CN118827504A - A method, device, equipment and medium for scheduling computing power network resources - Google Patents
A method, device, equipment and medium for scheduling computing power network resources Download PDFInfo
- Publication number
- CN118827504A CN118827504A CN202311016573.9A CN202311016573A CN118827504A CN 118827504 A CN118827504 A CN 118827504A CN 202311016573 A CN202311016573 A CN 202311016573A CN 118827504 A CN118827504 A CN 118827504A
- Authority
- CN
- China
- Prior art keywords
- computing network
- value
- network
- state
- computing
- 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
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L45/00—Routing or path finding of packets in data switching networks
- H04L45/02—Topology update or discovery
- H04L45/08—Learning-based routing, e.g. using neural networks or artificial intelligence
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L45/00—Routing or path finding of packets in data switching networks
- H04L45/12—Shortest path evaluation
- H04L45/124—Shortest path evaluation using a combination of metrics
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L45/00—Routing or path finding of packets in data switching networks
- H04L45/14—Routing performance; Theoretical aspects
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
本发明公开了一种算力网络资源的调度方法、装置、设备和介质，根据预设的业务需求，获取算力网络中的若干个算网关键因子，采用预设的神经网络模型计算算网路径和所述算网路径的总奖励值；其中，所述算网路径是由所有算网节点形成的路径；对所述神经网络模型的权值参数进行迭代寻优，直到所述神经网络模型的目标函数满足要求，得到训练完成的神经网络模型；其中，所述目标函数是根据所述算网路径的总奖励值定义的；根据所述训练完成的神经网络模型得到最优算网路径，以所述最优算网路径进行算力网络资源调度。采用本发明，其能够考虑算力网络中的多个算网关键因子，通过自主学习的方式迭代寻优最佳的算网路径，实现算力网络资源的融合调度。
The present invention discloses a scheduling method, device, equipment and medium for computing network resources. According to preset business requirements, several computing network key factors in the computing network are obtained, and a preset neural network model is used to calculate the computing network path and the total reward value of the computing network path; wherein the computing network path is a path formed by all computing network nodes; the weight parameters of the neural network model are iteratively optimized until the objective function of the neural network model meets the requirements, and a trained neural network model is obtained; wherein the objective function is defined according to the total reward value of the computing network path; the optimal computing network path is obtained according to the trained neural network model, and the computing network resources are scheduled with the optimal computing network path. The present invention can take into account multiple computing network key factors in the computing network, iteratively optimize the best computing network path by autonomous learning, and realize the integrated scheduling of computing network resources.
Description
Technical Field
The present invention relates to the field of communications technologies, and in particular, to a method, an apparatus, a device, and a medium for scheduling a computing power network resource.
Background
The computing power network is a novel information infrastructure oriented to computing network integration, and is based on a network by taking computing as a core, so that computing, storage and network resources are distributed among cloud, edges and ends according to requirements and flexibly scheduled. The computing network is a brand new network system, the computing network brain belongs to the arrangement management layer of the computing network system architecture, is a core link for unified management, arrangement, scheduling and operation and maintenance of computing network resources and capability elements, and is a management center of the whole computing network.
The computing power network has huge scale of resource types and resource quantity, and the computing network has rich service types, so the computing power network has high complexity. The computing network brain is used as a central decision system of the computing power network, and the decision capability of the computing network brain directly influences the operation efficiency of the whole computing power network. The current power network is separated from the power network in scheduling, so that a complete power network resource scheduling system cannot be formed, the global optimal solution of the power network resources cannot be achieved, the overall operation efficiency of the power network is affected, and the problem of how to realize the power network multi-element fusion scheduling is a great technical problem facing the power network.
Disclosure of Invention
The embodiment of the invention aims to provide a dispatching method, a device, equipment and a medium of a power network resource, which can consider a plurality of gateway key factors in the power network, and iteratively optimize an optimal network path in an autonomous learning mode, thereby realizing the fusion dispatching of the power network resource and improving the intellectualization of the power network resource dispatching.
To achieve the above object, an embodiment of the present invention provides a method for scheduling a computing power network resource, including:
acquiring a plurality of gateway key factors in a power calculation network according to preset service requirements;
calculating by adopting a preset neural network model according to the plurality of gateway key factors, and determining a network calculation path and a total rewarding value of the network calculation path; wherein the computational network path is a path formed by all computational network nodes;
iteratively optimizing the weight parameters of the neural network model until the objective function of the neural network model meets the requirement, thereby obtaining a trained neural network model; wherein the objective function is defined in terms of a total prize value for the network path;
and obtaining an optimal network path according to the trained neural network model, and carrying out calculation network resource scheduling according to the optimal network path.
