A slice orchestration method and device and a computer readable storage medium

Abstract

The application provides a slice arrangement method and device and a computer readable storage medium. The method comprises: mapping actual telecommunication identification network parameters to obtain a digital twin network; using an intelligent optimization arrangement algorithm based on deep learning to perform slice arrangement on the digital twin network to obtain target slice arrangement parameters; and transmitting the target slice arrangement parameters to the actual telecommunication identification network to perform actual slice arrangement. The method, device and medium can solve the problem that existing slice arrangement considering only a single dimension ignores the time delay caused by various parameters of the actual network structure.

Description

Technical Field

[0001] This invention relates to a slice arrangement method, apparatus, and computer-readable storage medium. Background Technology

[0002] In modern industrial production environments, higher computing power and lower latency are often required. However, existing slicing and orchestration methods are often limited by geographical distance, that is, simply providing computing power at the nearest distance to ensure lower latency. This is usually not optimal, as slicing and orchestration only considers a single dimension and ignores the latency caused by various parameters of the actual network structure. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to address the above-mentioned shortcomings of the prior art by providing a slicing arrangement method, apparatus and computer-readable storage medium. The method utilizes a digital twin network mapped from actual telecommunications network parameters to consider various parameters of the actual network when performing network slicing arrangement, thereby solving the latency problem caused by the existing slicing arrangement that only considers a single dimension and ignores various parameters of the actual network structure.

[0004] In a first aspect, the present invention provides a slice arrangement method, comprising:

[0005] Mapping actual telecommunications identification network parameters to obtain a digital twin network;

[0006] A deep learning-based intelligent optimization orchestration algorithm is used to slice and orchestrate the digital twin network to obtain the target slice orchestration parameters;

[0007] The target slice orchestration parameters are passed to the actual telecommunications identification network for actual slice orchestration.

[0008] Preferably, before mapping the actual telecommunications identification network parameters to obtain the digital twin network, the method further includes:

[0009] Obtain network data of the actual telecommunications identification network;

[0010] The network data is processed using deep packet inspection technology to obtain actual telecommunications identification network parameters.

[0011] Preferably, the digital twin network is a cargo transportation network, and the mapping of actual telecommunications identification network parameters to obtain the digital twin network specifically includes:

[0012] Map the data processing parameters of the actual telecommunications identification network to the revenue of nodes in the cargo transportation network;

[0013] Map the data occupancy parameters of the actual telecommunications identification network to the cost of nodes in the cargo transportation network;

[0014] The cost parameters of inter-node transmission in the actual telecommunications identification network are mapped to the distance of links in the freight transportation network.

[0015] Preferably, the data processing parameters include bandwidth, data storage capacity, data computing capacity, and latency;

[0016] The data usage parameters include resource usage, energy consumption, and data management costs;

[0017] The cost parameters for inter-node transmission include bandwidth usage and transmission fees.

[0018] Preferably, the mapping of data processing parameters of the actual telecommunications identification network to the revenue of nodes in the freight transport network satisfies the following formula:

[0019]

[0020] Where Prof(i) is the revenue of node i, Wid(i) is the bandwidth of node i, Mem(i) is the data storage capacity of node i, Cal(i) is the data computing capacity of node i, and Del(i) is the latency of node i. It is used to adjust the weights between parameters.

[0021] Preferably, the cost of mapping the data occupancy parameter of the actual telecommunications identification network to the cost of nodes in the freight transport network satisfies the following formula:

[0022]

[0023] Where Cost(i) is the cost of node i, Occ(i) is the resource consumption of node i, Enc(i) is the energy consumption of node i, and Man(i) is the data management cost of node i. It is used to adjust the weights between parameters.

[0024] Preferably, the cost parameter for transmission between nodes in the actual telecommunications identification network is mapped to the distance of a link in the freight transport network, satisfying the following formula:

[0025]

[0026] Where Dis(i,j) is the distance of the link between nodes i and j, and Cost(i,j) is the cost parameter for data transmission between nodes i and j. It is used to adjust the weights between parameters.

[0027] Preferably, the digital model of the cargo transportation network is as follows:

[0028] G = (V, E)

[0029]

[0030] Where G is the digital model, V is the set of all nodes in the digital model, and n k Let E be the encoding of the k-th node, and E be the set of all links in the digital model.

[0031] Preferably, the step of using a deep learning-based intelligent optimization orchestration algorithm to slice and orchestrate the digital twin network to obtain target slice orchestration parameters specifically includes:

[0032] Step S101: Initialize the neural network;

[0033] Step S102: Obtain the digital twin network parameter state S and initialize the experience replay set;

[0034] Step S103: Initialize the current level i of the digital twin network to 1;

[0035] Step S104: Determine the current node group for this operation based on the current level i of the digital twin network;

[0036] Step S105: Obtain the current level digital twin network state parameters S based on the digital twin network parameter state S. i And based on the current state of the digital twin network S i Obtain the corresponding network feature vector φ(S) i );

[0037] Step S106, the network feature vector φ(S) is... i The input is fed into the neural network to update the neural network after arranging it according to the current node group;

[0038] Step S107: Determine whether the current level i of the current digital twin network meets the digital twin network level threshold. If yes, proceed to step S108; otherwise, increment the current level i by 1 and return to step S104.

