:

The present invention provides a method and system for multi-task scheduling of IoT agents based on delay control, including: acquiring information about the arrival of multiple power services to the resource pool of IoT agents; inputting the information into a pre-built delay-controlled Solve the optimization problem model of multi-task deployment and resource allocation, and obtain the deployment method of the power service and the resource allocation method of the IoT agent; according to the deployment method of the power service and the resource allocation method of the IoT agent, the power service is scheduled; Wherein, the optimization problem model of delay-controlled multi-task deployment and resource allocation is constructed based on the goal of minimizing the total delay of multiple power services; by solving the optimization problem model of delay-controlled multi-task deployment and resource allocation , the present invention completes the coordinated deployment of multiple power tasks and realizes efficient allocation of resources oriented to time delay control.

Description of drawings

1 is a schematic flowchart of a method for scheduling multiple tasks of IoT agents based on delay control provided by the present invention;

Fig. 2 is the design flow schematic diagram of the multi-task deployment and resource allocation algorithm provided by the present invention;

3 is a schematic structural diagram of a multi-task scheduling system for IoT agents based on delay control provided by the present invention;

FIG. 4 is a schematic structural diagram of an example of a power IoT agent scheduling system provided by the present invention.

Detailed ways

The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

Example 1:

A schematic flowchart of a multi-task scheduling method for IoT agents based on delay control provided by the present invention is shown in FIG. 1 , including:

Step 1: Obtain the information that multiple power services reach the resource pool of the IoT agent;

Step 2: Input the information into the pre-built time-delay controlled multi-task deployment and resource allocation optimization problem model to solve, and obtain the deployment mode of the power service and the resource allocation mode of the IoT agent;

Step 3: according to the deployment mode of the power service and the resource allocation mode of the IoT agent, schedule the power service;

Among them, the multi-task deployment of delay control and the optimization problem model of resource allocation are constructed based on the goal of minimizing the total delay of multiple power services.

The invention utilizes the high abstraction of the lightweight container, and can realize the high reuse of hardware and software. As a limited resource, the IoT agent can play various roles in time-sharing and realize different functions, such as communication, computing, etc. The virtual switch software used for communication or a virtual network function can be placed in the remote warehouse as subtasks, and each edge IoT agent can download and run the subtasks to support edge power services. In order to conveniently describe the resource allocation process of the container, the present invention firstly constructs an edge node resource allocation model, and based on the model, establishes a delay-controlled multi-task deployment and resource allocation optimization problem model. For this model, the present invention proposes A MTDRA (Multi-task Deployment and Resource Allocation) algorithm is proposed. The design process of the algorithm is shown in Figure 2. Specifically, it includes: building delay-controlled multi-task deployment and resource allocation. The optimization problem model; the solution of the optimization problem model is decomposed into two sub-problems to be solved sequentially; the improved ant colony algorithm is designed based on the dynamic pheromone volatility coefficient to solve the multi-task deployment results; end strategy optimization) to design a resource allocation algorithm to allocate computing resources and communication resources (ie, bandwidth resources) of IoT agents.

First, the model used in the present invention will be described.

I-Power Edge Network Model.

The present invention divides the edge IoT agent resources into several public resource pools. The present invention uses z to represent the resource pool, and K _Z to represent the number of resource pools. Since the resource pools are connected by different data links, there are different uplink and downlink bandwidths, and C _z and B _z are respectively used to represent the amount of computing and bandwidth resources owned by the resource pool z. In the actual production situation, the combination of service requests in different time periods will change. When the power service (referred to as service) request s arrives in the dispatching system, K _S is used to represent the number of service requests in the same resource pool in the same time period.

The present invention assumes that requests arrive at the resource pool randomly in different time intervals, and the present invention uses Ω _s to represent the information that the service s arrives in the resource pool, where Ω _s =(N _s ,λ _s ,C _s,1 ,C _s,2 ,ω _s ), at this stage, the present invention regards the arrival process of service request flow data packets as a Poisson process with parameter λ _s , where N _s represents the sequence of containers required by the service request, |N _s | represents the number of containers, ω _s represents the priority of the power service s, C _s,1 represents the processing speed of the unit computing resource in the resource pool to the service s data packet, C _s,2 represents the transmission speed of the unit bandwidth resource in the resource pool to the s data, where

代表业务s在资源池z中的容器使用所有资源所得到的处理速度，使用

代表s使用资源池z中带宽资源得到的数据包传输速度。为了找到资源池资源分配的近似最优解，通过离散资源分配策略对资源池计算及其带宽资源进行分配，分配策略π(s,a)＝{κ_z,ν_z}，其中：Use of the present invention

Represents the processing speed obtained by using all the resources of the container of the business s in the resource pool z, using

Represents the data packet transmission speed obtained by s using the bandwidth resources in resource pool z. In order to find the approximate optimal solution of resource pool resource allocation, the resource pool calculation and its bandwidth resources are allocated through a discrete resource allocation strategy, the allocation strategy π(s,a)={κ _z ,ν _z }, where:

