Ultra-dense Internet of Things computing unloading method and system based on edge computing

Abstract

The invention discloses an edge computing-based ultra-dense Internet of Things computing unloading method and system, and relates to the Internet of Things and data processing technology, and the method comprises the steps: constructing a local execution model and a mobile edge execution model in a multi-user ultra-dense edge server scene; determining an optimal mobile edge computing unloading model; enumerating all selectable calculation offload decisions of all tasks, and selecting an offload decision contour for all MDs tasks, so that the offload decision contour has overall minimum calculation overhead under the condition of meeting given wireless channel constraints; and according to the selected unloading decision contour, determining an optimal unloading strategy by using a greedy approximate unloading method of the game theory. On the basis of traditional mobile edge computing unloading, the computing unloading method of the ultra-dense Internet of Things network is provided for a multi-user ultra-dense edge server scene, and the overall computing overhead of all tasks is minimized while the wireless performance is met.

Description

Technical Field

[0001] The present application relates to the technical fields of Internet of Things and data processing, and particularly relates to a method and system for ultra-dense Internet of Things computing offloading based on edge computing. Background Art

[0002] In recent years, with the emergence of Internet of Things technologies such as narrowband Internet of Things and intelligent Internet of Things, many new Internet of Things applications have emerged and attracted a great deal of attention. A large number of Internet of Things mobile devices, such as smart phones, smart bracelets, smart cameras, virtual reality, etc., have been widely used in life. On the other hand, with the rapid development of 5G communication technology, it is believed that the deployment of ultra-dense 5G will be the trend of future communication networks. The deployment of large-scale Internet of Things mobile devices (Mobile DeviceMD) and ultra-dense 5G cells jointly promotes the evolution of the Internet of Things towards future ultra-dense Internet of Things networks. Compared with existing Internet of Things, ultra-dense Internet of Things poses diverse requirements on communication networks in terms of quality of service (QoS) and quality of experience (QoE), and future ultra-dense Internet of Things faces unprecedented challenges.

[0003] Among these challenges, a fundamental challenge is how to resolve the conflict between resource-starved mobile applications and resource-constrained Internet of Things MDs. Many mobile applications requested by Internet of Things mobile devices MDs are computationally intensive and require high energy consumption, such as fingerprint recognition, face recognition, natural language processing, and interactive games. However, limited by the physical size of these lightweight Internet of Things MDs, they always have limited computing resources and battery life.

[0004] Most existing MECO research considers single-layer base station (Base station BS) scenarios, and the computing offloading scheme is very simple. In particular, MD either executes computing tasks locally on the CPU or offloads some computationally intensive tasks to the edge server of the MBS for execution. Considering the large number of Internet of Things MDs and mobile applications in ultra-dense Internet of Things networks, MBS is prone to congestion, seriously affecting QoS and QoE. In fact, we can deploy lightweight edge servers on small cells closer to Internet of Things MDs and offload some computationally intensive tasks to these edge servers, thereby reducing the burden on MBS. Most existing work considers simple computing task scenarios and ignores the situation where multiple computing tasks are randomly requested by MDs and the computing resources on edge servers change dynamically.

[0005] Although some studies have focused on the computational offloading problem of multiple edge servers in recent years, most of these studies consider simple computational task scenarios, that is, a fixed number of computational tasks are assumed in advance during a computational offloading period, and each MD has only one computational task to complete. In addition, the situation where MDs randomly request different types of computational tasks and the computational resources of edge servers change dynamically is also ignored in most existing works. Summary of the Invention

[0006] An embodiment of the present application provides a computational offloading method and system for ultra-dense Internet of Things based on edge computing. On the basis of traditional mobile edge computing offloading, for the scenario of multi-user ultra-dense edge servers, a computational offloading method for ultra-dense Internet of Things networks is proposed to minimize the overall computational overhead of all tasks while meeting wireless requirements.

