CN116471632A - Task migration method based on multi-point cooperation in mobile edge calculation - Google Patents
Task migration method based on multi-point cooperation in mobile edge calculation Download PDFInfo
- Publication number
- CN116471632A CN116471632A CN202310309080.8A CN202310309080A CN116471632A CN 116471632 A CN116471632 A CN 116471632A CN 202310309080 A CN202310309080 A CN 202310309080A CN 116471632 A CN116471632 A CN 116471632A
- Authority
- CN
- China
- Prior art keywords
- collaboration
- task
- decision
- user
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 48
- 230000005012 migration Effects 0.000 title claims abstract description 37
- 238000013508 migration Methods 0.000 title claims abstract description 37
- 238000004364 calculation method Methods 0.000 title claims abstract description 34
- 238000005457 optimization Methods 0.000 claims abstract description 19
- 238000004891 communication Methods 0.000 claims abstract description 13
- 230000006870 function Effects 0.000 claims description 25
- 230000009471 action Effects 0.000 claims description 22
- 230000008569 process Effects 0.000 claims description 21
- 230000005540 biological transmission Effects 0.000 claims description 19
- 230000008901 benefit Effects 0.000 claims description 18
- 238000003860 storage Methods 0.000 claims description 14
- 238000005265 energy consumption Methods 0.000 claims description 12
- 230000007246 mechanism Effects 0.000 claims description 8
- 230000000875 corresponding effect Effects 0.000 claims description 6
- 238000012549 training Methods 0.000 claims description 5
- 238000005070 sampling Methods 0.000 claims description 4
- 238000013528 artificial neural network Methods 0.000 claims description 3
- 238000011156 evaluation Methods 0.000 claims description 3
- 238000012546 transfer Methods 0.000 claims description 3
- 230000008859 change Effects 0.000 abstract description 13
- 238000010295 mobile communication Methods 0.000 abstract description 2
- 239000003795 chemical substances by application Substances 0.000 description 8
- 238000009826 distribution Methods 0.000 description 7
- 238000013468 resource allocation Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 230000011218 segmentation Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 2
- 230000007774 longterm Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 241000238371 Sepiidae Species 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 230000001413 cellular effect Effects 0.000 description 1
- 238000012733 comparative method Methods 0.000 description 1
- 230000001010 compromised effect Effects 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 239000002360 explosive Substances 0.000 description 1
- 238000005562 fading Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000002407 reforming Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Mobile Radio Communication Systems (AREA)
Abstract
The invention belongs to the technical field of mobile communication, and particularly relates to a task migration method based on multi-point cooperation in mobile edge calculation; the method comprises the following steps: constructing a network system model in an MEC scene; constructing a collaborative communication model, a task calculation model, a user mobility model and a load balancing model based on a network system model in an MEC scene; constructing an unloading decision and collaborative set association decision joint optimization problem according to a collaborative communication model, a task calculation model, a user mobility model and a load balancing model; node cooperation is carried out on a slow time scale for future load change, and a cooperation set is formed; solving an unloading decision and a collaboration set association decision joint optimization problem on a fast time scale according to the collaboration set to obtain an optimal unloading decision and a collaboration set association decision; the system carries out task migration according to the optimal unloading decision and the collaborative set association decision; the invention can effectively reduce the task mobility and the task execution time delay.
Description
Technical Field
The invention belongs to the technical field of mobile communication, and particularly relates to a task migration method based on multi-point cooperation in mobile edge calculation.
Background
With the continuous popularization of intelligent devices and the Internet of things, the demands of various new services on mobile data rates and server computing power are exponentially increased. The mobile edge computing (Mobile Edge Computing, MEC) sinks the computing resources and storage resources of the original cloud computing to the edge devices closer to the user side, enabling low latency and high reliability services to the user. In order to meet the explosive high-flow demand of users in a hot spot area, an ultra-dense edge computing network densely deploys a large number of base stations and servers in the hot spot area to provide services for the users. Due to the dense deployment of base stations and servers, the coverage area of the nodes is reduced, and in the calculation scenario of task transmission completed by a single node, a high-speed mobile user easily leaves the original unloading base station coverage area, which can cause frequent switching and task migration, so that tasks are interrupted or failed.
User mobility in the MEC environment is an important factor affecting user service quality, and problems such as task unloading failure, frequent switching, task interruption and the like may occur in the process of high-speed movement of the user. By considering the cooperation among a plurality of nodes, the wireless access points in the multi-point cooperation set simultaneously provide reliable communication service for users by means of the multi-point cooperation transmission technology, and the service range is enlarged. Task migration based on node cooperation in an MEC environment has been studied in many aspects, for example, aiming at the problem of frequent migration caused by high-speed movement of a user, a mobility sensing algorithm based on machine learning is provided, future service migration and node switching of the user are predicted, task migration is performed in advance, a multi-point cooperation set is established, and task interruption caused by user switching is reduced; aiming at the problem of resource allocation with CoMP receiving function in MEC environment, modeling the problem as a mixed integer nonlinear programming problem, a three-stage resource allocation algorithm is provided based on the concept of an interference diagram, and the three-stage resource allocation algorithm comprises the following steps: calculating resource allocation, subcarrier allocation, cell clustering, subcarrier re-use and cell clustering; aiming at the deployment problem of service function chains (Service Function Chain, SFC) in a 6G wireless edge network, the interference between service chains is reduced by utilizing a multipoint cooperation technology, and the SFC deployment problem with minimized long-term cost is solved by a deployment algorithm based on an Actor-Critic framework.
