Method and system for edge data migration based on SDWN and UAV edge in-network storage

Abstract

This invention discloses a method and system for migrating edge-based on-network storage data based on SDWN and UAVs. Under the SDWN architecture, multiple UAVs are used as destination nodes for data migration. The multi-UAV-assisted on-network storage data migration problem is modeled as an NP-hard combinatorial optimization model of many-to-many communication routing planning. A two-stage solution method is designed: first, a node similarity matrix is ​​obtained based on the average comprehensive network state score between nodes, and spectral clustering k-means is used to obtain reasonable data file allocation decision variables and UAV access node decision variables. Then, for the path selection of many-to-many communication routing planning when multiple data files are transmitted in parallel, the whale optimization algorithm is used to solve the problem, effectively reducing link contention and shortening the maximum migration time of the system. Experimental results show that, on different topologies, after determining reasonable parameter settings, the algorithm of this invention outperforms the benchmark method in data migration efficiency and has lower migration time overhead, demonstrating the effectiveness and applicability of the algorithm.

Description

Technical Field

[0001] This invention relates to the field of on-net storage data migration technology, and specifically to an edge on-net storage data migration method and system based on SDWN and UAV. Background Technology

[0002] With the development of technologies such as the Internet of Things (IoT) and edge computing, the scale of data generated by network edge devices such as various sensors, cameras, and mobile terminals continues to grow. Previously, this edge data was typically transmitted to cloud data centers for storage, but this resulted in problems such as high transmission latency, large bandwidth consumption, and vulnerabilities in data privacy and security. Deploying local storage servers at the edge for data storage can solve these problems, effectively shortening data transmission distance and latency while protecting user data privacy. Subsequently, the processed data can be migrated according to actual scenarios or user needs to alleviate the limitations of edge storage capacity. However, data generated by wirelessly connected edge networks needs to be transmitted to edge storage servers via multi-hop networks. Since the locations of deployed edge storage servers are relatively fixed, network nodes near the edge server locations experience rapid energy consumption. Furthermore, in field sensing network tasks such as disaster relief, emergency monitoring, and temporary inspections, the network topology may change frequently, and the link communication quality may be unstable due to interference. Deploying additional edge servers in fixed locations is not only inflexible, but also makes it difficult to guarantee the continuous and successful transmission of edge data to edge servers for storage when the edge network fails or is interrupted. This makes it difficult to meet the requirements of edge data for low latency, high reliability storage, and migration.

[0003] In response, some literature has proposed an integrated on-network storage technology that combines sensing, communication, and storage. This technology enables network nodes to not only forward data but also store data locally, storing edge data on network nodes based on network status and task requirements. Compared to edge storage, on-network storage reduces reliance on fixed-location edge servers, shortens data transmission distance and latency to storage locations, and reduces bandwidth consumption. When network communication quality is poor, or when link congestion or outages occur, edge data can be stored internally on network nodes. Once network communication is restored, data transmission can be completed according to the actual scenario or user needs, thereby improving the reliability of edge data storage. Like edge storage, on-network storage also faces storage capacity limitations, which can usually be addressed by migrating a portion of the stored data to free up storage space. Efficient on-network storage data migration solutions typically require minimizing transmission latency in the edge wireless network, which is related to the pre-selected destination node. Regarding the selection of the destination node for on-network storage data migration, most existing migration solutions use fixed edge servers as the destination node. However, in scenarios with complex terrain or limited infrastructure, there are also problems such as poor deployment flexibility and inconvenience in implementation. Summary of the Invention

[0004] The present invention addresses the problems of poor deployment flexibility and inconvenient implementation of existing on-net storage data migration methods, and provides an edge on-net storage data migration method and system based on SDWN and UAV.

[0005] To solve the above problems, the present invention is achieved through the following technical solution:

[0006] The edge-to-network storage data migration method based on SDWN and UAV includes the following steps:

[0007] Step 1: Based on the link attributes, determine the destination UAV decision variables, UAV access node decision variables, and path decision variables for the data files, and construct an optimization problem by minimizing the migration and transmission time of all data files;

[0008] ,

[0009] In the formula, To optimize the problem, The migration and transfer time for all data files is complete. , For data files Migration and transfer completion time; For UAV decision variables in data file destination, , Represents data file Should I select UAV? As the destination node; For UAV access node decision variables, , UAV Whether to connect to the node ; For path decision variables, , Indicates from the source node UAV to destination Choose a path ; For link For data files allocated bandwidth For link The maximum bandwidth limit; A collection of data files stored online. For link The collection of data files that pass through simultaneously For a set of nodes, For data files The set of nodes, For UAV set, For the source node UAV to destination The set of paths;

[0010] Step 2: Decompose the optimization problem constructed in Step 1 into two sub-problems: an optimization sub-problem concerning the UAV decision variables for the data file destination and the UAV access node decision variables, and an optimization sub-problem concerning the path decision variables.

[0011] Step 3: Use the spectral clustering k-means method to solve the optimization subproblem regarding the UAV decision variables for the data file destination and the UAV access node decision variables;

[0012] Step 4: Use the whale optimization method to solve the optimization subproblem concerning the path decision variables.

[0013] The edge-to-network storage data migration system based on SDWN and UAV to implement the method includes an SDWN network and multiple UAVs; the SDWN network consists of a data plane, a control plane and a knowledge plane.

[0014] The data plane is responsible for storing data on network nodes and forwarding data through network nodes. The data plane consists of two types of SDN switch nodes: integrated storage and communication SDN switches with storage capabilities and ordinary SDN switches that only handle data forwarding. Both types of switches are equipped with dual-band wireless communication modules. The low-frequency wireless channel serves as the control channel between the data plane and the control plane, while the high-frequency wireless channel is used to construct a wireless mesh network for data forwarding between data plane nodes. UAVs access the SDN switch nodes of the data plane through the high-frequency wireless channel, serving as the destination node for migrating stored data on the network.

[0015] The control plane is decoupled from the data plane through a low-frequency wireless channel as a southbound interface, enabling centralized control and global visibility. The control plane is based on the RYU controller and mainly includes a topology information perception module responsible for sensing and maintaining global network topology information, a network status measurement module responsible for sensing network operating status, a flow table installation and migration instruction issuance module responsible for converting the decisions of the control plane into specific execution rules of the data plane, and a UAV information processing module responsible for the unified management of UAV access and data migration processes by the control plane.

[0016] The knowledge plane is responsible for processing the network status information and topology node information obtained by the control plane, and performs unified caching and management of the network status information and node information. Based on this information, the destination UAV decision variables, UAV access node decision variables and path decision variables of the data file are obtained.