[0007] An embodiment of the present application provides a computational offloading method for ultra-dense Internet of Things based on edge computing, including:

[0008] In the scenario of multi-user ultra-dense edge servers, a local execution model and a mobile edge execution model are constructed, where in the scenario of multi-user ultra-dense edge servers, N co-located Internet of Things MDs are alternately covered by MBS and SSC;

[0009] According to the constructed local execution model and mobile edge execution model, an optimal mobile edge computing offloading model is determined;

[0010] By means of an enumeration method, all optional computational offloading decisions of all tasks are enumerated, and an offloading decision profile is selected for the tasks of all MDs, so that the offloading decision profile has the overall minimum computational overhead under the condition of meeting the given wireless channel constraints;

[0011] According to the selected offloading decision profile, using the greedy approximation offloading method of game theory, a globally optimal offloading decision profile is determined for all MDs by adopting a stochastic game theory strategy in the outer layer of GT-GAOA, and a locally optimal offloading strategy is determined for different types of computational tasks of MDi by adopting a greedy approximation method in the inner layer.

[0012] Optionally, in the scenario of multi-user ultra-dense edge servers, for an IoT MD with M types of computational tasks, in an ultra-dense network, communication with the MBS is achieved through dual connectivity or CoMP technology, and each type of computational task is atomic and cannot be further divided;

[0013] For any type-j computational task T of MD i i,j , satisfying T i,j =(m i,j , c i,j ), where m i,jis task T i,j is the input data size of i,j and c i,j is the number of CPU cycles required to complete T;

[0014] Define that MBS and each SC have K and L orthogonal radio frequency channels respectively. Let λ i,j ∈ {0, -1, 1, 2,..., S} represent the offloading decision of the j-th type of task at MD i.

[0015] Optionally, constructing the local execution model and the mobile edge execution model includes:

[0016] Constructing the local execution model:

[0017] In the case where MD i locally executes the j-th type of task, that is, when λ i,j = 0, calculate that its energy consumption and the processing time of T i,j satisfy:

[0018]

[0019] where f i and δ i represent the computing power of MD i and the energy consumed per CPU cycle respectively, are the energy consumption and processing time of locally executing the j-th type of task respectively;

[0020] Constructing the mobile edge execution model:

[0021] In the case of offloading task T i,J to the MEC server connected to MBS, the processing time of the type-j task of MD i mainly includes: task T i,J The transmission time from MD i to MBS through wireless access, T i,J The transmission time from MBS to the MEC server through the optical fiber link, task T i,J The queuing time and execution time on the MEC server. In this case, the processing time satisfies:

[0022]

[0023] where is the uplink data rate of transmitting T i,J from MD i to MBS through the wireless access channel, c is the uplink data rate of transmitting T i,J through the optical fiber link, and are the queuing time and execution time of task T i,J on the MEC server respectively.

[0024] Optionally, constructing the mobile edge execution model further includes:

[0025] Modeling the arrival and execution of task T on the MEC server as an M / M / 1 queue, and calculating the average arrival rate of tasks on the MEC server according to the service time on the MEC server following an exponential distribution i,J Satisfying: Satisfying:

[0026]

[0027] Calculating the average service time for completing task T i,J Satisfying:

[0028]

[0029] The energy consumed in this case is expressed as:

[0030]

[0031] Then the calculation overhead satisfies:

[0032]

[0033] Optionally, according to the constructed local execution model and mobile edge execution model, determining the optimal mobile edge computing offloading model includes:

[0034] Obtaining an optimal computing offloading profile for different types of computing tasks of the IoT MD based on the OMOU problem, satisfying:

[0035]

[0036] In the formula, the constraint condition C1 is used to ensure that the total number of tasks offloaded to the MEC server connected to the MBS is less than the number of wireless channels owned by the MBS, the constraint condition C2 is used to ensure that the total number of tasks offloaded to a certain SC k (1 ≤ k ≤ S) is not greater than the number of wireless channels owned by SC k, and for the constraint condition C3, the type-j task of MD i can choose any one of the S + 2 offloading decision options.

[0037] Optionally, through the enumeration method, enumerating all optional computing offloading decisions for all tasks and selecting an offloading decision profile for the tasks of all MDs includes:

[0038] Enumerating all optional offloading decision combinations for all MD computing tasks;

[0039] Discarding offloading decision combinations that do not satisfy the given wireless channel constraints of the MBS and SCs;

[0040] Calculate the overall computational overhead of the remaining offloading decision combinations to determine the overall offloading decision combination S with the minimum overall computational overhead Ψmin.

