Dynamic unloading method and device based on multi-user collaborative mobile edge computing system

Abstract

The present invention relates to a dynamic unloading method and system based on a multi-user collaborative mobile edge computing system. The method comprises: modeling based on the multi-user collaborative mobile edge computing system, establishing a system channel state model and a multi-user collaborative task model, and establishing a delay model and an energy model of the unloading task accordingly; using a topological sorting algorithm to sort the tasks in the multi-user collaborative task model to generate a task chain to be decided; based on the delay model and the energy model, referencing an attention mechanism to make an unloading decision on the task chain to be decided, and generating an unloading decision chain. The method of the present invention models multiple unloading tasks with task dependencies as a directed acyclic graph, topologically sorts the directed acyclic graph through a sorting algorithm, and proposes to refer to an attention mechanism to predict the unloading results of the task sequence, which can dynamically make unloading decisions according to different task dependencies and task amounts, effectively improving the unloading efficiency and unloading flexibility.

Description

Technical Field

[0001] The present invention relates to the technical field of dynamic offloading of mobile edge computing, and in particular to a dynamic offloading method and device based on a multi-user collaborative mobile edge computing system. Background Art

[0002] Mobile Edge Computing (MEC) is considered to be a promising technology in the development of wireless networks. MEC is an open architecture and platform that can be flexibly deployed in different locations to meet different services and needs. To this end, many researchers are committed to research in different aspects to give full play to the functions of mobile edge computing systems. For different offloading scenarios, the offloading environment is very different. For example, in some remote areas where there is a lack of infrastructure such as base stations, mobile edge computers can be deployed on satellites to assist ground communications. The transmission delay between satellites and the ground is large, but compared with ground facilities in remote areas, there are more abundant computing resources. For urban areas, computing resources are abundant, but different mobile edge computers have different load rates and delays, which need to be considered comprehensively.

[0003] For different deployment scenarios of MEC servers, most of the current research is based on the premise that the terminals are isolated from each other, each user has his own independent tasks, and there is no dependency between users. Therefore, when building the mobile edge computing system, the order of computing offload tasks and the interdependence between tasks are not considered. In fact, in real scenarios, due to the complex types of Internet services, the dependencies between services are highly correlated with different user identities (such as merchants and consumers, remote cooperation projects, etc.). Therefore, how to coordinate the scheduling of multiple users with task dependencies in real computing offload scenarios is a real problem that needs to be solved. Summary of the invention

[0004] The purpose of the present invention is to provide a dynamic unloading method and device based on a multi-user collaborative mobile edge computing system, so as to make dynamic task unloading decisions according to the dependencies between tasks and the task volume, and effectively improve the unloading efficiency and unloading flexibility of the mobile edge computing system.

[0005] To achieve the above object, the present invention provides the following solutions:

[0006] A dynamic unloading method based on a multi-user collaborative mobile edge computing system, comprising:

[0007] Modeling is performed based on a multi-user collaborative mobile edge computing system to establish a system channel state model and a multi-user collaborative task model;

[0008] Establishing a delay model and an energy model of the offloading task according to the system channel state model and the multi-user collaborative task model;

[0009] Using a topological sorting algorithm to sort the tasks in the multi-user collaborative task model to generate a task chain to be decided;

[0010] Based on the delay model and energy model of the offloaded task, an attention mechanism is used to make an offload decision on the task chain to be decided, and an offload decision chain is generated.

[0011] Optionally, the modeling based on the multi-user collaborative mobile edge computing system to establish a system channel state model specifically includes:

[0012] Establish a multi-user collaborative mobile edge computing system; the multi-user collaborative mobile edge computing system includes a ground layer, an air layer and a satellite layer; the ground layer includes a user terminal with an offloading task; the air layer includes an air base station with a task collection function; the satellite layer includes a LEO satellite on which a MEC server is deployed;

[0013] Establishing the system channel state model of the multi-user collaborative mobile edge computing system and Where R n，u and R n，d represent the uplink and downlink data rates of task n respectively; B n represents the transmission bandwidth allocated to task n; and They represent the uplink and downlink power margins of task n respectively.