[0017] Compared with existing technologies, this invention, under the SDWN architecture, uses multiple UAVs as destination nodes for data migration, which not only enables rapid and flexible deployment but also balances the transmission load of the migrated data. SDWN, by decoupling the control plane and data plane, can uniformly obtain status information such as network status, on-network storage nodes, and UAV locations, providing support for designing efficient on-network data migration strategies. The problem of multi-UAV-assisted on-network storage data migration is modeled as an NP-hard combinatorial optimization model of many-to-many communication routing planning. To address the difficulty of solving the coupling between the destination UAV node selection, UAV access network node selection, and path selection decision variables in this optimization model, a two-stage solution method is designed. First, the node similarity matrix is ​​obtained based on the average comprehensive network status score between nodes. Then, k-means algorithm is used to solve the data dimension reduction by combining spectral clustering to obtain reasonable data file allocation decision variables and UAV access node decision variables. Finally, the whale optimization algorithm is used to solve the path selection for many-to-many communication routing planning when multiple data files are transmitted in parallel, effectively reducing link contention and shortening the maximum migration time of the system. Experimental results show that, on different topologies, the algorithm of this invention outperforms the benchmark method in data migration efficiency and has lower migration time overhead after determining reasonable parameter settings. This demonstrates the effectiveness and applicability of the algorithm of this invention. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of multi-UAV-assisted data migration in on-net storage.

[0019] Figure 2 For the image Transform into For those with an average overall score greater than Virtual edges are added to the node pairs.

[0020] Figure 3 This is a multi-UAV-assisted on-network storage data migration system architecture based on SDWN.

[0021] Figure 4 This is the principle of out-of-band system design.

[0022] Figure 5 shows the parameters. The experimental results showing the impact on performance are shown in the figure, (a) 10-node topology. The impact of the selected values ​​on system performance, (b) 14-node topology The impact of the value on system performance, (c) 20-node topology The impact of the selected value on system performance.

[0023] Figure 6 shows the parameters. The experimental results showing the impact on performance are shown in the figure, (a) 10-node topology. The impact of the selected values ​​on system performance, (b) 14-node topology The impact of the value on system performance, (c) 20-node topology The impact of the selected value on system performance.

[0024] Figure 7 shows the comparative experimental results of different methods: (a) Comparison results of different algorithms for 10-node topology, (b) Comparison results of different algorithms for 14-node topology, and (c) Comparison results of different algorithms for 20-node topology. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific examples.

[0026] To achieve efficient data migration in edge scenarios, this invention proposes a data migration method for edge on-network storage systems based on software-defined wireless network (SDWN) and unmanned aerial vehicle (UAV) assistance. Addressing the problems of cumbersome configuration, inconvenient management, and high overhead in network status information and storage node measurement in traditional edge storage network architectures, this invention utilizes the SDWN architecture to obtain on-network storage network status information, storage node load information, and UAV location information, effectively solving the difficulties in obtaining and managing network and storage node status information, and providing strong support for designing and planning data migration routing strategies. Furthermore, addressing the issues of traditional storage data migration scenarios requiring fixed edge storage servers as destination nodes, which suffer from poor deployment flexibility and implementation in complex terrain and infrastructure-constrained scenarios, this invention uses UAVs as the data migration destination node carrier. This method offers quick and convenient deployment, flexibly adapting to special scenario requirements such as disaster relief, and the use of multiple UAVs can balance the data transmission network load, improving data migration efficiency.

[0027] The data migration problem discussed in this invention assumes that data is already distributed and stored across a certain number of ground network nodes. Using multiple UAVs (Unmanned Aerial Vehicles), the data to be migrated from the network nodes needs to be transmitted through multiple hops to the ground access node connected to the UAV to complete the migration. To improve migration efficiency, multiple data streams are typically transmitted in parallel, resulting in multiple data streams existing simultaneously in the network and bandwidth contention on shared links. Furthermore, the UAV access location affects the data convergence direction and transmission path, and the destination selection of the data also alters the load distribution on different links and UAVs. Figure 1As shown, there is a clear coupling relationship between the selection of the UAV access node, the selection of the destination UAV node for the data, and the path selection. Therefore, how to coordinate the above decisions within a unified framework to reduce link contention and overall migration latency during the parallel migration of multiple files is the core issue addressed by this invention.

[0028] Using undirected graphs This indicates the network topology stored on the network, where, Represents a set of network nodes, with the number of nodes being... , Denotes a set of links, where Represents a node and The links between them. The set of data files to be transmitted in the network storage network is... Number of files Assuming the data file The size is data file The network node is ,use Represents data file The set of network nodes. Nodes without data files to be transmitted are called ordinary forwarding nodes, represented by a set. This indicates that, thus having The link was measured using SDN technology. The link attribute vector is ,in Indicates link Remaining bandwidth For link The latency (which typically includes the sum of data propagation, processing, and queuing latency) For link The packet loss rate. During the migration of on-network stored data assisted by multiple UAVs, data files need to be transferred to the UAVs. Assume the set of UAVs is... ,in This refers to the number of UAVs. During data file migration, each UAV connects to only one ground network node and receives data files through that network node. These network nodes connected to by the UAVs are called the UAV access nodes.

[0029] To optimize the data file migration process, it is necessary to determine the UAV to be migrated, the network node to which the UAV is connected, and the routing path from the data source node where the file is located to the access node of the corresponding migrated UAV.

[0030] Using vectors The data file represents the destination UAV decision variable, where Represents data file Should I select UAV? As the destination node, satisfying

[0031] (1)

[0032] Because each data file Simply migrate to a single UAV, i.e.

[0033] (2)

[0034] Using vectors Denotes the decision variables for UAV access network nodes, where UAV Whether to connect to network nodes ,satisfy

[0035] (3)

[0036] Since each UAV connects to only one ground node, therefore:

[0037] (4)

[0038] Use sets Indicates from the source network node UAV to destination The set of paths, where Indicates from the source network node UAV to destination Path length sorted in ascending order Paths, using subsets Represents a file Determine the data source node and UAV access node The set of paths after, and using Indicates from node To the access node The number of paths.

[0039] Using vectors Denotes the path decision variables, where Indicates from the source network node UAV to destination Choose a path ,Right now

[0040] (5)

[0041] For data files Only one path needs to be selected for transmission, therefore there are

[0042] (6)

[0043] In scenarios involving parallel transmission of multiple files, this invention allocates the remaining bandwidth of the shared link proportionally to the file size; that is, the larger the file, the higher its initial bandwidth allocation on the competing link. Specifically, the link... The data files that pass through simultaneously are represented as

[0044] (7)

[0045] In the formula, Represented as a data file Select path and The above includes links .

[0046] according to ,link According to data file Bandwidth allocation is performed based on the size of the bandwidth, i.e.

[0047] (8)

[0048] Based on the above analysis, Together they determined the data file For ease of description, let's assume the path to the UAV is a data file. Determined path According to the bandwidth allocation in formula (8), the path The bottleneck bandwidth can be expressed as:

[0049] (9)

[0050] Based on link packet loss rate Determine the data file In the path packet loss rate

[0051] (10)

[0052] Based on link delay Calculate data file In the path latency

[0053] (11)

[0054] Finally, based on the bottleneck bandwidth Packet loss rate and latency Calculate data file Complete migration and transfer time

[0055] (12)

[0056] Therefore, the data file migration and transmission time is not only related to file size and path bottleneck bandwidth, but also closely related to path latency and link reliability. Complete all data files. The time required for parallel migration should be all The maximum value in, that is:

[0057] (13)

[0058] The objective of this invention is to base it on link attributes. Determine the destination UAV decision variables for the migration data files. UAV access network node decision variables and path decision variables To minimize the transfer time for migrating data files stored on the network, i.e.