As an improvement of the above solution, the iteratively optimizing the weight parameter of the neural network model until the objective function of the neural network model meets the requirement, to obtain a trained neural network model, includes:
Iteratively optimizing weight parameters of the neural network model by adopting a gradient ascending method until an objective function of the neural network model reaches a maximum value, so as to obtain a trained neural network model; wherein the objective function is such that the mathematical expectation of the total prize value is maximized.
As an improvement of the above solution, the calculating, according to the plurality of gateway key factors, by using a preset neural network model to perform calculation, determining a network path includes:
Selecting one computing network node as an access node at will;
Taking all the computing network nodes which pass at the current moment as the current state, constructing a multi-factor matrix according to the value of each computing gateway key factor under all the computing network nodes in the current state, inputting the multi-factor matrix into the preset neural network model for calculation, and determining the next passing computing network node to enter the next state;
and when all the calculation network nodes pass, obtaining the calculation network path.
As an improvement of the above solution, the determining the total prize value of the network path includes:
Calculating the total key factor value of each computing gateway key factor in each state according to the value of each computing gateway key factor in each computing network node in each state;
Calculating a reward value of each state according to the total key factor value of each gateway key factor in each state;
and calculating the total rewarding value of the network path according to the rewarding value of each state.
As an improvement of the above solution, the calculating the total key factor value of each gateway key factor in each state according to the value of each gateway key factor in each state under each node of each state includes:
determining a measurement factor to which each computation gateway key factor belongs;
And calculating the total key factor value of each gateway key factor in each state by adopting a calculation formula of the total key factor value corresponding to the preset value of the metric factor according to the value of each gateway key factor in each state.
As an improvement of the above, the metric factors include an additivity metric factor, a multiplicative metric factor, and a dishing metric factor;
The gateway key factors belonging to the additive metric factor comprise calculation factors, storage factors, delay factors, distance factors and energy consumption factors, the gateway key factors belonging to the multiplicative metric factor comprise reliability factors, and the gateway key factors belonging to the concave metric factor comprise bandwidth factors.
As an improvement of the above solution, said calculating the prize value of each state based on the total key factor value of each gateway key factor in each state includes:
Determining a weight coefficient of each gateway key factor according to the service demand;
calculating a unified measurement value of each state according to the weight coefficient of each gateway key factor and the total key factor value of each gateway key factor in each state;
and calculating the rewarding value of each state according to the unified metric value of each state.
As an improvement of the above solution, the calculating a unified metric value of each state according to the weight coefficient of each gateway key factor and the total key factor value of each gateway key factor in the state includes:
According to the weight coefficient of each gateway key factor and the total key factor value of each gateway key factor in each state, the following calculation formula is adopted to calculate the unified measurement value of each state:
Calculating the rewarding value of each state according to the unified metric value of each state, comprising:
According to the unified metric value of each state, calculating the rewarding value of each state by adopting the following calculation formula:
rk＝1/Mk；
Wherein M k is a unified metric for the kth state, r k is a prize value for the kth state, the kth state includes k passing nodes, delta i is a weight coefficient for the ith gateway key factor, For the total key factor value of the ith calculation gateway key factor in the kth state, n is the number of the calculation gateway key factors, and
As an improvement of the above solution, the calculating the total prize value of the network path according to the prize value in each state includes:
Performing average value removing treatment on the reward value of each state to obtain a treated reward value;
And obtaining the total prize value of the network path in a summation mode according to the processed prize value in each state:
Wherein R is a total prize value, R k is a prize value of a kth state, and K is the number of network nodes of the network path.
As an improvement of the above solution, after the obtaining a number of computing gateway key factors in the computing power network, the method further comprises:
And normalizing the values of the plurality of gateway key factors.
The embodiment of the invention also provides a scheduling device of the power network resources, which comprises the following steps:
the computing gateway key factor acquisition module is used for acquiring a plurality of computing gateway key factors in the computing power network according to preset service requirements;
The network path determining module is used for calculating by adopting a preset neural network model according to the plurality of gateway key factors to determine a network path and a total rewarding value of the network path; wherein the computational network path is a path formed by all computational network nodes;
The neural network model learning module is used for carrying out iterative optimization on the weight parameters of the neural network model until the objective function of the neural network model meets the requirement, so as to obtain a trained neural network model; wherein the objective function is defined in terms of a total prize value for the network path;
And the power calculation network resource scheduling module is used for obtaining an optimal network path according to the trained neural network model and performing power calculation network resource scheduling according to the optimal network path.