[0014] Optionally, the modeling based on the multi-user collaborative mobile edge computing system to establish a multi-user collaborative task model specifically includes:

[0015] The ground offloading task in the multi-user collaborative mobile edge computing system is modeled as a directed acyclic graph G=<T,E> As the multi-user collaborative task model, wherein T represents the task, and E represents the dependency relationship between tasks.

[0016] Optionally, establishing a delay model of the offloading task according to the system channel state model and the multi-user collaborative task model specifically includes:

[0017] Determine the uplink transmission delay between the ground user and the satellite according to the system channel state model and downlink transmission delay Where S n represents the data packet size of task n; η represents the ratio of the data size of the calculation result of task n to the data size before calculation;

[0018] According to the uplink transmission delay between the ground user and the satellite and downlink transmission delay Determine the actual uplink transmission delay during the transmission process and downlink transmission delay And the processing delay of task n where α n and β n It represents the decision made by the aerial base station on task n after collecting user tasks; represents the delay of task n when it is calculated in the user terminal; represents the delay of task n when it is calculated on the MEC server deployed by the satellite;

[0019] According to the actual uplink transmission delay during the transmission process Downlink transmission delay And the processing delay of task n Building a time delay model for offloading tasks RT n Indicates the actual completion time of task n.

[0020] Optionally, establishing an energy model of the offloading task according to the system channel state model and the multi-user collaborative task model specifically includes:

[0021] According to the data packet size S of the task n n Determine the energy consumption of task n when computing at the user terminal and at the MEC server deployed on the satellite and Where ∈ represents the processing density; represents the unit energy consumption when the user terminal performs task n; It represents the unit energy consumption of the satellite-deployed MEC server when executing task n;

[0022] Energy consumption when computing task n at the user terminal and at the MEC server deployed on the satellite and Building an energy model for offloading tasks Among them, RE n represents the actual energy consumption of task n.

[0023] Optionally, the adopting of a topological sorting algorithm to sort the tasks in the multi-user collaborative task model to generate a task chain to be decided specifically includes:

[0024] A topological sorting algorithm is used to sort the tasks in the multi-user collaborative task model, and the two-dimensional directed acyclic graph is converted into a one-dimensional linear sequence as the task chain to be decided.

[0025] Optionally, based on the delay model and energy model of the offloaded task, the attention mechanism is used to perform seq2seq offload decision on the task chain to be decided, and an offload decision chain is generated, which specifically includes:

[0026] A seq2seq model with an attention mechanism is referenced, and an encoder in the seq2seq model encodes the task chain to be decided into a semantic vector based on a delay model and an energy model of the offloaded task;

[0027] The decoder in the seq2seq model performs cyclic decoding according to the semantic vector to generate the offloading decision chain.

[0028] A dynamic unloading device based on a multi-user collaborative mobile edge computing system, comprising:

[0029] The system model building module is used to build a model based on a multi-user collaborative mobile edge computing system, and to establish a system channel state model and a multi-user collaborative task model;

[0030] A delay and energy model building module, used to build a delay model and an energy model of the offloading task according to the system channel state model and the multi-user collaborative task model;

[0031] A task topology sorting module is used to sort the tasks in the multi-user collaborative task model using a topology sorting algorithm to generate a task chain to be decided;

[0032] The task offloading decision module is used to make an offloading decision on the task chain to be decided based on the delay model and energy model of the offloading task and to generate an offloading decision chain by referencing the attention mechanism.

[0033] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0034] The present invention provides a dynamic unloading method and system based on a multi-user collaborative mobile edge computing system, the method comprising: modeling based on the multi-user collaborative mobile edge computing system, establishing a system channel state model and a multi-user collaborative task model; establishing a delay model and an energy model of the unloading task according to the system channel state model and the multi-user collaborative task model; using a topological sorting algorithm to sort the tasks in the multi-user collaborative task model to generate a task chain to be decided; based on the delay model and energy model of the unloading task, referencing an attention mechanism to make an unloading decision on the task chain to be decided, and generating an unloading decision chain. The method of the present invention models multiple unloading tasks with task dependencies as a directed acyclic graph, topologically sorts the directed acyclic graph through a sorting algorithm, and proposes to use an attention mechanism-based method to predict the unloading results of the task sequence, so that the multi-task mobile edge computing system can dynamically process according to different task dependencies and task amounts, effectively improving the unloading efficiency and unloading flexibility of the mobile edge computing system. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0036] Figure 1 This is a flow chart of a dynamic unloading method based on a multi-user collaborative mobile edge computing system according to an embodiment of the present invention;