[0059] (14)

[0060] The first three constraints ensure that the data file is selected by a unique UAV, the UAV is accessed only once, and the path is selected only once, respectively; the fourth constraint ensures that the link bandwidth allocation does not exceed the capacity limit.

[0061] The optimization problem corresponding to formula (14) Due to decision variables , and The problem involves mutual coupling, and optimizing the objective function constitutes a high-dimensional, nonlinear, complex, NP-hard combinatorial optimization problem. Solving for its global optimum using classical methods such as exhaustive search significantly increases computational and time costs. Therefore, we first define the optimization problem... To find its approximate optimal solution, we first need to determine the destination UAV decision variables of the migration data file. Decision variables for UAV access network nodes The optimization subproblem corresponding to the value of is then used to determine the path decision variables. The value of corresponds to the optimization subproblems. To reduce the significant network resource overhead and UAV migration data load balancing that often occurs when data files are transmitted simultaneously and in parallel on multiple source nodes, this invention solves the first optimization subproblem using the spectral clustering k-means method, and the second optimization subproblem using the whale optimization method.

[0062] (1) Purpose of data migration, UAV selection strategy and UAV node access network node selection strategy

[0063] To address the subproblem of UAV connection node selection and destination UAV selection decision for the data files to be migrated, and in order to improve the network load balancing of data migration, spectral clustering is used to reduce the dimensionality of the data and divide the ground network nodes into clusters. Each UAV is responsible for data migration within a non-overlapping network region. First, a similarity matrix is ​​constructed based on the average comprehensive communication status score between network nodes. Then, a low-dimensional feature matrix of the network nodes is constructed based on the similarity matrix, and k-means clustering is performed to divide the network into regions. Based on the region division results, the corresponding UAV access nodes for each region are obtained, and the destination UAV nodes for the data files to be migrated on the network nodes within the region are selected. The specific design process is as follows:

[0064] 1) Average integrated network state score among nodes

[0065] To cluster nodes with similar attributes into one class, a method for measuring the similarity between nodes needs to be defined first. Considering that this invention requires the highest possible latency efficiency for multi-source data transmission in the network storage data migration scenario, this invention first designs an evaluation scoring method for the communication quality of the path between any two nodes, and then designs a similarity calculation method between any two nodes based on this. Subsequently, a spectral clustering similarity matrix is ​​constructed using the node similarity values, as detailed below:

[0066] This approach comprehensively considers multiple state information of the inter-node links, including key attributes such as remaining bandwidth, latency, and packet loss rate. The comprehensive network state score (CNSS) of the inter-node paths is calculated using different weighting factors for evaluation. Before calculating the CNSS, the network state attributes of the nodes need to be normalized, as shown in the normalization formula below.

[0067] (15)

[0068] In the formula, For normalized attribute values, For a specific attribute value, such as the remaining bandwidth, latency, or packet loss rate of the link. and Network link set The maximum and minimum values ​​of this attribute are determined. Based on the normalized values ​​of the remaining bandwidth, latency, and packet loss rate of all links between nodes in the network and their neighboring nodes, the normalized values ​​of the remaining bandwidth, latency, and packet loss rate of the paths between each node can be obtained. , , .

[0069] For any node pair , record its first The path is .path The end-to-end normalized latency is:

[0070] (16)

[0071] path Normalized delay Further converted into time delay benefit items As shown below:

[0072] (17)

[0073] The total path delay can be seen. The smaller, The larger the value, the greater the total path delay. The larger, The smaller.

[0074] path bottleneck bandwidth :

[0075] (18)

[0076] path Success rate :

[0077] (19)

[0078] From equations (17), (18), and (19) above, we can derive... , and The range of values ​​is .

[0079] Arbitrary node pair Path between The network state score can be expressed as:

[0080] (20)

[0081] In the formula, , and Weighting factors for bandwidth, latency, and packet loss rate, respectively; weight matrix. Values Therefore, the overall network state score satisfies .

[0082] Since data migration in network storage involves many-to-many communication with parallel transmission of multiple data files, mutual interference occurs during transmission. Therefore, this invention considers that the communication quality between nodes should be fully characterized not only by the comprehensive network state score of a single path but also by the comprehensive network state score of multiple paths between nodes. Thus, this invention uses the average comprehensive network states score of multiple connected paths between nodes as the average comprehensive network states score (ACNSS) to characterize the communication quality between nodes. The ACNSS between nodes can be sorted by the path length between the nodes in ascending order. The average of the comprehensive network state scores for each path is taken. When the number of paths is less than [a certain value], [the score is determined by the average of the average network state scores for each path]. When repeated, repetition is allowed, as shown below:

[0083] (twenty one)

[0084] 2) Node similarity

[0085] Node similarity is the similarity between any two nodes in a network. and The higher the similarity value, the better the communication environment of the two nodes in the network, and the higher the probability that they will be classified into the same region during the k-means clustering process.

[0086] undirected graph Transform into a similarity graph The edge set was reconstructed in the similarity graph. And added a similarity matrix Specifically, for any two nodes in the graph... and The distance between the two points can be calculated according to equation (21). When two nodes are not adjacent nodes and When this happens, add a virtual edge to the node pair, such as... Figure 2 As shown. Among them. This represents the minimum acceptable integrated network state score threshold. The threshold is set to 0.25. This threshold is used to filter out node pairs with poor network conditions that are unlikely to form effective collaborative relationships during subsequent data migration, retaining only potential associations that are practically significant for region partitioning. After the virtual edges are constructed, similarity is calculated. :

[0087] (twenty two)

[0088] In the formula, For all The standard deviation of the average network state score between two points indicates that the higher the average integrated network state score, the greater the similarity and the more likely they are to be classified into the same region.

[0089] 3) Spectral clustering method

[0090] In graph partitioning, the similarity graph is divided into sections based on the node similarity matrix. A region, resulting in a region set. A certain area The sum of the weights of the edges connecting the graph to the other regions is denoted as a cut. The cut of the entire graph can then be represented as:

[0091] (twenty three)

[0092] In the formula, Indicates the area The sum of the weights of the edges connecting to the other regions, and the coefficients This is used to avoid double-counting edge weights. The goal of this graph partitioning is to maximize the similarity between nodes within the same region and minimize the similarity between different regions. However, using only cuts as the partitioning criterion can easily lead to some regions being too small, i.e., sacrificing region balance to obtain smaller cut values. To mitigate this bias, this invention uses Normalized Cut (NCut) as the optimization objective for spectral clustering, defined as:

[0093] (twenty four)

[0094] In the formula, Indicates the area The sum of the weights of all edges within the region, which is also the sum of the similarities of nodes within that region. This indicates the similarity between two regions. When... When the value is small, meaning the connections between nodes within the region are weak, the corresponding Ncut value will increase; conversely... When the value is large, the Ncut value will decrease. This processing requires that the connections between different regions be as weak as possible, i.e. Minimize the size. At the same time, it requires strong similarity within each region, i.e. The goal is to maximize the size of the node region to make the partitioning result more reasonable. Therefore, the node region partitioning problem can be transformed into the following optimization problem:

[0095] (25)

[0096] Due to the node-to-sameness matrix It is a symmetric matrix, defining the node similarity matrix. ,in For nodes The sum of similarities with all other nodes, i.e.