[0041] Optionally, according to the selected offloading decision profile, using the greedy approximation offloading method of game theory, a globally optimal offloading decision profile is determined for all MDs using a stochastic game theory strategy in the outer layer of GT-GAOA, and a locally optimal offloading strategy is determined for different types of computational tasks of MDi using the greedy approximation method in the inner layer, including:

[0042] Regard the IoT MD as a rational game player and model OMOU as a strategic game where the set N of MDs is the set of game players, is the strategy set of game player i, Ψ i (Λ i , Λ -i ) is the cost function that player i needs to minimize;

[0043] Solve the strategic game to determine a globally optimal offloading decision profile for all MDs using a stochastic game theory strategy in the outer layer of GT-GAOA and a locally optimal offloading strategy for different types of computational tasks of MDi using the greedy approximation method in the inner layer.

[0044] An embodiment of the present application also proposes an edge-computing-based ultra-dense IoT computing offloading system, including a processor and a memory, where a computer program is stored on the memory, and when the computer program is executed by the processor, the steps of the edge-computing-based ultra-dense IoT computing offloading method as described above are implemented.

[0045] Based on traditional mobile edge computing offloading, for the multi-user ultra-dense edge server scenario, an embodiment of the present application proposes a computing offloading method for an ultra-dense IoT network, which can minimize the overall computational overhead of all tasks while satisfying wirelessness.

[0046] The above description is only an overview of the technical solution of the present application. In order to be able to understand the technical means of the present application more clearly, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features and advantages of the present application more obvious and understandable, the specific embodiments of the present application are specifically described below. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] By reading the following detailed description of the preferred embodiments, various other advantages and benefits will become clear to those of ordinary skill in the art. The drawings are only for the purpose of showing the preferred embodiments and are not considered to be a limitation of the present application. And throughout the drawings, the same reference numerals are used to represent the same components. In the drawings:

[0048] Figure 1 This is a flowchart illustration of the edge - computing - based ultra - dense Internet of Things (IoT) computing offloading method according to an embodiment of the present application. Detailed implementation manners

[0049] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be completely conveyed to those skilled in the art.

[0050] Based on traditional mobile edge computing offloading, in view of the multi - user ultra - dense edge server scenario, an embodiment of the present application proposes a computing offloading method for an ultra - dense IoT network to minimize the overall computing overhead of all tasks while meeting wireless requirements. Specifically, an embodiment of the present application provides an edge - computing - based ultra - dense IoT computing offloading method, as Figure 1 shown, including the following steps:

[0051] In step S101, in a multi - user ultra - dense edge server scenario, a local execution model and a mobile edge execution model are constructed. In the multi - user ultra - dense edge server scenario, N co - located IoT MDs are alternately covered by MBS and SSC.

[0052] In some embodiments, a multi - user ultra - dense edge server scenario can be pre - constructed. An embodiment of the present application considers the collaborative MECO scenario in an ultra - dense IoT network. In this scenario, N co - located IoT MDs are alternately covered by MBS and SSC. Let N = {1, 2,.., N}, S = {1, 2,..., S}(S≥N), and M = {1, 2,..., M} respectively represent the sets of IoT MDs, SCs, and different types of computing tasks (e.g., face recognition, fingerprint recognition, natural language processing, and interactive games, etc.).

[0053] In a multi - user ultra - dense edge server scenario, for an IoT MD with M types of computing tasks, in an ultra - dense network, communication with MBS is achieved through dual - connection or CoMP technology, and the Internet is accessed simultaneously through nearby SCs. We assume that each type of computing task is atomic and cannot be divided.

[0054] To provide MEC services to MDs, one or more MEC servers are connected to MBS / SCs through fiber optic links. Without loss of generality, this application considers the case where the MEC server is connected to the MBS / SC. Therefore, for any Internet of Things MD i, it can choose at most three offloading strategies to complete its computing tasks, such as local computing on the CPU of the MD, offloading to the MEC server connected to the MBS, and offloading to the MEC server connected to the SC associated with MD i.

[0055] Specifically, for any type-j computing task T of MD i i,j , satisfying T i,j =(m i,j , c i,j ), where m i,j is the input data size of task T i,j , and c i,j is the number of CPU cycles required to complete T i,j . Denote P i,j as the probability that MD i requests a type-j task at a certain moment, 0 < P i,j < 1.