[0037] Figure 2 A schematic diagram of a multi-user collaborative mobile edge computing system provided by an embodiment of the present invention;

[0038] Figure 3 A schematic diagram of task dependency modeled as a directed acyclic graph provided in an embodiment of the present invention;

[0039] Figure 4 A schematic diagram of a task offloading strategy prediction model based on an attention mechanism provided in an embodiment of the present invention;

[0040] Figure 5 This is a structural diagram of a dynamic offloading system based on a multi-user collaborative mobile edge computing system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0041] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0042] The purpose of the invention is to provide a dynamic unloading method and device based on a multi-user collaborative mobile edge computing system, so as to make dynamic task unloading decisions according to the dependencies between tasks and the task volume, and effectively improve the unloading efficiency and unloading flexibility of the mobile edge computing system.

[0043] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0044] Figure 1 This is a flow chart of a dynamic unloading method based on a multi-user collaborative mobile edge computing system according to an embodiment of the present invention. Figure 1 The present invention provides a dynamic unloading method based on a multi-user collaborative mobile edge computing system, comprising:

[0045] Step 101: Modeling is performed based on a multi-user collaborative mobile edge computing system to establish a system channel state model and a multi-user collaborative task model.

[0046] Step 101 performs modeling based on a multi-user collaborative mobile edge computing system, including modeling the channel state of the mobile edge computing system and modeling the dependency relationships of multiple user tasks.

[0047] Figure 2 A schematic diagram of a multi-user collaborative mobile edge computing system provided by an embodiment of the present invention. Figure 2 , the multi-user collaborative mobile edge computing offloading system with the background of the integration of the earth and the sky includes the ground layer, the air layer and the satellite layer. Among them, the air layer is composed of air base stations with the function of task collection, which is responsible for collecting ground tasks and selecting them for offloading, which can be executed on the local user terminal (UE) or on the MEC server (S-MEC) deployed on the LEO satellite. The satellite layer is composed of LEO satellites (referred to as satellites) deploying MEC servers. The ground layer is composed of user terminals with offloading tasks (referred to as users). There is a mutual dependency between offloading tasks, which are modeled as a directed acyclic graph. The subtask can only be started when the parent task is completed.

[0048] First, the unloading system is modeled. The aerial base station is located at a lower position than the ground. In this model, the transmission delay between the aerial base station and the ground user is not considered for the sake of simplicity. c represents the speed of light, f cis the carrier frequency, h is used to represent the distance between the LEO satellite and the ground user task n, r is used to represent the distance between the location of task n and the center of the satellite coverage area, then the transmission path loss between the satellite and the ground can be expressed as:

[0049]

[0050] The transmission bandwidth allocated to task n can be expressed as:

[0051] B n =ρBξ n (2)

[0052] Where ρ is the throughput efficiency coefficient, B is the total bandwidth of satellite transmission, ξ n represents the channel utilization function, where ξ n It is a decreasing function of the number of tasks N.

[0053] The task uplink and downlink power headroom is expressed as:

[0054]

[0055]

[0056] in and represent the uplink and downlink transmission power of task n respectively, G sa represents the power gain, and represent the uplink and downlink noise power of task n respectively.

[0057] Therefore, the mission uplink and downlink data rates can be expressed as:

[0058]

[0059]

[0060] Different scenarios have different channel states and different path losses, so they need to be modeled according to the actual offloading system. At the same time, the corresponding task relationships under different models are also different. For offloading task n, different tasks have a mutually dependent relationship. It is modeled as a directed acyclic graph, and the execution of subtasks depends on the results of the parent task. Only when the parent task is completed can it be executed. In addition, there is no direct correspondence between the number of users and the number of tasks. A user may have multiple tasks and be associated with the tasks of other users. Therefore, the present invention uses the number of tasks as the main calculation standard.

