Virtual network mapping method, device and equipment based on unmanned aerial vehicle cluster scene

Abstract

The invention provides a virtual network mapping method, device and equipment based on an unmanned aerial vehicle cluster scene, and the method comprises the steps: obtaining a topological graph sequence of an unmanned aerial vehicle cluster network when a virtual network embedding request is received; extracting space-time dynamic features among the topological graphs in the topological graph sequence based on a pre-constructed feature extraction network; inputting the space-time dynamic characteristics into a pre-trained node strategy generation network to generate a node mapping strategy; mapping to-be-embedded virtual nodes to unmanned aerial vehicle nodes of an unmanned aerial vehicle cluster network based on a node mapping strategy; and after all the virtual nodes are successfully mapped, mapping the virtual link to be embedded into the shortest physical link. According to the embodiment of the invention, the virtual mapping requirement in a network dynamic scene can be met; and the virtual nodes and the virtual links are mapped in sequence by adopting a two-section type mapping method, so that the nodes and the links can be successfully mapped, and the mapping success rate is improved.

Description

Technical Field

[0001] The present invention relates to the field of communication technology, and in particular to a virtual network mapping method, device and equipment based on a drone cluster scenario. Background Art

[0002] VNE (Virtual Network Embedding) is a virtual network resource allocation technology that achieves more flexible, efficient and reliable network services by mapping virtual nodes to physical nodes and virtual links to physical links on the basis of meeting the service quality requirements of the virtual network. Efficient virtual network mapping can improve the utilization of physical network resources.

[0003] In the prior art, virtual networks are mapped to physical networks. Some are based on heuristic or meta-heuristic algorithms, which gradually search and optimize mapping solutions through a series of heuristic strategies and rules. However, heuristic algorithms usually make decisions based on experience and heuristic rules, and may not guarantee the global optimal solution. Others use reinforcement learning algorithms, but existing reinforcement learning online algorithms are based on the static topology of the physical network, without considering the dynamic changes of the physical network, and cannot meet the virtual mapping requirements of dynamic network scenarios. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide a virtual network mapping method, device and equipment based on a drone cluster scenario to meet the virtual mapping requirements of dynamic network scenarios.

[0005] In a first aspect, a virtual network mapping method based on a drone cluster scenario is provided, the method comprising: Upon receiving a virtual network embedding request, a topology map sequence of the drone cluster network is obtained; the virtual network embedding request includes at least virtual nodes and virtual links to be embedded; the topology map sequence is composed of topology maps captured at each snapshot moment; Extracting spatiotemporal dynamic features between topological graphs in a topological graph sequence based on a pre-built feature extraction network; Input the spatiotemporal dynamic features into the pre-trained node strategy generation network to generate a node mapping strategy; Based on the node mapping strategy, the virtual nodes to be embedded are mapped to the drone nodes of the drone cluster network; After all virtual nodes are mapped successfully, the shortest physical link between the mapped drone nodes is determined based on the Floyd algorithm, and the virtual link to be embedded is mapped to the shortest physical link.

[0006] Optionally, the feature extraction network is composed of a graph convolutional network and a Transformer-XL network; extracting spatiotemporal dynamic features between topological graphs in a topological graph sequence based on a pre-built feature extraction network includes: Extract the spatial features of each topological graph one by one based on the graph convolutional network; Determine whether the number of topology maps in the topology map sequence exceeds a preset number threshold; If the number exceeds the preset threshold, the spatial features of the topological graphs with the preset threshold are input into the Transformer-XL network to extract the spatiotemporal dynamic features between the topological graphs with the preset threshold. If the preset number threshold is not exceeded, all spatial features extracted by the graph convolutional network are input into the Transformer-XL network to extract the spatiotemporal dynamic features between the topological graphs.

[0007] Optionally, the graph convolutional network includes two hidden layers; extracting spatial features of each topological graph one by one based on the graph convolutional network includes: Determine, for each topological graph, a first adjacency matrix of the drone nodes in the topological graph and an attribute feature matrix of the drone cluster network; Input the first adjacency matrix and attribute feature matrix into the first hidden layer to obtain the initial spatial features; The initial spatial features are input into the second hidden layer to extract higher-dimensional feature information and obtain the final spatial features of the topological map.