[0056] Define that the MBS and each SC have K and L orthogonal radio frequency channels respectively, and let λ i,j ∈{0, -1, 1, 2,..., S} represent the offloading decision of the j-th type of task at MD i. Assume that during a computing offloading period, the MDs are still covered by the MBS and SCs.

[0057] In step S102, according to the constructed local execution model and mobile edge execution model, determine the optimal mobile edge computing offloading model.

[0058] In step S103, through the enumeration method, enumerate all optional computing offloading decisions for all tasks, and select an offloading decision profile for the tasks of all MDs, so that the offloading decision profile has the overall minimum computing overhead under the given wireless channel constraints.

[0059] In step S104, according to the selected offloading decision profile, use the greedy approximation offloading method of game theory. In the outer layer of GT-GAOA, adopt a stochastic game theory strategy to determine an overall optimal offloading decision profile for all MDs, and in the inner layer, adopt the greedy approximation method to determine a local optimal offloading strategy for different types of computing tasks of MDi.

[0060] In the method of the embodiment of the present application, in the scenario of a multi-user ultra-dense edge server, the case where MD randomly requests different types of computing tasks is considered, and the case where tasks randomly arrive at the edge server and the computing resources of the edge server change dynamically is considered. A system model is established according to the energy consumption of the Internet of Things MD and the processing delay of computing tasks, and the optimal MECO problem in the ultra-dense Internet of Things is modeled as a constrained optimization problem, which can minimize the overall computing overhead while satisfying the given wireless channel constraints.

[0061] In some embodiments, constructing the local execution model and the mobile edge execution model includes:

[0062] Constructing the local execution model:

[0063] In the case where MD i locally executes the j-th type of task, that is, when λ i,j = 0, its energy consumption and the processing time of T i,j satisfy:

[0064]

[0065] where f i and δ i respectively represent the computing power of MD i and the energy consumed per CPU cycle. are respectively the energy consumption and processing time of locally executing the j-th type of task.

[0066] The corresponding computing overhead is obtained by adopting local computing:

[0067]

[0068] where are respectively the weight coefficients of the processing time and energy consumption for MD i to make an offloading decision. Obviously, there is If that is, the type-j task of MD i is more sensitive to processing delay. These applications include face recognition, fingerprint recognition, etc. Otherwise, if MD i is in a very low battery state, in practical applications, it can be set even

[0069] Constructing the mobile edge execution model:

[0070] For the computing offloading scenario in the ultra-dense IoT network, in addition to locally computing on its own CPU, MD i also needs to offload the task Ti,J to the MEC server connected to the MBS. For the case of offloading the task T i,J to the MEC server connected to the MBS, the processing time of the type-j task of MD i mainly includes: the task Ti,J The transmission time, T, from the MD i to the MBS via wireless access i,J The transmission time of the task T from the MBS to the MEC server via the optical fiber link i,J The queuing time and execution time on the MEC server. In this case, the processing time satisfies:

[0071]

[0072] where, is the uplink data rate for transmitting T from the MD i via the wireless access channel, and c is the uplink data rate for transmitting T via the optical fiber link i,J to the MBS, and i,J are the queuing time and execution time of the task T on the MEC server respectively. For most mobile applications, such as face recognition, interactive games, etc., the calculation result is usually much smaller than the input data volume. In the embodiments of the present application, the time cost of sending the calculation result from the MEC server back to the MD is ignored. and respectively represent i,J the queuing time and execution time of the task T on the MEC server. For most mobile applications, such as face recognition, interactive games, etc., the calculation result is usually much smaller than the input data volume. In the embodiments of the present application, the time cost of sending the calculation result from the MEC server back to the MD is ignored.

[0073] The data rate of the uplink can be obtained by calculation:

[0074]

[0075] where, is the wireless channel bandwidth of the MBS, is the MD i and the channel gain between the MBS.