是一个有限的离散集合，表示电力业务在资源池中可获得的资源量集合，具体来说，将资源分配划分为10个离散的资源块，这时

这使得在定义资源分配动作时大大减少动作空间的大小，降低算法收敛难度。σ代表业务请求s在资源池可以得到的最大资源量，这样做的目的使避免某一请求占用过多资源影响其他业务的运行。根据资源分配策略可得到当前业务请求的处理和传输速率。In the above formula, K _S represents the number of service requests in the same resource pool in the same time period, _KB represents the total number of power services deployed on the resource pool, κ _z , ν _z represent the resource allocation of resource pool z to the service request chain, respectively, κ _z,i represents the ratio of computing resources allocated by resource pool z to power service i, and ν _z,i represents the ratio of bandwidth resources allocated by resource pool z to power service i. in

is a limited discrete set, which represents the set of resources available for power services in the resource pool. Specifically, the resource allocation is divided into 10 discrete resource blocks. At this time,

This greatly reduces the size of the action space and reduces the difficulty of algorithm convergence when defining resource allocation actions. σ represents the maximum amount of resources available to the service request s in the resource pool. The purpose of this is to prevent a request from occupying too many resources and affecting the operation of other services. According to the resource allocation strategy, the processing and transmission rate of the current service request can be obtained.

表示进行资源分配后资源池z中的容器针对业务s的处理速度，

表示进行资源分配后资源池z中的容器针对业务s的传输速度。In the above formula,

Indicates the processing speed of the container in the resource pool z for the business s after resource allocation,

Indicates the transmission speed of the container in the resource pool z for the service s after resource allocation.

In the same service request, the data packet processing speed and transmission speed in the same resource pool should satisfy c _s,1 =c _s,2 as much as possible. Only in this way, the data packet can be processed in the resource pool without queuing time. It is transmitted through the data link. In resource allocation, both the maximization of resource allocation and the balance of resource allocation should be considered to avoid unnecessary waste of resources. Therefore, conditions must be met when allocating resources in resource pools.

II-Delay model, which is an optimization problem model for multi-task deployment and resource allocation of delay control.

In order to accurately describe the end-to-end delay in the multi-task collaboration scenario, the present invention first establishes a delay model based on the serial queuing model, which improves the accuracy of network delay evaluation.

The present invention will separately discuss the process that the data packet of the service request passes through the container 1 and passes through the subsequent containers. The present invention first analyzes the leaving process of the first container, and in this process, the present invention focuses on the time interval when two adjacent data packets in the same service leave. When the arrival rate of service request flow data packets is much smaller than the processing rate of the resource pool, the time interval of the data packets leaving the resource pool conforms to the Poisson process with parameter λ _s , and when the arrival rate is equal to the processing rate of the resource pool, the data packets leave the resource pool. The time interval of the pool is approximated by a deterministic process with parameter c _s,1 . The present invention obtains the complete container data packet processing and serial queuing of data packet transmission by analyzing the processing delay, transmission delay, and queuing backlog delay caused by different arrival rates and processing rates in the container 1. In the model, the average delay of data packets at the first container of service request flow S is as follows:

In the formula, c _s,1 represents the processing speed of the power service s after resource allocation, c _s,2 represents the transmission speed of the power service s after resource allocation, and P _s,1 represents the power service s in the container 1 The probability of not being empty in the processing queue.

Based on the above analysis, the present invention continues to analyze the link transmission delay after data leaves container 1. If container 1 and container 2 are in the same resource pool, the data link transmission delay does not need to be considered. If container 2 is in another resource pool , then the transmission delay is mainly determined by the switches and the number of links passed by the link. Define K _n as the number of switches passed through, and the number of links is also K _n , then the link transmission delay is:

In the formula, n ₁ represents the size of the data to be transmitted at container 1.

Next, the present invention studies the sum of the average delay of data packets in the container after passing through the container 1, because the arrival process of the data packet in the container 2 is affected by the processing rate and the transmission rate of the container 1. It is difficult to analyze the time interval Z _i during which the data packets of the traffic flow s continuously arrive at the container 2 . In order to eliminate the influence of the processing and transmission rate of each stream data packet at container 1 on the process of reaching container 2, the present invention approximates the process of data flow S of a service reaching container 2 as a Poisson process with arrival rate λ _s , and build the same M/D/1 queuing model. According to the obtained data packet arrival process, the average delay sum of data packets at container 2 is as follows:

In the formula, c _s,1 represents the processing speed of the power service s after resource allocation, and c _s,2 represents the transmission speed of the power service s after resource allocation, when c _s,1 <c _s,2 , due to the slow processing rate of data packets in container 2, resulting in a backlog of queue delay at container 2, when c _s,1 ≥c _s,2 , the data packets pass through container 2 at the processing rate of c _s,1 , and leaves the edge computing node at a transmission rate of c _s,2 . Therefore, the total delay can be calculated separately from the resource pool and data link through which the service request data flow passes:

表示资源池z后的数据链路传输时延，D_s,z和

的计算式如下：In the above formula, D _s,z represents the processing rate and transmission rate of the resource pool z (that is, the average delay of the power service s at the resource pool z),

represents the data link transmission delay after resource pool z, D _{s, z} and

The calculation formula is as follows:

In the formula, n _z represents the size of the data to be transmitted in the resource pool z.