[0008] Optionally, the node strategy generation network adopts a deep reinforcement learning network based on the A3C algorithm, and the deep reinforcement learning network based on the A3C algorithm includes an Actor network and a Critic network; the pre-training process of the node strategy generation network includes: Obtain the current environmental status of multiple drone cluster networks; the current environmental status includes at least the CPU capacity, remaining CPU capacity, bandwidth capacity, remaining bandwidth capacity and maximum network delay of the drone node; Based on the current environment status and virtual mapping request, the mapped drone node is selected in the drone cluster network through the Actor network to obtain the node mapping strategy; The policy gradient algorithm is used to calculate the gradient update direction of the Actor network, and the advantage function value of the Actor network is calculated along the gradient update direction until the advantage function value is minimized. Evaluate the value of the node mapping strategy generated by the Actor network in the current environment state based on the Critic network; The Critic network is trained by minimizing the TD error until the TD error is minimized.

[0009] Optionally, mapping the virtual node to be embedded to a drone node of the drone cluster network based on the node mapping strategy includes: Sort the virtual nodes to be embedded from large to small based on the CPU demand of the virtual nodes in the virtual embedding request; Map the first-ranked virtual node to the drone node of the drone cluster network according to the node mapping strategy; Determine whether the first virtual node is mapped successfully; If the mapping is successful, the next virtual node will be mapped to the drone node of the drone cluster network according to the node mapping strategy according to the sorting until all virtual nodes are mapped successfully.

[0010] Optionally, determining whether the first virtual node is mapped successfully includes: Obtain the CPU demand and network delay of the first virtual node; and the CPU capacity and network delay of the drone node to which the first virtual node is mapped; Determine whether the CPU capacity and network delay of the drone node are greater than the CPU demand and network delay of the first virtual node; If it is greater, the mapping is successful.

[0011] Optionally, the method further comprises: Determine whether all virtual links are mapped successfully; If any virtual link mapping fails, the virtual network embedding request is rejected.

[0012] In a second aspect, a virtual network mapping device based on a drone cluster scenario is provided, the device comprising: The acquisition unit is used to acquire a topology map sequence of the drone cluster network when receiving a virtual network embedding request; the virtual network embedding request includes at least a virtual node and a virtual link to be embedded; the topology map sequence is composed of topology maps captured at each snapshot moment; An extraction unit, used for extracting spatiotemporal dynamic features between topological graphs in a topological graph sequence based on a pre-built feature extraction network; A generation unit, used for inputting the spatiotemporal dynamic features into a pre-trained node strategy generation network to generate a node mapping strategy; A node mapping unit, used for mapping the virtual node to be embedded into the drone node of the drone cluster network based on the node mapping strategy; The link mapping unit is used to determine the shortest physical link between the mapped drone nodes based on the Floyd algorithm after all virtual nodes are successfully mapped, and map the virtual link to be embedded into the shortest physical link.

[0013] In a third aspect, an electronic device is provided, comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; Memory, used to store computer programs; The processor is used to implement any method step described in the first aspect when executing the program stored in the memory.

[0014] According to a fourth aspect, a computer-readable storage medium is provided, wherein a computer program is stored in the computer-readable storage medium, and when the computer program is executed by a processor, any method step described in the first aspect is implemented.

[0015] The virtual network mapping method, device and equipment provided by the present invention based on the drone cluster scenario, when receiving a virtual network embedding request, obtains a topology map sequence of the drone cluster network; extracts the spatiotemporal dynamic features between each topology map in the topology map sequence based on a pre-built feature extraction network; inputs the spatiotemporal dynamic features into a pre-trained node strategy generation network to generate a node mapping strategy; maps the virtual nodes to be embedded to the drone nodes of the drone cluster network based on the node mapping strategy; after all virtual nodes are successfully mapped, determines the shortest physical link between the mapped drone nodes based on the Floyd algorithm, and maps the virtual links to be embedded to the shortest physical link. The embodiment of the present invention generates a node mapping strategy and the shortest mapping path based on the spatiotemporal dynamic features, which can meet the virtual mapping requirements in the network dynamic scenario; and adopts a two-stage mapping method to map virtual nodes and virtual links in sequence, which can ensure that both nodes and links are successfully mapped, and improve the mapping success rate.