[0039] Step S108: Obtain the target slice arrangement parameters based on the updated neural network.

[0040] Preferably, in step S106, the network feature vector φ(S) is... i The input to the neural network is used to update the neural network after arranging the neural network according to the current node group, including:

[0041] Step S201, the network feature vector φ(S) is... iInput the data into the neural network to obtain the current node group and the connection matrix of the links in the current node group;

[0042] Step S202: Initialize the current level k of the neural network to 1;

[0043] Step S203: Arrange the connection matrix in the neural network at the current level k to obtain the arranged current node and links in the current node group;

[0044] Step S204: Evaluate the current node and the links in the current node group after the arrangement to obtain the current evaluation value and the current reward value;

[0045] Step S205, the network feature vector φ(S) is... i The current node and the links in the current node group, the current evaluation value, and the current reward value are stored in the experience replay set and labeled with the current level k of the neural network;

[0046] Step S206: Determine whether the current neural network level k meets the neural network level threshold. If yes, proceed to step S207. If no, increment the current level k by 1 and return to step S203.

[0047] Step S207: Extract m samples from the experience replay set;

[0048] Step S208: Calculate the target values ​​for m samples in the target network;

[0049] Step S209: Update the neural network according to the target value.

[0050] Preferably, the arrangement of the connection matrix to obtain the arranged current node and links in the current node group satisfies the following formula:

[0051] G V (n s ) = SF V (n v )

[0052]

[0053] Where, n v It is a set of underlying physical nodes that satisfy the classification of identification data in the Telecommunication Identification Network (TIN), n s It is a set of arranged slice nodes, G V (n s ) is the node network after node orchestration, SF v It is an algorithm for intelligent node orchestration. It is the path connecting physical nodes, and m is the virtual link l. s The number of physical paths, G C (l s ) is a sliced ​​network after link orchestration, SF C It is an algorithm for intelligent link orchestration.

[0054] Preferably, the evaluation of the current node and the links in the current node group after the arrangement to obtain the current evaluation value and the current reward value satisfies the following formula:

[0055]

[0056] Rew=∑Prof(i,j)-∑Cost(i,j)

[0057] Where Cost(i) and Cost(j) are the costs of nodes i and j, Up(i,j) is the unit price between nodes (i,j), Dis(i,j) is the distance of the link between nodes i and j, Cost(i,j) is the cost parameter for data transmission between nodes i and j, Prof(i,j) is the revenue from transmission between nodes i and j, Rew is the reward value, and Eva is the evaluation value.

[0058] Preferably, the target network is a copy of the current neural network;

[0059] The target value for m samples in the target network is calculated according to the following formula:

[0060]

[0061] Among them, Rew j It is the reward for sample j. It is the maximum value of the evaluation value obtained by the target network, γ is the attenuation coefficient, and y j It is the target value of sample j.

[0062] Preferably, updating the neural network based on the target value includes:

[0063] Calculate the mean squared error loss for each sample;

[0064] The neural network is updated based on the mean squared error loss of each sample.

[0065] Preferably, the calculation of the mean squared error loss for each sample satisfies the following formula:

[0066]

[0067] Among them, Eva j It is the evaluation value after arranging the neural network and inputting the parameters of the target neural network, Lj It is the mean squared error loss.

[0068] Preferably, updating the neural network based on the mean squared error loss of each sample includes:

[0069] The mean squared error loss of each sample is used to update the neural network using gradient backpropagation.

[0070] In a second aspect, the present invention also provides a slicing and arranging device, comprising:

[0071] The mapping module is used to map actual telecommunications identification network parameters to obtain a digital twin network;

[0072] The orchestration module, connected to the mapping module, is used to perform slice orchestration on the digital twin network using a deep learning-based intelligent optimization orchestration algorithm to obtain target slice orchestration parameters.

[0073] The transmission module, connected to the orchestration module, is used to transmit the target slice orchestration parameters to the actual telecommunications identification network for actual slice orchestration.

[0074] Thirdly, the present invention also provides a slicing arrangement apparatus, including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to implement the slicing arrangement method described in the first aspect above.

[0075] Fourthly, the present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the slicing and arrangement method described in the first aspect above.

