CN117290104A - Edge computing method, device and equipment - Google Patents
Edge computing method, device and equipment Download PDFInfo
- Publication number
- CN117290104A CN117290104A CN202311277627.7A CN202311277627A CN117290104A CN 117290104 A CN117290104 A CN 117290104A CN 202311277627 A CN202311277627 A CN 202311277627A CN 117290104 A CN117290104 A CN 117290104A
- Authority
- CN
- China
- Prior art keywords
- calculation
- model
- computing
- edge
- models
- 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.)
- Granted
Links
- 238000004364 calculation method Methods 0.000 title claims abstract description 447
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 95
- 230000005540 biological transmission Effects 0.000 claims abstract description 24
- 238000012163 sequencing technique Methods 0.000 claims abstract description 12
- 230000006870 function Effects 0.000 claims description 75
- 238000013528 artificial neural network Methods 0.000 claims description 15
- 238000004590 computer program Methods 0.000 claims description 13
- 238000005457 optimization Methods 0.000 claims description 12
- 238000005094 computer simulation Methods 0.000 claims description 9
- 238000010276 construction Methods 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 6
- 238000011478 gradient descent method Methods 0.000 claims description 3
- 239000003795 chemical substances by application Substances 0.000 description 14
- 238000010586 diagram Methods 0.000 description 14
- 238000012545 processing Methods 0.000 description 14
- 230000005611 electricity Effects 0.000 description 13
- 238000000034 method Methods 0.000 description 13
- 238000010248 power generation Methods 0.000 description 12
- 238000005259 measurement Methods 0.000 description 6
- 230000008569 process Effects 0.000 description 6
- 238000011161 development Methods 0.000 description 4
- 238000004519 manufacturing process Methods 0.000 description 3
- 238000003860 storage Methods 0.000 description 3
- 238000010521 absorption reaction Methods 0.000 description 2
- 230000029087 digestion Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 239000002360 explosive Substances 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000002787 reinforcement Effects 0.000 description 1
- 238000013468 resource allocation Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5061—Partitioning or combining of resources
- G06F9/5072—Grid computing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/48—Program initiating; Program switching, e.g. by interrupt
- G06F9/4806—Task transfer initiation or dispatching
- G06F9/4843—Task transfer initiation or dispatching by program, e.g. task dispatcher, supervisor, operating system
- G06F9/4881—Scheduling strategies for dispatcher, e.g. round robin, multi-level priority queues
-
- 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
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to an edge computing method, device and equipment, and belongs to the technical field of edge computing of the Internet of things. Generating a data transmission component, an algorithm component and an operation component based on the N calculation tasks and the calculated amount of each calculation task; generating a plurality of calculation models required by each calculation task by utilizing a data transmission component, an algorithm component and an operation component, and constructing a model group corresponding to each calculation task based on the plurality of calculation models; constructing an edge calculation model to schedule and optimize all calculation models in N model groups corresponding to N calculation tasks; and solving the edge computing models by using a Hungary algorithm, a CPOP algorithm and a DQN algorithm, distributing edge computing nodes for each computing model in the N model groups, and sequencing the execution sequence of all computing models to solve N computing tasks. The invention can still ensure the efficiency of edge calculation under the condition of more calculation tasks, so that all calculation tasks can be operated efficiently.
Description
Technical Field
The invention relates to the technical field of edge computing of the internet of things, in particular to an edge computing method, an edge computing device and edge computing equipment.
Background
With the development, application and popularization of new technologies such as the internet and AI, big data enter an explosive growth era. In processing these large data, the data processing often takes up 60% to 80% of the time. For example, large amounts of data in the industrial field come directly from growth equipment, and the types of growth equipment involved in the production of products are as many as hundreds, and industrial data types are numerous, often with high complexity and inefficiency in processing and computing these data. In order to solve the problem, an edge computing mode is used at present, computing tasks are distributed to edge computing equipment close to a data source, data transmission and processing time is reduced, computing efficiency is improved, and the burden of a central processing unit is reduced.
In the prior art, aiming at different computing tasks in different industry fields, a task code is often needed, the code is also required to be changed along with the change of an application scene and the computing task, even the computing tasks using the same computing method in different industry fields are also required to be written with different codes to be realized, and repeated opening of the same algorithm brings a plurality of invariants to programming technicians; in addition, as the computing tasks increase, a large amount of edge computing equipment and computing power are required to realize real-time response and data stream processing even if edge computing is used, if the computing tasks in the edge computing are not scheduled and optimized, the computing efficiency cannot be improved, so that some computing tasks with strong real-time performance cannot be realized. For example, a photovoltaic power plant needs to adjust a power generation plan by monitoring the daily rate in real time, and if the calculation tasks of the power plant on the same day are too many to cause low edge calculation efficiency, the daily rate of the power plant on the same day cannot be obtained in real time.
In summary, the existing edge computing method has strong task pertinence, and the same algorithm cannot be shared by different industries, so that the same algorithm is repeatedly developed, and labor and time are wasted; and under the condition of more calculation tasks, the calculation tasks cannot be effectively scheduled and optimized, so that the calculation efficiency is low.
Disclosure of Invention
Therefore, the invention aims to solve the technical problems that the edge computing method in the prior art has stronger task pertinence, and the same algorithm cannot be shared by different industries, so that the same algorithm is repeatedly developed, and the labor and time are wasted; and under the condition of more calculation tasks, the calculation tasks cannot be effectively scheduled and optimized, so that the problem of low calculation efficiency is caused.
In order to solve the above technical problems, the present invention provides an edge computing method, including:
acquiring data to be calculated, generating N calculation tasks and the calculated amount of each calculation task based on the data to be calculated, and generating a data transmission assembly, an algorithm assembly and an operation assembly by using graphical programming based on the calculated amount of each calculation task; the calculated amount comprises the number of intermediate values required to be calculated for completing the calculation task and a calculation method of each intermediate value;
Generating a plurality of calculation models required by each calculation task based on the calculated amount of each calculation task by using the data transmission component, the algorithm component and the calculation component, and constructing a model group corresponding to each calculation task based on the plurality of calculation models;
establishing an edge calculation objective function based on the execution time and minimization of N model groups corresponding to N calculation tasks, respectively establishing an edge calculation constraint function based on the execution time constraint condition of each calculation model, the execution sequence constraint condition of each calculation model and the calculation resource constraint condition of each edge calculation node, and establishing an edge calculation model based on the edge calculation objective function and the edge calculation constraint function;
solving the edge computing models by using a Hungary algorithm, and distributing edge computing nodes for each computing model in each model group; solving the edge calculation models by using a CPOP algorithm and a DQN algorithm, and sequencing the execution sequence of all calculation models in the N model groups;
and obtaining an edge computing optimization scheme based on the edge computing node of each computing model in the N model groups and the execution sequence of all computing models so as to solve the N computing tasks.