The above collaboration aggregation forming mechanism directly centering on the user needs to continuously predict the movement track of the user, but in a multi-user scene, the controller needs to collect and store a large amount of user track information, which brings great prediction cost. In addition, the collaboration set is built by taking the users as the center, the number of the collaboration sets can be increased along with the increase of the number of the users, and the calculation decision cost is too high.
Disclosure of Invention
Aiming at the defects existing in the prior art, the invention provides a task migration method based on multi-point cooperation in mobile edge calculation, which comprises the following steps:
s1: constructing a network system model in an MEC scene;
s2: constructing a collaborative communication model, a task calculation model, a user mobility model and a load balancing model based on a network system model in an MEC scene;
s3: constructing an unloading decision and collaborative set association decision joint optimization problem according to a collaborative communication model, a task calculation model, a user mobility model and a load balancing model;
s4: solving an unloading decision and collaboration set association decision joint optimization problem by adopting a two-time scale decision model to obtain an optimal unloading decision and collaboration set association decision; and the system performs task migration according to the optimal unloading decision and the collaborative set association decision.
Preferably, the network system model in the MEC scene specifically includes: u base stations and M users, wherein each base station is provided with an MEC server; the set of base stations is denoted asThe user set is expressed asThe task of the user is expressed as->Wherein (1)>Representing task data size, +.>Representing the number of CPU cycles required per bit of task, and (2)>Representing the maximum tolerable delay of a task, < >>Representing the moving speed of the user u in t time slots; user selection collaboration set +.>Offloading tasks, define->Representing collaboration set association variable, +.>Representing user selection in collaboration set +.>Execution task->Indicating that the user does not select the collaboration set +.>Definitions->Representing task offloading decision variables, +.>Indicating task offloading to edge side execution, +.>Indicating that the task is executing locally.
Preferably, the user mobility model includes: the residence time of the user is used as an evaluation index of whether the task can be successfully transmitted in the cell switching process; if the user remains in the collaboration set for a timeLess than task transfer time->The task transmission fails; the probability of task transmission failure is predicted based on the stay time.
Preferably, the load balancing model comprises:
calculating the predicted calculation resource occupation amount and the predicted storage resource occupation amount of the server by adopting the LSTM network, and calculating the expected load of the server according to the predicted calculation resource occupation amount and the predicted storage resource occupation amount; setting an expected load threshold, and taking the server as a high load node when the expected load of the server is larger than the expected load threshold, otherwise, taking the server as a low load node;
the high-load node is used for evacuating redundant computing tasks to other adjacent nodes and not receiving computing tasks of other nodes, and the low-load node is used for receiving computing loads of other nodes and does not need to unload computing tasks to other nodes; and sharing computing resources between the high-load nodes and the low-load nodes to form a collaboration set, and defining expected load deviation of the collaboration set.
Preferably, the unloading decision and collaborative set association decision joint optimization problem is expressed as:
wherein,,user delay cost representing successful execution of task for user u at time slot, < >>Represents a set of time slots, T represents a system time period, < ->Representing user set->Representing a collaboration set,/->Indicating the maximum tolerable delay of the task for user u in t time slots,/->Task offloading decision variable representing user u at time slot, < ->A collaboration set association variable representing t-slot user u,indicating that t-slot user u receives +.>Signal-to-interference-and-noise ratio, p 0 Representing the signal-to-interference-and-noise ratio threshold,representing t time slot user equipment energy consumption, +.>Representing the user equipment energy consumption budget.
Preferably, the process of solving the unloading decision and collaborative set association decision joint optimization problem comprises: the two-time scale decision model comprises a slow time scale model and a fast time scale model; the slow time scale model updates the collaboration set by adopting a collaboration mechanism based on the FCM algorithm and alliance game according to the load information, and the fast time scale model solves the optimal unloading decision and collaboration set association decision by adopting an Actor-Critic-based migration decision algorithm according to the collaboration set and user task information.
Further, the process of updating the collaboration set by adopting the collaboration mechanism based on the FCM algorithm and the alliance game comprises the following steps:
clustering all server nodes by adopting an FCM algorithm to obtain K collaboration sets;
repeatedly executing the merging and splitting processes on the K collaboration sets until the final collaboration set is not changed any more;
combining: setting a alliance utility function according to the expected load deviation of the collaboration set, comparing the pareto advantages of the combined set and the collaboration sets according to the alliance utility function, and taking the combined set as a new collaboration set when the pareto advantages of the combined set are larger than the pareto advantages of the collaboration sets, otherwise, reserving the collaboration sets; the merging set is a union set of a plurality of collaboration sets;
splitting: comparing the pareto advantages of the segmented sub-sets with those of the original collaboration sets, and taking the segmented sub-sets as new collaboration sets when the pareto advantages of the segmented sub-sets are larger than those of the original collaboration sets, otherwise, reserving the original collaboration sets; wherein the partitioned subset is a subset of the original collaboration set.