[0076] The present invention provides a slicing orchestration method, apparatus, and computer-readable storage medium that maps actual telecommunications identification network parameters to obtain a digital twin network. A deep learning-based intelligent optimization orchestration algorithm is used to slice and orchestrate the digital twin network to obtain target slice orchestration parameters. These target slice orchestration parameters are then transmitted to the actual telecommunications identification network for actual slice orchestration. Because the present invention utilizes the digital twin network mapped from actual telecommunications network parameters, it considers various parameters of the actual network during network slice orchestration, thereby solving the latency problem caused by existing slicing orchestration methods that only consider a single dimension and ignore various parameters of the actual network structure. Attached Figure Description

[0077] Figure 1 This is a flowchart of a slice arrangement method according to Embodiment 1 of the present invention;

[0078] Figure 2 for Figure 1 Flowchart of step S2;

[0079] Figure 3 for Figure 2 Flowchart of step S106;

[0080] Figure 4 This is a schematic diagram of the structure of a slicing and arranging device according to Embodiment 2 of the present invention;

[0081] Figure 5 This is a schematic diagram of a slicing and arranging device according to Embodiment 3 of the present invention. Detailed Implementation

[0082] To enable those skilled in the art to better understand the technical solution of the present invention, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0083] It is understood that the specific embodiments and accompanying drawings described herein are merely for explaining the invention and are not intended to limit the invention.

[0084] It is understood that, without conflict, the various embodiments and features in the embodiments of the present invention can be combined with each other.

[0085] It is understood that, for ease of description, only the parts related to the present invention are shown in the accompanying drawings, while the parts unrelated to the present invention are not shown in the drawings.

[0086] It is understood that each unit or module involved in the embodiments of the present invention may correspond to only one entity structure, or may be composed of multiple entity structures, or multiple units or modules may be integrated into one entity structure.

[0087] It is understood that, without conflict, the functions and steps marked in the flowcharts and block diagrams of this invention may occur in a different order than that marked in the accompanying drawings.

[0088] It is understood that the flowcharts and block diagrams of this invention illustrate the possible architecture, functions, and operations of systems, apparatuses, devices, and methods according to various embodiments of this invention. Each block in the flowchart or block diagram may represent a unit, module, program segment, or code, containing executable instructions for implementing the specified function. Furthermore, each block or combination of blocks in the block diagram and flowchart can be implemented using a hardware-based system to achieve the specified function, or using a combination of hardware and computer instructions.

[0089] It is understood that the units and modules involved in the embodiments of the present invention can be implemented by software or by hardware. For example, the units and modules can be located in a processor.

[0090] Example 1:

[0091] This embodiment provides a slice arrangement method, such as Figure 1 As shown, the method includes:

[0092] Step S1: Map the actual telecommunications identification network parameters to obtain a digital twin network.

[0093] In this embodiment, there is a large amount of identification code data in the actual telecommunications identification network. Each identification code has a storage code to record the storage location of the identification code. The storage code consists of the national central node code, the regional central / provincial capital node code, the local core / edge node code, and the edge internal node code. In order to build the intelligent slicing and orchestration algorithm of TIN, it is necessary to first build the digital twin network of TIN.

[0094] Optionally, before mapping the actual telecommunications identification network parameters to obtain the digital twin network, the method further includes:

[0095] Obtain network data of the actual telecommunications identification network;

[0096] The network data is processed using deep packet inspection technology to obtain actual telecommunications identification network parameters.

[0097] In this embodiment, network data acquisition and processing components can be installed at each node of the actual telecommunications network. After acquiring the actual telecommunications network data, deep packet inspection technology is used to analyze and summarize the data, obtain various parameters, and form data records.

[0098] Optionally, the digital twin network is a cargo transportation network, and the mapping of actual telecommunications identification network parameters to obtain the digital twin network specifically includes:

[0099] Map the data processing parameters of the actual telecommunications identification network to the revenue of nodes in the cargo transportation network;

[0100] Map the data occupancy parameters of the actual telecommunications identification network to the cost of nodes in the cargo transportation network;

[0101] The cost parameters of inter-node transmission in the actual telecommunications identification network are mapped to the distance of links in the freight transportation network.

[0102] In this embodiment, since a fully digital twin of a real telecommunications network requires the creation of a high-fidelity virtual model to realistically reproduce the structure, function, real-time operating status, performance parameters, network management and maintenance of the physical network, the implementation is difficult and costly. Therefore, in combination with the characteristics and requirements of the TIN network, and referring to the standard model architecture of digital twins, the complex TIN network parameters are normalized and simplified. The digital twin network is a freight transportation network, and the various identification parameters of the TIN are mapped to different types of freight transportation network parameters, thereby avoiding the complexity and high cost of a fully digital twin of the actual telecommunications identification network.

[0103] Optionally, the data processing parameters include bandwidth, data storage capacity, data computing capacity, and latency;

[0104] The data usage parameters include resource usage, energy consumption, and data management costs;

[0105] The cost parameters for inter-node transmission include bandwidth usage and transmission fees.

[0106] Optionally, the mapping of data processing parameters of the actual telecommunications identification network to the revenue of nodes in the freight transport network satisfies the following formula:

[0107]

[0108] Where Prof(i) is the revenue of node i, Wid(i) is the bandwidth of node i, Mem(i) is the data storage capacity of node i, Cal(i) is the data computing capacity of node i, and Del(i) is the latency of node i. It is used to adjust the weights between parameters.

[0109] Optionally, the cost of mapping the data occupancy parameter of the actual telecommunications identification network to the cost of nodes in the freight transport network satisfies the following formula:

[0110]

[0111] Where Cost(i) is the cost of node i, Occ(i) is the resource consumption of node i, Enc(i) is the energy consumption of node i, and Man(i) is the data management cost of node i. It is used to adjust the weights between parameters.