[0061] Taking a group of tasks with task association as an example, the ground tasks are modeled as a directed acyclic graph G =<T,E> As a multi-user collaborative task model, T represents the task set, E represents the dependency set between tasks, |T| = N, where N represents the total number of tasks in the directed acyclic graph. Each directed edge in E is represented by Indicates that t i represents the parent task, t j Represents a subtask, the calculation of which depends on the calculation result of the parent task.

[0062] Figure 3 The task dependency diagram modeled as a directed acyclic graph provided in the embodiment of the present invention. The modeling of ground tasks needs to be modeled according to the specific task conditions collected by the air base station, such as Figure 3 As shown, the task is modeled as a directed acyclic graph, G =<T,E> , where |T| in the three task graphs is 3, 5, and 8 respectively, and each directed edge in E is Indicates that t i represents the parent task, t j Indicates a subtask. For example Figure 3 In (2), t1 is the parent task of t3, and t3 is the parent task of t5. The child task can only start executing when the parent task is completed.

[0063] Therefore, the step 101 is based on the multi-user collaborative mobile edge computing system to establish a system channel state model, which specifically includes:

[0064] Establish a multi-user collaborative mobile edge computing system; the multi-user collaborative mobile edge computing system includes a ground layer, an air layer and a satellite layer; the ground layer includes a user terminal with an offloading task; the air layer includes an air base station with a task collection function; the satellite layer includes a LEO satellite on which a MEC server is deployed;

[0065] Establishing the system channel state model of the multi-user collaborative mobile edge computing system and Where R n，u and R n，d represent the uplink and downlink data rates of task n respectively; B n represents the transmission bandwidth allocated to task n; and They represent the uplink and downlink power margins of task n respectively.

[0066] The ground offloading task in the multi-user collaborative mobile edge computing system is modeled as a directed acyclic graph G=<T,E> As the multi-user collaborative task model, wherein T represents the task, and E represents the dependency relationship between tasks.

[0067] Step 102: Establish a delay model and an energy model of the offloading task according to the system channel state model and the multi-user collaborative task model.

[0068] Step 102 models the energy and delay model of the user task, wherein the transmission delay needs to fully consider the transmission delay of the predecessor task set and the maximum transmission delay of the system, and the energy consumption is the total energy consumption of the system.

[0069] Since the factors that affect the offloading decision need to be considered separately in different scenarios, the influence of latency and energy consumption on the offloading decision needs to be fully considered to assign appropriate weights. In the process of modeling as a directed acyclic graph for task offloading, the subtask must obtain the calculation result with the largest latency among all parent tasks before it can be executed. If the number of subtasks of a task is zero, it is called an exit task, and the maximum task execution time of the entire directed acyclic graph is the maximum exit task completion time. The energy model has a certain impact on the offloading decision-making process, but it is not the ultimate goal of system optimization.

[0070] The process of establishing the delay model is as follows.

[0071] Use S n Indicates the packet size (Mbits) of task n. and They represent the computing power allocated to task n by the satellite-deployed MEC server (S-MEC) and the user terminal UE respectively.

[0072] The uplink transmission delay and downlink transmission delay between the ground user and the satellite are:

[0073]

[0074]

[0075] Where η represents the ratio of the data size of the calculation result of task n to the data size before calculation. The delay of the task in S-MEC calculation is expressed as:

[0076]

[0077] The delay of the task when it is calculated on the UE side can be expressed as:

[0078]

[0079] Use α n and β n It represents the decision made by the aerial base station on task n after collecting user tasks. When the decision task is processed locally, α n =1; decision is made when S-MEC is processing β n=1, where α n , β n ∈{0,1},α n +β n =1.

[0080] The actual uplink transmission delay and downlink transmission delay of task n during the transmission process can be expressed as:

[0081]

[0082]

[0083] The processing delay of task n is expressed as:

[0084]

[0085] Then the actual completion time of task n can be expressed as:

[0086]

[0087] The completion time of task n can be expressed as:

[0088]

[0089] in, Indicates the maximum completion time of the parent task of task n.