[0097] (26)

[0098] Based on this, the Laplace matrix of the graph is introduced.

[0099] (27)

[0100] To describe the region partitioning results in matrix form, each node is assigned a vector. To indicate the category feature of the node, with a length of , The stacking of vectors from nodes can form an indicator matrix. .definition

[0101] (28)

[0102] The normalized cut objective can then be written as

[0103] (29)

[0104] The original problem is transformed into

[0105] (30)

[0106] Since the aforementioned discrete-constraint optimization problem is an NP-hard problem, it is difficult to directly obtain the global optimal solution. Therefore, this invention employs a spectral relaxation method to relax the discrete values ​​in the indicator matrix to the real number domain. Let...

[0107] (31)

[0108] The problem is then relaxed to

[0109] (32)

[0110] make

[0111] (33)

[0112] The optimization problem can then be written as:

[0113] (34)

[0114] because and It is symmetric, therefore the matrix Since the matrix is ​​real symmetric, according to spectral theory, the optimal solution of the above equation can be obtained from... The former The eigenvectors corresponding to the smallest eigenvalues ​​are used to construct the matrix. That is, only the normalized Laplacian matrix needs to be obtained. The former The eigenvectors corresponding to the smallest eigenvalues ​​are arranged into a matrix by their columns. This allows us to obtain the spectral embedding representation of the node in the low-dimensional space. Furthermore, from...

[0115] (35)

[0116] The low-dimensional feature representation matrix of the nodes can be obtained. .matrix Each row corresponds to a network node. The coordinates in the feature space comprehensively reflect the relative positional relationship of nodes in the global topology and overall network state. If two nodes are close in this feature space, it indicates that they have higher similarity in the original network and are more suitable to be assigned to the same region.

[0117] 4) Data file destination node selection strategy and UAV node access node selection strategy

[0118] Obtain the spectral clustering representation matrix between nodes. Then, k-means is used for region partitioning. Classical k-means only aims to minimize intra-class distances, without considering load balancing across classes. This invention incorporates both node distances within a region and regional data size balancing into the objective function. Let the... The set of nodes in each region is ,node The spectral coordinates are The coordinates of its class center are ,node Data files to be migrated Size is ,have

[0119] (36)

[0120] Therefore, the first The total amount of data in each region is

[0121] (37)

[0122] The total amount of data to be migrated across the entire network is recorded as follows:

[0123] (38)

[0124] The objective function for constructing and adjusting nodes is then:

[0125] (39)

[0126] In the formula, To balance the parameters, the first term of the objective function characterizes the distance from a node to its class center; the second term characterizes the balance of data size across regions; if... If the value is increased, the algorithm will be more biased towards load balancing; if... If the value is reduced, the algorithm will place more emphasis on the communication quality between nodes within the same class.

[0127] In each k-means iteration, the initial node classification is performed according to the minimum distance principle. Before updating the class centers, a class adjustment based on simulated annealing is added. Specifically, the current classification result is first taken as the current optimal solution; then, nodes are randomly selected and attempted to be adjusted to other classes. If the adjustment reduces the objective function value, it is accepted; otherwise, according to the probability acceptance mechanism of simulated annealing, the inferior solution is accepted with a certain probability, thereby enhancing the algorithm's ability to escape local optima. With the temperature parameter... As the number of iterations decreases, the probability of accepting suboptimal solutions gradually diminishes, ultimately retaining the best classification result obtained in that round of iterations. The algorithm terminates if the classification result remains unchanged for two consecutive iterations or if the maximum number of iterations is reached.

[0128] After obtaining the final region division results, for the first... For each of the 10 cluster regions, the actual ground network node closest to the cluster center within that region is selected as the 10th cluster. UAV The access node. Therefore, UAV... Access node decision variables are

[0129] (40)

[0130] In the formula, Represents a node Coordinates in spectral embedding space, Indicates the first The cluster center coordinates of each region can be obtained. Therefore, the access point decision variables for all UAVs are... .

[0131] At the same time, if the data file The source node satisfies Then the data files to be migrated on that node will be... Assigned to the UAV Responsible for, i.e., data files The target UAV decision variables are

[0132] (41)

[0133] Therefore, the destination UAV decision variables for all data files are: At this point, the decision variables... and This can be accomplished simultaneously using a clustering optimization algorithm. The specific algorithm is as follows:

[0134]

[0135] As shown in Algorithm 1, the network topology is... Link attributes The set of data files to be transmitted UAV quantity The data file serves as the input to the algorithm, and the output is the destination UAV decision variables. Decision variables of UAV connection nodes The first line normalizes the three attributes of the link. Lines 3 through 6 calculate the average network state score between any two node pairs, and then calculate the similarity between the node pairs based on the scores, providing a foundation for subsequent spectral clustering. Lines 7 through 10 are the core steps of spectral clustering: constructing the node pair similarity matrix, the node similarity matrix, and the normalized Laplacian matrix, and obtaining their respective frontiers. The eigenvectors corresponding to the smallest eigenvalues ​​are used to obtain the low-dimensional spectral embedding representation matrix of the nodes. The 11th row in the matrix Upper initialization Given cluster centers, obtain the cluster center coordinate matrix. Lines 12 to 28 execute the k-means clustering process. Lines 13 to 15 classify nodes based on their distances to cluster centers. Lines 17 to 26 introduce simulated annealing in each k-means clustering iteration to fine-tune the current classification results, enhancing the algorithm's ability to escape local optima through objective function calculation and classification result updates. Line 27 recalculates cluster centers based on the updated classification results. Lines 29 to 37 determine the access node decision variables for all UAVs based on the clustering results. The purpose of the data file is UAV decision variables. .

[0136] (2) Many-to-many communication routing optimization strategy based on whale optimization algorithm

[0137] UAV decision variables for determining the purpose of the data file Decision variables for UAV access nodes After that, the data file can be confirmed. Corresponding path set In fact, all that is needed next is to determine the data file. exist Path selection in this context means that only the decision variables need to be determined. In The rest It has no practical significance; let's assume they are all... .

[0138] 1) IWOA Algorithm

[0139] Since the many-to-many communication routing planning problem during parallel transmission of multiple files is a discrete, nonlinear, and strongly coupled combinatorial optimization problem, this invention employs the Improved Whale Optimization Algorithm (IWOA) for solution. IWOA utilizes a nonlinear shrinkage factor and adaptive inertia weight design to achieve a balance between global exploration and local exploitation. It is highly adaptable to handling high-dimensional, discontinuous discrete combinatorial optimization problems and is suitable for solving discrete optimization problems in many-to-many communication routing planning.

[0140] 2) Path selection scheme coding and fitness function design

[0141] This invention represents an individual whale as all data files. The path joint selection result, i.e.