In one embodiment of the invention, an edge calculation objective function is built for the goal based on the execution time and minimization of all calculation models in the N model groups, the edge calculation objective function being:
TotalTime＝min(∑ i ∑ j c i,j (∑ r x i,j,r ))，
wherein total time represents the sum of the execution times of all calculation models in the N model groups, and sigma r x i,j,r =1 means that the ith computation model is executed sequentially at the jth edge computation node, c i,j Representing the time required for the ith computing model to execute at the jth edge computing node, i representing the sequence number of the computing model, j representing the jth edge computing node of the edge computation.
In one embodiment of the present invention, the constructing the edge computing constraint function based on the execution time constraint condition of each computing model, the execution sequence constraint condition of the computing model, and the computing resource constraint condition of each edge computing node includes:
constructing a first constraint function based on the execution time of each calculation model being less than or equal to a preset deadline, wherein the execution time comprises the sum of the time of executing the current calculation model and the waiting time of the current calculation model; the first constraint function is:
wherein ET (p) represents the execution time of the p-th calculation model, DT (p) represents the preset cutoff time of the p-th calculation model;
Constructing a second constraint function based on the execution sequence constraint condition of the calculation model, wherein the second constraint function is as follows:
wherein EST (q) represents the earliest time when the qth calculation model starts to execute, CT (p) represents the completion time of the qth calculation model, and p- > q represents the result data of the qth calculation model required when the qth calculation model is executed;
constructing a third constraint function based on the computing resource constraint condition of each edge computing node, wherein the third constraint function is as follows:
∑ i (∑ r x i,j,r )·a i,j ≤A j ，
wherein, sigma r x i,j,r =1 means that the ith computation model is executed sequentially at the jth edge computation node, a i,j Representing the computational resources occupied by the ith computational model at the jth edge computational node, A j Representing the total amount of computing resources of the jth edge computing node.
In one embodiment of the present invention, the solving the edge computing model by using the hungarian algorithm, assigning an edge computing node to each computing model in each model group includes:
calculating the calculation resource ratio of each calculation model in each model group occupying each edge calculation node, wherein the calculation formula of the calculation resource ratio is as follows:
wherein A is j Representing the total amount of computing resources of the jth edge computing node, Representing the g calculation model in the k model group, wherein N represents N model groups corresponding to N calculation tasks;
constructing a similarity matrix based on the computing resource ratio of each computing model in each model group occupying each edge computing node;
and solving the similarity matrix by using a Hungary algorithm, and distributing edge computing nodes for each computing model in each model group.
In one embodiment of the present invention, the solving the edge calculation model by using CPOP algorithm and DQN algorithm, and ordering the execution sequence of all calculation models in the N model groups includes:
calculating the priority of each calculation model in N model groups based on the execution sequence constraint condition of the calculation model by using a CPOP algorithm, and taking the priority of each calculation model as a rewarding value of each calculation model;
ordering the execution order of all the calculation models in the N model groups based on the reward value of each calculation model by using an DQN algorithm.
In one embodiment of the present invention, the calculating the priority of each calculation model in the N model groups based on the execution sequence constraint condition of the calculation model using the CPOP algorithm includes:
Assigning a downward-sorting priority and an upward-sorting priority to each calculation model in the N model groups based on an execution sequence constraint condition of the calculation model;
and obtaining the priority of each calculation model based on the upward-ordered priority and the downward-ordered priority corresponding to each calculation model, wherein the calculation formula is as follows:
rank(p)＝rank d (p)+rank u (p)，
wherein rank (p) represents the priority of the p-th calculation model, rank d (p) represents the downward-ordered priority, rank of the p-th computational model u (p) represents the upward-ordered priority of the p-th computational model.
In one embodiment of the present invention, the ordering the execution order of all the calculation models in the N model groups based on the reward value of each calculation model by using the DQN algorithm includes:
step 1: taking the priority of each calculation model as a reward value of the calculation model, acquiring initial parameters of a neural network and initial state values of an intelligent agent, and calculating an initial Q value of each calculation model based on the initial parameters of the neural network, the initial state values of the intelligent agent and the reward value of each calculation model;
step 2: randomly selecting one calculation model in N model groups based on a greedy strategy, updating the state value of the intelligent agent based on the rewarding value of the calculation model and the current state value of the intelligent agent, and updating the Q value of each calculation model based on the updated state value of the intelligent agent, the rewarding value of each calculation model and the neural network parameter;
Step 3: constructing a loss function based on the updated Q value and the initial Q value corresponding to the calculation model with the maximum updated Q value, and updating the neural network parameters by using a gradient descent method until the loss function is minimum;
step 4: and (3) eliminating the calculation model with the largest Q value after updating to obtain N updated model groups, returning to the execution step (2) until the number of calculation models in the N model groups is 0, and taking the eliminating sequence of the calculation models as the execution sequence of the calculation models.
In one embodiment of the invention, the loss function is:
L＝1/2[y t -Q(s t ,a,ω)] 2 ，
y t ＝r a +γ*max a Q(s t+1 ,a,ω)，
wherein Q(s) t A, ω) represents the initial Q value, s, of the calculation model a t Represents the current state value of the intelligent agent, omega represents the neural network parameter, r a Represents the prize value of the calculation model a, gamma represents the decay factor, max a Represents selecting the calculation model with the highest Q value after updating, Q (s t+1 A, ω) represents the updated Q value, s, of the calculation model a t+1 Representing the updated state value of the agent.