[0090] Formula (14) is the established delay model of offloading task n.

[0091] The process of establishing the system energy model is as follows.

[0092] use represents the unit energy consumption when the user terminal performs task n, represents the unit energy consumption when S-MEC executes task n, ∈ represents the processing density, then the energy consumption of the task when the local UE and S-MEC are calculated can be expressed as:

[0093]

[0094]

[0095] Therefore, after the aerial base station makes a decision, the actual energy consumption of task n can be expressed as:

[0096]

[0097] Formula (18) is the established energy model of offloading task n.

[0098] Therefore, the step 102 establishes a delay model of the offload task according to the system channel state model and the multi-user collaborative task model, and specifically includes:

[0099] Determine the uplink transmission delay between the ground user and the satellite according to the system channel state model and downlink transmission delay Where S n represents the data packet size of task n; η represents the ratio of the data size of the calculation result of task n to the data size before calculation;

[0100] According to the uplink transmission delay between the ground user and the satellite and downlink transmission delay Determine the actual uplink transmission delay during the transmission process and downlink transmission delay And the processing delay of task n where α n and β n It represents the decision made by the aerial base station on task n after collecting user tasks; represents the delay of task n when it is calculated in the user terminal; represents the delay of task n when it is calculated on the MEC server deployed by the satellite;

[0101] According to the actual uplink transmission delay during the transmission process Downlink transmission delay And the processing delay of task n Building a time delay model for offloading tasks RT n represents the actual completion time of task n;

[0102] According to the data packet size S of the task n n Determine the energy consumption of task n when computing at the user terminal and at the MEC server deployed on the satellite and Where ∈ represents the processing density; represents the unit energy consumption when the user terminal performs task n; It represents the unit energy consumption of the satellite-deployed MEC server when executing task n;

[0103] Energy consumption when computing task n at the user terminal and at the MEC server deployed on the satellite and Building an energy model for offloading tasks Among them, RE n represents the actual energy consumption of task n.

[0104] Step 103: Use a topological sorting algorithm to sort the tasks in the multi-user collaborative task model to generate a task chain to be decided.

[0105] Step 103 topologically sorts the directed acyclic graph and converts the complex task graph with task dependencies into a linear sequence. That is, a topological sorting algorithm is used to sort the tasks in the multi-user collaborative task model and convert the two-dimensional directed acyclic graph into a one-dimensional linear sequence as the task chain to be decided.

[0106] The specific algorithm for sorting a directed acyclic graph (DAG) using a topological sorting algorithm is as follows:

[0107]

[0108] Through the above algorithm 1, no matter how many tasks there are, they can be converted into a one-dimensional linear sequence X as the task chain to be decided, which is convenient for seq2seq calculation.

[0109] The step 103 converts the two-dimensional task dependency graph into a one-dimensional task sequence, which can effectively reduce the computational complexity.

[0110] As a specific embodiment, by applying Algorithm 1 to perform topological sorting, Figure 3 The three directed acyclic graphs can be represented as: (1) {t1, t2, t3}; (2) {t1, t2, t4, t3, t5}; (3) {t1, t2, t3, t4, t5, t6, t7, t8}. After the directed acyclic graphs are topologically sorted, they can be reassigned to the sequence X = {x1, x2, ..., x i}.like Figure 3 (1) can be re-expressed as {x1, x2, x3}, Figure 3 (2) can be re-expressed as {x1, x2, x3, x4, x5}, Figure 3 (3) can be re-expressed as {x1, x2, x3, x4, x5, x6, x7, x8}. In which, each state x i It contains the state information of the task to facilitate the seq2seq operation based on the attention mechanism.

[0111] Step 104: Based on the delay model and energy model of the offloaded task, an attention mechanism is used to make an offload decision on the task chain to be decided, and an offload decision chain is generated.

[0112] Step 104 uses the attention mechanism to make an unloading decision on the task sequence, and introduces the seq2seq model of the attention mechanism, whose input end is the task chain that needs to be decided (i.e., the task chain to be decided). After the calculation of the RNN neural network inside seq2seq, the output end is the unloading decision chain made.