[0016] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0018] Figure 1 A flowchart of a virtual network mapping method based on a drone cluster scenario provided by an embodiment of the present invention is shown; Figure 2 A schematic diagram of the model structure of a feature extraction network and a node strategy generation network provided in an embodiment of the present invention is shown; Figure 3 A comparison chart showing the results of the long-term average benefits of four algorithms in the simulation experiment provided by the embodiment of the present invention is shown; Figure 4 A comparison chart showing the results of the long-term average benefit-cost ratio of four algorithms in the simulation experiment provided by the embodiment of the present invention is shown; Figure 5 A comparison chart showing the request acceptance rates of four algorithms in a simulation experiment provided by an embodiment of the present invention is shown; Figure 6 A schematic diagram of the structure of a virtual network mapping device based on a drone cluster scenario provided by an embodiment of the present invention is shown; Figure 7 A schematic structural diagram of an electronic device provided by an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present invention.

[0020] Considering that the existing online reinforcement learning algorithms are based on the static topological structure of the physical network without considering the dynamic change characteristics of the physical network, they cannot meet the mapping requirements of dynamic network scenarios.

[0021] Based on this, an embodiment of the present invention provides a virtual network mapping method and device based on a drone cluster scenario, which is described below through an embodiment.

[0022] The embodiment of the present invention provides a virtual network mapping method based on a drone cluster scenario, such as Figure 1 As shown, the method comprises the following steps: Step S101: upon receiving a virtual network embedding request, a topology map sequence of the drone cluster network is obtained.

[0023] In this step, the virtual network embedding request includes at least the attribute information of the virtual node, virtual link and virtual network to be embedded. The attribute information of the virtual network includes one or more of the following: the number of virtual nodes, the distribution range of the virtual node CPU capacity and the distribution range of the virtual link bandwidth.

[0024] In the embodiment of the present invention, the topology sequence is composed of topology maps captured at each snapshot moment. Specifically, the topology sequence is topology maps captured sequentially at preset snapshot intervals between the current virtual network embedding request moment and the previous virtual network embedding request moment.

[0025] At each snapshot moment, the topological structure information and network status information of the drone cluster network corresponding to that moment are captured. The topological structure includes drone nodes and the edges between drone nodes. The network status information includes the status of drone nodes and the edges between nodes, such as the number of drone nodes, the number of drone physical links, the distribution range of drone node CPU capacity, the distribution range of physical link bandwidth, and the distribution range of drone node maximum delay.

[0026] In one example, the topological graph sequence of a drone cluster collected through snapshots is represented as ; Where T represents the snapshot time; Represents the topological graph collected at snapshot time T; in, ; Snapshot moment The set of drone nodes at that time; Snapshot moment The UAV link set at that time; Snapshot moment The network state set of the drone cluster at the time.

[0027] Step S102: extracting spatiotemporal dynamic features between topological graphs in the topological graph sequence based on a pre-constructed feature extraction network.

[0028] In this step, the pre-constructed feature extraction network consists of a graph convolutional network and a Transformer-XL network. The spatial features between drone nodes are extracted through the graph convolutional network; the dynamic features of drone nodes that change over time are extracted through the Transformer-XL network to obtain the spatiotemporal dynamic features of the drone cluster network. Transformer generally processes text sequence data, so the embodiment of the present invention encodes the topological map of the drone into a topological map sequence format in the form of snapshots, so as to extract the dynamic characteristics of the topological map between different times through the Transformer-XL network.

[0029] Step S103: input the spatiotemporal dynamic features into the pre-trained node strategy generation network to generate a node mapping strategy.

[0030] This spatiotemporal dynamic feature reflects the dynamic characteristics of the UAV cluster network in time and space. The node mapping strategy can be generated based on this dynamic feature, greatly improving the accuracy of node mapping.

[0031] Step S104: Map the virtual node to be embedded to the drone node of the drone cluster network based on the node mapping strategy.

[0032] The specific mapping method will be described in the following embodiments and will not be repeated here.

[0033] Step S105: After all virtual nodes are mapped successfully, the shortest physical link between the mapped drone nodes is determined based on the Floyd algorithm, and the virtual link to be embedded is mapped to the shortest physical link.