The invention also provides an edge computing device, which comprises:
the data acquisition and component drawing module is used for acquiring data to be calculated, generating N calculation tasks and the calculated amount of each calculation task based on the data to be calculated, and generating a data transmission component, an algorithm component and an operation component by utilizing graphical programming based on the calculated amount of each calculation task; the calculated amount comprises the number of intermediate values required to be calculated for completing the calculation task and a calculation method of each intermediate value;
The computing model and model group generating module is used for generating a plurality of computing models required by each computing task based on the computing amount of each computing task by utilizing the data transmission component, the algorithm component and the computing component, and constructing a model group corresponding to each computing task based on the plurality of computing models;
the edge calculation model construction module is used for constructing an edge calculation objective function based on the execution time and minimization of N model groups corresponding to N calculation tasks, constructing an edge calculation constraint function based on the execution time constraint condition of each calculation model, the execution sequence constraint condition of the calculation model and the calculation resource constraint condition of each edge calculation node, and constructing an edge calculation model based on the edge calculation objective function and the edge calculation constraint function;
the edge computing model solving module is used for solving the edge computing models by using the Hungary algorithm and distributing edge computing nodes for each computing model in each model group; solving the edge calculation models by using a CPOP algorithm and a DQN algorithm, and sequencing the execution sequence of all calculation models in the N model groups;
and the computing task solving module is used for obtaining an edge computing optimization scheme based on the edge computing node of each computing model in the N model groups and the execution sequence of all computing models so as to solve the N computing tasks.
The invention also provides an edge computing device comprising:
a memory for storing a computer program;
and the processor is used for realizing the steps of the edge calculation method when executing the computer program.
Compared with the prior art, the invention has the following beneficial effects:
1. the data transmission component, the algorithm component and the operation component are generated through graphical programming, corresponding components are selected to form a calculation model based on a calculation method needed to be used in a calculation task, the same calculation model can be used for the calculation task using the same calculation method, different codes do not need to be written, manpower and time consumption caused by repeated development of the algorithm are reduced, a user does not need to have programming skills, only the components need to be dragged, and the technical threshold is reduced; in addition, the execution time of all calculation tasks is reduced and the objective function is minimized by constructing the execution time and the execution sequence of each calculation model and the calculation resources of each edge calculation node are constrained, so that the resource scheduling in the edge calculation process is realized; finally, allocating edge computing nodes for each computing model by using a Hungary algorithm, and sequencing the execution sequences of all computing models by combining a CPOP algorithm and a DQN algorithm, so that the efficiency of edge computing can be ensured under the condition of more computing tasks, and all computing tasks can be efficiently operated;
2. The CPOP algorithm is utilized to calculate the priority of each calculation model based on the execution sequence constraint condition of the calculation model, the priority is used as the rewarding value of each calculation model to be applied to the DQN algorithm, when all calculation models are ordered, the execution sequence constraint condition among the calculation models is considered, and the rewarding value of each calculation model in different states is not required to be repeatedly calculated in the DQN algorithm, so that the complexity of the DQN algorithm is reduced, and the convergence rate of the DQN algorithm is improved.
Drawings
In order that the invention may be more readily understood, a more particular description of the invention will be rendered by reference to specific embodiments thereof that are illustrated in the appended drawings, in which
FIG. 1 is a schematic flow chart of an edge computing method according to the present invention;
FIG. 2 is a schematic diagram of a component library provided by the present invention;
FIG. 3 is a schematic diagram of a calculation model and model set according to the present invention;
FIG. 4 is a schematic diagram of an edge computing model construction principle provided by the invention;
FIG. 5 is a schematic diagram of a model set for calculating a daily rate of consumption according to an embodiment of the present invention;
FIG. 6 is a schematic diagram of a calculation model for calculating a daily rate of consumption according to an embodiment of the present invention;
Fig. 7 is a schematic structural diagram of an edge computing device according to the present invention.
Detailed Description
The present invention will be further described with reference to the accompanying drawings and specific examples, which are not intended to be limiting, so that those skilled in the art will better understand the invention and practice it.
Referring to fig. 1, fig. 1 is a flowchart of an edge computing method provided in the present application, which specifically includes:
s10: acquiring data to be calculated, generating N calculation tasks and the calculated amount of each calculation task based on the data to be calculated, and generating a data transmission assembly, an algorithm assembly and an operation assembly by utilizing graphical programming based on the calculated amount of each calculation task; the calculated amount comprises the number of intermediate values required to be calculated for completing the calculation task and a calculation method of each intermediate value;
s20: generating a plurality of calculation models required by each calculation task based on the calculated amount of each calculation task by utilizing a data transmission component, an algorithm component and an operation component, and constructing a model group corresponding to each calculation task based on the plurality of calculation models;
s30: an edge calculation objective function is built based on the execution time and minimization of N model groups corresponding to N calculation tasks, an edge calculation constraint function is built based on the execution time constraint condition of each calculation model, the execution sequence constraint condition of the calculation model and the calculation resource constraint condition of each edge calculation node, and an edge calculation model is built based on the edge calculation objective function and the edge calculation constraint function;
S40: solving the edge computing models by using a Hungary algorithm, and distributing edge computing nodes for each computing model in each model group; solving the edge calculation models by using a CPOP algorithm and a DQN algorithm, and sequencing the execution sequence of all calculation models in the N model groups;
s50: and obtaining an edge computing optimization scheme based on the edge computing node of each computing model in the N model groups and the execution sequence of all computing models so as to solve the N computing tasks.
According to the edge computing method, the data transmission component, the algorithm component and the operation component are generated through graphical programming, corresponding components are selected to form a computing model based on the computing method needed to be used in the computing task, the same computing model can be used for the computing task using the same computing method, different codes do not need to be written, manpower and time consumption caused by repeated development of the algorithm are reduced, a user does not need to have programming skills, only the components need to be dragged, and the technical threshold is reduced; in addition, the execution time of all calculation tasks is minimized by constructing an objective function, the calculation time is reduced, and the execution time and the execution sequence of each calculation model and the calculation resources of each edge calculation node are constrained, so that the resource scheduling in the edge calculation process is realized; and finally, allocating edge computing nodes for each computing model by using a Hungary algorithm, and sequencing the execution sequences of all computing models by combining a CPOP algorithm and a DQN algorithm, so that the efficiency of edge computing can be ensured under the condition of more computing tasks, and all computing tasks can be efficiently operated.
Specifically, referring to fig. 2, fig. 2 shows a component library provided in the present application, which includes a data transmission component, an algorithm component, and an operation component. The data transmission assembly comprises an input assembly and a data output assembly, and the operation assembly comprises a logic operation assembly, an arithmetic operation assembly and a relation operation assembly. In some embodiments of the present application, various commonly used data transmission components, algorithm components and operation components may be generated in advance by using graphical programming, and when in use, whether components needed to be used for completing a calculation task exist in a component library may be searched first, if yes, dragging may be directly performed, and components meeting the calculation requirement may be called out.