[0113] Figure 4 A schematic diagram of a task offloading strategy prediction model (seq2seq model) based on an attention mechanism provided in an embodiment of the present invention. Since the size and complexity of the directed acyclic graph of tasks are different, the length of the sequence generated after topological sorting is also different. Therefore, a seq2seq model with an attention mechanism is referenced to generate decision queues of different lengths according to the number of tasks.

[0114] See also Figure 4 , seq2seq is a kind of Encoder-Decoder structure. The task sequence after topological sorting is input into the seq2seq network based on the attention mechanism for calculation. Among them, the encoder Encoder encodes all input vectors into a unified semantic vector C, which is then decoded by the decoder Decoder.

[0115] The unified semantic vector C means that the processing weights of all inputs are the same. The attention mechanism cited in the present invention expands the semantic representation C and defines C as C = {c1, c2, ..., c i},c i Different representations have different processing weights on the input vector, indicating different concentrations of attention.

[0116] In the present invention, the attention mechanism focuses on the information that is more critical to the current task, such as latency or energy consumption, among a large amount of input information, and reduces the attention paid to other information.

[0117] In the Encoder part, the hidden layer state h i According to the state h of the previous step i-1 and the task input state x i It turns out that:

[0118] h i =f1(h i-1 , x i ) (19)

[0119] Where f1 is the conversion function of Encoder; x i is the i-th element in the task chain X to be decided, x i The relationship between the delay model and the energy consumption model can be expressed as x i = {RE i ,RT i},REi and RT i are the actual completion time and actual energy consumption of task i, respectively.

[0120] During the decoding process of the Decoder, the output of the previous state is continuously used as the input of the next state, and the information of all hidden layers is Perform weighted summation and cyclic decoding until the output is a stop symbol, and finally output the offloading decision chain.

[0121] Decoder will input a semantic vector c each time i , the new state of the hidden layer According to the status of the previous step and a i 、c i A nonlinear function f2 of the three gives:

[0122]

[0123] From this formula, we can conclude that and a i are the previous state of the Decoder stage and the predicted value of the previous output, respectively. i Corresponding to the probability of attention allocation in different states.

[0124] At the decoder end, since the output of the previous state affects the input of the next state, the decision y at the output end i All previous output decisions and semantic vector c i Jointly determined, and distance y i The closer the decision is to y i The greater the influence of i The further the decision is, the more important it is for y i The smaller the impact of . That is, the process of generating target offloading decisions can be expressed as follows:

[0125] y1=f2(c1) (21)

[0126] y2=f2(c2,y1) (22)

[0127] y3=f2(c3,y1,y2) (23)

[0128] …

[0129] y i =f2(c i , y1, y2, ..., y i-1 ) (twenty four)

[0130] Where f2 is the nonlinear transformation function of the Decoder.i The probability of attention allocation corresponding to different states is calculated as follows:

[0131]

[0132] Where N represents the total number of tasks in the directed acyclic graph, a ij For each h j The corresponding weight value, f1 represents the conversion function of Encoder, RE j and RT j are the actual completion time and the actual energy consumption of task j respectively. In the present invention:

[0133] h j =f1(h j-1 ,RT j ,RE j ) (26)

[0134] a ij It can be regarded as a probability, reflecting h j C i The importance of can be expressed using softmax:

[0135]

[0136] Where N represents the length of the task sequence, e ij The calculation formula is:

[0137] e ij =v T tanh(W a x i-1 +U a h j ) (28)

[0138] Among them, v, W a and U a are the model parameters to be learned.

[0139] Therefore, the step 104 refers to the attention mechanism to perform seq2seq offloading decision on the task chain to be decided based on the delay model and energy model of the offloading task, and generates an offloading decision chain, which specifically includes:

[0140] A seq2seq model with an attention mechanism is referenced, and an encoder in the seq2seq model encodes the task chain to be decided into a semantic vector based on a delay model and an energy model of the offloaded task;

[0141] The decoder in the seq2seq model performs cyclic decoding according to the semantic vector to generate the offloading decision chain.