[0034] In this step, the Floyd algorithm is an existing technology and will not be described here. In an example, for example, there are four drone nodes A, B, C, and D, and three virtual nodes a, b, and c. The generated node mapping strategy is aA, bB, and cD. The paths between the mapped drone nodes A, B, and D are ABD and ABCD. The shortest path ABD is selected through the Floyd algorithm, and the link of the virtual node is mapped to the shortest path ABD, thereby reducing the consumption of bandwidth resources.

[0035] The present invention can extract the spatiotemporal dynamic characteristics of the network in the drone cluster network scenario through a feature extraction network; generate a node mapping strategy and the shortest mapping path based on the spatiotemporal dynamic characteristics, which can meet the virtual mapping requirements in the network dynamic scenario; and adopt a two-stage mapping method to map virtual nodes and virtual links in sequence, which can ensure that both nodes and links are mapped successfully and improve the mapping success rate.

[0036] Based on the above embodiments, Figure 2 As shown in the figure, the feature extraction network consists of a graph convolutional network GCN and a Transformer-XL network; the spatiotemporal dynamic features between the topological graphs in the topological graph sequence extracted based on the pre-built feature extraction network include: Step S1021: Extract the spatial features of each topological graph one by one based on the graph convolutional network.

[0037] In this step, the graph convolutional network includes two hidden layers.

[0038] In a feasible implementation, extracting the spatial features of each topological graph includes: Step A: For each topological graph, determine the first adjacency matrix of the drone nodes in the topological graph and the attribute feature matrix of the drone cluster network.

[0039] In this step, for each topology , based on node sets Link Set Determine the adjacency matrix ; Based on the network status set Determine the attribute characteristic matrix .

[0040] Step B: Input the first adjacency matrix and attribute feature matrix into the first hidden layer to obtain the initial spatial features.

[0041] Step C: Input the initial spatial features into the second hidden layer to extract higher-dimensional feature information to obtain the final spatial features of the topological map.

[0042] In one example, the layer-by-layer propagation formula of a graph convolutional network is: (1); in, Represents the input vector of the graph convolutional network, that is, the attribute feature matrix, whose dimension is ; is the number of drone nodes in the drone cluster network, is the dimension of multiple state information of drone nodes, is the parameter of the graph convolutional network, and its dimension is , is the number of neurons in the hidden layer of the graph convolutional network; is the normalized adjacency matrix of drone nodes; is the degree matrix; is the spatial feature of the output, with dimension .

[0043] in, Indicates that the adjacency matrix and degree matrix are normalized. The normalized matrix is ​​set to , then the layer-by-layer propagation formula of the graph convolutional network can be simplified to: (2); in, Represents the normalized matrix of adjacency matrix and degree matrix; Indicates The spatial feature matrix of the layer; express The spatial feature matrix of the layer output; Represents the parameters of the graph convolutional network.

[0044] By introducing two hidden layers, the embodiment of the present invention can enable the graph convolutional network GCN to learn more complex and abstract node feature representations. The first hidden layer can capture the information of direct neighbors, while the second layer can further integrate information from more distant neighbors, which helps to understand long-distance dependencies in the network structure. In addition, compared with deeper networks, two hidden layers can reduce the number of parameters to a certain extent, thereby reducing the risk of overfitting. Two hidden layers require fewer computing resources, including memory and computing time, which is especially important for processing large-scale graph data. Due to its relatively simple structure, the GCN with two hidden layers can converge to a satisfactory solution more quickly, accelerating the speed of feature extraction.

[0045] Step S1022: determining whether the number of topology graphs in the topology graph sequence exceeds a preset number threshold.

[0046] In this step, a topological map sequence may include 20 or even more topological maps. Inputting a large number of topological maps into the Transformer-XL network will greatly reduce the efficiency of feature extraction and increase the amount of calculation. Therefore, a preset number threshold is set to control the amount of data input into the Transformer-XL network.

[0047] In one example, the preset quantity threshold is set to 5, that is, the maximum input is the spatial features of the topological graph of 5 sequences.

[0048] Step S1023: If the number exceeds the preset threshold, the spatial features of the topological graphs of the preset threshold that are captured most recently are input into the Transformer-XL network to extract the spatiotemporal dynamic features Zt between the topological graphs of the preset threshold.

[0049] In this step, for example, there are a total of 10 topological map sequences collected at snapshot moments. In chronological order, the spatial features of the topological maps at the most recent five moments are collected as the input of the Transformer-XL network.