Because the corresponding algorithm components and the corresponding operation components are generated by using graphical programming, even if task calculation in different fields involves the same algorithm, the components in the component library can be used without repeated programming, thereby reducing the burden of programmers and avoiding the manpower resource waste caused by repeated development of the same algorithm; in addition, various component order drag generated by using graphical programming can form a calculation model, and even users without programming skills can use the calculation model, so that the technical threshold is reduced.
Referring to fig. 3, fig. 3 is a schematic diagram of a relationship between a computing model and a model set provided in the present application, where each computing model is formed by using components in a component library, and a plurality of computing models corresponding to the same computing task form a model set, so that each model set corresponds to one computing task.
For example, if the obtained data to be calculated 1 is the electricity consumption of a week of the factory, the data to be calculated 2 is the electricity consumption of a Saturday of the factory, the data to be calculated 3 is the electricity consumption of all days from Monday to Friday of the factory, and the calculation task is the electricity consumption of the day of the factory. The calculation model 1 is used for calculating the total electricity consumption of the factories from monday to friday based on the daily electricity consumption of the factories from monday to friday; the calculation model 2 is used for calculating the total electricity consumption of the factories from monday to friday based on the total electricity consumption of the factories from monday to friday and the electricity consumption of friday; the calculation model 3 is used to calculate the electricity consumption of the plant sunday based on the electricity consumption of the plant for one week and the total amount of electricity consumption of monday to Saturday. Then computing model 1, computing model 2, and computing model 3 are combined as a model set for a corresponding one of the computing tasks.
Because each calculation task corresponds to a model group which comprises a plurality of calculation models, when the calculation tasks are relatively more, the calculation models which need to be executed are correspondingly increased, each calculation model needs to be calculated by a corresponding edge calculation node, and the execution of some calculation models also needs the data output by another calculation model, therefore, the edge calculation models need to be established to schedule and optimize all model groups and calculation models in the edge calculation process, so that the high efficiency of edge calculation is ensured.
Referring to fig. 4, fig. 4 is a schematic diagram illustrating a construction principle of an edge calculation model provided in the present application; taking real-time requirements of all calculation tasks into consideration, the execution time and minimization of the calculation models in all model groups are taken as optimization targets; for each calculation model in the N model groups, the execution time of the calculation model is required to be less than or equal to the preset cut-off time, so that the execution time constraint of the calculation model is required to be established; in addition, for all the calculation models in the same model group, the result data of the preamble calculation model is needed by part of calculation models, so that the constraint of the execution sequence exists among the calculation models; finally, when performing edge computation, each computation model needs to have a corresponding edge computation node to perform computation, and therefore, the computation resources of each edge computation node need to be constrained.
Based on the above-mentioned edge calculation model construction principle, the present application constructs an edge calculation model, specifically, step S30 includes:
establishing an edge computing objective function based on the execution time and minimization of all computing models in the N model groups;
specifically, the edge calculation objective function is:
TotalTime＝min(∑ i ∑ j c i,j (∑ r x i,j,r ))，
Wherein, total time represents the execution time sum of all calculation models in the N model groups,
∑ r x i,j,r =1 means that the ith computation model is executed sequentially at the jth edge computation node, c i,j Representing the time required for the ith computing model to execute at the jth edge computing node, i representing the sequence number of the computing model, j representing the jth edge computing node of the edge computation.
Constructing a first constraint function based on the execution time of each calculation model being less than or equal to a preset cutoff time, wherein the execution time of a calculation model comprises the sum of the time of executing the calculation model and the waiting time of the calculation model;
specifically, the first constraint function is:
wherein ET (p) represents the execution time of the p-th calculation model, and DT (p) represents the preset cutoff time of the p-th calculation model.
Constructing a second constraint function based on the execution sequence constraint condition of the calculation model;
specifically, the second constraint function is:
wherein EST (q) represents the earliest time when the qth calculation model starts to execute, CT (p) represents the completion time of the qth calculation model, and p→q represents the result data of the qth calculation model required when the qth calculation model is executed.
Constructing a third constraint function based on the computing resource constraint condition of each edge computing node;
Specifically, the third constraint function is:
∑ i (∑ r x i,j,r )·a i,j ≤A j ，
wherein, sigma r x i,j,r =1 means that the ith computation model is executed sequentially at the jth edge computation node, a i,j Representing the computational resources occupied by the ith computational model at the jth edge computational node, A j Representing the total amount of computing resources of the jth edge computing node.
And obtaining an edge calculation constraint function based on the first constraint function, the second constraint function and the third constraint function, and constructing an edge calculation model based on the edge calculation objective function and the edge calculation constraint function.
Based on the edge calculation model, the method adopts a heuristic method and reinforcement learning to solve the model: for the problem of edge computing node allocation (resource allocation), considering that a plurality of model groups can be allocated to different edge computing nodes for computing, according to the computing resource constraint of the edge computing nodes and the number of computing models contained in the model groups, the edge computing nodes are allocated to the computing models in each model group by using a Hungary algorithm; for the task scheduling problem, the execution sequence of all the calculation models is ordered by combining the DQN algorithm and the CPOP algorithm, so that the execution time of each calculation model is optimal under the condition of meeting the resource constraint of the edge calculation nodes.
Specifically, in some embodiments of the present application, solving the edge computation model in step S40 using the hungarian algorithm, assigning an edge computation node to each computation model in each model group includes:
calculating the calculation resource ratio of each calculation model in each model group occupying each edge calculation node, specifically, the calculation formula of the calculation resource ratio is:
wherein A is j Representing the total amount of computing resources of the jth edge computing node,representing the g calculation model in the k model group, wherein N represents N model groups corresponding to N calculation tasks;
constructing a similarity matrix based on the computing resource ratio of each computing model in each model group occupying each edge computing node;
and solving the similarity matrix by using a Hungary algorithm, and distributing edge computing nodes for each computing model in each model group.
In some embodiments of the present application, solving the edge calculation models in step S40 using the CPOP algorithm and the DQN algorithm, and ordering the execution order of all calculation models in the N model groups includes:
calculating the priority of each calculation model in the N model groups based on the execution sequence constraint condition of the calculation model by using a CPOP algorithm, and taking the priority of each calculation model as a reward value of each calculation model;
Specifically, each of the N model groups is assigned a downward-ordered priority and an upward-ordered priority based on execution order constraints of the computing models, respectively;
and obtaining the priority of each calculation model based on the upward-ordered priority and the downward-ordered priority corresponding to each calculation model, wherein the calculation formula is as follows:
rank(p)＝rank d (p)+rank u (p)，
wherein rank (p) represents the priority of the p-th calculation model, rank d (p) represents the downward-ordered priority, rank of the p-th computational model u (p) represents the upward-ordered priority of the p-th computational model.