Further, the process of solving the optimal unloading decision and the collaborative set association decision by adopting an Actor-Critic based migration decision algorithm comprises the following steps: abstracting the unloading decision and the collaborative set association decision joint optimization problem into a Markov decision process, using a base station as an agent, and constructing a corresponding state space, action space and rewarding function; each agent has an Actor network and a Critic network; the Actor network and the Critic network are composed of two neural networks with the same structure; the Actor network generates corresponding actions according to the current local observation state of the single intelligent agent and updates the reward function according to the actions to enter the next state; the Critic network takes the action output by the Actor network as input, and adjusts the action of the Actor network through the output strategy gradient; generating experience information according to the current state, the next state, the action and the rewarding value; sampling a plurality of pieces of experience information to train an Actor network and a Critic network, and updating network parameters to obtain the trained Actor network and Critic network; and obtaining an optimal unloading decision and a collaboration set association decision according to the Actor network training result.
Further, the reward function is:
wherein R is t Representing the prize value obtained by the action performed by the t-slot user,user delay cost representing successful execution of task for user u at time slot, < >>Representing a set of users.
The beneficial effects of the invention are as follows: the invention provides a task migration method based on multi-point cooperation in mobile edge calculation based on a high-speed mobile scene, which is characterized in that a cooperative communication model, a task calculation model, a user mobility model and a load balancing model are constructed to further construct a joint optimization problem of unloading decisions and cooperative set association decisions, and a two-time-scale decision model is designed to respectively solve the node cooperation problem and the user task migration problem in two time scales, namely, the node cooperation is performed on a slow time scale for future load change to form a cooperative set; solving an unloading decision and collaborative set association decision joint optimization problem on a fast time scale according to the collaborative set; compared with the prior art, the invention establishes the collaboration set for future load change, and avoids the ping-pong effect generated by the prior online collaboration mechanism; in addition, task migration is performed based on multi-point cooperation, so that the service range and the aggregate computing capacity can be enlarged, and the task mobility and the task execution time delay are effectively reduced.
Drawings
FIG. 1 is a flow chart of a task migration method based on multi-point cooperation in mobile edge calculation;
fig. 2 is a schematic diagram of a network system model in the MEC scenario of the present invention;
FIG. 3 is a block diagram of a two-time-scale decision model according to the present invention;
FIG. 4 is a graph showing the average execution time delay of tasks at different node distribution densities for the present invention and the comparative method.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The invention provides a task migration method based on multi-point cooperation in mobile edge calculation, as shown in fig. 1, the method comprises the following steps:
s1: and constructing a network system model in the MEC scene.
The invention can be used for a high-speed moving scene, wherein U base stations and M users exist in a network system model under the MEC scene constructed by the invention, each base station is provided with one MEC server, which is called an edge service node, and the computing capacities of different servers are heterogeneous; as shown in fig. 2, when the user is at the location 1, tasks are transmitted to the cooperation set consisting of BS1, BS2 and BS3 by means of the coordinated multi-point transmission technique, and they together provide communication services for the user. Servers deployed on the base stations will also cooperate together to provide computing services for users, and computing resources can be shared between servers in the cooperation set to ensure load balancing, and server overload is avoided. The set of base stations in the network is denoted asThe user set is expressed asThe time period discretization of the whole system is expressed as +.>Each slot has a length τ.
Considering the random arrival of user tasks, the user tasks are expressed asWherein (1)>Representing task data size (bits), +.>Representing the number of CPU cycles (cycles/bit) required per bit of task, is->Representing the maximum tolerable delay of a task, < >>Representing the moving speed of the user u in t time slots; considering the scenario that the task arrives randomly and the statistical property is unpredictable, when the t time slot does not arrive, the user is +.>
Assuming t slots, a total of K cooperating sets may be formed, the kth cooperating set may be expressed asThe user can select the collaboration set to offload tasks, define +.>Representing collaboration set association variable, when +.>Representing user selection in collaboration set +.>Execution task->Indicating that the user does not select the collaboration set +.>Definitions->Representing task offloading decision variables, +.>Indicating task offloading to edge side execution, +.>Indicating that the task is executing locally.
S2: and constructing a collaborative communication model, a task calculation model, a user mobility model and a load balancing model based on the network system model in the MEC scene.
Collaborative communication model:
in the system, the users in the same cell adopt the orthogonal frequency division multiple access technology, the interference among the users in the same cell is ignored, and the possible interference among the users in different cells is considered. Let the ue transmit power be fixed at P u . t-slot user u receives data from the collaboration setThe signal-to-interference-and-noise ratio of a signal can be expressed as:
wherein P is j Representing interference user transmit power, D u,m,t Representing the distance between user u and the serving access point m, alpha is the path loss constant,the fast fading coefficient of the channel allocated to user u by the t-slot node m obeys the standard gaussian distribution.Representing except collaboration set->Other collaboration sets than N 0 Representing the noise power.
t-slot user u will calculate the taskSend to collaboration set->Wireless transmission rate +.>The method comprises the following steps:
where B is the channel bandwidth of the user.