[0142] The present invention provides a dynamic unloading method based on a multi-user collaborative mobile edge computing system, which transforms the research focused on isolated unloading tasks of a single user into multi-user collaborative task unloading, and combines the attention mechanism in the field of natural language processing to use the attention mechanism to predict unloading actions. The method effectively reduces the complexity by topologically sorting the directed acyclic graph of tasks, while breaking the limitation of task interdependence; by combining the attention mechanism for action prediction, the limitation on the number of tasks with dependencies can be broken; and the introduction of the attention mechanism can focus on the more critical information of the current task, reduce the attention to other information, and effectively solve the problem of information overload. Therefore, the method of the present invention can effectively improve the unloading efficiency and improve the unloading flexibility of the mobile edge computing system.

[0143] Based on the method provided by the present invention, the present invention also provides a dynamic unloading device based on a multi-user collaborative mobile edge computing system, see Figure 5 , the system comprising:

[0144] The system model building module 501 is used to build a system channel state model and a multi-user collaborative task model based on a multi-user collaborative mobile edge computing system;

[0145] A delay and energy model building module 502, used to build a delay model and an energy model of the offloading task according to the system channel state model and the multi-user collaborative task model;

[0146] The task topology sorting module 503 is used to sort the tasks in the multi-user collaborative task model using a topology sorting algorithm to generate a task chain to be decided;

[0147] The task offloading decision module 504 is used to make an offloading decision on the task chain to be decided based on the delay model and energy model of the offloading task and to generate an offloading decision chain by referencing the attention mechanism.

[0148] The beneficial effects of the dynamic unloading method and device based on a multi-user collaborative mobile edge computing system of the present invention are:

[0149] 1. The present invention introduces a directed acyclic graph and performs topological sorting, so that the mobile edge computing system is no longer restricted by the limitations of task dependencies, and transforms the two-dimensional task relationship graph into a one-dimensional sequence, effectively expanding the service scope of the mobile edge computing system.

[0150] 2. The present invention models multiple user tasks with task dependencies as a directed acyclic graph, topologically sorts the directed acyclic graph through a sorting algorithm, and proposes a method based on an attention mechanism to predict the offloading results of the task sequence, so that the multi-task mobile edge computing system can dynamically process tasks according to different task dependencies and task amounts, effectively improving the ability of the mobile edge computing system to handle dynamic tasks, effectively improving the offloading flexibility of the mobile edge computing system, and reducing the complexity of the problem.

[0151] 3. The present invention introduces a seq2seq model combined with an attention mechanism. On the one hand, it can get rid of the limitation of the number of tasks of different lengths and generate an action sequence of corresponding length according to the sequence length of the input state; on the other hand, the attention mechanism can allocate different attention according to the importance of different parts of the state information, focusing on more important information in the current state, and effectively improving the unloading efficiency and accuracy of the mobile edge computing system.

[0152] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0153] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A dynamic unloading method based on a multi-user collaborative mobile edge computing system, characterized in that: include: Modeling is performed based on a multi-user collaborative mobile edge computing system to establish a system channel state model and a multi-user collaborative task model; The multi-user collaborative mobile edge computing system is used to model and establish a multi-user collaborative task model, specifically including: The ground offloading task in the multi-user collaborative mobile edge computing system is modeled as a directed acyclic graph G=<T,E> As the multi-user collaborative task model; wherein T represents the task, and E represents the dependency relationship between tasks; Establishing a delay model and an energy model of the offloading task according to the system channel state model and the multi-user collaborative task model; Using a topological sorting algorithm to sort the tasks in the multi-user collaborative task model to generate a task chain to be decided; Based on the delay model and energy model of the offloading task, an attention mechanism is used to make an offloading decision on the task chain to be decided, and an offloading decision chain is generated; Based on the delay model and energy model of the offload task, the attention mechanism is used to perform seq2seq offload decision on the task chain to be decided, and an offload decision chain is generated, which specifically includes: A seq2seq model with an attention mechanism is referenced, and an encoder in the seq2seq model encodes the task chain to be decided into a semantic vector based on a delay model and an energy model of the offloaded task; The decoder in the seq2seq model performs cyclic decoding according to the semantic vector to generate the offloading decision chain.