Ordering the execution order of all the calculation models in the N model groups based on the reward value of each calculation model by using the DQN algorithm, wherein the method specifically comprises the following steps:
step 1: taking the priority of each calculation model as a reward value of the calculation model, acquiring initial parameters of the neural network and initial state values of the intelligent agent, and calculating an initial Q value of each calculation model based on the initial parameters of the neural network, the initial state values of the intelligent agent and the reward value of each calculation model;
step 2: randomly selecting one calculation model in N model groups based on a greedy strategy, updating the state value of the intelligent agent based on the rewarding value of the calculation model and the current state value of the intelligent agent, and updating the Q value of each calculation model based on the updated state value of the intelligent agent, the rewarding value of each calculation model and the neural network parameter;
Step 3: constructing a loss function based on the updated Q value and the initial Q value corresponding to the calculated model with the maximum updated Q value, and updating the neural network parameters by using a gradient descent method until the loss function is minimum;
specifically, the loss function is:
L＝1/2[y t -Q(s t ,a,ω)] 2 ，
y t ＝r a +γ*max a Q(s t+1 ,a,ω)，
wherein Q(s) t A, ω) represents the initial Q value, s, of the calculation model a t Represents the current state value of the intelligent agent, omega represents the neural network parameter, r a Represents the prize value of the calculation model a, represents the decay factor, max a Indicating that the Q value after the selection update is the largestCalculation model, Q(s) t+1 A, ω) represents the updated Q value, s, of the calculation model a t+1 Representing the updated state value of the agent;
step 4: and (3) eliminating the calculation model with the largest Q value after updating to obtain N updated model groups, returning to the execution step (2) until the number of calculation models in the N model groups is 0, and taking the eliminating sequence of the calculation models as the execution sequence of the calculation models.
In addition, the priority of each calculation model is calculated by using the CPOP algorithm and applied to the DQN algorithm as a reward value, so that the constraint condition of the execution sequence among the calculation models is considered, the reward value of each calculation model is not required to be repeatedly calculated in the ordering process by using the DQN algorithm, the complexity of the DQN algorithm is reduced, and the convergence speed of a Q network in the DQN algorithm is improved.
By sequencing the execution sequence of all the calculation models by using the CPOP algorithm and the DQN algorithm, the calculation model sequencing result with the shortest execution time under the condition of meeting the calculation resource constraint is obtained. And when the calculation is performed, each calculation model is sequentially executed according to the execution sequence after the optimization, so that the important calculation model is ensured to be processed preferentially, the calculation efficiency is improved, and the accuracy and timeliness of the calculation model are ensured.
The above-described edge calculation method is further explained below by taking an example of application of the edge calculation method to the industrial field:
since a large amount of data in the industrial field comes directly from the growing equipment, and the growing equipment types involved in the production process of the product are too many, the industrial data are too many, the complexity and the efficiency of processing and calculating the data are high, and a large amount of edge calculating equipment and calculation force are required even if the edge calculation is used, therefore, if the dispatching and optimizing are not performed on a plurality of calculating tasks in the edge calculation process, the calculating efficiency cannot be improved, and thus, the calculating tasks with strong instantaneity cannot be realized, for example, the daily rate of absorption of photovoltaic power plants in the industrial field is caused.
According to the demand and supply conditions of the electric power market, the photovoltaic power plant can reasonably adjust the daily power generation plan by calculating the daily consumption rate; and through real-time monitoring of the daily rate of consumption, the photovoltaic power plant can know the sales condition of the photovoltaic power plant, including the proportion of actual sales electric quantity and power generation capacity, so that the sales condition can be evaluated in real time, and sales problems and potential sales opportunities can be found in time. Therefore, the daily rate of consumption is an index with strong real-time performance for the photovoltaic power plant, and the calculation efficiency is improved by carrying out scheduling optimization on the calculation tasks in the edge calculation, so that the timeliness of the index is ensured.
Specifically, calculating the daily rate of absorption of the photovoltaic power plant specifically comprises the following steps:
1. obtaining data to be calculated (an accumulated value of the generated energy of the power generating equipment, the installed capacity of the generator and an accumulated value of the online electric quantity) in the photovoltaic power plant, and generating a calculation task and a calculation amount based on the data to be calculated, wherein the calculation task is the daily rate of the photovoltaic power plant, the calculation amount comprises 7 intermediate values which are required to be calculated for completing the calculation task, and the calculated values are respectively as follows: the method comprises the steps of first data of the total power generation amount of the day, average value of the total power generation amount of the previous day, first data of the total power generation amount of the current day, average value of the total power generation amount of the previous day, daily power generation amount of the current day and daily power consumption rate.
The chinese and english control abbreviations for each calculation are shown in table 1:
TABLE 1
2. And drawing out corresponding data transmission components, logic operation components and arithmetic operation components in the component library based on the calculated quantities, forming a calculation model corresponding to each calculated quantity based on the data transmission components, the logic operation components and the arithmetic operation components, and constructing a model group for calculating the daily consumption rate based on all calculation models, as shown in fig. 5.