The transmission time of the task is related to the channel transmission rate and the task data size, and the t-slot user u-direction collaboration setTask transmission delay->Can be expressed as:
when a user sends a task to a node, the user needs to consume own energy, and the energy consumption of the part is not negligible due to the limited energy of user equipment. At this time, the taskTransmission energy consumption of->Can be expressed as:
task computing model:
when the task is executed locallyTime delay of local calculationCan be expressed as:
wherein,,is the computing power of the user equipment u.
The energy consumption of the user equipment only comprises transmission energy consumption and local calculation energy consumption, and tasksLocally calculated energy consumption->Can be expressed as:
wherein v is 0 Is an effective energy cost factor associated with chip architecture.
Each base station is provided with a maximum CPU frequency F m And provides computing services to the user. When a plurality of users are unloaded onto one server for calculation, the server provides services for the plurality of users simultaneously through processor sharing. For convenience of presentation, assuming that the computing resources allocated to each user by the server are the same, the computing resources allocated to each user by the server m may be expressed asSince the slot length is relatively short, assume +.>The time-slots are not changed in each time-slot,+.>Possibly changing. />Expressed in collaboration set +.>The number of servers in the whole collaboration set, average computational performance +.>Can be expressed as:
when tasks are offloaded to a collaboration setDuring calculation, servers in the collaboration set share calculation resources to provide services for users, and task calculation time delay is +.>Can be expressed as:
the whole edge execution stage comprises three parts, namely a user sends tasks to a collaboration set, the collaboration set completes task calculation, and a calculation result is sent to the user. The third part of time delay is negligible because the task output result is often smaller and the downlink transmission rate is faster. Considering the task transmission delay and the execution delay on the collaboration set, the total delay of the task execution at the edgeCan be expressed as:
the total time delay of task execution of the user is expressed as:
user mobility model:
in order to measure the mobility speed of a user in a system, the invention adopts the residence time of the user as a mobility evaluation index in the cell switching process, and the residence time distribution obeys an exponential distribution, which can be expressed as follows:
in the method, in the process of the invention,mean residence time is indicated for measuring mobility strength,/->The smaller the value, the stronger the user mobility, and the specific value can be obtained by analyzing the historical data through a machine learning method.
If the user remains in a certain collection for a timeLess than task transfer time->The task transmission fails. Predicting the probability of task transmission failure based on the stay time can be expressed as:
the probability of successful task transmission is
Load balancing model:
calculating the predicted calculation resource occupation amount and the predicted storage resource occupation amount of the server by adopting an LSTM network; the LSTM network is a prediction algorithm based on time sequence, and because the node load change has periodicity, the future load change condition of the node can be obtained according to the historical load change information through the LSTM network. And inputting the historical CPU and storage occupation information of the nodes in the network into the LSTM network, and outputting the predicted value of the CPU and storage resource occupation.
In order to measure the load state of the server, the expected load of the server m is defined by considering the utilization rate of the computing resource and the storage resource of the serverThe method comprises the following steps:
wherein,,and->The predicted computation resource occupancy and storage resource occupancy of the server m at time t are represented, respectively. />Representing the maximum computing power of the server m, < >>Representing the maximum storage capacity of the server m; omega 1 Representing computing resource weights, ω 2 Representation ofThe weight of the storage resource, the computing resource and the storage resource are valued according to the importance degree, preferably, the computing resource and the storage resource can be equally important, thus omega can be taken 1 ＝ω 2 ＝0.5。
Setting an expected load threshold ΓL m The high load node set is defined asExpected load value->When this node is considered a high load node, it needs to evacuate redundant computational tasks to other neighboring nodes and not accept the computational tasks of other nodes. The low load node is defined as +.>Expected load value->When the node is considered a low-load node, the computational load of other nodes can be received, and the computational tasks do not need to be offloaded to other nodes.
In order to measure the load deviation degree between nodes in the formed collaboration set, the sharing of computing resources between high-load nodes and low-load nodes is encouraged to form the collaboration set; defining a collaboration setExpected load deviation->The method comprises the following steps:
wherein n represents a collaboration setThe number of servers in->Representing collaboration set +.>Average load of all nodes in the network. When the nodes form cooperation, in order to avoid uneven load distribution, the smaller the expected load deviation of the cooperation set is, the better the expected load deviation is.
S3: and constructing an unloading decision and collaborative set association decision joint optimization problem according to the collaborative communication model, the task calculation model, the user mobility model and the load balancing model.