[0142] (42)

[0143] In the formula, Indicates the first Individual whale; This indicates the data file under this individual. The selected path number. Therefore, an individual This corresponds to a set of files to be migrated. Path decision variable scheme According to formula (13), the fitness function is constructed as follows:

[0144] (43)

[0145] In the formula, Indicates by individual The specified file path The extremely small positive number introduced to prevent the denominator from being zero. Clearly, The smaller, The larger the fitness value, the more efficient the algorithm in this invention is in minimizing the maximum migration completion time of the system by maximizing fitness equivalence.

[0146] 3) PWLCM Chaotic Map Initialization

[0147] To improve the uniformity of the initial population distribution, a piecewise linear chaotic mapping (PWLCM) is used to generate the initial individuals:

[0148] (44)

[0149] In the formula, For control parameters, when hour PWLCM exhibits good chaotic properties and uniform distribution effect. In this invention, firstly, a... The continuous chaotic variables in the interval are then mapped to the path number space to obtain discrete initial solutions:

[0150] (45)

[0151] In the formula, The first generation generated for PWLCM The first in the individual Chaotic variables in each file. Truncation to the legal range:

[0152] (46)

[0153] 4) IWOA Location Update Mechanism

[0154] The whale optimization algorithm achieves its search by simulating the hunting behavior of humpback whales and bubble web attacks. Its core includes three mechanisms: prey encirclement, spiral update, and random search.

[0155] a) Encirclement of prey mechanism

[0156] Let the position of the globally optimal individual be... The update method for other whales to move closer to the optimal individual is as follows:

[0157] (47)

[0158] (48)

[0159] in and All are coefficients. , ,and Use random numbers; to overcome the linear shrinkage factor in WOA. To address the issue of insufficient later-stage search capabilities, IWOA will reduce the shrinkage factor. Designed in a non-linear form:

[0160] (49)

[0161] in, This represents the current iteration number. The maximum number of iterations, To adjust the parameters.

[0162] b) Spiral Steam Drum Network Attack Mechanism

[0163] The whale swims in a spiral around the optimal solution, and its update method is as follows:

[0164] (50)

[0165] (51)

[0166] in, It is a constant. The numbers are random. Furthermore, the encircling update and spiral update are based on probability. To switch, i.e.

[0167] (52)

[0168] c) Random search mechanism

[0169] when At this time, the algorithm tends to perform global exploration, in which case individuals circle around random whales. renew:

[0170] (53)

[0171] (54)

[0172] To further adjust the search step size, IWOA introduces a sigmoid-type adaptive inertia weight:

[0173] (55)

[0174] in, and This is the weight adjustment parameter. After introducing inertia weights, when... The position update formula for the partial development phase is rewritten as follows:

[0175] (56)

[0176] when At that time, the global exploration phase is updated to

[0177] (57)

[0178] The specific algorithm implementation is as follows:

[0179]

[0180] As shown in Algorithm 2, the target UAV decision variables of the data file obtained by Algorithm 1 are... Decision variables for UAV access nodes The population size and PWLCM parameters are taken as inputs, and the optimal path selection scheme is output. The first line initializes the population using PWLCM; the second line evaluates each individual based on the fitness function and determines the current optimal solution; lines 3 to 20 execute the whale optimization process, where lines 4 and 5 calculate the nonlinear convergence factor and adaptive inertia weight, respectively; lines 6 to 16 update the position of each individual and execute encirclement, spiral attack, or random search strategies based on different conditions; line 18 recalculates the fitness value of the updated individuals; and line 19 updates the global optimal solution; finally, the path decision variables of all data files are output. .

[0181] In summary, the edge-to-network storage data migration method based on SDWN and UAV proposed in this invention includes the following steps:

[0182] Step 1: Based on the link attributes, determine the destination UAV decision variables, UAV access node decision variables, and path decision variables for the data files, and construct an optimization problem by minimizing the migration and transmission time of all data files;

[0183] ,

[0184] In the formula, To optimize the problem; link Link attributes ,in Indicates link The remaining bandwidth, For link The time delay, For link The packet loss rate; The migration and transfer time for all data files is complete. , For data files Complete migration and transfer time, , For data files size, For data files Determined path The bottleneck bandwidth For data files Determined path packet loss rate For data files Determined path The delay; For UAV decision variables in data file destination, , Represents data file Should I select UAV? As the destination node; For UAV access node decision variables, , UAV Whether to connect to the node ; For path decision variables, , Indicates from the source node UAV to destination Choose a path ; For link For data files allocated bandwidth For link The maximum bandwidth limit; A collection of data files stored online. For link The collection of data files that pass through simultaneously For a set of nodes, For data files The set of nodes, For UAV set, For the source node UAV to destination The set of paths.

[0185] Step 2: Decompose the optimization problem constructed in Step 1 into two sub-problems: an optimization sub-problem concerning the UAV decision variables for the data file destination and the UAV access node decision variables, and an optimization sub-problem concerning the path decision variables.

[0186] Step 3: Use the spectral clustering k-means method to solve the optimization subproblem regarding the UAV decision variables for the data file destination and the UAV access node decision variables.

[0187] Step 3.1: Calculate the network state score for each path between every two nodes in the SDWN network topology, and take the TOP path with the longest path. The average of the network state scores for each path is used to obtain the comprehensive network state score:

[0188] Node pairs path Network state score for:

[0189] ,

[0190] Node pairs Comprehensive network state score for:

[0191] ,

[0192] In the formula, , and Each is a node pair path The normalized values ​​of bandwidth, latency, and packet loss rate. , and These are the weighting factors for bandwidth, latency, and packet loss rate, respectively. This is the set value.

[0193] Step 3.2: When two nodes in the SDWN network topology are non-neighboring nodes and their combined network state score is greater than the set combined network state score threshold, add a virtual edge to these two nodes to obtain a new SDWN network topology.

[0194] Step 3.3: Calculate the node pair similarity between every two nodes in the new SDWN network topology, and sum the node pair similarities of each node with all other nodes to obtain the node similarity score.

[0195] Node pairs Similarity for:

[0196] ,

[0197] node Similarity for:

[0198] ,

[0199] In the formula, For node pairs The overall network state score, For all node pairs The standard deviation of the overall network state score, A set of virtual edges; The number of nodes.

[0200] Step 3.4: Subtract the node similarity matrix constructed from node pair similarity from the node similarity matrix constructed from node similarity to obtain the Laplace matrix.

[0201] Step 3.5: Calculate the normalized Laplace matrix:

[0202] ,

[0203] In the formula, For the normalized Laplace matrix, This is the node similarity matrix.

[0204] Step 3.6: Take the first normalized Laplace matrix. The eigenvectors corresponding to the smallest eigenvalues ​​are then arranged column-wise to form an intermediate matrix, thus obtaining the low-dimensional feature representation matrix of the nodes:

[0205] .

[0206] In the formula, For the intermediate matrix, is the low-dimensional feature representation matrix of the nodes.

[0207] Step 3.7: First, initialize the low-dimensional feature representation matrix of the nodes. The system identifies cluster centers and obtains a distance-to-center coordinate matrix. Then, it performs a k-means clustering process to classify nodes based on their distances to each cluster center, resulting in the final cluster region partitioning.