The first calculation model in the model group is used for calculating first data of the total power generation amount of the current day, and a specific calculation formula is as follows:
If T GENT ＝T DFIRST Dfirst=dfirst,
if T GENT ≠T DFIRST Dfirst=gent,
wherein T is GENT Represents the measurement point time of GENT, T DFIRST The measurement point time of DFIRST is represented;
the second calculation model in the model group is used for calculating the average value of the total power generation amount of the previous day, and the specific calculation formula is as follows:
DAVG＝V GENT ，
wherein V is GENT Mean value of the day before the measurement point of GENT;
the third calculation model in the model group is used for calculating the first data of the total current network electricity quantity, and the specific calculation formula is as follows:
if T OGDFIRST ＝T OGGENT Ogdfirst=ogdfirst,
if T OGDFIRST ≠T OGGENT Ogdfirst=oggent,
wherein T is OGDFIRST Represents the measurement point time, T, of OGDFIRRT OGGENT The measuring point time of OGGENT is represented;
the fourth calculation model in the model group is used for calculating the average value of the previous-day internet surfing electric quantity, and the specific calculation formula is as follows:
OGDFIRST＝V OGGENT ，
wherein V is OGGENT Mean value of day before measuring point of OGGENT;
the fifth calculation model in the model group is used for calculating the daily power generation amount, and the specific calculation formula is as follows:
if the DAVG measurement point state is overtime and GENT-DFIRST < CAPACITY 10, gend=gent-DFIRST;
if the DAVG measurement point state is not overtime and GENT-DFIRST > CAPACITY 10, then
GEND＝GENT-DAVG；
The sixth calculation model in the model group is used for calculating the daily internet power, and the specific calculation formula is as follows:
If the state of the OGGENT measuring point is overtime, oggent=oggent-OGDFIRST;
if the OGGENT measuring point state is not overtime, OGGENT=OGGENT-ODGAVG;
the seventh calculation model in the model group is used for calculating the daily rate of consumption, and the specific calculation formula is as follows:
DABSORBRATIO＝(GEND-OGGEND)*100/GEND。
fig. 6 shows a seventh calculation model in the calculation daily rate model set provided in the embodiment of the present application, where the daily rate is calculated by using a data transmission component and an operation component in the component library based on the daily power generation output by the fifth calculation model and the daily power consumption output by the sixth calculation model.
3. N model groups of N photovoltaic power plants for calculating the daily digestion rate are obtained, an edge calculation objective function is built based on the execution time and minimization of the N model groups, an edge calculation constraint function is built based on the execution time constraint condition of each calculation model, the execution sequence constraint condition of the calculation model and the calculation resource constraint condition of each edge calculation node, and an edge calculation model is built based on the edge calculation objective function and the edge calculation constraint function.
4. Solving the edge computing models by using a Hungary algorithm, and distributing edge computing nodes for each computing model in the N model groups; and solving the edge calculation models by using a CPOP algorithm and a DQN algorithm, and sequencing the execution sequence of all calculation models in the N model groups.
5. And obtaining an edge calculation optimization scheme based on the edge calculation nodes of each calculation model in the N model groups and the execution sequence of all calculation models so as to solve the daily digestion rates of the N photovoltaic power plants in the shortest time.
According to the edge computing method provided by the embodiment of the application, the implementation personnel do not need to have programming skills, the computing model and the model group are built in a dragging assembly mode, the technical threshold is greatly reduced, and the operation and management personnel of the power plant can directly participate in the processing and optimization of the computing task of the power plant, so that the long-term operation and maintenance of the power plant are facilitated; the data and index calculation in the power plant are processed in an edge calculation mode, so that the time consumption of data transmission and processing is reduced, and the burden of a central processing unit is lightened; in addition, the method and the device improve the calculation efficiency by carrying out scheduling optimization on a plurality of calculation tasks in the power plant, so that indexes and data with stronger timeliness can be realized.
Based on the above-mentioned edge computing method, the embodiment of the present application further provides an edge computing device, as shown in fig. 7, which specifically includes:
the data acquisition and component drawing module 10 is used for acquiring data to be calculated, generating N calculation tasks and the calculation amount of each calculation task based on the data to be calculated, and generating a data transmission component, an algorithm component and an operation component by using graphical programming based on the calculation amount of each calculation task; the calculated amount comprises the number of intermediate values required to be calculated for completing the calculation task and a calculation method of each intermediate value;
A calculation model and model group generation module 20, configured to generate a plurality of calculation models required for each calculation task based on the calculation amount of each calculation task by using the data transmission component, the algorithm component and the calculation component, and construct a model group corresponding to each calculation task based on the plurality of calculation models;
the edge computing model construction module 30 is configured to construct an edge computing objective function based on the execution time and minimization of N model groups corresponding to the N computing tasks, construct an edge computing constraint function based on the execution time constraint condition of each computing model, the execution sequence constraint condition of the computing model, and the computing resource constraint condition of each edge computing node, and construct an edge computing model based on the edge computing objective function and the edge computing constraint function;
an edge computing model solving module 40, configured to solve the edge computing models by using the hungarian algorithm, and allocate edge computing nodes to each computing model in each model group; solving the edge calculation models by using a CPOP algorithm and a DQN algorithm, and sequencing the execution sequence of all calculation models in the N model groups;
the computing task solving module 50 is configured to obtain an edge computing optimization scheme based on the edge computing node of each computing model in the N model groups and the execution sequence of all computing models, so as to solve the N computing tasks.
The embodiment of the application also provides edge computing equipment, which comprises:
a memory for storing a computer program;
and the processor is used for realizing the steps of the edge calculation method when executing the computer program.
It will be appreciated by those skilled in the art that embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
It is apparent that the above examples are given by way of illustration only and are not limiting of the embodiments. Other variations and modifications of the present invention will be apparent to those of ordinary skill in the art in light of the foregoing description. It is not necessary here nor is it exhaustive of all embodiments. While still being apparent from variations or modifications that may be made by those skilled in the art are within the scope of the invention.
Claims (10)
1. An edge computing method, comprising:
acquiring data to be calculated, generating N calculation tasks and the calculated amount of each calculation task based on the data to be calculated, and generating a data transmission assembly, an algorithm assembly and an operation assembly by using graphical programming based on the calculated amount of each calculation task; the calculated amount comprises the number of intermediate values required to be calculated for completing the calculation task and a calculation method of each intermediate value;
generating a plurality of calculation models required by each calculation task based on the calculated amount of each calculation task by using the data transmission component, the algorithm component and the calculation component, and constructing a model group corresponding to each calculation task based on the plurality of calculation models;
establishing an edge calculation objective function based on the execution time and minimization of N model groups corresponding to N calculation tasks, respectively establishing an edge calculation constraint function based on the execution time constraint condition of each calculation model, the execution sequence constraint condition of each calculation model and the calculation resource constraint condition of each edge calculation node, and establishing an edge calculation model based on the edge calculation objective function and the edge calculation constraint function;
solving the edge computing models by using a Hungary algorithm, and distributing edge computing nodes for each computing model in each model group; solving the edge calculation models by using a CPOP algorithm and a DQN algorithm, and sequencing the execution sequence of all calculation models in the N model groups;
And obtaining an edge computing optimization scheme based on the edge computing node of each computing model in the N model groups and the execution sequence of all computing models so as to solve the N computing tasks.