[0112] Optionally, the cost parameter for transmission between nodes in the actual telecommunications identification network is mapped to the distance of a link in the freight transport network, satisfying the following formula:

[0113]

[0114] Where Dis(i,j) is the distance of the link between nodes i and j, and Cost(i,j) is the cost parameter for data transmission between nodes i and j. It is used to adjust the weights between parameters.

[0115] Optionally, the digital model of the cargo transportation network is:

[0116] G = (V, E)

[0117]

[0118] Where G is the digital model, V is the set of all nodes in the digital model, and n k Let E be the encoding of the k-th node, and E be the set of all links in the digital model.

[0119] In this embodiment, n k It consists of national central node codes, regional central / provincial capital node codes, local core / edge node codes, and edge internal node codes, which are used to quickly identify node levels and the relationships between nodes. Various benefits of the digital twin network can be settled at the nodes.

[0120] Step S2: Use a deep learning-based intelligent optimization orchestration algorithm to slice and orchestrate the digital twin network to obtain the target slice orchestration parameters.

[0121] In this embodiment, the slice orchestration algorithm is based on Deep Q-Learning and combines the characteristics of TIN networks to achieve intelligent orchestration.

[0122] Optionally, such as Figure 2 As shown, the method of using a deep learning-based intelligent optimization orchestration algorithm to slice and orchestrate the digital twin network to obtain target slice orchestration parameters specifically includes:

[0123] Step S101: Initialize the neural network;

[0124] Step S102: Obtain the digital twin network parameter state S and initialize the experience replay set;

[0125] Step S103: Initialize the current level i of the digital twin network to 1;

[0126] Step S104: Determine the current node group for this operation based on the current level i of the digital twin network;

[0127] Step S105: Obtain the current level digital twin network state parameters S based on the digital twin network parameter state S. i And based on the current state of the digital twin network S iObtain the corresponding network feature vector φ(S) i );

[0128] Step S106, the network feature vector φ(S) is... i The input is fed into the neural network to update the neural network after arranging it according to the current node group;

[0129] Step S107: Determine whether the current level i of the current digital twin network meets the digital twin network level threshold. If yes, proceed to step S108; otherwise, increment the current level i by 1 and return to step S104.

[0130] Step S108: Obtain the target slice arrangement parameters based on the updated neural network.

[0131] In this embodiment, the neural network continuously optimizes the cargo transportation network through continuous updates. The level of the digital twin network is determined based on the node encoding, and the digital twin network parameter state S is the real-time state of various parameters that determine costs and benefits.

[0132] Optionally, such as Figure 3 As shown, in step S106, the network feature vector φ(S) is... i The input to the neural network is used to update the neural network after arranging the neural network according to the current node group, including:

[0133] Step S201, the network feature vector φ(S) is... i Input the data into the neural network to obtain the current node group and the connection matrix of the links in the current node group;

[0134] Step S202: Initialize the current level k of the neural network to 1;

[0135] Step S203: Arrange the connection matrix in the neural network at the current level k to obtain the arranged current node and links in the current node group;

[0136] Step S204: Evaluate the current node and the links in the current node group after the arrangement to obtain the current evaluation value and the current reward value;

[0137] Step S205, the network feature vector φ(S) is... i The current node and the links in the current node group, the current evaluation value, and the current reward value are stored in the experience replay set and labeled with the current level k of the neural network;

[0138] Step S206: Determine whether the current neural network level k meets the neural network level threshold. If yes, proceed to step S207. If no, increment the current level k by 1 and return to step S203.

[0139] Step S207: Extract m samples from the experience replay set;

[0140] Step S208: Calculate the target values ​​for m samples in the target network;

[0141] Step S209: Update the neural network according to the target value.

[0142] In this embodiment, samples drawn from the experience replay set are used to continuously optimize the neural network. A first part of the samples can be randomly drawn from the experience replay set, and a second part of the samples can be selected based on the evaluation values ​​of the samples. By adjusting the ratio of the first part of the samples and the second part of the samples, it is possible to both ensure the independence of the samples and select better-performing samples from historical experience.

[0143] Optionally, the arrangement of the connection matrix to obtain the arranged current node and links in the current node group satisfies the following formula:

[0144] G V (n s ) = SF V (n v )

[0145]

[0146] Where, n v It is a set of underlying physical nodes that satisfy the classification of Telecommunication Identification Network (TIN) identifier data, n s It is a set of arranged slice nodes, G V (n s ) is the node network after node orchestration, SF v It is an algorithm for intelligent node orchestration. It is the path connecting physical nodes, and m is the virtual link l. s The number of physical paths, G C (l s ) is a sliced ​​network after link orchestration, SF C It is an algorithm for intelligent link orchestration.