[0050] Step S1024: If the preset quantity threshold is not exceeded, all spatial features extracted by the graph convolutional network are input into the Transformer-XL network to extract the spatiotemporal dynamic features Zt between each topological graph.

[0051] Continuing with the previous example, if there are less than 5 topological graphs, for example 4, the spatial features of these 4 topological graphs can be directly input into the Transformer-XL network.

[0052] The Transformer-XL network is a neural network with a multi-head attention mechanism. In the embodiment of the present invention, four attention heads are set, and the expression formula is: (3); in, - represents the attention head; represents a fully connected layer, which connects four attention heads to a 512-dimensional feature space.

[0053] (4); (5); in, Indicates Attention head; Indicates The query vector of the attention head; Indicates The key vector of the attention heads; Indicates The value vector of the attention heads; represents the dimension of the key vector; , , Represent the weights of each vector respectively.

[0054] From Formula 4 of the attention head, we can see that the working principle of each attention head is: first, the similarity score between the query and the key vector is calculated, and it is scaled by dividing by the dimension of the key vector. Then, the softmax function is used to convert these scores into attention weights in the form of probability distribution. Finally, these weights are applied to the value vector to obtain the weighted summed output.

[0055] Compared with the traditional Transformer, Transformer-XL introduces a memory mechanism. This mechanism allows the Transformer-XL network to process sequence data without defining a fixed window size or distance limit in advance. By retaining the hidden state of the previous window, it can automatically learn the dependencies between different positions and capture long-range dependencies across windows. By retaining the hidden state data of the previous window, there is no need to repeat the calculation for each sequence, which greatly reduces the amount of calculation and improves the efficiency of feature extraction, thereby improving the embedding efficiency of the virtual network and reducing large-scale virtual request migration.

[0056] Based on the above embodiment, the node strategy generation network adopts a deep reinforcement learning network based on the A3C algorithm, such as Figure 2As shown in the figure, the deep reinforcement learning network based on the A3C algorithm includes an Actor network and a Critic network; the pre-training process of the node strategy generation network includes: Step a: Obtain the current environment status of multiple drone cluster networks.

[0057] The current environment status includes at least the CPU capacity, remaining CPU capacity, bandwidth capacity, remaining bandwidth capacity and maximum network delay of the drone node.

[0058] Step b: Based on the current environment status and virtual mapping request, the mapped drone node is selected in the drone cluster network through the Actor network to obtain the node mapping strategy.

[0059] In this step, all possible node mapping actions in the current environment state are obtained based on the virtual mapping request; for example, virtual nodes a, b, c; drone nodes include A, B, C, D, and three virtual nodes are mapped to four drone nodes, with a total of 24 possible node mapping strategies. The reward value of each node mapping strategy can be obtained through the Actor network, and the node mapping strategy corresponding to the larger reward value can be applied to the actual scene, and the next environment state after application can be obtained, and then training can be continued for all possible node mapping actions in the next environment state until the training is completed.

[0060] like Figure 2 As shown in , after the training is completed, when the application is actually deployed, the dynamic characteristics of the drone cluster network output by the Transformer-XL network can be used as the input of the Actor network. The dynamic characteristics of the drone cluster network are equivalent to the environmental state of the drone cluster network.

[0061] Step c: Calculate the gradient update direction of the Actor network through the policy gradient algorithm, and calculate the advantage function value of the Actor network along the gradient update direction until the advantage function value is minimized.

[0062] Among them, the policy gradient algorithm is as follows: (6); Where: is the gradient parameter of the Actor network; Is the current environment status All actions Node mapping strategy; is the log-likelihood ratio of the policy distribution with respect to the parameter The gradient of , also called the score function; The decay parameter β is set large at the beginning and gradually decreases during the training phase; is the learning rate. In order to prevent premature convergence to the local optimal solution and encourage exploration, the present invention adds an entropy regularization term to the policy gradient algorithm. , this term is multiplied by a small decay parameter , and is subtracted from the policy gradient, thereby gradually reducing its influence during training.

[0063] The Actor network is mainly used to generate node mapping strategies; the goal of the Critic network is to learn the value function , mainly used to estimate the given state Follow the strategy The Critic network learns by minimizing the square of the TD error. Except for the output layer, the two networks share a similar structure.