To sum up, the purpose of the present invention is to optimize the end-to-end delay of the container cluster service request flow data packets. The optimization problem model established by the present invention is shown in the following formula:

III-Delay Controlled Multitask Deployment and Algorithm for Dynamic Allocation of CPU Computing Resources.

Aiming at the complex environment and different services of the power Internet of things, the present invention firstly proposes a multi-task deployment algorithm of delay control based on the improved ant colony algorithm.

A business consists of one or more tasks, and the user will start the business by submitting some tasks to the scheduling module. The present invention designs and improves the ant colony algorithm (Improved Ant Colony Optimization, IACO), and selects a group of nodes that meet the specified constraints through the scheduling module, and deploys tasks to these nodes (ie, containers in the resource pool).

The goal of the scheduling module is to place tasks on available resources. Available resources are used at each scheduling time. An artificial ant randomly searches for resources by observing the pheromone trajectory of each resource. The calculation formula of each computing resource of a node is:

其中，n为待分配内存的任务个数，ψ_i表示分配内存的比率。In the above formula: R(j) is the resource of node j, r' _m,j is the available memory of node j, r _m,j is the total memory of the node, rp _,j ' is the available CPU of the node, rp _{, j} is the total CPU of the node, ψ _m is the memory size of all nodes, ψ _p is the CPU size of all nodes,

Among them, n is the number of tasks to be allocated memory, and ψ _i represents the ratio of allocated memory.

To initialize the pheromone trajectory of each node, R(j) is used in the cyclic greedy algorithm, which simply places each task on a node in the cyclic pattern, τ _0,j =R(j) for each node the starting pheromone. The calculation formula for the probability of selecting the current node for task deployment is:

In the above formula, η _e is the heuristic value of node e, α is the information heuristic factor, β is the expected heuristic factor, and m is the total number of nodes.

Calculate the deployment probability p(t,j) of each node, and select the next node j.

where j∈P(k), q ₀ is the preset exploration rate threshold, and q is the exploration rate of the ant colony algorithm. P(k) represents the resource nodes that can choose to deploy tasks at the k-th iteration. N is the set of all nodes.

The next step is to calculate the disappearance of the pheromone. The formula for calculating the disappearance of the pheromone trajectory of each node is:

In the formula, τ(t,j) represents the disappearance of the pheromone trajectory corresponding to resource pool j at time t, τ _0,j represents the initialized pheromone corresponding to resource pool j, Δ represents the change of pheromone, and ρ represents the volatilization of dynamic pheromone coefficient. When a task is placed on a specific resource by the scheduler, the value of Δ is always less than 0. Pheromone is calculated using the above formula, in order to significantly reduce the pheromone level of the selected node, making it less important for the next task, so that the remaining tasks are allocated to the whole resource.

The value of the pheromone volatility coefficient in the ant colony algorithm plays a key role in the convergence of the algorithm and the performance of the search. The value of ρ in the traditional ant colony algorithm is generally a fixed value between 0 and 1. Because the pheromone concentration on the path is relatively small at the beginning of the iteration, as the number of iterations increases, even if the pheromone on the path is volatilized every time, the pheromone concentration will continue to increase, so if a fixed information volatilization coefficient is used It will lead to excessive volatilization of pheromone in the early stage and less volatilization of pheromone in the later stage, which will seriously restrict the performance of the ant colony optimization algorithm. The function is monotonically increasing, and it maps the value of ρ to (0,1), which will weaken the volatilization of the path pheromone concentration in the early stage and increase the volatilization of the pheromone concentration in the later stage. The value formula of ρ becomes:

In the above formula, k represents the number of iterations of the algorithm.

The best plan uses the following formula to select the most frequent plan.

In the formula, x is the value when p(k) reaches the maximum value, and this value is assigned to p _w .

By improving the traditional ant colony algorithm, the invention solves the poor performance caused by the traditional ant colony algorithm using a fixed information volatilization coefficient, and completes the coordinated deployment of multiple tasks.

Secondly, after completing the deployment of multi-task services on the container, the present invention designs a resource allocation algorithm oriented to delay control to guide the dynamic allocation of CPU computing resources and bandwidth resources in the IoT agent, thereby ensuring the service delay. demand.

Since the environment of edge computing network is complex and changeable, in order to learn in this challenging environment, it is necessary to use a reliable and highly scalable learning algorithm, since the PPO algorithm works by binding the range of parameter updates to the trust region. To ensure stability, the present invention considers using this algorithm to complete resource allocation. In order to be able to use the deep reinforcement learning algorithm to solve the problem obtained in the previous section, the present invention first converts the problem into an MDP (Markov decision processes, Markov decision) process. It contains four elements: agent, input, action, reward.

Agent: In the present invention, the agent is the edge IoT agent.

式中，以

为例，

表示资源池z在t时刻针对任务1的资源分配，

为所有资源池的资源分配策略。Input: After the environmental information changes, the system state is expressed as x _t =[N _t,z ,A _t ], the current system state is input into the model, where N _z,t represents the container flow group of resource pool z

In the formula, with

For example,

represents the resource allocation of resource pool z for task 1 at time t,

Resource allocation policy for all resource pools.