[0147] Optionally, the evaluation of the current node and the links in the current node group after the arrangement is performed to obtain the current evaluation value and the current reward value, satisfying the following formula:

[0148]

[0149] Rew=∑Prof(i,j)-∑Cost(i,j)

[0150] Where Cost(i) and Cost(j) are the costs of nodes i and j, Up(i,j) is the unit price between nodes (i,j), Dis(i,j) is the distance of the link between nodes i and j, Cost(i,j) is the cost parameter for data transmission between nodes i and j, Prof(i,j) is the revenue from transmission between nodes i and j, Rew is the reward value, and Eva is the evaluation value.

[0151] Optionally, the target network is a copy of the current neural network;

[0152] The target value for m samples in the target network is calculated according to the following formula:

[0153]

[0154] Among them, Rew j It is the reward for sample j. It is the maximum value of the evaluation value obtained by the target network, γ is the attenuation coefficient, and y j It is the target value of sample j.

[0155] In this embodiment, the target network is used to evaluate the benefits as a learning objective, and the value of γ can be between 0 and 1.

[0156] Optionally, updating the neural network based on the target value includes:

[0157] Calculate the mean squared error loss for each sample;

[0158] The neural network is updated based on the mean squared error loss of each sample.

[0159] Optionally, the calculation of the mean squared error loss for each sample satisfies the following formula:

[0160]

[0161] Among them, Eva j It is the evaluation value after arranging the neural network and inputting the parameters of the target neural network, L j It is the mean squared error loss.

[0162] Optionally, updating the neural network based on the mean squared error loss of each sample includes:

[0163] The mean squared error loss of each sample is used to update the neural network using gradient backpropagation.

[0164] In this embodiment, the mean squared error loss represents the difference between the sample and the optimal target. The set of parameters with the minimum mean squared error loss can be updated in the neural network.

[0165] Step S3: Transmit the target slice orchestration parameters to the actual telecommunications identification network for actual slice orchestration.

[0166] In this embodiment, by continuously learning from samples through a neural network, a set of target slice arrangement parameters can be obtained. This set of target slice arrangement parameters is then transmitted to the actual telecommunications network for actual slice arrangement.

[0167] The slicing orchestration method provided in this invention maps actual telecommunications identification network parameters to obtain a digital twin network, employs a deep learning-based intelligent optimization orchestration algorithm to slice and orchestrate the digital twin network to obtain target slice orchestration parameters, and transmits the target slice orchestration parameters to the actual telecommunications identification network for actual slice orchestration. Because this invention utilizes the digital twin network mapped from actual telecommunications network parameters, it considers various parameters of the actual network when performing network slice orchestration, thereby solving the latency problem caused by existing slice orchestration methods that only consider a single dimension and ignore various parameters of the actual network structure.

[0168] Example 2:

[0169] like Figure 4 As shown, this embodiment provides a slice arrangement apparatus for performing the above-described slice arrangement method, including:

[0170] Mapping module 11 is used to map actual telecommunications identification network parameters to obtain a digital twin network;

[0171] The orchestration module 12, connected to the mapping module 11, is used to perform slice orchestration on the digital twin network using a deep learning-based intelligent optimization orchestration algorithm to obtain target slice orchestration parameters.

[0172] The transmission module 13, connected to the orchestration module 12, is used to transmit the target slice orchestration parameters to the actual telecommunications identification network for actual slice orchestration.

[0173] Preferably, the device further includes:

[0174] The network data module is used to acquire network data of the actual telecommunications identification network;

[0175] The data processing module is used to process the network data using deep packet inspection technology to obtain actual telecommunications identification network parameters.

[0176] Optionally, the mapping module 11 includes:

[0177] A revenue mapping unit is used to map the data processing parameters of the actual telecommunications identification network to the revenue of nodes in the cargo transportation network.

[0178] The cost mapping unit is used to map the data occupancy parameters of the actual telecommunications identification network to the cost of nodes in the cargo transportation network.

[0179] The distance mapping unit is used to map the cost parameters of inter-node transmission in the actual telecommunications identification network to the distance of links in the freight transportation network.

[0180] Optionally, the data processing parameters include bandwidth, data storage capacity, data computing capacity, and latency;

[0181] The data usage parameters include resource usage, energy consumption, and data management costs;

[0182] The cost parameters for inter-node transmission include bandwidth usage and transmission fees.

[0183] Optionally, the revenue mapping unit satisfies the following formula:

[0184]

[0185] Where Prof(i) is the revenue of node i, Wid(i) is the bandwidth of node i, Mem(i) is the data storage capacity of node i, Cal(i) is the data computing capacity of node i, and Del(i) is the latency of node i. It is used to adjust the weights between parameters.

[0186] Optionally, the cost mapping unit satisfies the following formula:

[0187]

[0188] Where Cost(i) is the cost of node i, Occ(i) is the resource consumption of node i, Enc(i) is the energy consumption of node i, and Man(i) is the data management cost of node i. It is used to adjust the weights between parameters.

[0189] Optionally, the distance mapping unit satisfies the following formula:

[0190]

[0191] Where Dis(i,j) is the distance of the link between nodes i and j, and Cost(i,j) is the cost parameter for data transmission between nodes i and j. It is used to adjust the weights between parameters.