[0076] In some embodiments, constructing the mobile edge execution model further includes:

[0077] Modeling the arrival and execution of the task T on the MEC server as an M / M / 1 queue. In this case, the service time on the MEC server follows an exponential distribution. In this embodiment, according to the service time on the MEC server following an exponential distribution, let i,J be used to represent the offloading decision. When λ =-1, it means that the average arrival rate i,j of the computing task on the MEC server satisfies:

[0078]

[0079] Calculating that the average service time for completing the task T i,J satisfies:

[0080]

[0081] The energy consumed in this case is expressed as:

[0082]

[0083] Then the computational overhead satisfies:

[0084]

[0085] For the MECO scenario in ultra-dense IoT networks, any type-j task of MD i has S + 2 offloading decision choices, and it is easy to select one of them with the minimum computational overhead. As more types of computational tasks are offloaded to the MEC server, the wireless uplink data rate will drop sharply due to the wireless interference model, and the average service time to complete task T i,j will increase synchronously, and the overall computational overhead obtained will increase significantly. Therefore, there is a trade-off relationship between the overall computational overhead achieved and the number of tasks offloaded to the MEC server. The problem of cooperative mobile edge computing offloading in ultra-dense Internet of Things networks is to find an optimal computational offloading profile for the tasks of all MDs, while minimizing the overall computational overhead subject to the wireless channel constraints of the MBS and SCs. In some embodiments, according to the constructed local execution model and mobile edge execution model, determining the optimal mobile edge computing offloading model includes:

[0086] Given the initial information of all IoT MDs and their computational tasks, the computational capabilities of the MEC servers connecting the MBS and SCs, the number of wireless channels, and the uplink data rates of the wireless and fiber-optic networks, the OMOU problem is to obtain an optimal computational offloading profile for different types of computational tasks of IoT MDs, while minimizing the overall computational overhead in terms of processing time and energy consumption subject to the wireless channel constraints of the MBS and SCs.

[0087] Obtaining an optimal computational offloading profile for different types of computational tasks of IoT MDs based on the OMOU problem satisfies:

[0088]

[0089] In the formula, the constraint condition C1 is used to ensure that the total number of tasks offloaded to the MEC server connected to the MBS is less than the number of wireless channels owned by the MBS, the constraint condition C2 is used to ensure that the total number of tasks offloaded to a certain SC k (1 ≤ k ≤ S) is not greater than the number of wireless channels owned by SC k, and for the constraint condition C3, any one of the S + 2 offloading decision choices can be selected for the type-j task of MD i. For example, local CPU computing of MD i, offloading to the MEC server connected to the MBS, offloading to the MEC server connected to the first SC, offloading to the MEC server connected to the second SC, etc.

[0090] To solve the above OMOU problem, all possible offloading decisions for all MDs tasks should be examined, and a globally optimal offloading decision profile should be selected for the tasks to achieve the overall minimum computational overhead while satisfying the given wireless channel constraints. In some embodiments, by an enumeration method, namely the Optimal Enumeration Offloading Algorithm (OEOA), all optional computational offloading decisions for all tasks are enumerated, and the offloading decision profile selected for the tasks of all MDs includes:

[0091] Enumerate all optional offloading decision combinations for all MD computational tasks;

[0092] Discard the offloading decision combinations that do not satisfy the given wireless channel constraints of the MBS and SCs;

[0093] Calculate the overall computational overhead of the remaining offloading decision combinations to determine the overall offloading decision combination S with the minimum overall computational overhead Ψmin.

[0094] Enumerate all optional offloading decision choices for all MD computational tasks and traverse the entire solution space. Therefore, if there is an optimal solution to this problem, the optimal solution to the OMOU problem can surely be obtained. Considering that OEOA has a very high computational complexity and cannot handle a large number of computational tasks, OEOA is only used as a benchmark to verify the effectiveness of the subsequent solutions.

[0095] Since OEOA is not feasible in the practical application of a large number of computational tasks, game theory, as an effective method, is widely used to solve the decision-making problems among multiple game players with different objectives. In a finite-player game, at each step, a rational player will react to the actions of other players in the previous step and make a locally optimal decision. After several steps, these players self-organize into a mutually balanced state, namely the Nash equilibrium, in which no player can further reduce his cost by unilaterally changing his strategy. In some embodiments, according to the selected offloading decision profile, using the greedy approximation offloading method of game theory (GT-GAOA), a globally optimal offloading decision profile is determined for all MDs by adopting a stochastic game theory strategy in the outer layer of GT-GAOA, and a locally optimal offloading strategy is determined for different types of computational tasks of MDi by adopting a greedy approximation method in the inner layer, including:

[0096] Regard the IoT MD as a rational game player and model OMOU as a strategic game where the set N of MDs is the set of game players, is the strategy set of game player i, Ψi (Λ i , Λ -i ) is the cost function that the player i needs to minimize.