The invention considers the influence of mobility on the successful completion of the task, while the influence of user mobility on the completion of the task is mainly in the task transmission process, so the invention successfully executes the task of the user u with t time slots at the user delay costModeling is as follows:
because the user moves in the network at a high speed and needs to carry out unloading decision and collaborative set association decision in real time, the invention takes the minimum task completion delay cost as an objective function, and the unloading decision and collaborative set association decision joint optimization problem can be modeled as follows:
wherein,,represents a set of time slots, T represents a system time period, < ->Representing a collaboration set, p 0 Representing a preset signal-to-interference-and-noise ratio threshold, < + >>Indicating the energy consumption of the user equipment->Representing a user equipment energy consumption budget; constraint condition C1 ensures that the total time delay of task execution does not exceed the maximum tolerance time delay, C2 and C3 respectively ensure the values of an unloading decision variable and a collaboration set association variable, and C4 indicates that the signal-to-interference-and-noise ratio of a user and a collaboration set should be greater than a preset minimum threshold p 0 So as to ensure that the task can be successfully transmitted, and C5 is the energy consumption constraint of the user equipment.
S4: solving an unloading decision and collaboration set association decision joint optimization problem by adopting a two-time scale decision model to obtain an optimal unloading decision and collaboration set association decision; and the system performs task migration according to the optimal unloading decision and the collaborative set association decision.
In view of the fact that in a high-speed moving scene, computing tasks are continuously generated in the moving process of a user, unloading and migration decisions need to be carried out in real time. For example, the time scale of task generation in road safety scenes is often in the order of hundreds of milliseconds, so a fast time scale decision mode is adopted, that is, each time slot selects a proper collaboration set according to the position of the time slot to unload. According to the size of the sampling time scale of the data set in the real world, the node load is often changed greatly in the second or minute level, the generation speed is slower relative to the user task, and the node load change trend is considered when the invention performs collaboration, so the node collaboration decision is considered as a decision on a slow time scale.
As shown in fig. 3, the present invention proposes a two-time scale decision model to solve the problem of joint optimization of unloading decisions and collaborative set association decisions; specific:
defining consecutive T time slots as a time scale, and representing the time slot set of the kth time scale as T k = { kT, kt+1, (k+1) T1 }. Two time scales of speed and slow are defined, and decisions are respectively made on the two time scales, and the specific definition is as follows:
1) Slow time scale: and when each time scale starts, updating load information according to a load prediction algorithm, and reforming a new collaboration set by utilizing a collaboration mechanism based on FCM and alliance game.
2) Fast time scale: and according to the user task information and the collaboration set, searching an optimal unloading decision and collaboration set association decision for the user by using an Actor-Critic-based migration decision algorithm in each time slot.
The process of updating the collaboration set by adopting the collaboration mechanism based on the FCM algorithm and the alliance game comprises the following steps:
if the current load condition of the nodes is only considered to form a collaboration set, a ping-pong effect can be caused, the formed collaboration set is unstable, the load state of the nodes is time-varying at different moments, and a game model is required to be continuously established for solving. Therefore, the long-term load change condition of the nodes is comprehensively considered to form a stable collaboration set. And predicting the future load change condition of the node according to a load prediction algorithm, and forming a stable collaboration set according to the provided collaboration algorithm.
In order to expand the service scope of the collaboration collection, when the collaboration collection is formed, the collaboration needs to be performed by combining the relative distances between nodes in addition to the condition of load change of the nodes in the future. The clusters formed by the traditional distance-based data clustering algorithm, such as a K-means algorithm, are usually fixed when calculating the mass center, and cannot form effective dynamic cooperation. The method comprises the steps of pre-segmenting nodes by adopting a Fuzzy C-Means clustering algorithm (FCM), and then performing alliance game by combining the expected load state of the nodes to form a stable collaboration set; specific:
the FCM algorithm acts as a data clustering algorithm, by assigning a weight to each data point and cluster, to indicate the degree of membership to the cluster that the current object belongs to. The cluster center is obtained by minimizing the cluster loss function. FCM divides a plurality of data points into a plurality of fuzzy clusters, x is the invention i Representing the node location. The cluster loss function is expressed as:
wherein K is the number of collaboration sets, n is the number of collaboration points in the collaboration sets, and c j Mu, being the center of the collaboration set ij Is membership degree; τ is a membership index used to control the ambiguity of the partitioned collaboration sets.
In order to minimize the cluster loss function, collaboration center c j The update rule of (2) is:
membership mu ij The update rule of (2) is:
collaboration center c j And degree of membership mu ij Respectively by the two modesAnd updating the generation until the FCM algorithm converges to obtain K collaboration sets.