Action: At time t, resource pool z can only use one resource allocation strategy a _t,z =[κ _t,z ,ν _t,z ], where

Reward function: When the agent chooses a certain action, the corresponding system state will change and the reward value of the strategy will be obtained. In the present invention, the end-to-end delay of data packets of the container service request flow in the current state and the next state is used. difference as a reward. As shown in the following formula:

In the above formula, S and S' represent the power services processed by the container groups of the resource pool services in the current state and the next state, respectively. When the end-to-end delay obtained by the next state is lower, a positive reward value is obtained, otherwise, a negative reward value is obtained.

The PPO algorithm is a deep reinforcement learning algorithm based on the actor-critic algorithm framework. The PPO architecture designed by the present invention includes two Actor networks, Actor1 and Actor2. Actor1 represents the current up-to-date policy π and interacts with the edge network environment, which selects tasks to deploy actions based on the current environment state. The critic evaluates the current policy based on the reward obtained after performing the deployment action, and updates the parameters in the critic network through back-propagation of the loss function. Actor2 represents the old strategy π _old . Every time the agent is trained for a period of time, it uses the parameters in Actor1 to update Actor2, and the above process is repeated until the PPO algorithm converges. At this time, a trained edge multitasking based on the AC framework is obtained. Deploy the model.

The PPO algorithm needs to limit the similarity value r _t (θ) between the old and new strategies.

L ^CLIP (θ)=E _t [min(r _t (θ)B _t ,clip(r _t (θ)),1-ε,1+ε)B _t ]

In the above formula, ε∈[0,1] is a hyperparameter, and clip() constrains the value of r _t (θ) within the interval [1-ε,1+ε].

表示新旧策略的相似度，上式中，π_θ(a|s)为使用策略π在电力业务s对应状态下采取动作a的概率，在代入计算时，将r_t(θ)带入上式。π_θ(a|s)无具体的计算公式，具体结果由神经网络输出；B_t为t时刻的优势函数，E_t为t时刻的期望值。

Represents the similarity between the old and new strategies. In the above formula, π _θ(a|s) is the probability of using the strategy π to take action a in the corresponding state of the power business s. When substituting it into the calculation, bring r _t (θ) into the above formula . There is no specific calculation formula for π _θ(a|s) , and the specific results are output by the neural network; B _t is the advantage function at time t, and E _t is the expected value at time t.

The schedulable resources in the edge network include the available computing resources and available bandwidth resources of the edge IoT agent. The algorithm generates an instant reward for the agent that takes a certain action through the reward function, and the algorithm continuously learns according to the reward obtained to obtain the optimal resource allocation. Strategy.

The actor network of the PPO algorithm consists of two neural networks Actor1 and Actor2. Actor1 guides the agent to interact with the environment, obtain transfer samples, and cache them. The strategy parameter in Actor2 represents the old strategy. After each iteration, the parameters in Actor1 will be used to update the parameters in Actor2. The critic network consists of a neural network. The specific training steps of the task deployment model are as follows:

a) Input the current state into the Actor1 network, the agent selects an action based on the policy π _old , ie a _l =π(s _l ). Repeating the above process, the agent continuously interacts with the edge network for T time steps, collects historical interaction information and caches it.

b) Calculate the advantage function for each time step using the following equation, where γ is the discount factor, V is the state value function, φ is the critic network parameter, and l is the time step.

B _t =∑ _l>t γ ^lt rw _l -V _φ (s _t )

c) Calculate the loss function of the critic network using the following formula, and update the critic network parameter φ according to the back-propagation of this function.

d) Use L ^CLIP (θ) and the advantage function to update the parameters of the actor network.

e) Repeat step d, and use the network parameters in Actor1 to update the parameters of Actor2 after a certain step.

f) Cycle a-f steps.

The PPO algorithm based on the Actor-Critic framework is obtained, and then the agent outputs the next action according to the actor network according to the given task queue, and the critic network gives the corresponding evaluation, and iterates continuously until the allocation of resources for time-sensitive tasks is completed. .

The present invention adopts the resource allocation algorithm based on PPO to realize the efficient allocation of time delay control-oriented calculation and communication resources.

Example 2:

Based on the same inventive concept, the present invention also provides a multi-task scheduling system for IoT agents based on time delay control.

Among them, the data collection module is used to obtain the information that multiple power services reach the resource pool of the IoT agent;

The solving module is used to input the information into the pre-built time-delay controlled multi-task deployment and resource allocation optimization problem model for solving, and obtain the deployment mode of the power service and the resource allocation mode of the IoT agent;

The scheduling module is used to schedule the power service according to the deployment method of the power service and the resource allocation method of the IoT agent;

Among them, the multi-task deployment of delay control and the optimization problem model of resource allocation are constructed based on the goal of minimizing the total delay of multiple power services.