[0064] Step d: Evaluate the value of the node mapping strategy generated by the Actor network under the current environment state based on the Critic network; Step e: Train the Critic network by minimizing the TD error until the TD error is minimized.

[0065] The TD error is calculated as follows: (7); Where: Indicates status Estimation of the value function under ; is the parameter of the Critic network; is the learning rate of the Critic network; It’s an instant reward; It is the discount factor that brings up the importance of adjusting future rewards.

[0066] The A3C algorithm combines the updates of these two networks and uses multi-threaded asynchronous updates to learn in parallel in different environment instances, thereby improving learning efficiency and stability.

[0067] Based on the above embodiment, mapping the virtual node to be embedded to the drone node of the drone cluster network based on the node mapping strategy includes: Sort the virtual nodes to be embedded from large to small based on the CPU demand of the virtual nodes in the virtual embedding request; Map the first-ranked virtual node to the drone node of the drone cluster network according to the node mapping strategy; Determine whether the first virtual node is mapped successfully.

[0068] In a feasible real-time manner, determining whether the first virtual node is successfully mapped includes: Obtain the CPU demand and network delay of the first virtual node; and the CPU capacity and network delay of the drone node to which the first virtual node is mapped; Determine whether the CPU capacity and network delay of the drone node are greater than the CPU demand and network delay of the first virtual node; If it is greater, the mapping is successful.

[0069] If the mapping is successful, the next virtual node will be mapped to the drone node of the drone cluster network according to the node mapping strategy according to the sorting until all virtual nodes are mapped successfully.

[0070] If a virtual node mapping fails, the virtual network embedding request is rejected.

[0071] Based on the above embodiment, the method further includes: Determine whether all virtual links are mapped successfully.

[0072] In a feasible implementation, whether the link mapping is successful is determined by judging whether the bandwidth and maximum network delay of the physical link between the drone nodes meet the bandwidth and maximum network delay of the virtual link. If both are met, the link mapping is successful.

[0073] If any virtual link mapping fails, the virtual network embedding request is rejected.

[0074] After the node mapping strategy and link mapping path are generated, when mapping nodes through the node mapping strategy and mapping virtual links through the shortest path, the accuracy of mapping is further guaranteed by judging again whether the mapping is successful.

[0075] Finally, in order to verify the performance of the feature extraction network composed of the graph convolutional network GCN and the Transformer-XL network provided in the embodiment of the present invention to extract the dynamic features of the network and then generate the node mapping strategy through the deep reinforcement learning network, a set of experiments is used to illustrate. For simplicity, the feature extraction network composed of the GCN and Transformer-XL network is referred to as GCFormer below.

[0076] The simulation experiment was conducted by generating a network using the simulation software GT-ITM to achieve the initialization experimental data required for setting up the simulation experiment: the experimental topology contains 550 physical links and 100 physical nodes. The CPU capacity of each physical node is evenly distributed between (50, 100), and the bandwidth is evenly distributed between (20, 50). Two virtual network request sets are generated, each containing 1000 virtual embedding requests. One set is used for training and the other is used for testing. The number of nodes for each virtual request is evenly distributed between (2, 10), and the CPU of the virtual request is evenly distributed between (0, 50). The bandwidth resources of each virtual connection are evenly distributed between (0, 50). The connection probability between virtual links is set to 0.5. Among them, the arrival time of the virtual embedding request follows the Poisson distribution. On average, 4 virtual embedding requests arrive within 100 time units. The duration of each virtual embedding request is exponentially distributed, with an average duration of 1000 time units.

[0077] On this basis, the simulation results of the heuristic algorithm NodeRank algorithm, RLVNE algorithm and GCN-VNE algorithm are compared. Figure 3 , Figure 4 , Figure 5 As shown, they are the evaluation indicators of the four algorithms; Figure 3 It is the long-term average return indicator; Figure 4 It is an indicator of long-term benefit-cost ratio; Figure 5 It is the request acceptance rate indicator.

[0078] It can be found that the heuristic-based NodeRank algorithm gives priority to heavyweight nodes, so the initial performance is reasonable. However, from the perspective of long-term performance, dynamic multi-domain networks are difficult to effectively characterize and have poor performance. RLVNE and GCN-VNE outperform NodeRank in all evaluation indicators, which fully demonstrates that the introduction of the AI ​​algorithm significantly improves the performance of VNE. In addition, GCFormer outperforms RLVNE and GCNVNE in the long run. This is because the embodiments of the present invention better understand the dynamic characteristics of nodes in a dynamic environment and can better focus on the properties of the dynamic physical environment.