[0208] In each round of k-means clustering iteration, a simulated annealing strategy is introduced to fine-tune the current classification results. The algorithm’s ability to escape local optima is enhanced by adjusting the objective function calculation and updating the classification results through node adjustment. The cluster centers are recalculated based on the updated classification results.

[0209] The above nodes adjust the objective function for:

[0210] ,

[0211] In the formula, For the first The total number of data files for each cluster region This represents the total number of data files to be migrated across the entire network. For the number of UAVs, For nodes Spectral coordinates The coordinates of the cluster center are, To balance the parameters, The number of nodes.

[0212] Step 3.8: After obtaining the final clustering region division results, the node closest to the cluster center in each cluster region is taken as the access node of the UAV corresponding to that cluster region, thus obtaining the UAV access node decision variable; the data files of the nodes in each cluster region are migrated to the UAV corresponding to that cluster region, thus obtaining the data file destination UAV decision variable;

[0213] Step 4: Use the whale optimization method to solve the optimization subproblem concerning the path decision variables.

[0214] Step 4.1: Represent an individual whale as the result of the joint selection of paths for all data files, and randomly generate an initial individual;

[0215] Step 4.2: Evaluate each individual based on the fitness function and determine the current optimal solution;

[0216] The above fitness function for:

[0217] ,

[0218] In the formula, For individuals Determined path The time required for all data files to be migrated and transferred. The extremely small positive numbers introduced to prevent the denominator from being zero;

[0219] Step 4.3: Execute the whale optimization process to obtain the global optimal solution, which is the path decision variable.

[0220] To achieve the above-mentioned on-network storage data file migration function, this invention proposes an edge on-network storage data migration system architecture based on SDWN and UAV, such as... Figure 3 As shown, it includes the data plane, control plane, and knowledge plane, which are described below:

[0221] (1) Data plane

[0222] The data plane is responsible for data storage within network nodes and data forwarding through these nodes. It consists of two types of SDN switch nodes: integrated storage and communication SDN switches with storage capabilities and ordinary SDN switches that only handle data forwarding. Both types of switches are equipped with dual-band Wi-Fi communication modules at 2.4 GHz and 5 GHz. The 2.4 GHz channel serves as the control channel between the data plane and the control plane, while the 5 GHz band is used to build a wireless mesh network for data forwarding between data plane nodes. Furthermore, each SDN switch in the data plane can act as a wireless access point (AP) for UAVs. UAVs can access any AP node in the data plane via the 5 GHz Wi-Fi channel, serving as the destination node for migrating stored data.

[0223] The data plane not only enables efficient forwarding and storage of data stored on the network, but also provides flexible and programmable underlying network support for multi-UAV assisted data migration under the unified scheduling of the control plane.

[0224] (2) Control plane

[0225] The control plane is decoupled from the data plane via a 2.4G wireless channel as a southbound interface, enabling centralized control and global visibility. The control plane, based on the RYU controller, mainly comprises four functional modules: a topology information sensing module, a network status measurement module, a flow table installation and migration command issuance module, and a UAV information processing module.

[0226] 1) Topology Information Awareness Module: This module is responsible for the awareness and maintenance of global network topology information. The controller collects information such as network status, storage resource status, and data distribution in the network via out-of-band TCP / IP protocol on the 2.4GHz wireless control channel. Simultaneously, the controller periodically acquires switch characteristic information and port information (such as ports) through the OpenFlow mechanism, and combines this with a link discovery mechanism (usually based on LLDP link detection) to identify the connection relationships between switches, thereby constructing a network topology map and dynamically updating it.

[0227] 2) Network Status Measurement Module: This module is mainly responsible for sensing the network's operating status. The controller obtains information such as port statistics, port status, and flow table statistics through OpenFlow statistical requests and event callback mechanisms, thereby comprehensively acquiring information on the current network resource utilization and link status changes.

[0228] 3) Flow Table Installation and Migration Command Issuance Module: This module is responsible for translating control plane decisions into specific execution rules for the data plane. On one hand, the controller forms the routing policies generated by the knowledge plane into corresponding forwarding table entries and sends them to relevant switches via OpenFlow protocol messages, achieving unified control over the data plane forwarding paths. On the other hand, for integrated communication and storage network nodes with data migration requirements, the controller issues migration start commands via TCP / IP protocol, thereby coordinating and triggering the data migration task.

[0229] 4) UAV Information Processing Module: This module is primarily responsible for the unified management of UAV access and data migration processes by the control plane. At the start of a migration task, the controller determines the UAV access node based on migration requirements and sends the target access node location information to the UAV via a 2.4GHz TCP / IP communication link, guiding the UAV to move to the designated access area. During this process, the controller continuously receives the location and access status reported by the UAV, and after confirming that it has completed access to the target node, it issues a data migration task command.

[0230] (3) Knowledge plane

[0231] The knowledge plane is responsible for processing the network state and topology node information acquired by the control plane, and calculating information such as remaining bandwidth, latency, and packet loss rate of network links. Based on this, the knowledge plane performs unified caching and management of network state and node information, and uses this information for clustering optimization to obtain the target UAV decision variables for the data file. Decision variables for UAV access network nodes The path optimization algorithm further combines clustering results and network state information to obtain the decision variables for many-to-many communication routing paths in data migration. .

[0232] System implementation and testing:

[0233] (1) Experimental environment and hardware design

[0234] This system includes one controller, ten integrated communication and storage switches, ten ordinary switches, and two drones. The system's software environment is listed in the table.

[0235] Table 1 System Software Version

[0236]

[0237] This invention selects the N5105 model soft router from Changwang Microcontroller Co., Ltd. as the RYU controller. This soft router has an Intel quad-core processor, DDR4 dual memory and wireless Wi-Fi function, and is small in size, making it suitable for the design of on-network storage data migration systems in complex and dynamic scenarios.

[0238] This invention selects the AX3000 router as the SDN integrated communication and storage switch. The router's main controller is the MT7981BA, with a CPU frequency of 1.3GHz, 1GB DDR4 RAM, and 8GB eMMC Flash. It also features a 1TB NVMe SSD for data storage and uses the MT7976 wireless chip to achieve dual-band Wi-Fi with a wireless transmission rate of up to 2400Mbps, making it suitable for establishing wireless mesh links in data migration systems and enabling high-bandwidth, high-speed transmission. The Redmi AX6000 router was selected as the SDN general-purpose switch. This router has a CPU frequency of up to 2GHz, 512MB of RAM, and 128MB of Flash, and also uses the MT7976 wireless chip to achieve dual-band Wi-Fi.

[0239] This invention configures a router as a wireless SDN switch. The process includes: compiling OpenWrt firmware and integrating necessary software and drivers under a Linux system; configuring the Open vSwitch plugin to enable the router to have SDN switch functionality; establishing a wireless mesh network based on the 802.11s protocol in the 5GHz band, and configuring automatic hard drive mounting and vsFTP service to achieve self-organizing network communication, data storage, and data migration.