[0097] After transforming the OMOU problem into a multi-user computing offloading game problem, the OMOU problem can be solved.

[0098] All MDs first calculate their local optimal offloading decision profiles in future time slots by calling, that is, obtaining the optimal associated SC k for any MD i and the computing offloading decision list for all its tasks with the minimum computing offloading overhead. Then, the MDs decide whether their offloading profiles should be updated. If the answer is yes, then these MDs will compete for the opportunity to update their offloading decision profiles, otherwise, on the contrary, only one MD can obtain the update opportunity in each time slot, such as MD j. Then, MD j updates its offloading decision profile and broadcasts the number of tasks to be offloaded to the edge server connecting the MBS and the SC. At the same time, the offloading decision profiles of other MDs remain unchanged. According to the existence theorem of Nash equilibrium, our multi-user computing offloading game, as a finite-player game, where each MD can choose a pure strategy from a finite set of offloading strategies, has a Nash equilibrium.

[0099] Solving the strategy game Thus, a global optimal offloading decision profile is determined for all MDs by adopting a stochastic game theory strategy in the outer layer of GT-GAOA, and a local optimal offloading strategy is determined for different types of computing tasks of MDi by adopting a greedy approximation method in the inner layer.

[0100] The method of this application proposes an optimal enumeration offloading scheme to solve the optimal MECO problem in the ultra-dense IoT network, which can achieve the optimal solution of the problem. This application further proposes a two-layer game theory greedy approximation offloading scheme with higher computing efficiency, in which collaborative computing offloading is carried out among IoT MDs, the edge server connecting the MBS, and the SCs.

[0101] The embodiment of this application also proposes an ultra-dense IoT computing offloading system based on edge computing, including a processor and a memory, and a computer program is stored on the memory. When the computer program is executed by the processor, the steps of the ultra-dense IoT computing offloading method based on edge computing as described above are implemented.

[0102] It should be noted that in the embodiments of the present application, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of additional identical elements in the process, method, article or device including that element.

[0103] The serial numbers of the embodiments of the present application above are only for description and do not represent the superiority or inferiority of the embodiments.

[0104] Through the description of the above embodiments, those skilled in the art can clearly understand that the above-described embodiment methods can be implemented by means of software plus a necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions for causing a terminal (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of the present application.

[0105] The embodiments of the present application have been described above in conjunction with the accompanying drawings. However, the present application is not limited to the above specific embodiments. The above specific embodiments are merely illustrative and not restrictive. Under the inspiration of the present application, those of ordinary skill in the art can also make many forms without departing from the purpose of the present application and the scope protected by the claims. All of these are within the protection scope of the present application.

Claims

1. A method for offloading ultra-dense Internet of Things computing based on edge computing, characterized in that: include: In the multi-user ultra-dense edge server scenario, a local execution model and a mobile edge execution model are constructed, where N co-located IoT MDs are alternately covered by MBS and SSC in the multi-user ultra-dense edge server scenario; Determine the optimal mobile edge computing offloading model based on the constructed local execution model and mobile edge execution model; By enumerating all optional computational offloading decisions for all tasks, and selecting offloading decision profiles for tasks of all mobile devices MDs, such that the offloading decision profiles have an overall minimum computational overhead while satisfying given wireless channel constraints; According to the selected offloading decision profile, the greedy approximate offloading method of game theory is used. In the outer layer of the greedy approximate offloading method (GT-GAOA), a random game theory strategy is adopted to determine an overall optimal offloading decision profile for all MDs, and in the inner layer, a greedy approximate method is used to determine a local optimal offloading strategy for different types of computing tasks of MDi.

2. The ultra-dense Internet of Things computing offloading method based on edge computing according to claim 1, characterized in that: In the multi-user ultra-dense edge server scenario, for IoT mobile devices IoT MD with M types of computing tasks, in the ultra-dense network, dual connectivity or CoMP technology is used to achieve communication with MBS, and each type of computing task is atomic and cannot be subdivided; For any type-j computing task T of MD i i,j , satisfying T i,j =(m i,j ,c i,j ), where m i,j It is task T i,j The input data size, c i,j Yes Complete T i,j The number of CPU cycles required; Define that MBS and each SC have K and L orthogonal radio frequency channels respectively, let λ i,j ∈{0,-1,1,2,...,S} represents the offloading decision of the j-th task at MD i.