2. The edge computing method according to claim 1, wherein an edge computing objective function is constructed for the purpose of minimizing based on the execution time of all the computing models in the N model groups, the edge computing objective function being:
TotalTime＝min(∑ i ∑ j c i,j (∑ r x i,j,r ))，
wherein total time represents the sum of the execution times of all calculation models in the N model groups, and sigma r x i,j,r =1 means that the ith computation model is executed sequentially at the jth edge computation node, c i,j Representing the time required for the ith computing model to execute at the jth edge computing node, i representing the sequence number of the computing model, j representing the jth edge computing node of the edge computation.
3. The edge computing method of claim 1, wherein constructing the edge computing constraint function based on the execution time constraint of each computing model, the execution order constraint of the computing model, and the computing resource constraint of each edge computing node, respectively, comprises:
constructing a first constraint function based on the execution time of each calculation model being less than or equal to a preset deadline, wherein the execution time comprises the sum of the time of executing the current calculation model and the waiting time of the current calculation model; the first constraint function is:
Wherein ET (p) represents the execution time of the p-th calculation model, DT (p) represents the preset cutoff time of the p-th calculation model;
constructing a second constraint function based on the execution sequence constraint condition of the calculation model, wherein the second constraint function is as follows:
wherein EST (q) represents the earliest time when the qth calculation model starts to execute, CT (p) represents the completion time of the qth calculation model, and p- > q represents the result data of the qth calculation model required when the qth calculation model is executed;
constructing a third constraint function based on the computing resource constraint condition of each edge computing node, wherein the third constraint function is as follows:
∑ i (∑ r x i，j，r )·a i,j ≤A j ，
wherein, sigma r x i，j，r =1 means that the ith computation model is executed sequentially at the jth edge computation node, a i，j Representing the computational resources occupied by the ith computational model at the jth edge computational node, A j Representing the total amount of computing resources of the jth edge computing node.
4. The edge computing method of claim 1, wherein solving the edge computing models using a hungarian algorithm, assigning an edge computing node to each computing model in each model group comprises:
calculating the calculation resource ratio of each calculation model in each model group occupying each edge calculation node, wherein the calculation formula of the calculation resource ratio is as follows:
Wherein A is j Representing the total amount of computing resources of the jth edge computing node,representing the g calculation model in the k model group, wherein N represents N model groups corresponding to N calculation tasks;
constructing a similarity matrix based on the computing resource ratio of each computing model in each model group occupying each edge computing node;
and solving the similarity matrix by using a Hungary algorithm, and distributing edge computing nodes for each computing model in each model group.
5. The edge computing method of claim 1, wherein solving the edge computing model using CPOP algorithm and DQN algorithm, ordering the execution order of all computing models in the N model groups comprises:
calculating the priority of each calculation model in N model groups based on the execution sequence constraint condition of the calculation model by using a CPOP algorithm, and taking the priority of each calculation model as a rewarding value of each calculation model;
ordering the execution order of all the calculation models in the N model groups based on the reward value of each calculation model by using an DQN algorithm.
6. The edge computing method of claim 5, wherein computing the priority of each computing model in the N model groups based on the execution order constraint of the computing models using a CPOP algorithm comprises:
Assigning a downward-sorting priority and an upward-sorting priority to each calculation model in the N model groups based on an execution sequence constraint condition of the calculation model;
and obtaining the priority of each calculation model based on the upward-ordered priority and the downward-ordered priority corresponding to each calculation model, wherein the calculation formula is as follows:
rank(p)＝rank d (p)+rank u (p)，
wherein rank (p) represents the priority of the p-th calculation model, rank d (p) represents the downward-ordered priority, rank of the p-th computational model u (p) represents the upward-ordered priority of the p-th computational model.
7. The edge computing method of claim 6, wherein the ordering of execution order of all computing models in the N model groups based on the reward value of each computing model using DQN algorithm comprises:
step 1: taking the priority of each calculation model as a reward value of the calculation model, acquiring initial parameters of a neural network and initial state values of an intelligent agent, and calculating an initial Q value of each calculation model based on the initial parameters of the neural network, the initial state values of the intelligent agent and the reward value of each calculation model;
step 2: randomly selecting one calculation model in N model groups based on a greedy strategy, updating the state value of the intelligent agent based on the rewarding value of the calculation model and the current state value of the intelligent agent, and updating the Q value of each calculation model based on the updated state value of the intelligent agent, the rewarding value of each calculation model and the neural network parameter;
Step 3: constructing a loss function based on the updated Q value and the initial Q value corresponding to the calculation model with the maximum updated Q value, and updating the neural network parameters by using a gradient descent method until the loss function is minimum;
step 4: and (3) eliminating the calculation model with the largest Q value after updating to obtain N updated model groups, returning to the execution step (2) until the number of calculation models in the N model groups is 0, and taking the eliminating sequence of the calculation models as the execution sequence of the calculation models.
8. The edge computing method of claim 7, wherein the loss function is:
L＝1/2[y t -Q(s t ,a,ω)] 2 ，
y t ＝r a +γ*max a Q(s t+1 ,a,ω)，
wherein Q(s) t A, ω) represents the initial Q value, s, of the calculation model a t Represents the current state value of the intelligent agent, omega represents the neural network parameter, r a Represents the prize value of the calculation model a, gamma represents the decay factor, max a Represents selecting the calculation model with the highest Q value after updating, Q (s t+1 A, ω) represents the updated Q value, s, of the calculation model a t+1 Representing the updated state value of the agent.
9. An edge computing device, comprising:
the data acquisition and component drawing module is used for acquiring data to be calculated, generating N calculation tasks and the calculated amount of each calculation task based on the data to be calculated, and generating a data transmission component, an algorithm component and an operation component by utilizing graphical programming based on the calculated amount of each calculation task; the calculated amount comprises the number of intermediate values required to be calculated for completing the calculation task and a calculation method of each intermediate value;
The computing model and model group generating module is used for generating a plurality of computing models required by each computing task based on the computing amount of each computing task by utilizing the data transmission component, the algorithm component and the computing component, and constructing a model group corresponding to each computing task based on the plurality of computing models;
the edge calculation model construction module is used for constructing an edge calculation objective function based on the execution time and minimization of N model groups corresponding to N calculation tasks, constructing an edge calculation constraint function based on the execution time constraint condition of each calculation model, the execution sequence constraint condition of the calculation model and the calculation resource constraint condition of each edge calculation node, and constructing an edge calculation model based on the edge calculation objective function and the edge calculation constraint function;
the edge computing model solving module is used for solving the edge computing models by using the Hungary algorithm and distributing edge computing nodes for each computing model in each model group; solving the edge calculation models by using a CPOP algorithm and a DQN algorithm, and sequencing the execution sequence of all calculation models in the N model groups;
and the computing task solving module is used for obtaining an edge computing optimization scheme based on the edge computing node of each computing model in the N model groups and the execution sequence of all computing models so as to solve the N computing tasks.