[0240] This invention uses two quadcopter UAVs as the carriers of the data migration system. The core component of the flight control system is the IMU module, which is responsible for performing sensor data fusion, data filtering, attitude calculation, control algorithms, and PWM signal output from the motors. Simultaneously, the UAVs are equipped with a self-developed integrated communication and storage router, which also has dual-band Wi-Fi to communicate with both the control plane and the data plane simultaneously, thereby receiving control commands and collecting data. In this architecture, the UAVs' general-purpose MCU receives commands from the controller via the onboard router through a serial port, thus realizing complete flight control functions.

[0241] (2) System out-of-band design

[0242] Control plane management of data plane switches typically employs two methods: out-of-band management and in-band management. In-band management means that the control plane's control flow and the data plane's service data flow share the same physical link, eliminating the need for a dedicated management channel. However, in-band management usually requires the data plane's link to be loop-free; otherwise, control flows from the controller may trigger broadcast storms, leading to control failures on the data plane. Furthermore, in-band management suffers from high latency. For example, during flow table updates, switches located far from the controller often need to traverse multiple links to receive control messages, resulting in poor real-time control performance—a problem particularly pronounced in wireless link environments.

[0243] Out-of-band management mode uses an additional channel to manage the data plane, is not limited by loops, and has good real-time performance. The out-of-band management mode designed in this invention is as follows: Figure 4 As shown.

[0244] In its implementation, this invention uses the 2.4 GHz Wi-Fi interface of the data plane switch as a wireless client, connecting to the 2.4 GHz wireless network of the main router via a relay bridging method. Upon successful connection, the device automatically generates a wwan interface (Wireless Wide Area Network interface) to carry uplink communication between the switch and the main router. Simultaneously, the LAN-side interface br-lan of the main router serves as a local area network bridging interface, uniformly assigning IP addresses within the same local area network segment to devices connected to the wireless network. The RYU controller can connect to the same local area network via the 2.4 GHz or 5 GHz wireless network of the main router, thereby establishing out-of-band control connections with each data plane switch. The UAV side also uses the same method to connect to the network where the main router is located, maintaining control communication with the controller to receive task commands and report status information. The reason for using 2.4 GHz as the out-of-band control channel and 5 GHz as the data plane transmission link is primarily due to their different performance characteristics. In comparison, 2.4 GHz has stronger coverage and penetration capabilities, making it more suitable for carrying control flows that have lower bandwidth requirements but higher connection stability requirements; while 5 GHz has greater link bandwidth, which can better meet the needs of large-scale data transmission in the data plane.

[0245] Because the integrated communication and storage switch designed in this invention not only has data forwarding capabilities but also provides data storage and migration functions, the system further creates a virtual bridge br-ovs and configures it as the bridging interface of the Open vSwitch instance to carry the virtual switching and logical interconnection functions within the switch. By configuring br-ovs with IP addresses in a unified network segment, basic communication capabilities between switches can be achieved.

[0246] (3) Experimental results and analysis

[0247] To verify the effectiveness of the proposed method, this invention conducted two-UAV-assisted data migration experiments on network topologies with 10, 14, and 20 nodes, respectively. The amount of data to be migrated on the integrated communication and storage switch nodes in the network nodes was set to 200MB to 400MB. Each experiment was repeated 10 times, and the maximum migration time across all data files was used as the performance evaluation metric.

[0248] 1) The impact of parameter settings on performance

[0249] This invention tests the construction parameters of the average integrated network state score between nodes and the trade-off parameters in k-means, comparing the maximum migration time of data files corresponding to different parameter settings under 10, 14, and 20 node topologies.

[0250] In the calculation of the average integrated network state score among nodes, the top [number] nodes are selected respectively. The average of the comprehensive network state scores for each path is calculated, where... Take 1, 2, 3, and 4, and set the formula. Parameters in The experimental results are shown in Figure 5. For all three topologies, the maximum migration time generally decreases as the number of paths involved in the averaging increases, and the fluctuations in the results also decrease. This indicates that in parallel data transmission, the communication environment between nodes cannot be fully characterized by only a small number of paths. The mean of the comprehensive state scores based on multiple paths can construct a node similarity matrix that is more suitable for parallel data transmission, thereby obtaining better decision variables for UAV access nodes and destination UAV decision variables for data files.

[0251] Trade-off parameters in k-means Used to coordinate the similarity between nodes within a region and to balance the migration data load between different UAVs. For comparing different... The values ​​are set to reflect the changes in the system's maximum migration time. , , and Set the formula Parameters in The experimental results are shown in Figure 6. The results indicate that under the three topologies, when the parameters... When the parameter is set to a moderate value, the system performance is optimal; when the parameter is set to a moderate value, the system performance is optimal When the parameters are too small, clustering algorithms emphasize the similarity within a cluster, which can easily lead to some UAVs handling an excessive amount of migration data, thereby increasing link contention near the access node and the UAV write load; when the parameters are too small... While excessively high values ​​can aid in load balancing, they can also reduce the similarity in network states among nodes within a region. Therefore, selecting appropriate trade-off parameters is more beneficial for improving the overall performance of multi-UAV assisted data migration.

[0252] 2) Comparative experiments of different methods

[0253] After the parameters were determined, the method of this invention was further compared with two baseline methods. The method of this invention is denoted as Cluster_Iwoa_Heuristic, which first designs a k-means method for data dimensionality reduction using spectral clustering to complete the decision variables for UAV access nodes. With regard to the purpose of the data file, UAV decision variables After determining the variables, the whale optimization method is used for many-to-many communication routing path planning during parallel transmission of multiple files. Two baseline methods are Full_Iwoa_Heuristic and All_Random. The former directly addresses the decision variables. , and routing planning path decision variables A joint heuristic search was performed, in which a randomized strategy was applied to all three decision variables. The experimental results are shown in Figure 7. The All_Random method performed the worst under different topologies, indicating that if the topology is not properly optimized... , and Optimizing these three decision variables appropriately will significantly increase the maximum migration time of the system. Both the Cluster_Iwoa_Heuristic and Full_Iwoa_Heuristic methods can reduce the maximum migration time, but the Cluster_Iwoa_Heuristic method of this invention shows better adaptability as the topology size and complexity increase. In a 10-node topology, due to the small network size and relatively limited solution space, both the method of this invention and direct joint search can obtain good results, so the difference between the two is not significant. However, when the topology size increases to 14 and 20 nodes, the coupling between the three decision variables strengthens, and direct search is easily affected by complex coupling relationships, leading to local optima problems and increased result fluctuations. In contrast, the method of this invention first obtains the similarity between nodes through the average comprehensive state score between nodes, and then performs clustering based on this. Furthermore, the clustering process considers the impact of parallel transmission of multiple data files and the load impact of UAV migration data, thus obtaining a reasonable... and The decision variables allow for a more rational aggregation of data traffic within the network and the data load of UAV migration during on-network storage data migration. Based on this, many-to-many communication routing planning is then performed, thus reducing link contention caused by parallel transmission of multiple files in large-scale topologies, resulting in a shorter maximum system migration time and better stability.

[0254] It should be noted that although the embodiments described above are illustrative, they are not intended to limit the invention. Therefore, the invention is not limited to the specific embodiments described above. Any other embodiments obtained by those skilled in the art under the guidance of this invention without departing from its principles are considered to be within the protection scope of this invention.