The embodiment of the invention also provides a scheduling device of the power network resources, which comprises a processor, a memory and a computer program stored in the memory and configured to be executed by the processor, wherein the scheduling method of the power network resources is realized by the processor when the computer program is executed.
The embodiment of the invention also provides a computer readable storage medium, which comprises a stored computer program, wherein the computer program controls equipment where the computer readable storage medium is located to execute the dispatching method of the computing power network resource according to any one of the above when running.
Compared with the prior art, the method, the device, the equipment and the medium for scheduling the computing power network resources acquire a plurality of computing gateway key factors in the computing power network according to the preset service requirements; calculating by adopting a preset neural network model according to the plurality of gateway key factors, and determining a network calculation path and a total rewarding value of the network calculation path; wherein the computational network path is a path formed by all computational network nodes; iteratively optimizing the weight parameters of the neural network model until the objective function of the neural network model meets the requirement, thereby obtaining a trained neural network model; wherein the objective function is defined in terms of a total prize value for the network path; and obtaining an optimal network path according to the trained neural network model, and carrying out calculation network resource scheduling according to the optimal network path. By adopting the invention, various indexes such as calculation power, network and benefit factors in the calculation power network are comprehensively considered, the self-learning capability of the intelligent body is utilized, and the optimal calculation network path is iteratively optimized by adopting a machine learning method, so that the integral arrangement and fusion scheduling of calculation network resources are realized, the splitting of calculation power and the network can be effectively avoided, the utilization efficiency of the calculation network resources is improved, and the intellectualization of the calculation power network resource scheduling is improved.
Drawings
Fig. 1 is a schematic flow chart of a scheduling method of computing power network resources according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of determining a network path according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a neural network model in an embodiment of the invention;
fig. 4 is a schematic structural diagram of a scheduling device for computing power network resources according to an embodiment of the present invention;
fig. 5 is a schematic structural diagram of a scheduling device for computing power network resources according to an embodiment of the present invention.
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.
Referring to fig. 1, a flow chart of a method for scheduling computing power network resources according to an embodiment of the present invention is provided, and the embodiment of the present invention provides a method for scheduling computing power network resources, including steps S11 to S14:
s11, acquiring a plurality of calculation gateway key factors in a calculation power network according to preset service requirements;
S12, calculating by adopting a preset neural network model according to the plurality of gateway key factors, and determining a network calculation path and a total rewarding value of the network calculation path; wherein the computational network path is a path formed by all computational network nodes;
S13, carrying out iterative optimization on the weight parameters of the neural network model until the objective function of the neural network model meets the requirements, and obtaining a trained neural network model; wherein the objective function is defined in terms of a total prize value for the network path;
s14, obtaining an optimal network path according to the trained neural network model, and carrying out calculation network resource scheduling according to the optimal network path.
In the embodiment of the invention, according to different business and SLA (SERVICE LEVEL AGREEMENT ) requirements, various factors in the computing power network are considered, including but not limited to computing power factors, network factors and benefit factors, and a plurality of computing gateway key factors are screened out to form a computing gateway key factor set.
By way of example, the gateway key factors relied upon in the power network decisions are shown in Table 1 below:
TABLE 1
It will be appreciated that the above type of gateway key factor is only an alternative embodiment, and in the practical application process, the corresponding gateway key factor may be obtained according to different service requirements, which does not limit the present invention.
Preferably, in step S11, that is, after the acquiring a number of gateway key factors in the computing power network, the method further includes:
S11', carrying out normalization processing on the values of the plurality of gateway key factors.
In the embodiment of the invention, in order to facilitate optimization training, normalization processing is performed on each calculation gateway key factor. Optionally, the normalization is performed by using a Z-score method, and the normalization formula is as follows:
Where X i is the value of the gateway key factor, X i is the normalized value of the gateway key factor, μ is the average value of all the gateway key factors X i, σ is the standard deviation of all the gateway key factors X i.
It should be noted that, since the network paths will pass through a plurality of network nodes, the normalized gateway key factor value under each network node is shown in table 2:
TABLE 2
Further, referring to fig. 2, a schematic diagram of determining a network path according to an embodiment of the present invention is shown, where a neural network model is preset in the embodiment of the present invention, so as to generate a network path from an access node to a target node under the condition of comprehensively considering multiple gateway key factors.