[0192] Optionally, the digital model of the cargo transportation network is:

[0193] G = (V, E)

[0194]

[0195] Where G is the digital model, V is the set of all nodes in the digital model, and n k Let E be the encoding of the k-th node, and E be the set of all links in the digital model.

[0196] Optionally, the orchestration module 12 includes:

[0197] The first initialization unit is used to initialize the neural network;

[0198] The replay set unit is used to obtain the digital twin network parameter state S and initialize the experience replay set;

[0199] The second initialization unit is used to initialize the current level i of the digital twin network to 1;

[0200] The node group unit is used to determine the current node group for this operation based on the current level i of the digital twin network.

[0201] Feature vector unit, used to obtain the current level digital twin network state parameters S based on the digital twin network parameter state S. i And based on the current state of the digital twin network S i Obtain the corresponding network feature vector φ(S) i );

[0202] The update unit is used to update the network feature vector φ(S) i The input is fed into the neural network to update the neural network after arranging it according to the current node group;

[0203] The first judgment unit is used to determine whether the current level i of the current digital twin network meets the digital twin network level threshold. If it does, the parameter unit is executed; otherwise, the current level i is incremented by 1 and the process is returned to the node group unit.

[0204] The parameter unit is used to obtain the target slice arrangement parameters based on the updated neural network.

[0205] Optionally, the update unit includes:

[0206] Input unit, used to input the network feature vector φ(S) i Input the data into the neural network to obtain the current node group and the connection matrix of the links in the current node group;

[0207] The third initialization unit is used to initialize the current level k of the neural network to 1;

[0208] An orchestration unit is used to orchestrate the connection matrix in the neural network at the current level k to obtain the orchestrated current node and links in the current node group;

[0209] The evaluation unit is used to evaluate the current node and the links in the current node group after the arrangement to obtain the current evaluation value and the current reward value;

[0210] The annotation unit is used to annotate the network feature vector φ(S) i The current node and the links in the current node group, the current evaluation value, and the current reward value are stored in the experience replay set and labeled with the current level k of the neural network;

[0211] The second judgment unit is used to determine whether the current neural network level k meets the neural network level threshold. If yes, the extraction unit is executed; otherwise, the current level k is incremented by 1 and the process is returned to the orchestration unit.

[0212] Extraction unit, used to extract m samples from the experience replay set;

[0213] A computational unit is used to compute the target value for m samples in the target network;

[0214] A target value unit is used to update the neural network based on the target value.

[0215] Optionally, the arrangement unit satisfies the following formula:

[0216] G V (n s ) = SF V (n v )

[0217]

[0218] Where, n v It is a set of underlying physical nodes that satisfy the classification of Telecommunication Identification Network (TIN) identifier data, n s It is a set of arranged slice nodes, G N (n s ) is the node network after node orchestration, SF v It is an algorithm for intelligent node orchestration. It is the path connecting physical nodes, and m is the virtual link l. s The number of physical paths, G C (l s ) is a sliced ​​network after link orchestration, SF C It is an algorithm for intelligent link orchestration.

[0219] Optionally, the evaluation unit satisfies the following formula:

[0220]

[0221] Rew=∑Prof(i,j)-∑Cost(i,j)

[0222] Where Cost(i) and Cost(j) are the costs of nodes i and j, Up(i,j) is the unit price between nodes (i,j), Dis(i,j) is the distance of the link between nodes i and j, Cost(i,j) is the cost parameter for data transmission between nodes i and j, Prof(i,j) is the revenue from transmission between nodes i and j, Rew is the reward value, and Eva is the evaluation value.

[0223] Optionally, the target network is a copy of the current neural network;

[0224] The calculation unit satisfies the following formula:

[0225]

[0226] Among them, Rew j It is the reward for sample j. It is the maximum value of the evaluation value obtained by the target network, γ is the attenuation coefficient, and y j It is the target value of sample j.

[0227] Optionally, the target value unit includes:

[0228] The loss unit is used to calculate the mean squared error loss for each sample.

[0229] The mean squared error unit is used to update the neural network based on the mean squared error loss of each sample.

[0230] Optionally, the loss unit satisfies the following formula:

[0231]

[0232] Among them, Eva j It is the evaluation value after arranging the neural network and inputting the parameters of the target neural network, L j It is the mean squared error loss.

[0233] Optionally, the mean squared error unit is also used to update the neural network using gradient backpropagation on the mean squared error loss of each sample.

[0234] Example 3:

[0235] like Figure 5As shown, this embodiment provides a slicing arrangement apparatus for executing the above-described slicing arrangement method, including a memory 21 and a processor 22. The memory 21 stores a computer program, and the processor 22 is configured to run the computer program to execute the slicing arrangement method in Embodiment 1.

[0236] The memory 21 is connected to the processor 22. The memory 21 can be a flash memory, a read-only memory or other memory, and the processor 22 can be a central processing unit or a microcontroller.

[0237] Example 4:

[0238] This embodiment provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the slicing arrangement method in Embodiment 1 above.