Claims

1. A method for migrating edge-to-network storage data based on SDWN and UAV, characterized by: The steps include the following: Step 1: Based on the link attributes, determine the destination UAV decision variables, UAV access node decision variables, and path decision variables for the data files, and construct an optimization problem by minimizing the migration and transmission time of all data files; ， In the formula, To optimize the problem, The migration and transfer time for all data files is complete. , For data files Migration and transfer completion time; For UAV decision variables in data file destination, , Represents data file Should I select UAV? As the destination node; For UAV access node decision variables, , UAV Whether to connect to the node ; For path decision variables, , Indicates from the source node UAV to destination Choose a path ; For link For data files allocated bandwidth For link The maximum bandwidth limit; A collection of data files stored online. For link The collection of data files that pass through simultaneously For a set of nodes, For data files The set of nodes, For UAV set, For the source node UAV to destination The set of paths; Step 2: Decompose the optimization problem constructed in Step 1 into two sub-problems: an optimization sub-problem concerning the UAV decision variables for the data file destination and the UAV access node decision variables, and an optimization sub-problem concerning the path decision variables. Step 3: Use the spectral clustering k-means method to solve the optimization subproblem regarding the UAV decision variables for the data file destination and the UAV access node decision variables; Step 4: Use the whale optimization method to solve the optimization subproblem concerning the path decision variables.

2. The edge-to-network storage data migration method based on SDWN and UAV according to claim 1, characterized in that, link Link attributes ,in Indicates link The remaining bandwidth, For link The time delay, For link The packet loss rate.

3. The edge-to-network storage data migration method based on SDWN and UAV according to claim 1, characterized in that, Data file Complete migration and transfer time for: ， In the formula, For data files size, For data files Determined path The bottleneck bandwidth For data files Determined path packet loss rate For data files Determined path The time delay.

4. The edge-to-network storage data migration method based on SDWN and UAV according to claim 1, characterized in that, The specific process of step 3 is as follows: Step 3.1: Calculate the network state score for each path between every two nodes in the SDWN network topology, and take the TOP path with the longest path. The average of the network state scores for each path is used to obtain the comprehensive network state score: Step 3.2: When two nodes in the SDWN network topology are non-neighboring nodes and their combined network state score is greater than the set combined network state score threshold, add a virtual edge to these two nodes to obtain a new SDWN network topology. Step 3.3: Calculate the node pair similarity of every two nodes in the new SDWN network topology, and sum the node pair similarities of each node with all other nodes to obtain the node similarity: Step 3.4: Subtract the node pair similarity matrix constructed from node pair similarity from the node similarity matrix constructed from node similarity to obtain the Laplace matrix; Step 3.5: Calculate the normalized Laplace matrix: ， Step 3.6: Take the first normalized Laplace matrix. The eigenvectors corresponding to the smallest eigenvalues ​​are then arranged column-wise to form an intermediate matrix, thus obtaining the low-dimensional feature representation matrix of the nodes: ， Step 3.7: First, initialize the low-dimensional feature representation matrix of the nodes. The system identifies cluster centers and obtains a distance-to-center coordinate matrix. Then, it performs a k-means clustering process to classify nodes based on their distances to each cluster center, resulting in the final cluster region partitioning. Step 3.8: After obtaining the final clustering region division results, the node closest to the cluster center in each clustering region is taken as the access node of the UAV corresponding to that clustering region, thereby obtaining the UAV access node decision variables; The data files of nodes within each cluster region are migrated to the corresponding UAV of that cluster region, thereby obtaining the decision variables of the destination UAV for the data files; In the formula, For the normalized Laplace matrix, This is the node similarity matrix. For the intermediate matrix, The low-dimensional feature representation matrix for nodes; This represents the number of UAVs.

5. The edge-to-network storage data migration method based on SDWN and UAV according to claim 4, characterized in that, In step 3.1, Node pairs path Network state score for: ， Node pairs Comprehensive network state score for: ， In the formula, , and Each is a node pair path The normalized values ​​of bandwidth, latency, and packet loss rate. , and These are the weighting factors for bandwidth, latency, and packet loss rate, respectively. This is the set value.

6. The edge-to-network storage data migration method based on SDWN and UAV according to claim 4, characterized in that, In step 3.3, Node pairs Similarity for: ， node Similarity for: ， In the formula, For node pairs The overall network state score, For all node pairs The standard deviation of the overall network state score, A set of virtual edges; The number of nodes.

7. The edge-to-network storage data migration method based on SDWN and UAV according to claim 4, characterized in that, Step 3.7 further includes: in each round of k-means clustering iteration, a simulated annealing strategy is introduced to fine-tune the current classification result, and the algorithm's ability to escape local optima is enhanced by adjusting the objective function calculation and updating the classification result through node adjustment; and the cluster centers are recalculated based on the updated classification result. The above nodes adjust the objective function for: ， In the formula, For the first The total number of data files for each cluster region This represents the total number of data files to be migrated across the entire network. For the number of UAVs, For nodes Spectral coordinates The coordinates of the cluster center are, To balance the parameters, The number of nodes.

8. The edge-to-network storage data migration method based on SDWN and UAV according to claim 1, characterized in that, The specific process of step 4 is as follows: Step 4.1: Represent an individual whale as the result of the joint selection of paths for all data files, and randomly generate an initial individual; Step 4.2: Evaluate each individual based on the fitness function and determine the current optimal solution; The above fitness function for: ， In the formula, For individuals Determined path The time required for all data files to be migrated and transferred. The extremely small positive numbers introduced to prevent the denominator from being zero; Step 4.3: Execute the whale optimization process to obtain the global optimal solution, which is the path decision variable.

9. An edge-to-network storage data migration system based on SDWN and UAV that implements the method of claim 1, characterized in that, This includes the SDWN network and multiple UAVs; the SDWN network consists of a data plane, a control plane, and a knowledge plane. The data plane is responsible for storing data within network nodes and forwarding data through network nodes. The data plane consists of two types of SDN switch nodes: integrated storage and communication SDN switches with storage capabilities and ordinary SDN switches that only handle data forwarding. Both types of switches are equipped with dual-band wireless communication modules. The low-frequency wireless channel serves as the control channel between the data plane and the control plane, while the high-frequency wireless channel is used to construct a wireless mesh network for data forwarding between data plane nodes. UAVs access the SDN switch nodes of the data plane through the high-frequency wireless channel, serving as the destination node for migrating stored data on the network. The control plane is decoupled from the data plane through a low-frequency wireless channel as a southbound interface, enabling centralized control and global visibility. The control plane is based on the RYU controller and mainly includes a topology information sensing module responsible for sensing and maintaining global network topology information, a network status measurement module responsible for sensing network operating status, a flow table installation and migration instruction issuance module responsible for converting the decisions of the control plane into specific execution rules of the data plane, and a UAV information processing module responsible for the unified management of UAV access and data migration processes by the control plane. The knowledge plane is responsible for processing the network status information and topology node information obtained by the control plane, and performs unified caching and management of the network status information and node information. Based on this information, the destination UAV decision variables, UAV access node decision variables and path decision variables of the data file are obtained.

Images