Wherein, the solving module is further configured to construct an objective function with the goal of minimizing the total time delay under the deployment of multiple power services in the resource pool of each IoT agent and the resource allocation mode;

Build constraints based on resource constraints of IoT agents;

Build an optimization problem model for delay-controlled multi-task deployment and resource allocation based on objective functions and constraints;

The resources include computing resources and bandwidth resources; the information includes at least one or more of the following: the sequence of containers in the resource pool required by the power service request, the priority of the power service, and the data packets of the power service leaving the resource pool The time interval distribution corresponds to the parameters of the Poisson process, the processing speed of the power service by the unit computing resource of the resource pool, and the transmission speed of the power service by the unit bandwidth resource of the resource pool.

Among them, the calculation formula of the objective function constructed by the solving module is as follows:

In the formula, F represents the total delay of the deployment and resource allocation of multiple power services in the resource pool of each IoT agent, K _S represents the number of power services in the same resource pool in the same time period, ω _s represents the power service The priority of s, D _s represents the total delay of the power service s, and S represents the set of all power services;

The calculation formula of the total delay D _s of the power service s is as follows:

表示电力业务s在离开资源池z后的链路时延；In the formula, |N _s | represents the number of containers in the resource pool that process the power service s, D _s,z represents the average delay of the power service s at the resource pool z,

represents the link delay of the power service s after leaving the resource pool z;

The calculation formula of the average delay D _s,z of the power service s at the resource pool z is as follows:

In the formula, c _s,1 represents the processing speed of the power service s after resource allocation, c _s,2 represents the transmission speed of the power service s after resource allocation, and P _s,z represents the power service s in the resource pool. The non-empty probability in the processing queue of z, λ _s represents the parameters of the Poisson process corresponding to the time interval distribution of the data packets of the power service s leaving the resource pool;

的计算式如下：Link delay of power service s after leaving resource pool z

The calculation formula is as follows:

In the formula, n _z represents the size of the data to be transmitted in the resource pool z.

Among them, the calculation formula for solving the resource constraints constructed by the module is as follows:

表示电力业务在资源池中可获得的资源量集合，

表示电力业务在资源池z中的容器使用所有计算资源时的处理速度，

表示电力业务在资源池z中的容器使用所有带宽资源时的传输速度。In the formula, _KB represents the total number of power services deployed on the resource pool, κ _z,i represents the proportion of computing resources allocated by resource pool z to power service i, and ν _z,i represents the bandwidth resources allocated by resource pool z to power service i , σ represents the maximum amount of resources available to the power business in the resource pool,

Represents the set of resources available for the power business in the resource pool,

represents the processing speed of the power business when the containers in the resource pool z use all computing resources,

Indicates the transmission speed of the power service when all the bandwidth resources are used by the containers in the resource pool z.

Among them, the solving module is specifically used for:

Input the information into the pre-built delay-controlled multi-task deployment and resource allocation optimization problem model, and use the improved ant colony algorithm based on the dynamic pheromone volatility coefficient to optimize the delay-controlled multi-task deployment and resource allocation problem model. Solve to get the deployment mode of each container in the resource pool of the IoT agent for multiple power tasks;

Based on the deployment methods of multiple power tasks in the resource pool of the IoT agent, the near-end policy optimization algorithm is used to solve the optimization problem model of the multi-task deployment and resource allocation of delay control, and the resources of the IoT agent are obtained. How the pool allocates resources to each container.

Among them, the solving module uses the improved ant colony algorithm based on the dynamic pheromone volatility coefficient to solve the optimization problem model of the multi-task deployment and resource allocation of delay control, and obtains the multiple power tasks in each container in the resource pool of the IoT agent. deployment methods, including:

Use the computing resources in the resource pool of the IoT agent to initialize the pheromone of the ant colony algorithm;

Calculate the disappearance of the pheromone trajectory of each resource pool based on the dynamic pheromone volatilization coefficient;

Calculate the probability of selecting each resource pool for task deployment based on the disappearance of the pheromone trajectory, and select the next resource pool to deploy the power task based on the probability;

Based on the disappearance of pheromone trajectory and the selection of the resource pool for the next deployment power task, the ant colony is used for iterative calculation, and the most frequent plan in the ant colony path planning is obtained as the deployment of each container in the resource pool of the IoT agent as multiple power tasks Way.

Among them, the calculation formula of the pheromone of the solving module to initialize the ant colony algorithm is as follows:

In the formula, τ _0,j represents the initialized pheromone corresponding to resource pool j, r' _m,j represents the available memory of resource pool j, subscript m represents memory, r _m,j represents the total memory of resource pool j, ψ _m represents the total memory of all resource pools, r' _p,j represents the available CPU of resource pool j, subscript p represents CPU, r _p,j represents the total CPU of resource pool j, ψ _p represents the total CPU of all resource pools.