The FCM algorithm only considers the position relation among the nodes, does not consider the future load state among the nodes, and further adjusts the cooperation set according to the expected load condition of the nodes on the basis of the pre-segmentation of the FCM algorithm. Alliance gaming concerns which alliances can be formed, and collective and individual benefits generated during the formation of the alliances, which can be modeled as alliance gaming for the collaborative set adjustment problem in the present invention, specific to the invention:
the invention defines a alliance game Is the set of participants, i.e., the set of all nodes in the network, Q represents the utility function value available to the federation. The goal of the coalition game is to maximize the revenue value within each collaboration pool, which may represent the load fairness of the collaboration pool. Each node acts as a game participant and decides whether to join or leave the collaboration pool. A federation formed by multiple nodes should be targeted to maximize the federation utility function. Taking the expected load bias value of the collaboration set +.>Standard deviation of>Is the inverse of the utility function of the federation->Expressed as:
since multiple distinct federations may be formed between nodes, to describe which federation a node is more prone to select,the concept of a stable federation is defined, i.e., a stable federation is formed when no node is willing to leave an existing federation to make up a new federation. At this point, all nodes in the federation may obtain higher utility than leaving the current federation. To compare the merits of the formed alliances, the concept of pareto dominance was introduced: for two setsAnd->They are all federations made up of the same subset of nodes. If and only if->For some nodes, there is +.>Pareto is superior to->Represented asPareto advantage suggests a comparison to the collection +.>The presence of partial nodes is more prone to join +.>To obtain a greater utility value; specific:
repeatedly executing the merging and splitting processes on the K collaboration sets until the final collaboration set is not changed any more;
combining: comparing the pareto advantages of the combined set and the plurality of collaboration sets, when the pareto advantages of the combined set are greater than those of the plurality of collaboration sets, using the combined set as a final collaboration set, otherwise, reserving the plurality of collaboration sets; the merging set is a union set of a plurality of collaboration sets;
splitting: when the pareto advantage of the segmentation sub-set is larger than that of the original set, the original set is split to obtain a higher utility value, the segmentation sub-set is used as a new collaboration set, and otherwise, the original collaboration set is reserved.
The two actions of merging and splitting can be expressed as:
combining: when (when)Pareto is superior to the subset->When merge subset->
Splitting: when subsetPareto is superior to->When in use, split->Is a subset
Wherein l is [1, K ].
By performing a merge, a group of nodes can run and form a larger federation if and only if such a combination increases the utility of at least one node without decreasing the utility of any other participants. Thus, the merge decision ensures that the benefits of all relevant nodes are not compromised, as well as one federation can decide to split, if the benefits that can be achieved by splitting are greater, it can split into smaller federations; by the method, a stable collaboration set can be formed.
OptimizationProblem(s)Is a mixed integer nonlinear programming problem and is difficult to directly solve. Meanwhile, considering time variability of channel states and positions in the moving process of a user, the invention provides an Actor-Critic-based migration decision algorithm. The process of solving the optimal unloading decision and the collaborative set association decision by adopting an Actor-Critic based migration decision algorithm comprises the following steps:
under the condition of not losing generality, assuming that the wireless environment change and the available computing resource change have Markov, abstracting the problem of joint optimization of the unloading decision and the collaboration set association decision into a Markov decision process, and enabling a base station to serve as an agent, wherein each agent has an Actor network and a Critic network; the Actor network and the Critic network consist of two neural networks with the same structure; constructing a state space, an action space and a reward function, and specifically:
state space:
the state space is the environmental information around the users and nodes observed by the agent, including collaborative set computing power, channel state, user tasks, defined as:
wherein,,representing the average computational performance of all cooperating sets in the network, H (t) representing the channel gain matrix, +.>Representing a set of user tasks.
Action space:
the action space is used for reflecting a set of actions that an agent can take, including offloading decisions, server association policies, defined as:
bonus function: at each time slot, the controller responsible for making the offloading decisions and server association policies obtains a corresponding reward based on the action state. And finding the optimal decision by maximizing the reward function value. In order to minimize the average task execution delay, the present invention will optimize the problemThe reciprocal of the objective function of (2) is set to a prize value, expressed as:
the Actor network generates corresponding actions according to the current local observation state of the single intelligent agent and updates the reward function according to the actions to enter the next state; the Critic network takes the action output by the Actor network as input, and adjusts the action of the Actor network through the output strategy gradient; generating experience information according to the current state, the next state, the action and the rewarding value; and (3) sampling a plurality of pieces of experience information to train an Actor network and a Critic network, and alternately executing through the Actor network and the Critic network, wherein the Actor network with sufficient training can generate an optimal unloading decision and a collaboration set association decision. And the system performs task migration according to the optimal unloading decision and the collaborative set association decision.
The invention was evaluated:
the load prediction algorithm adopts the CPU and the stored usage data of the server in the google cluster load data set for training, wherein the first 80% of the data set is used for training, and the second 20% is used for testing. When the FCM algorithm is adopted to pre-cluster the nodes, an EUA data set containing the geographic positions of all Base Stations (BS) in the CBD area of the Australian cuttlefish is adopted as an edge server position data set; fig. 4 shows the variation of the user average task completion delay at different node distribution densities, expressed as the ratio of the number of nodes to the number of users in the area. As can be seen from fig. 4, as the distribution density of nodes increases, more nodes can provide services for users, and the average completion time delay of tasks gradually decreases. And compared with the latest CoMP cooperation scheme (SUN W, LIU J.2-to-Mcoordinated multipoint-based uplink transmission in ultra-dense cellular networks [ J ]. IEEE Transactions on Wireless Communications,2018,17 (12): 8342-8356.) and the CoMP-free scheme, the scheme provided by the invention has lower average time delay of task completion, because the node expected load change is taken into consideration when the scheme provided by the invention forms the cooperation set, and the calculation performance of the cooperation set can be better ensured.