[0079] In summary, in dynamic network scenarios, GCFormer has better performance and better reduces the fragmentation of physical resources.

[0080] Based on the same inventive concept, a virtual network mapping device based on a drone cluster scenario is provided, such as Figure 6 As shown, the device comprises: The acquisition unit 601 is used to obtain a topology map sequence of the drone cluster network when receiving a virtual network embedding request.

[0081] The virtual network embedding request at least includes the virtual nodes and virtual links to be embedded; the topology map sequence is composed of topology maps captured at each snapshot moment.

[0082] The extraction unit 602 is used to extract the spatiotemporal dynamic features between the topological graphs in the topological graph sequence based on a pre-built feature extraction network.

[0083] The generating unit 603 is used to input the spatiotemporal dynamic features into the pre-trained node strategy generating network to generate a node mapping strategy.

[0084] The node mapping unit 604 is used to map the virtual node to be embedded into the drone node of the drone cluster network based on the node mapping strategy.

[0085] The link mapping unit 605 is used to determine the shortest physical link between the mapped drone nodes based on the Floyd algorithm after all virtual nodes are successfully mapped, and map the virtual link to be embedded into the shortest physical link.

[0086] Based on the same technical concept, an embodiment of the present invention further provides an electronic device, such as Figure 7 As shown, it includes a processor 701 , a communication interface 702 , a memory 703 and a communication bus 704 , wherein the processor 701 , the communication interface 702 , and the memory 703 communicate with each other via the communication bus 704 .

[0087] Memory 703, used for storing computer programs; The processor 701 is used to implement the steps of the virtual network mapping method based on the drone cluster scenario when executing the program stored in the memory 703.

[0088] The communication bus mentioned in the above electronic device can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0089] The communication interface is used for communication between the above electronic device and other devices.

[0090] The memory may include a random access memory (RAM) or a non-volatile memory (NVM), such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0091] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0092] The computer program product for performing virtual network mapping based on a drone cluster scenario provided in an embodiment of the present invention includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the method described in the previous method embodiment. The specific implementation can be found in the method embodiment, which will not be repeated here.

[0093] The device for virtual network mapping based on the drone cluster scenario provided in the embodiment of the present invention can be specific hardware on the device or software or firmware installed on the device. The implementation principle and technical effects of the device provided in the embodiment of the present invention are the same as those of the aforementioned method embodiment. For the sake of brief description, the parts not mentioned in the device embodiment can refer to the corresponding contents in the aforementioned method embodiment. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the systems, devices and units described above can all refer to the corresponding processes in the aforementioned method embodiment, and will not be repeated here.

[0094] In the embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0095] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0096] In addition, each functional unit in the embodiment provided by the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0097] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0098] It should be noted that similar numbers and letters represent similar items in the following figures. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. In addition, the terms "first", "second", "third", etc. are only used to distinguish the description and are not to be understood as indicating or implying relative importance.

[0099] Finally, it should be noted that the above-described embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention is described in detail with reference to the above-mentioned embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-mentioned embodiments within the technical scope disclosed by the present invention, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. They should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A virtual network mapping method based on a drone cluster scenario, characterized in that: The method comprises: Upon receiving a virtual network embedding request, a topology map sequence of the drone cluster network is obtained; the virtual network embedding request includes at least virtual nodes and virtual links to be embedded; the topology map sequence is composed of topology maps captured at each snapshot moment; Extracting spatiotemporal dynamic features between topological graphs in the topological graph sequence based on a pre-built feature extraction network; Inputting the spatiotemporal dynamic features into a pre-trained node strategy generation network to generate a node mapping strategy; Mapping the virtual node to be embedded to a drone node of the drone cluster network based on the node mapping strategy; After all virtual nodes are mapped successfully, the shortest physical link between the mapped drone nodes is determined based on the Floyd algorithm, and the virtual link to be embedded is mapped to the shortest physical link.