Specifically, initializing a neural network model, calculating the plurality of gateway key factors by adopting the neural network model to obtain a network path, and calculating the total rewarding value of the network path. And calculating an objective function of the neural network model according to the total rewarding value, if the objective function does not meet the requirement, updating weight parameters of the neural network model to obtain an updated neural network model, calculating the plurality of gateway key factors by adopting the updated neural network model to obtain a new network calculation path, calculating the total rewarding value of the network calculation path, calculating the objective function of the neural network model according to the total rewarding value, and continuing to judge whether the objective function meets the requirement or not, so as to realize iterative optimization of weight parameters of the neural network model until the objective function of the neural network model meets the requirement, obtaining a trained neural network model, and determining to obtain an optimal network calculation path. The optimal network path is the optimal route from the access node to the target node under the condition of considering a plurality of gateway key factors corresponding to the current service requirement. And further, performing power network resource scheduling according to the optimal network path.
By adopting the technical means of the embodiment of the invention, various indexes such as calculation power, network and benefit factors in the calculation power network are comprehensively considered, the self-learning capacity of the intelligent body is utilized, the optimal calculation network path is iteratively optimized by adopting a machine learning method, the integral arrangement and fusion scheduling of calculation network resources are realized, the splitting of calculation power and the network can be effectively avoided, the utilization efficiency of the calculation network resources is improved, and the intellectualization of the calculation power network resource scheduling is improved.
As a preferred implementation manner, the embodiment of the present invention is further implemented on the basis of the foregoing embodiment, and in step S12, the calculating, according to the plurality of gateway key factors, by using a preset neural network model to perform calculation, to determine a network path includes:
s121, arbitrarily selecting one computing network node as an access node;
s122, taking all the computing network nodes which pass at the current moment as the current state, constructing a multi-factor matrix according to the value of each computing gateway key factor under all the computing network nodes in the current state, inputting the multi-factor matrix into the preset neural network model for calculation, and determining the next passing computing network node to enter the next state;
And S123, when all the calculation network nodes pass, obtaining the calculation network path.
Preferably, in step S12, the determining the total prize value of the network path includes:
s124, calculating the total key factor value of each gateway key factor in each state according to the value of each gateway key factor in each state;
S125, calculating a reward value of each state according to the total key factor value of each gateway key factor under each state;
and S126, calculating the total rewards value of the network calculation path according to the rewards value of each state.
Referring to fig. 3, which is a schematic diagram of a neural network model in the embodiment of the present invention, n computing gateway key factors and K computing network nodes are taken as examples, in the embodiment of the present invention, one computing network node is arbitrarily selected as an access node, where the node corresponds to an initial state S 1, and a multi-factor matrix under a construction state S 1 is:
Initializing a neural network model pi θ, wherein θ is a weight parameter of the neural network model, inputting a multi-factor matrix corresponding to the initial state S 1 to the neural network model to obtain an action a 1 (θ), and obtaining a reward value r 1 (θ), where the reward value r 1 (θ) is calculated according to values of one computing network node (i.e., the access node) corresponding to the n computing gateway key factors in the state S 1. Then, according to action a 1 (θ), path selection is performed, the next computing network node is determined, state S 2 is entered, and the multi-factor matrix in state S 2 is constructed as follows:
Inputting the multi-factor matrix corresponding to the initial state S 2 into the neural network model to obtain an action a 2 (theta) and a reward value r 2 (theta), wherein the reward value r 2 (theta) is obtained by calculating the values of the two network computing nodes corresponding to the state S 2 of the n network computing key factors to obtain n total key factor values and calculating the n total key factors. Next, a path is selected according to the operation a 2 (θ), the next node is determined, and the process advances to the state S 3. And so on, the multi-factor matrix in the final state S K is obtained as follows:
And continuously inputting the multi-factor matrix corresponding to the final state S K into the neural network model pi θ, obtaining a reward r K (theta), stopping the operation of the action a K (theta) as the whole path reaches the target node, ending the flow, obtaining the network path, and calculating the total reward value of the network path according to the reward values in the K states.
Preferably, the calculation of the prize value R k for each of the states, and the total prize value R for the network path, is explained below.
Step S124, namely, calculating a total key factor value of each gateway key factor in each state according to the value of each gateway key factor in each state under each node of each state, including:
S1241, determining a measurement factor of each calculation gateway key factor;
S1242, calculating the total key factor value of each gateway key factor in each state by adopting a calculation formula of the total key factor value corresponding to the preset value of the metric factor according to the value of each gateway key factor in each state.