10. An edge computing device, comprising:
a memory for storing a computer program;
a processor for implementing the steps of the edge calculation method of any one of claims 1-8 when executing said computer program.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311277627.7A CN117290104B (en) | 2023-09-28 | Edge computing method, device and equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311277627.7A CN117290104B (en) | 2023-09-28 | Edge computing method, device and equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117290104A true CN117290104A (en) | 2023-12-26 |
| CN117290104B CN117290104B (en) | 2024-05-31 |
Family
ID=
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109918182A (en) * | 2019-01-23 | 2019-06-21 | 中国人民解放军战略支援部队信息工程大学 | More GPU task dispatching methods under virtualization technology |
| CN112130993A (en) * | 2020-09-07 | 2020-12-25 | 国网江苏省电力有限公司信息通信分公司 | Power edge Internet of things agent edge calculation method and system based on graphical modeling |
| US20220255790A1 (en) * | 2021-01-29 | 2022-08-11 | Beijing University Of Posts And Telecommunications | Deep reinforcement learning-based information processing method and apparatus for edge computing server |
| CN115766875A (en) * | 2022-11-16 | 2023-03-07 | 中国电信股份有限公司 | Edge computing power resource scheduling method, device, system, electronic equipment and medium |
| CN116501483A (en) * | 2022-12-14 | 2023-07-28 | 天翼电子商务有限公司 | Vehicle edge calculation task scheduling method based on multi-agent reinforcement learning |
| CN116700931A (en) * | 2023-06-21 | 2023-09-05 | 中国科学院空天信息创新研究院 | Multi-target edge task scheduling method, device, equipment, medium and product |
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109918182A (en) * | 2019-01-23 | 2019-06-21 | 中国人民解放军战略支援部队信息工程大学 | More GPU task dispatching methods under virtualization technology |
| CN112130993A (en) * | 2020-09-07 | 2020-12-25 | 国网江苏省电力有限公司信息通信分公司 | Power edge Internet of things agent edge calculation method and system based on graphical modeling |
| US20220255790A1 (en) * | 2021-01-29 | 2022-08-11 | Beijing University Of Posts And Telecommunications | Deep reinforcement learning-based information processing method and apparatus for edge computing server |
| CN115766875A (en) * | 2022-11-16 | 2023-03-07 | 中国电信股份有限公司 | Edge computing power resource scheduling method, device, system, electronic equipment and medium |
| CN116501483A (en) * | 2022-12-14 | 2023-07-28 | 天翼电子商务有限公司 | Vehicle edge calculation task scheduling method based on multi-agent reinforcement learning |
| CN116700931A (en) * | 2023-06-21 | 2023-09-05 | 中国科学院空天信息创新研究院 | Multi-target edge task scheduling method, device, equipment, medium and product |
Non-Patent Citations (3)
| Title |
|---|
| C QUADRI: "Multi-user edge service orchestration based on Deep Reinforcement Learning", 《SEMANTICSCHOLAR》, 1 March 2023 (2023-03-01) * |
| NAN XU: "Modeling and Implementation of Distributed Environment Based on Pi-Calculus and Agent", 《2018 14TH INTERNATIONAL CONFERENCE ON NATURAL COMPUTATION, FUZZY SYSTEMS AND KNOWLEDGE DISCOVERY》, 30 July 2018 (2018-07-30) * |
| 程君翊: "基于强化学习的动态任务调度算法及平台", 《中国优秀硕士学位论文全文数据库信息科技辑》, 15 November 2022 (2022-11-15) * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Shen et al. | Mathematical modeling and multi-objective evolutionary algorithms applied to dynamic flexible job shop scheduling problems | |
| CN108984284A (en) | DAG method for scheduling task and device based on off-line calculation platform | |
| CN114756358B (en) | DAG task scheduling method, device, equipment and storage medium | |
| CN112685153A (en) | Micro-service scheduling method and device and electronic equipment | |
| CN106371924B (en) | A kind of method for scheduling task minimizing MapReduce cluster energy consumption | |
| El Mouayni et al. | Jobs scheduling within Industry 4.0 with consideration of worker’s fatigue and reliability using Greedy Randomized Adaptive Search Procedure | |
| CN114625517A (en) | DAG graph computation distributed big data workflow task scheduling platform | |
| CN106293947B (en) | GPU-CPU (graphics processing Unit-Central processing Unit) mixed resource allocation system and method in virtualized cloud environment | |
| CN111967642A (en) | Resource constraint type dynamic ship scheduling method and scheduling platform based on time window strategy | |
| Rauchecker et al. | Using high performance computing for unrelated parallel machine scheduling with sequence-dependent setup times: Development and computational evaluation of a parallel branch-and-price algorithm | |
| CN106934539A (en) | It is a kind of with limited and expense restriction workflow schedule method | |
| Maghsoudlou et al. | A framework for preemptive multi-skilled project scheduling problem with time-of-use energy tariffs | |
| CN113010296B (en) | Formalized model based task analysis and resource allocation method and system | |
| CN114662932A (en) | Node-hierarchical workflow timing task scheduling method | |
| CN117290104B (en) | Edge computing method, device and equipment | |
| CN109634714A (en) | A kind of method and device of intelligent scheduling | |
| CN112488542A (en) | Intelligent building site material scheduling method and system based on machine learning | |
| Singh et al. | Cuckoo search based workflow scheduling on heterogeneous cloud resources | |
| CN117290104A (en) | Edge computing method, device and equipment | |
| Qi et al. | Solving resource availability cost problem in project scheduling by pseudo particle swarm optimization | |
| Chen et al. | Flow shop scheduling of hybrid make-to-stock and make-to-order in a distributed precast concrete production system | |
| Piramuthu et al. | Information-based dynamic manufacturing system scheduling | |
| Wang et al. | An Organizational Structure and Self-Adaptive Mechanism for Holonic Multi-Agent Systems | |
| CN109783189A (en) | A kind of quiescent operation stream scheduling method and device | |
| Baumgärtel et al. | Combining multi-agent systems and constraint techniques in production logistics |
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 | ||
| GR01 | Patent grant |