2. The method according to claim 1, characterized in that: The feature extraction network is composed of a graph convolutional network and a Transformer-XL network; the extraction of spatiotemporal dynamic features between topological graphs in the topological graph sequence based on the pre-built feature extraction network includes: Extracting spatial features of each topological graph one by one based on the graph convolutional network; Determining whether the number of topology maps in the topology map sequence exceeds a preset number threshold; If the number exceeds the preset threshold, the spatial features of the topological graphs of the preset threshold are input into the Transformer-XL network to extract the spatiotemporal dynamic features between the topological graphs of the preset threshold; If the preset number threshold is not exceeded, all spatial features extracted by the graph convolutional network are input into the Transformer-XL network to extract the spatiotemporal dynamic features between the topological graphs.

3. The method according to claim 2, characterized in that The graph convolution network includes two hidden layers; and extracting the spatial features of each topological graph one by one based on the graph convolution network includes: Determine, for each topological graph, a first adjacency matrix of the drone nodes in the topological graph and an attribute feature matrix of the drone cluster network; Inputting the first adjacency matrix and attribute feature matrix into the first hidden layer to obtain initial spatial features; The initial spatial features are input into the second hidden layer to extract feature information of higher dimensions, thereby obtaining the final spatial features of the topological map.

4. The method according to claim 1, characterized in that The node strategy generation network adopts a deep reinforcement learning network based on the A3C algorithm, and the deep reinforcement learning network based on the A3C algorithm includes an Actor network and a Critic network; The pre-training process of the node strategy generation network includes: Obtaining the current environmental status of multiple drone cluster networks; the current environmental status includes at least the CPU capacity, remaining CPU capacity, bandwidth capacity, remaining bandwidth capacity and maximum network delay of the drone nodes; Based on the current environment state and the virtual mapping request, a mapped drone node is selected in the drone cluster network through the Actor network to obtain a node mapping strategy; Calculate the gradient update direction of the Actor network by using a policy gradient algorithm, and calculate the advantage function value of the Actor network along the gradient update direction until the advantage function value is minimized; Evaluate the value of the node mapping strategy generated by the Actor network in the current environment state based on the Critic network; The Critic network is trained by minimizing the TD error until the TD error is minimized.

5. The method according to claim 1, characterized in that Mapping the virtual node to be embedded to the drone node of the drone cluster network based on the node mapping strategy includes: Sort the virtual nodes to be embedded from large to small based on the CPU demand of the virtual nodes in the virtual embedding request; Mapping the first-ranked virtual node to a drone node in the drone cluster network according to the node mapping strategy; Determine whether the first virtual node is mapped successfully; If the mapping is successful, the next virtual node will be mapped to the drone node of the drone cluster network according to the node mapping strategy according to the sorting until all virtual nodes are mapped successfully.

6. The method according to claim 5, characterized in that The step of determining whether the first virtual node is successfully mapped includes: Obtain the CPU demand and network delay of the first virtual node; and the CPU capacity and network delay of the drone node to which the first virtual node is mapped; Determine whether the CPU capacity and network delay of the drone node are greater than the CPU demand and network delay of the first virtual node; If it is greater, the mapping is successful.

7. The method according to claim 1, characterized in that The method further comprises: Determining whether all the virtual links are mapped successfully; If any virtual link mapping fails, the virtual network embedding request is rejected.

8. A virtual network mapping device based on a drone cluster scenario, characterized in that: The device comprises: An acquisition unit is used to acquire a topology map sequence of the drone cluster network upon receiving a virtual network embedding request; the virtual network embedding request includes at least virtual nodes and virtual links to be embedded; the topology map sequence is composed of topology maps captured at each snapshot moment; An extraction unit, used for extracting spatiotemporal dynamic features between topological graphs in the topological graph sequence based on a pre-built feature extraction network; A generating unit, configured to input the spatiotemporal dynamic features into a pre-trained node strategy generating network to generate a node mapping strategy; A node mapping unit, configured to map the virtual node to be embedded into a drone node of the drone cluster network based on the node mapping strategy; The link mapping unit is used to determine the shortest physical link between the mapped drone nodes based on the Floyd algorithm after all virtual nodes are successfully mapped, and map the virtual link to be embedded into the shortest physical link.

9. An electronic device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; Memory, used to store computer programs; A processor, for implementing the method steps described in any one of claims 1 to 7 when executing a program stored in a memory.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method steps described in any one of claims 1 to 7 are implemented.

Images