Among them, the calculation formula of the solving module to calculate the disappearance of the pheromone trajectory is as follows:

In the formula, τ(t,j) represents the disappearance of the pheromone trajectory corresponding to resource pool j at time t, τ _0,j represents the initialized pheromone corresponding to resource pool j, Δ represents the change of pheromone, and P(k) represents the first pheromone. The resource pool that can be selected in k iterations, ρ represents the dynamic pheromone volatility coefficient;

The calculation formula of dynamic pheromone volatilization coefficient ρ is as follows:

Example 3:

The power IoT agent dispatching system constructed by the present invention is shown in FIG. 4 . Business terminals include various smart meters, photovoltaic panels and charging piles and other power business equipment. The power communication network includes industrial Ethernet, EPON (Ethernet Passive Optical Network, Ethernet Passive Optical Network), 2G/3G/4G, wireless local area network, wireless wide area network, etc. The IoT agent layer is composed of a software layer and a hardware layer, which is used to connect the service terminal and the platform layer, and realize various functions such as data collection, message adaptation, edge computing, and transmission of the terminal. Based on the lightweight container technology, the IoT agent completes the flexible scheduling of edge-side virtualized computing and network resources through the built-in resource scheduling engine, and realizes unified access, multi-dimensional perception, unified modeling, and resource scheduling. The IoT management and control platform manages and controls the IoT agent through Restful/OpenFlow/SNMP.

Example 4:

Based on the same inventive concept, the present invention also provides a computer device, the computer device includes a processor and a memory, the memory is used for storing a computer program, the computer program includes program instructions, and the processor is used for executing the Program instructions stored by the computer storage medium. The processor may be a central processing unit (Central Processing Unit, CPU), or other general-purpose processors, digital signal processors (Digital Signal Processor, DSP), application specific integrated circuits (Application Specific Integrated Circuit, ASIC), off-the-shelf programmable gate array ( Field-Programmable GateArray, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., which are the computing core and control core of the terminal, which are suitable for implementing one or more instructions, specifically suitable for loading And execute one or more instructions in the computer storage medium to realize the corresponding method process or corresponding function, so as to realize the steps of the multi-task scheduling method for IoT agents based on delay control in the above embodiment.

Example 5:

Based on the same inventive concept, the present invention also provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device for storing programs and data. It can be understood that, the computer-readable storage medium here may include both a built-in storage medium in a computer device, and certainly also an extended storage medium supported by the computer device. The computer-readable storage medium provides storage space in which the operating system of the terminal is stored. In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space, and these instructions may be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory, or a non-volatile memory (non-volatile memory), such as at least one disk memory. One or more instructions stored in the computer-readable storage medium can be loaded and executed by the processor to implement the steps of the method for scheduling multiple tasks of an IoT agent based on delay control in the foregoing embodiment.

As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block in the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to the processor of a general purpose computer, special purpose computer, embedded processor or other programmable data processing device to produce a machine such that the instructions executed by the processor of the computer or other programmable data processing device produce Means for implementing the functions specified in a flow or flow of a flowchart and/or a block or blocks of a block diagram.

These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory result in an article of manufacture comprising instruction means, the instructions The apparatus implements the functions specified in the flow or flow of the flowcharts and/or the block or blocks of the block diagrams.

These computer program instructions can also be loaded on a computer or other programmable data processing device to cause a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process such that The instructions provide steps for implementing the functions specified in the flow or blocks of the flowcharts and/or the block or blocks of the block diagrams.

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 the scope of its protection. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand: Those skilled in the art can still make various changes, modifications or equivalent replacements to the specific embodiments of the application after reading the present disclosure, but these changes, modifications or equivalent replacements are all within the protection scope of the pending claims of the application.

Claims (18)

1. a method for multi-task scheduling of IoT agents based on time delay control, is characterized in that, comprising: Obtain the information that multiple power services arrive at the resource pool of the IoT agent; Inputting the information into a pre-built time-delay controlled multi-task deployment and resource allocation optimization problem model to solve, to obtain the deployment mode of the power service and the resource allocation mode of the IoT agent; According to the deployment method of the power service and the resource allocation method of the IoT agent, the power service is scheduled; Wherein, the multi-task deployment and resource allocation optimization problem model of delay control is constructed based on the goal of minimizing the total delay of multiple power services. 2. The method according to claim 1, wherein the construction of the optimization problem model of the multi-task deployment and resource allocation of the delay control comprises: The objective function is constructed with the goal of minimizing the total delay in the deployment of multiple power services in the resource pool of each IoT agent and the resource allocation mode; Build constraints based on resource constraints of IoT agents; Build an optimization problem model of delay-controlled multi-task deployment and resource allocation based on the objective function and constraints; Wherein, the resources include computing resources and bandwidth resources; the information includes at least one or more of the following: the sequence of containers in the resource pool required by the power service request, the priority of the power service, and the data of the power service The time interval distribution of packets leaving the resource pool corresponds to the parameters of the Poisson process, the processing speed of the unit computing resource of the resource pool to the power service, and the transmission speed of the unit bandwidth resource of the resource pool to the power service. 3. method as claimed in claim 2, is characterized in that, the computational formula of described objective function is as follows:

In the formula, F represents the total delay of the deployment and resource allocation of multiple power services in the resource pool of each IoT agent, K _S represents the number of power services in the same resource pool in the same time period, ω _s represents the power service The priority of s, D _s represents the total delay of the power service s, and S represents the set of all power services; The calculation formula of the total delay D _s of the power service s is as follows:

In the formula, |N _s | represents the number of containers in the resource pool that process the power service s, D _s,z represents the average delay of the power service s at the resource pool z,

represents the link delay of the power service s after leaving the resource pool z; The calculation formula of the average delay D _s,z of the power service s at the resource pool z is as follows:

In the formula, c _s,1 represents the processing speed of the power service s after resource allocation, c _s,2 represents the transmission speed of the power service s after resource allocation, and P _s,z represents the power service s in the resource pool. The non-empty probability in the processing queue of z, λ _s represents the parameters of the Poisson process corresponding to the time interval distribution of the data packets of the power service s leaving the resource pool; The link delay of the power service s after leaving the resource pool z

The calculation formula is as follows:

In the formula, n _z represents the size of the data to be transmitted in the resource pool z. 4. The method of claim 2, wherein the calculation formula of the resource constraint is as follows:

In the formula, _KB represents the total number of power services deployed on the resource pool, κ _z,i represents the proportion of computing resources allocated by resource pool z to power service i, and ν _z,i represents the bandwidth resources allocated by resource pool z to power service i , σ represents the maximum amount of resources available to the power business in the resource pool,

Represents the set of resources available for the power business in the resource pool,

represents the processing speed of the power business when the containers in the resource pool z use all computing resources,

Indicates the transmission speed of the power service when all the bandwidth resources are used by the containers in the resource pool z. 5. The method according to claim 2, wherein the information is input into a pre-built time-delay controlled multi-task deployment and an optimization problem model of resource allocation to solve, to obtain a deployment mode of the power service and The resource allocation method of the IoT agent, including: The information is input into a pre-built optimization problem model of delay-controlled multi-task deployment and resource allocation, and an improved ant colony algorithm based on dynamic pheromone volatility coefficient is used to optimize the delay-controlled multi-task deployment and resource allocation. The optimization problem model is solved to obtain the deployment mode of each container in the resource pool of the IoT agent for multiple power tasks; Based on the deployment methods of multiple power tasks in the resource pool of the IoT agent, the near-end strategy optimization algorithm is used to solve the optimization problem model of the multi-task deployment and resource allocation for delay control, and the IoT agent is obtained. The resource allocation method of the resource pool to each container. 6. The method according to claim 5, characterized in that, the improved ant colony algorithm based on dynamic pheromone volatility coefficient is used to solve the optimization problem model of the multi-task deployment and resource allocation of the time delay control, Get the deployment methods of multiple power tasks in each container in the IoT agent's resource pool, including: Use the computing resources in the resource pool of the IoT agent to initialize the pheromone of the ant colony algorithm; Calculate the disappearance of the pheromone trajectory of each resource pool based on the dynamic pheromone volatilization coefficient; Calculate the probability of selecting each resource pool for task deployment based on the disappearance of the pheromone trajectory, and select the next resource pool to deploy the power task based on the probability; Based on the disappearance of the pheromone trajectory and the selected resource pool for the next deployment power task, the ant colony is used for iterative calculation, and the most frequent planning in the ant colony path planning is obtained as multiple power tasks in each container in the resource pool of the IoT agent deployment method. 7. The method according to claim 6, wherein the calculation formula of the pheromone of the initialization ant colony algorithm is as follows:

In the formula, τ _0,j represents the initialized pheromone corresponding to resource pool j, r′ _m,j represents the available memory of resource pool j, the subscript m represents the memory, r _m,j represents the total memory of resource pool j, ψ _m represents the total memory of all resource pools, r′ _p,j represents the available CPU of resource pool j, subscript p represents CPU, r _p,j represents the total CPU of resource pool j, ψ _p represents the total CPU of all resource pools. 8. The method of claim 6, wherein the calculation formula for the disappearance of the pheromone track is as follows:

In the formula, τ(t,j) represents the disappearance of the pheromone trajectory corresponding to resource pool j at time t, τ _0,j represents the initialized pheromone corresponding to resource pool j, Δ represents the change of pheromone, and P(k) represents the first pheromone. The resource pool that can be selected in k iterations, ρ represents the dynamic pheromone volatility coefficient; The calculation formula of the dynamic pheromone volatilization coefficient ρ is as follows:

9. A IoT agent multi-task scheduling system based on time delay control, characterized in that it comprises: a data acquisition module, a solution module and a scheduling module; The data acquisition module is used to acquire the information that multiple power services reach the resource pool of the IoT agent; The solving module is used for inputting the information into a pre-built time-delay controlled multi-task deployment and resource allocation optimization problem model for solving, to obtain the deployment mode of the power service and the resource allocation mode of the IoT agent; The scheduling module is used to schedule the power service according to the deployment mode of the power service and the resource allocation mode of the IoT agent; Wherein, the multi-task deployment and resource allocation optimization problem model of delay control is constructed based on the goal of minimizing the total delay of multiple power services. 10 . The system according to claim 9 , wherein the solving module is further configured to target the minimum total time delay in the deployment of multiple power services in the resource pool of each IoT agent and the resource allocation mode. 11 . Build the objective function; Build constraints based on resource constraints of IoT agents; Build an optimization problem model of delay-controlled multi-task deployment and resource allocation based on the objective function and constraints; Wherein, the resources include computing resources and bandwidth resources; the information includes at least one or more of the following: the sequence of containers in the resource pool required by the power service request, the priority of the power service, and the data of the power service The time interval distribution of packets leaving the resource pool corresponds to the parameters of the Poisson process, the processing speed of the unit computing resource of the resource pool to the power service, and the transmission speed of the unit bandwidth resource of the resource pool to the power service. 11. system as claimed in claim 10, is characterized in that, the computational formula of the objective function that described solving module builds is as follows:

In the formula, F represents the total delay of the deployment and resource allocation of multiple power services in the resource pool of each IoT agent, K _S represents the number of power services in the same resource pool in the same time period, ω _s represents the power service The priority of s, D _s represents the total delay of the power service s, and S represents the set of all power services; The calculation formula of the total delay D _s of the power service s is as follows:

In the formula, |N _s | represents the number of containers in the resource pool that process the power service s, D _s,z represents the average delay of the power service s at the resource pool z,

represents the link delay of the power service s after leaving the resource pool z; The calculation formula of the average delay D _s,z of the power service s at the resource pool z is as follows:

In the formula, c _s,1 represents the processing speed of the power service s after resource allocation, c _s,2 represents the transmission speed of the power service s after resource allocation, and P _s,z represents the power service s in the resource pool. The non-empty probability in the processing queue of z, λ _s represents the parameters of the Poisson process corresponding to the time interval distribution of the data packets of the power service s leaving the resource pool; The link delay of the power service s after leaving the resource pool z

The calculation formula is as follows:

In the formula, n _z represents the size of the data to be transmitted in the resource pool z. 12. The system according to claim 10, wherein the calculation formula of the resource constraint constructed by the solving module is as follows:

In the formula, _KB represents the total number of power services deployed on the resource pool, κ _z,i represents the proportion of computing resources allocated by resource pool z to power service i, and ν _z,i represents the bandwidth resources allocated by resource pool z to power service i , σ represents the maximum amount of resources available to the power business in the resource pool,

Represents the set of resources available for the power business in the resource pool,

represents the processing speed of the power business when the containers in the resource pool z use all computing resources,

Indicates the transmission speed of the power service when all the bandwidth resources are used by the containers in the resource pool z. 13. The system of claim 10, wherein the solving module is specifically used for: The information is input into a pre-built optimization problem model of delay-controlled multi-task deployment and resource allocation, and an improved ant colony algorithm based on dynamic pheromone volatility coefficient is used to optimize the delay-controlled multi-task deployment and resource allocation. The optimization problem model is solved to obtain the deployment mode of each container in the resource pool of the IoT agent for multiple power tasks; Based on the deployment methods of multiple power tasks in the resource pool of the IoT agent, the near-end strategy optimization algorithm is used to solve the optimization problem model of the multi-task deployment and resource allocation for delay control, and the IoT agent is obtained. The resource allocation method of the resource pool to each container. 14. The system according to claim 13, wherein the solving module adopts an improved ant colony algorithm based on dynamic pheromone volatilization coefficient to carry out the optimization problem model of the multi-task deployment and resource allocation of the delay control. Solve to get the deployment mode of each container in the resource pool of the IoT agent for multiple power tasks, including: Use the computing resources in the resource pool of the IoT agent to initialize the pheromone of the ant colony algorithm; Calculate the disappearance of the pheromone trajectory of each resource pool based on the dynamic pheromone volatilization coefficient; Calculate the probability of selecting each resource pool for task deployment based on the disappearance of the pheromone trajectory, and select the next resource pool to deploy the power task based on the probability; Based on the disappearance of the pheromone trajectory and the selected resource pool for the next deployment power task, the ant colony is used for iterative calculation, and the most frequent planning in the ant colony path planning is obtained as multiple power tasks in each container in the resource pool of the IoT agent deployment method. 15. The system of claim 14, wherein the calculation formula of the pheromone of the ant colony algorithm initialized by the solution module is as follows:

In the formula, τ _0,j represents the initialized pheromone corresponding to resource pool j, r′ _m,j represents the available memory of resource pool j, the subscript m represents the memory, r _m,j represents the total memory of resource pool j, ψ _m represents the total memory of all resource pools, r′ _p,j represents the available CPU of resource pool j, subscript p represents CPU, r _p,j represents the total CPU of resource pool j, ψ _p represents the total CPU of all resource pools. 16. The system of claim 14, wherein the calculation formula for calculating the disappearance of the pheromone trajectory by the solving module is as follows:

In the formula, τ(t,j) represents the disappearance of the pheromone trajectory corresponding to resource pool j at time t, τ _0,j represents the initialized pheromone corresponding to resource pool j, Δ represents the change of pheromone, and P(k) represents the first pheromone. The resource pool that can be selected in k iterations, ρ represents the dynamic pheromone volatility coefficient; The calculation formula of the dynamic pheromone volatilization coefficient ρ is as follows:

17. A computer device, comprising: one or more processors; memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the method for scheduling multiple tasks of IoT agents based on delay control according to any one of claims 1 to 8 is implemented. 18. A computer-readable storage medium, characterized in that a computer program is stored thereon, and when the computer program is executed, the delay control-based IoT according to any one of claims 1 to 8 is realized. Proxy multitasking scheduling method.

Images