3. The ultra-dense Internet of Things computing offloading method based on edge computing as claimed in claim 2, characterized in that: Building a local execution model and a mobile edge execution model includes: Build a local execution model: In the case where MD i is the local execution of the jth type of task, that is, λ i,j = 0, calculate its energy consumption and T i,j The processing time meets: Among them, f i and δ i They represent the computing power of MD i and the energy consumed per CPU cycle, are the energy consumption and processing time of executing the jth type of task locally, respectively; Building a mobile edge execution model: In the task T i,J When offloading to the MEC server connected to the MBS, the processing time of the type-j task of MD i mainly includes: task T i,J Transmission time from MD i to MBS via wireless access, T i,J The transmission time from MBS to MEC server through the optical fiber link, task T i,J The queuing time and execution time on the MEC server, in this case the processing time satisfies: in, To transmit T from MD i via the wireless access channel i,J The uplink data rate to MBS is T i,J Uplink data rate, and Task T i,J Queueing time and execution time on MEC servers.

4. The ultra-dense Internet of Things computing offloading method based on edge computing as claimed in claim 3, characterized in that: Building a mobile edge execution model also includes: Set the task T on the MEC server i,J The arrival and execution of tasks are modeled as an M / M / 1 queue. The service time on the MEC server follows an exponential distribution and the average arrival rate of tasks on the MEC server is calculated. satisfy: Calculate the completed task T i,J The average service time meets: The energy consumed in this case is expressed as: The computational cost satisfies: 。 5. The ultra-dense Internet of Things computing offloading method based on edge computing as claimed in claim 4, characterized in that: According to the constructed local execution model and mobile edge execution model, the optimal mobile edge computing offloading model is determined including: Based on the optimal mobile edge computing offloading OMOU problem, an optimal computing offloading profile is obtained for different types of computing tasks in the Internet of Things MD to meet the following requirements: Wherein, constraint C1 is used to ensure that the total number of tasks offloaded to the MEC server connected to the MBS is less than the number of wireless channels owned by the MBS, constraint C2 is used to ensure that the total number of tasks offloaded to a certain SC k (1≤k≤S) is not greater than the number of wireless channels owned by SC k, and constraint C3 is used for the type-j task of MD i to select any of the S+2 offloading decision options.

6. The ultra-dense Internet of Things computing offloading method based on edge computing according to claim 5, characterized in that: Through the enumeration method, all optional computation offloading decisions for all tasks are enumerated, and the offloading decision contours for the tasks of all MDs are selected, including: Enumerate all optional offloading decision combinations for all MD computing tasks; Discard the offloading decision combinations that do not satisfy the given wireless channel constraints of MBS and SCs; The overall computational cost of the remaining offloading decision combinations is calculated to determine the overall offloading decision combination S with the minimum overall computational cost Ψmin.

7. The ultra-dense Internet of Things computing offloading method based on edge computing according to claim 6, characterized in that: According to the selected offloading decision profile, the greedy approximate offloading method of game theory is used to determine an overall optimal offloading decision profile for all MDs using a random game theory strategy in the outer layer of GT-GAOA, and a greedy approximate method is used in the inner layer to determine a local optimal offloading strategy for different types of computing tasks of MDi, including: Consider IoT MD as a rational player and model OMOU as a strategic game The set N of MDs is the set of people in the game. is the strategy set of player i in the game, Ψ i (Λ i ,Λ -i ) is the cost function that player i needs to minimize; Solving Strategy Games The random game theory strategy is adopted in the outer layer of GT-GAOA to determine an overall optimal offloading decision contour for all MDs, and the greedy approximation method is adopted in the inner layer to determine a local optimal offloading strategy for different types of computing tasks of MDi.

8. Ultra-dense IoT computing offloading system based on edge computing, characterized in that: It comprises a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the steps of the ultra-dense Internet of Things computing offloading method based on edge computing as described in any one of claims 1 to 7 are implemented.

Images