Specifically, attribute labeling is performed on the plurality of computing gateway key factors according to different measurement factors in the computing power network. The metric factors are divided into three different attributes, including an additivity metric factor, a multiplicative metric factor, and a dishing metric factor.
The gateway key factors belonging to the additive measurement factors comprise calculation factors, storage factors, time delay factors, distance factors and energy consumption factors, the gateway key factors belonging to the multiplicative measurement factors comprise reliability factors, and the gateway key factors belonging to the concave measurement factors comprise bandwidth factors.
For the additivity metric factor, the calculation formula of the corresponding key factor value is as follows:
for the multiplicative metric, the calculation formula of the corresponding key factor value is as follows:
For the concavity metric factor, the calculation formula of the corresponding key factor value is as follows:
by way of example, taking the 7 gateway key factors mentioned in Table 1 above as examples, for the gateway key factor corresponding to the additivity metric factor:
The total calculated factor value of the calculation factor x 1 in the kth state is:
the total storage factor value of storage factor x 2 in the kth state is:
the total delay factor value of the delay factor x 4 in the kth state is:
the total distance factor value of distance factor x 5 in the kth state is:
the total energy consumption factor value of the energy consumption factor x 7 in the kth state is:
For the computing gateway key factor corresponding to the multiplicative metric factor:
The total reliability factor value of the reliability factor x 6 in the kth state is:
for the computational gateway key factor corresponding to the concavity metric factor:
the total wideband factor value for wideband factor x 3 in the kth state is:
further, step S125, namely calculating a prize value for each state according to the total key factor value of each gateway key factor in each state, includes:
s1251, determining a weight coefficient of each gateway key factor according to the service requirement;
s1252, calculating a unified measurement value of each state according to the weight coefficient of each gateway key factor and the total key factor value of each gateway key factor in each state;
s1253, calculating the rewarding value of each state according to the unified measurement value of each state.
Specifically, in the embodiment of the invention, a unified metric is designed to be used for characterizing the performance of the route. According to the current service demand, determining the importance degree of each gateway key factor, thereby setting the weight coefficient of each gateway key factor, wherein the weight coefficient is delta 1,δ2, delta n, and the formula of the unified metric value is as follows:
and calculating the rewarding value of the state according to the unified measurement value, wherein the formula is as follows:
rk＝1/Mk；
Wherein M k is a unified metric for the kth state, r k is a prize value for the kth state, the kth state includes k passing nodes, delta i is a weight coefficient for the ith gateway key factor, For the total key factor value of the ith calculation gateway key factor in the kth state, n is the number of the calculation gateway key factors, and
Further, step S126, namely calculating a total prize value of the network path according to the prize value in each state, includes:
S126, carrying out average value removing treatment on the rewarding value of each state to obtain a treated rewarding value;
and S127, obtaining the total rewarding value of the network calculation path in a summation mode according to the processed rewarding value in each state.
In the embodiment of the invention, in order to avoid that the reward value in each state can always be a positive value, the effective implementation of a punishment mechanism is ensured, and the average value removing operation is carried out on the reward value of each state. Further, the total prize value of the network path is:
Wherein R is a total prize value, R k is a prize value of a kth state, and K is the number of network nodes of the network path.
By adopting the technical means of the embodiment of the invention, various indexes such as calculation power, network and benefit factors in a calculation power network are comprehensively considered, unified measurement value parameters are designed, a machine learning method is adopted, a neural network model of states, actions and rewards values is constructed through the mapping relation between the unified measurement values and rewards values, and the self-learning ability of an intelligent body is utilized to carry out iterative optimization so as to obtain the optimal network path. The invention utilizes the mapping relation between the unified measurement value and the rewarding value to apply the unified measurement value to the whole optimizing strategy, thereby ensuring the final achievement of the target, and providing the average value removing operation aiming at the rewarding value, ensuring the combined action of rewarding and punishment and improving the efficiency of data training. The invention can effectively realize the integral arrangement and fusion scheduling of the computational network resources, avoids the splitting of the computational power and the network, comprehensively considers the influence of multiple factors, reduces the complexity of the system, improves the utilization efficiency of the computational network resources, improves the intellectualization of the computational power network resource scheduling, and has strong practicability.
As a preferred implementation manner, the embodiment of the present invention is further implemented on the basis of any one of the foregoing embodiments, and step S13, that is, performing iterative optimization on the weight parameters of the neural network model until the objective function of the neural network model meets the requirement, obtains a trained neural network model, includes:
Iteratively optimizing weight parameters of the neural network model by adopting a gradient ascending method until an objective function of the neural network model reaches a maximum value, so as to obtain a trained neural network model; wherein the objective function is such that the mathematical expectation of the total prize value is maximized.
In the embodiment of the invention, for the network path determined by adopting the neural network model pi θ of the weight parameter theta, the reward value of each state of the network path is expressed as R k (theta), and the total reward value of the network path is expressed as R (theta):
And determining an objective function of the neural network model according to the total rewarding value, and performing iterative optimization on the weight parameters of the neural network model by adopting a gradient ascent method so as to obtain an optimal network path.
In an alternative embodiment, an objective function is definedThe weighting parameter θ is calculated so that the mathematical expectation of the prize value of the whole process is maximized.
According to the big number theorem, get: where N represents the number of random trials.
And (3) based on a gradient ascending strategy, carrying out weight parameter optimization according to a gradient formula to obtain an optimal path. The gradient formula based on the big theorem in the embodiment of the invention is as follows:
Updating weight parameters And performing iterative optimization until J (theta) is maximized, determining a weight parameter theta at the moment, and obtaining a trained neural network model, thereby determining an optimal network path and completing the optimization operation of the network path.
By adopting the technical means of the embodiment of the invention, various indexes such as calculation power, network and benefit factors in the calculation power network are comprehensively considered, a unified measurement value parameter is designed, a neural network model of states, actions and rewards values is constructed through the mapping relation between the unified measurement value and rewards values, and an objective function J (theta) is introduced to give a gradientBy utilizing a gradient ascending algorithm, the iterative optimization of the link weight parameters of the computing network route is realized, the global solution of the computing network route is obtained, the invention can effectively realize the integral arrangement and fusion scheduling of the computing network resources, the influence of multiple factors is comprehensively considered, the complexity of the system is reduced, the utilization efficiency of the computing network resources is improved, the intellectualization of the computing network resource scheduling is improved, the perception of users is ensured, and the method has strong feasibility.
It should be understood that the specific calculation process involved in iterating the weight parameters of the neural network model to obtain the optimal solution is only a preferred embodiment, and in the practical application process, the iteration of the weight parameters may be implemented by using a gradient-increasing method with other similar functions to obtain the optimal solution, which does not limit the present invention.
Referring to fig. 4, a schematic structural diagram of a scheduling apparatus for computing power network resources according to an embodiment of the present invention is provided, and an embodiment of the present invention provides a scheduling apparatus 20 for computing power network resources, including:
The computing gateway key factor obtaining module 21 is configured to obtain a plurality of computing gateway key factors in the computing power network according to a preset service requirement;
The network path determining module 22 is configured to determine a network path and a total prize value of the network path by performing calculation with a preset neural network model according to the plurality of gateway key factors; wherein the computational network path is a path formed by all computational network nodes;
The neural network model learning module 23 is configured to iteratively optimize the weight parameters of the neural network model until the objective function of the neural network model meets the requirement, thereby obtaining a trained neural network model; wherein the objective function is defined in terms of a total prize value for the network path;
and the power calculation network resource scheduling module 24 is used for obtaining an optimal network path according to the trained neural network model and performing power calculation network resource scheduling according to the optimal network path.
By adopting the technical means of the embodiment of the invention, various indexes such as calculation power, network and benefit factors in the calculation power network are comprehensively considered, the self-learning capacity of the intelligent body is utilized, the optimal calculation network path is iteratively optimized by adopting a machine learning method, the integral arrangement and fusion scheduling of calculation network resources are realized, the splitting of calculation power and the network can be effectively avoided, the utilization efficiency of the calculation network resources is improved, and the intellectualization of the calculation power network resource scheduling is improved.
Preferably, the apparatus further comprises:
and the normalization processing module is used for carrying out normalization processing on the values of the plurality of gateway key factors.
As a preferred embodiment, the algorithm path determining module 22 is specifically configured to:
Selecting one computing network node as an access node at will; taking all the computing network nodes which pass at the current moment as the current state, constructing a multi-factor matrix according to the value of each computing gateway key factor under all the computing network nodes in the current state, inputting the multi-factor matrix into the preset neural network model for calculation, and determining the next passing computing network node to enter the next state; and when all the calculation network nodes pass, obtaining the calculation network path.
Calculating the total key factor value of each computing gateway key factor in each state according to the value of each computing gateway key factor in each computing network node in each state; calculating a reward value of each state according to the total key factor value of each gateway key factor in each state; and calculating the total rewarding value of the network path according to the rewarding value of each state.
Preferably, the calculating the total key factor value of each gateway key factor in each state according to the value of each gateway key factor in each state under each network node of each state comprises:
determining a measurement factor to which each computation gateway key factor belongs;
And calculating the total key factor value of each gateway key factor in each state by adopting a calculation formula of the total key factor value corresponding to the preset value of the metric factor according to the value of each gateway key factor in each state.
Preferably, the metric factors include an additivity metric factor, a multiplicative metric factor, and a dishing metric factor; the gateway key factors belonging to the additive metric factor comprise calculation factors, storage factors, delay factors, distance factors and energy consumption factors, the gateway key factors belonging to the multiplicative metric factor comprise reliability factors, and the gateway key factors belonging to the concave metric factor comprise bandwidth factors.
Preferably, the calculating the reward value of each state according to the total key factor value of each gateway key factor in each state comprises:
Determining a weight coefficient of each gateway key factor according to the service demand;
calculating a unified measurement value of each state according to the weight coefficient of each gateway key factor and the total key factor value of each gateway key factor in each state;
and calculating the rewarding value of each state according to the unified metric value of each state.
Preferably, the calculating a unified metric value of each state according to the weight coefficient of each gateway key factor and the total key factor value of each gateway key factor in the state comprises:
According to the weight coefficient of each gateway key factor and the total key factor value of each gateway key factor in each state, the following calculation formula is adopted to calculate the unified measurement value of each state:
Calculating the rewarding value of each state according to the unified metric value of each state, comprising:
According to the unified metric value of each state, calculating the rewarding value of each state by adopting the following calculation formula:
rk＝1/Mk；
Wherein M k is a unified metric for the kth state, r k is a prize value for the kth state, the kth state includes k passing nodes, delta i is a weight coefficient for the ith gateway key factor, For the total key factor value of the ith calculation gateway key factor in the kth state, n is the number of the calculation gateway key factors, and
Preferably, the calculating the total prize value of the network path according to the prize value in each state includes:
Performing average value removing treatment on the reward value of each state to obtain a treated reward value;
And obtaining the total prize value of the network path in a summation mode according to the processed prize value in each state:
Wherein R is a total prize value, R k is a prize value of a kth state, and K is the number of network nodes of the network path.
As a preferred embodiment, the neural network model learning module 23 is specifically configured to:
Iteratively optimizing weight parameters of the neural network model by adopting a gradient ascending method until an objective function of the neural network model reaches a maximum value, so as to obtain a trained neural network model; wherein the objective function is such that the mathematical expectation of the total prize value is maximized.
It should be noted that, the scheduling device for computing power network resources provided by the embodiment of the present invention is configured to execute all the flow steps of the scheduling method for computing power network resources in the foregoing embodiment, and the working principles and beneficial effects of the two correspond to each other one by one, so that the description is omitted.
Referring to fig. 5, which is a schematic structural diagram of a scheduling device for computing power network resources according to an embodiment of the present invention, the embodiment of the present invention further provides a scheduling device 30 for computing power network resources, including a processor 31, a memory 32, and a computer program stored in the memory and configured to be executed by the processor, where the processor implements the scheduling method for computing power network resources according to any one of the foregoing embodiments when executing the computer program.
The embodiment of the invention also provides a computer readable storage medium, which comprises a stored computer program, wherein when the computer program runs, the equipment where the computer readable storage medium is located is controlled to execute the method for scheduling the power network resources according to any one of the embodiments.
Those skilled in the art will appreciate that implementing all or part of the above-described methods in accordance with the embodiments may be accomplished by way of a computer program stored on a computer readable storage medium, which when executed may comprise the steps of the embodiments of the methods described above. The storage medium may be a magnetic disk, an optical disk, a Read-Only Memory (ROM), a random-access Memory (Random Access Memory, RAM), or the like.
While the foregoing is directed to the preferred embodiments of the present invention, it will be appreciated by those skilled in the art that changes and modifications may be made without departing from the principles of the invention, such changes and modifications are also intended to be within the scope of the invention.
Claims (13)
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311016573.9A CN118827504A (en) | 2023-08-11 | 2023-08-11 | A method, device, equipment and medium for scheduling computing power network resources |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311016573.9A CN118827504A (en) | 2023-08-11 | 2023-08-11 | A method, device, equipment and medium for scheduling computing power network resources |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN118827504A true CN118827504A (en) | 2024-10-22 |
Family
ID=93069061
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311016573.9A Pending CN118827504A (en) | 2023-08-11 | 2023-08-11 | A method, device, equipment and medium for scheduling computing power network resources |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN118827504A (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111602149A (en) * | 2018-01-30 | 2020-08-28 | D5Ai有限责任公司 | Self-Organizing Partially Ordered Network |
| CN115550370A (en) * | 2022-12-01 | 2022-12-30 | 浩鲸云计算科技股份有限公司 | Computing power resource optimal scheduling allocation method based on multi-factor strategy |
| CN115942498A (en) * | 2022-11-29 | 2023-04-07 | 浪潮通信技术有限公司 | Route distribution method, device, equipment and storage medium for computing power awareness network |
| CN116248577A (en) * | 2022-12-08 | 2023-06-09 | 中国联合网络通信集团有限公司 | A method and device for determining computing power nodes |
-
2023
- 2023-08-11 CN CN202311016573.9A patent/CN118827504A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111602149A (en) * | 2018-01-30 | 2020-08-28 | D5Ai有限责任公司 | Self-Organizing Partially Ordered Network |
| CN115942498A (en) * | 2022-11-29 | 2023-04-07 | 浪潮通信技术有限公司 | Route distribution method, device, equipment and storage medium for computing power awareness network |
| CN115550370A (en) * | 2022-12-01 | 2022-12-30 | 浩鲸云计算科技股份有限公司 | Computing power resource optimal scheduling allocation method based on multi-factor strategy |
| CN116248577A (en) * | 2022-12-08 | 2023-06-09 | 中国联合网络通信集团有限公司 | A method and device for determining computing power nodes |
Non-Patent Citations (1)
| Title |
|---|
| 周俊: "算网大脑智能化建模和决策方法研究", 电信科学, no. 9, 8 September 2023 (2023-09-08) * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111507601A (en) | Resource optimization allocation decision-making method based on deep reinforcement learning and blockchain consensus | |
| CN116523079A (en) | Reinforced learning-based federal learning optimization method and system | |
| CN113824489B (en) | Satellite network resource dynamic allocation method, system and device based on deep learning | |
| CN111405569A (en) | Method and device for computing offloading and resource allocation based on deep reinforcement learning | |
| CN113918240B (en) | Task offloading method and device | |
| CN114584951B (en) | A joint computation offloading and resource allocation method based on multi-agent DDQN | |
| CN115202400B (en) | Unmanned aerial vehicle cluster task planning method based on self-adaptive punishment TAEA (total automatic analysis) | |
| CN115358831A (en) | User bidding method and device based on multi-agent reinforcement learning algorithm under federal learning | |
| CN117687714A (en) | A mine emergency response task offloading method based on multi-agent reinforcement learning | |
| CN114172558B (en) | Task unloading method based on edge calculation and unmanned aerial vehicle cluster cooperation in vehicle network | |
| CN117669741A (en) | Dynamic collaborative reasoning method for UAV cluster size model based on genetic algorithm | |
| CN117675823A (en) | Task processing method and device of computing power network, electronic equipment and storage medium | |
| CN115562746A (en) | A MEC associated task offloading method with limited computing resources | |
| CN115499440A (en) | A Serverless Edge Task Offloading Method Based on Experience Sharing Deep Reinforcement Learning | |
| CN119323240A (en) | Time delay and energy consumption optimization method based on federal learning | |
| CN118170524A (en) | Task scheduling method, device, equipment, medium and product based on reinforcement learning | |
| CN118093141A (en) | Edge computing self-adaptive task scheduling method based on meta reinforcement learning | |
| CN113722112A (en) | Service resource load balancing processing method and system | |
| Hossain et al. | Edge orchestration based computation peer offloading in MEC-enabled networks: a fuzzy logic approach | |
| CN109375999A (en) | A Bayesian Network-based Random Task Transfer Method for MEC | |
| Gu et al. | DRL-based joint task scheduling and trajectory planning method for UAV-assisted MEC scenarios | |
| CN118827504A (en) | A method, device, equipment and medium for scheduling computing power network resources | |
| CN115118591A (en) | Cluster federation learning method based on alliance game | |
| CN119011583A (en) | Edge computing unloading method and system for graph neural network assisted reinforcement learning | |
| CN116880540B (en) | Heterogeneous unmanned aerial vehicle group task allocation method based on alliance game formation |
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 |