While the foregoing is directed to embodiments, aspects and advantages of the present invention, other and further details of the invention may be had by the foregoing description, it will be understood that the foregoing embodiments are merely exemplary of the invention, and that any changes, substitutions, alterations, etc. which may be made herein without departing from the spirit and principles of the invention.
Claims (9)
1. The task migration method based on multi-point cooperation in mobile edge calculation is characterized by comprising the following steps:
s1: constructing a network system model in an MEC scene;
s2: constructing a collaborative communication model, a task calculation model, a user mobility model and a load balancing model based on a network system model in an MEC scene;
s3: constructing an unloading decision and collaborative set association decision joint optimization problem according to a collaborative communication model, a task calculation model, a user mobility model and a load balancing model;
s4: solving an unloading decision and collaboration set association decision joint optimization problem by adopting a two-time scale decision model to obtain an optimal unloading decision and collaboration set association decision; and the system performs task migration according to the optimal unloading decision and the collaborative set association decision.
2. The task migration method based on multi-point collaboration in mobile edge computing according to claim 1, wherein the method is characterized in that under the MEC sceneSpecifically, the network system model of (1) comprises: u base stations and M users, wherein each base station is provided with an MEC server; the set of base stations is denoted asUser set is denoted +.>The task of the user is expressed as->Wherein (1)>Representing task data size, +.>Representing the number of CPU cycles required per bit of task, and (2)>Representing the maximum tolerable delay of a task, < >>Representing the moving speed of the user u in t time slots; user selection collaboration set +.>Offloading tasks, define->Representing collaboration set association variable, +.>Representing user selection in collaboration set +.>Execution task->Indicating that the user does not select the collaboration set +.>Definitions->Representing task offloading decision variables, +.>Indicating task offloading to edge side execution, +.>Indicating that the task is executing locally.
3. The method for task migration based on multi-point collaboration in mobile edge computing according to claim 1, wherein the user mobility model comprises: the residence time of the user is used as an evaluation index of whether the task can be successfully transmitted in the cell switching process; if the user remains in the collaboration set for a timeLess than task transfer time->The task transmission fails; the probability of task transmission failure is predicted based on the stay time.
4. The method for task migration based on multi-point collaboration in mobile edge computing according to claim 1, wherein the load balancing model comprises:
calculating the predicted calculation resource occupation amount and the predicted storage resource occupation amount of the server by adopting the LSTM network, and calculating the expected load of the server according to the predicted calculation resource occupation amount and the predicted storage resource occupation amount; setting an expected load threshold, and taking the server as a high load node when the expected load of the server is larger than the expected load threshold, otherwise, taking the server as a low load node;
the high-load node is used for evacuating redundant computing tasks to other adjacent nodes and not receiving computing tasks of other nodes, and the low-load node is used for receiving computing loads of other nodes and does not need to unload computing tasks to other nodes; and sharing computing resources between the high-load nodes and the low-load nodes to form a collaboration set, and defining expected load deviation of the collaboration set.
5. The task migration method based on multi-point collaboration in mobile edge computing according to claim 1, wherein the joint optimization problem of unloading decision and collaboration set association decision is expressed as:
wherein,,user delay cost representing successful execution of task for user u at time slot, < >>Represents a set of time slots, T represents a system time period, < ->Representing user set->Representing a collaboration set,/->Indicating the maximum tolerable delay for the task of user u at time slot t,task offloading decision variable representing user u at time slot, < ->A collaboration set association variable representing t-slot user u,indicating that t-slot user u receives +.>Signal-to-interference-and-noise ratio, p 0 Representing the signal-to-interference-and-noise ratio threshold,representing t time slot user equipment energy consumption, +.>Representing the user equipment energy consumption budget.
6. The method for task migration based on multi-point collaboration in mobile edge computing according to claim 1, wherein the process of solving the joint optimization problem of unloading decision and collaboration set association decision comprises:
the two-time scale decision model comprises a slow time scale model and a fast time scale model; the slow time scale model updates the collaboration set by adopting a collaboration mechanism based on the FCM algorithm and alliance game according to the load information, and the fast time scale model solves the optimal unloading decision and collaboration set association decision by adopting an Actor-Critic-based migration decision algorithm according to the collaboration set and user task information.
7. The method for task migration based on multi-point coordination in mobile edge computing according to claim 6, wherein the process of updating the coordination set by adopting a coordination mechanism based on FCM algorithm and alliance gaming comprises:
clustering all server nodes by adopting an FCM algorithm to obtain K collaboration sets;
repeatedly executing the merging and splitting processes on the K collaboration sets until the final collaboration set is not changed any more;
combining: setting a alliance utility function according to the expected load deviation of the collaboration set, comparing the pareto advantages of the combined set and the collaboration sets according to the alliance utility function, and taking the combined set as a new collaboration set when the pareto advantages of the combined set are larger than the pareto advantages of the collaboration sets, otherwise, reserving the collaboration sets; the merging set is a union set of a plurality of collaboration sets;
splitting: comparing the pareto advantages of the segmented sub-sets with those of the original collaboration sets, and taking the segmented sub-sets as new collaboration sets when the pareto advantages of the segmented sub-sets are larger than those of the original collaboration sets, otherwise, reserving the original collaboration sets; wherein the partitioned subset is a subset of the original collaboration set.