[0239] The computer-readable storage medium includes volatile or non-volatile, removable or non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, computer program modules, or other data). Computer-readable storage media include, but are not limited to, RAM (Random Access Memory), ROM (Read-Only Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory), flash memory or other memory technologies, CD-ROM (Compact Disc Read-Only Memory), DVD or other optical disc storage, cartridges, magnetic tapes, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer.

[0240] The slicing orchestration apparatus and computer-readable storage medium provided in Examples 2 to 4 map actual telecommunications identification network parameters to obtain a digital twin network. A deep learning-based intelligent optimization orchestration algorithm is used to slice and orchestrate the digital twin network to obtain target slice orchestration parameters. These target slice orchestration parameters are then transmitted to the actual telecommunications identification network for actual slice orchestration. Because this invention utilizes the digital twin network mapped from actual telecommunications network parameters, it considers various parameters of the actual network during network slice orchestration, thereby solving the latency problem caused by existing slice orchestration methods that only consider a single dimension and ignore various parameters of the actual network structure.

[0241] It is understood that the above embodiments are merely exemplary implementations used to illustrate the principles of the present invention, and the present invention is not limited thereto. For those skilled in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also considered to be within the scope of protection of the present invention.

Claims

1. A method for arranging slices, characterized in that, include: Mapping actual telecommunications identification network parameters to obtain a digital twin network; A deep learning-based intelligent optimization orchestration algorithm is used to slice and orchestrate the digital twin network to obtain the target slice orchestration parameters; The target slice orchestration parameters are passed to the actual telecommunications identification network for actual slice orchestration. The digital twin network is a cargo transportation network, and the mapping of actual telecommunications identification network parameters to obtain the digital twin network specifically includes: The data processing parameters of the actual telecommunications identification network are mapped to the revenue of nodes in the cargo transportation network. The data processing parameters include bandwidth, data storage capacity, data computing capacity, and latency. The data occupancy parameters of the actual telecommunications identification network are mapped to the cost of nodes in the cargo transportation network. The data occupancy parameters include resource consumption, energy consumption, and data management costs. The cost parameters of inter-node transmission in the actual telecommunications identification network are mapped to the distance of the link in the cargo transportation network. The cost parameters of inter-node transmission include bandwidth usage and transmission fees. The mapping of data processing parameters of the actual telecommunications identification network to the revenue of nodes in the freight transportation network satisfies the following formula: Where Prof(i) is the revenue of node i, Wid(i) is the bandwidth of node i, Mem(i) is the data storage capacity of node i, Cal(i) is the data computing capacity of node i, and Del(i) is the latency of node i. These are the weights used to adjust the weights between parameters; The process of mapping the data occupancy parameters of the actual telecommunications identification network to the cost of nodes in the freight transport network satisfies the following formula: Where Cost(i) is the cost of node i, Occ(i) is the resource consumption of node i, Enc(i) is the energy consumption of node i, and Man(i) is the data management cost of node i. These are the weights used to adjust the weights between parameters; The mapping of cost parameters for transmission between nodes in the actual telecommunications identification network to the distance of links in the freight transportation network satisfies the following formula: Where Dis(i,j) is the distance of the link between nodes i and j, and Cost(i,j) is the cost parameter for data transmission between nodes i and j. It is used to adjust the weights between parameters.

2. The slice arrangement method according to claim 1, characterized in that, Before mapping the actual telecommunications identification network parameters to obtain the digital twin network, the process also includes: Obtain network data of the actual telecommunications identification network; The network data is processed using deep packet inspection technology to obtain actual telecommunications identification network parameters.

3. The slice arrangement method according to claim 1, characterized in that, The digital model of the cargo transportation network is as follows: G = (V, E) Where G is the digital model, V is the set of all nodes in the digital model, and n k Let E be the encoding of the k-th node, and E be the set of all links in the digital model.

4. The slice arrangement method according to claim 1, characterized in that, The method of employing a deep learning-based intelligent optimization orchestration algorithm to slice and orchestrate the digital twin network to obtain target slice orchestration parameters specifically includes: Step S101: Initialize the neural network; Step S102: Obtain the digital twin network parameter state S and initialize the experience replay set; Step S103: Initialize the current level i of the digital twin network to 1; Step S104: Determine the current node group for this operation based on the current level i of the digital twin network; Step S105: Obtain the current level digital twin network state parameters S based on the digital twin network parameter state S. i And based on the current state of the digital twin network S i Obtain the corresponding network feature vector φ(S) i ); Step S106, the network feature vector φ(S) is... i The input is fed into the neural network to update the neural network after arranging it according to the current node group; Step S107: Determine whether the current level i of the current digital twin network meets the digital twin network level threshold. If yes, proceed to step S108; otherwise, increment the current level i by 1 and return to step S104. Step S108: Obtain the target slice arrangement parameters based on the updated neural network.

5. The slice arrangement method according to claim 4, characterized in that, Step S106, the network feature vector φ(S) is... i The input to the neural network is used to update the neural network after arranging the neural network according to the current node group, including: Step S201, the network feature vector φ(S) is... i Input the data into the neural network to obtain the current node group and the connection matrix of the links in the current node group; Step S202: Initialize the current level k of the neural network to 1; Step S203: Arrange the connection matrix in the neural network at the current level k to obtain the arranged current node and links in the current node group; Step S204: Evaluate the current node and the links in the current node group after the arrangement to obtain the current evaluation value and the current reward value; Step S205, the network feature vector φ(S) is... i The current node and the links in the current node group, the current evaluation value, and the current reward value are stored in the experience replay set and labeled with the current level k of the neural network; Step S206: Determine whether the current neural network level k meets the neural network level threshold. If yes, proceed to step S207. If no, increment the current level k by 1 and return to step S203. Step S207: Extract m samples from the experience replay set; Step S208: Calculate the target values ​​for m samples in the target network; Step S209: Update the neural network according to the target value.

6. The slice arrangement method according to claim 5, characterized in that, The connection matrix is ​​arranged to obtain the current node and the links in the current node group, satisfying the following formula: G V (n s )=SF V (n v ) Where, n v It is a set of underlying physical nodes that satisfy the classification of Telecommunication Identification Network (TIN) identifier data, n s It is a set of arranged slice nodes, G V (n s ) is the node network after node orchestration, SF v It is an algorithm for intelligent node orchestration. It is the path connecting physical nodes, and m is the virtual link l. s The number of physical paths, G C (l s ) is a sliced ​​network after link orchestration, SF C It is an algorithm for intelligent link orchestration.

7. The slice arrangement method according to claim 5, characterized in that, The evaluation of the current node and the links in the current node group after the arrangement is performed to obtain the current evaluation value and the current reward value, satisfying the following formula: Rew = ΣProf(i,j) - ΣCost(i,j) Where Cost(i) and Cost(j) are the costs of nodes i and j, Up(i,j) is the unit price between nodes (i,j), Dis(i,j) is the distance of the link between nodes i and j, Cost(i,j) is the cost parameter for data transmission between nodes i and j, Prof(i,j) is the revenue from transmission between nodes i and j, Rew is the reward value, and Eva is the evaluation value.

8. The slice arrangement method according to claim 5, characterized in that, The target network is a copy of the current neural network; The target value for m samples in the target network is calculated according to the following formula: Among them, Rew j It is the reward for sample j. It is the maximum value of the evaluation value obtained by the target network, γ is the attenuation coefficient, and y j It is the target value of sample j.

9. The slice arrangement method according to claim 8, characterized in that, Updating the neural network based on the target value includes: Calculate the mean squared error loss for each sample; The neural network is updated based on the mean squared error loss of each sample.

10. The slice arrangement method according to claim 9, characterized in that, The calculation of the mean squared error loss for each sample satisfies the following formula: Among them, Eva j It is the evaluation value of sample j after arranging the neural network and inputting the parameters of the target neural network. j It is the mean squared error loss of sample j.

11. The slice arrangement method according to claim 9, characterized in that, The step of updating the neural network based on the mean squared error loss of each sample includes: The mean squared error loss of each sample is used to update the neural network using gradient backpropagation.

12. A slicing and arranging device, characterized in that, include: The mapping module is used to map actual telecommunications identification network parameters to obtain a digital twin network; The orchestration module, connected to the mapping module, is used to perform slice orchestration on the digital twin network using a deep learning-based intelligent optimization orchestration algorithm to obtain target slice orchestration parameters. The transmission module, connected to the orchestration module, is used to transmit the target slice orchestration parameters to the actual telecommunications identification network for actual slice orchestration. The mapping module includes: The revenue mapping unit is used to map the data processing parameters of the actual telecommunications identification network to the revenue of nodes in the cargo transportation network. The data processing parameters include bandwidth, data storage capacity, data computing capacity, and latency. The cost mapping unit is used to map the data occupancy parameters of the actual telecommunications identification network to the cost of nodes in the cargo transportation network. The data occupancy parameters include resource occupancy, energy consumption, and data management consumption. The distance mapping unit is used to map the cost parameters of inter-node transmission in the actual telecommunications identification network to the distance of the link in the cargo transportation network. The cost parameters of inter-node transmission include bandwidth usage and transmission fees. The revenue mapping unit satisfies the following formula: Where Prof(i) is the revenue of node i, Wid(i) is the bandwidth of node i, Mem(i) is the data storage capacity of node i, Cal(i) is the data computing capacity of node i, and Del(i) is the latency of node i. These are the weights used to adjust the weights between parameters; The cost mapping unit satisfies the following formula: Where Cost(i) is the cost of node i, Occ(i) is the resource consumption of node i, Enc(i) is the energy consumption of node i, and Man(i) is the data management cost of node i. These are the weights used to adjust the weights between parameters; The distance mapping unit satisfies the following formula: Where Dis(i,j) is the distance of the link between nodes i and j, and Cost(i,j) is the cost parameter for data transmission between nodes i and j. It is used to adjust the weights between parameters.

13. A slicing and arranging device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to implement the slicing arrangement method as described in any one of claims 1-11.

14. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the slicing arrangement method as described in any one of claims 1-11.

Images