2. The method according to claim 1, characterized in that The modeling is performed based on a multi-user collaborative mobile edge computing system to establish a system channel state model, specifically including: Establish a multi-user collaborative mobile edge computing system; the multi-user collaborative mobile edge computing system includes a ground layer, an air layer and a satellite layer; the ground layer includes a user terminal with an offloading task; the air layer includes an air base station with a task collection function; the satellite layer includes a LEO satellite on which a MEC server is deployed; Establishing the system channel state model of the multi-user collaborative mobile edge computing system and Where R n，u and R n，d represent the uplink and downlink data rates of task n respectively; B n represents the transmission bandwidth allocated to task n; and They represent the uplink and downlink power margins of task n respectively.

3. The method according to claim 2, characterized in that The delay model of the offloading task is established according to the system channel state model and the multi-user collaborative task model, specifically including: Determine the uplink transmission delay between the ground user and the satellite according to the system channel state model and downlink transmission delay Where S n represents the data packet size of task n; η represents the ratio of the data size of the calculation result of task n to the data size before calculation; According to the uplink transmission delay between the ground user and the satellite and downlink transmission delay Determine the actual uplink transmission delay during the transmission process and downlink transmission delay And the processing delay of task n where α n and β n It represents the decision made by the aerial base station on task n after collecting user tasks; represents the delay of task n when it is calculated in the user terminal; represents the delay of task n when it is calculated on the MEC server deployed by the satellite; According to the actual uplink transmission delay during the transmission process Downlink transmission delay And the processing delay of task n Building a time delay model for offloading tasks RT n Indicates the actual completion time of task n.

4. The method according to claim 3, characterized in that The step of establishing an energy model of the offloading task according to the system channel state model and the multi-user collaborative task model specifically includes: According to the data packet size S of the task n n Determine the energy consumption of task n when computing at the user terminal and at the MEC server deployed on the satellite and Where ∈ represents the processing density; represents the unit energy consumption when the user terminal performs task n; It represents the unit energy consumption of the satellite-deployed MEC server when executing task n; Energy consumption when computing task n at the user terminal and at the MEC server deployed on the satellite and Building an energy model for offloading tasks Among them, RE n represents the actual energy consumption of task n.

5. The method according to claim 4, characterized in that The topological sorting algorithm is used to sort the tasks in the multi-user collaborative task model to generate a task chain to be decided, specifically including: A topological sorting algorithm is used to sort the tasks in the multi-user collaborative task model, and the two-dimensional directed acyclic graph is converted into a one-dimensional linear sequence as the task chain to be decided.

6. A dynamic unloading device based on a multi-user collaborative mobile edge computing system, characterized in that: include: The system model building module is used to build a model based on a multi-user collaborative mobile edge computing system, and to establish a system channel state model and a multi-user collaborative task model; The multi-user collaborative mobile edge computing system is used to model and establish a multi-user collaborative task model, specifically including: The ground offloading task in the multi-user collaborative mobile edge computing system is modeled as a directed acyclic graph G=<T,E> As the multi-user collaborative task model; wherein T represents the task, and E represents the dependency relationship between tasks; A delay and energy model building module, used to build a delay model and an energy model of the offloading task according to the system channel state model and the multi-user collaborative task model; A task topology sorting module is used to sort the tasks in the multi-user collaborative task model using a topology sorting algorithm to generate a task chain to be decided; A task offloading decision module is used to make an offloading decision on the task chain to be decided based on the delay model and energy model of the offloading task and to generate an offloading decision chain by referencing the attention mechanism; Based on the delay model and energy model of the offload task, the attention mechanism is used to perform seq2seq offload decision on the task chain to be decided, and an offload decision chain is generated, which specifically includes: A seq2seq model with an attention mechanism is referenced, and an encoder in the seq2seq model encodes the task chain to be decided into a semantic vector based on a delay model and an energy model of the offloaded task; The decoder in the seq2seq model performs cyclic decoding according to the semantic vector to generate the offloading decision chain.

Images