8. The method for task migration based on multi-point collaboration in mobile edge computing according to claim 6, wherein the process of solving the optimal offloading decision and the collaborative set association decision by using an Actor-Critic based migration decision algorithm comprises: abstracting the unloading decision and the collaborative set association decision joint optimization problem into a Markov decision process, using a base station as an agent, and constructing a corresponding state space, action space and rewarding function; each agent has an Actor network and a Critic network; the Actor network and the Critic network are composed of two neural networks with the same structure; the Actor network generates corresponding actions according to the current local observation state of the single intelligent agent and updates the reward function according to the actions to enter the next state; the Critic network takes the action output by the Actor network as input, and adjusts the action of the Actor network through the output strategy gradient; generating experience information according to the current state, the next state, the action and the rewarding value; sampling a plurality of pieces of experience information to train an Actor network and a Critic network, and updating network parameters to obtain the trained Actor network and Critic network; and obtaining an optimal unloading decision and a collaboration set association decision according to the Actor network training result.
9. The method for task migration based on multi-point collaboration in mobile edge computing according to claim 8, wherein the reward function is:
wherein R is t Representing the prize value obtained by the action performed by the t-slot user,the user delay cost for successful execution of the task for the t-slot user u, u representing the user set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310309080.8A CN116471632A (en) | 2023-03-27 | 2023-03-27 | Task migration method based on multi-point cooperation in mobile edge calculation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310309080.8A CN116471632A (en) | 2023-03-27 | 2023-03-27 | Task migration method based on multi-point cooperation in mobile edge calculation |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116471632A true CN116471632A (en) | 2023-07-21 |
Family
ID=87183411
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310309080.8A Pending CN116471632A (en) | 2023-03-27 | 2023-03-27 | Task migration method based on multi-point cooperation in mobile edge calculation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116471632A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118102393A (en) * | 2024-04-25 | 2024-05-28 | 江西师范大学 | Multi-user service migration method based on mobile edge calculation |
-
2023
- 2023-03-27 CN CN202310309080.8A patent/CN116471632A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118102393A (en) * | 2024-04-25 | 2024-05-28 | 江西师范大学 | Multi-user service migration method based on mobile edge calculation |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Liu et al. | Deep reinforcement learning for offloading and resource allocation in vehicle edge computing and networks | |
| Wang et al. | Mobility-aware task offloading and migration schemes in fog computing networks | |
| CN108809695B (en) | Distributed uplink unloading strategy facing mobile edge calculation | |
| CN109947545B (en) | Task unloading and migration decision method based on user mobility | |
| CN107819840B (en) | Distributed mobile edge computing unloading method in ultra-dense network architecture | |
| CN111800828B (en) | Mobile edge computing resource allocation method for ultra-dense network | |
| CN112020103A (en) | Content cache deployment method in mobile edge cloud | |
| CN112367683B (en) | Network selection method based on improved deep Q learning | |
| Khan et al. | Fuzzy based multi-criteria vertical handover decision modeling in heterogeneous wireless networks | |
| Chen et al. | An MDP-based vertical handoff decision algorithm for heterogeneous wireless networks | |
| Zhang et al. | Joint communication and computation resource allocation in fog-based vehicular networks | |
| Dai et al. | Multi-objective intelligent handover in satellite-terrestrial integrated networks | |
| Wang et al. | Task allocation mechanism of power internet of things based on cooperative edge computing | |
| CN116471632A (en) | Task migration method based on multi-point cooperation in mobile edge calculation | |
| Li et al. | Joint computation offloading and service caching for MEC in multi-access networks | |
| CN106060876B (en) | A kind of method of heterogeneous wireless network equally loaded | |
| Ekpenyong et al. | Handoffs decision optimization of mobile celular networks | |
| Wei et al. | Resource allocation and power control policy for device-to-device communication using multi-agent reinforcement learning | |
| Luo et al. | Joint game theory and greedy optimization scheme of computation offloading for UAV-aided network | |
| Su et al. | A power allocation scheme based on deep reinforcement learning in HetNets | |
| Kai et al. | Deep reinforcement learning based user association and resource allocation for D2D-enabled wireless networks | |
| Liu et al. | Mobility-aware task offloading and migration schemes in scns with mobile edge computing | |
| CN113821346A (en) | Computation uninstalling and resource management method in edge computation based on deep reinforcement learning | |
| Tsukamoto et al. | User-centric AP Clustering with Deep Reinforcement Learning for Cell-Free Massive MIMO | |
| Mehbodniya et al. | An intelligent vertical handoff algorithm for next generation wireless networks |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |