Adaptive adjustment method and system for network routing

Abstract

The present application relates to the technical field of network routing, and discloses a method and system for adaptively adjusting network routing, the method comprising: performing detection data packet transmission and hierarchical clustering on a network interface to obtain a path cluster set; performing double-circular matrix encoding on the path cluster set to obtain a routing path encoding set; constructing a spatiotemporal feature tensor based on the routing path encoding set, and performing tensor decomposition and feature pattern extraction to obtain a time-varying weight coefficient; performing sparse-aware proportional iterative optimization calculation based on the spatiotemporal feature tensor and the time-varying weight coefficient to obtain an optimal routing configuration scheme, thereby achieving differentiated processing of different business flows and improving network resource utilization efficiency.

Description

Technical Field

[0001] The present application relates to the technical field of network routing, and in particular to a method and system for adaptively adjusting network routing. Background Art

[0002] High-speed routers face the challenges of dynamic changes in network topology and unbalanced traffic load. Due to the complexity of network topology and the suddenness of business traffic, traditional static routing methods are difficult to adapt to the rapid changes in network status, resulting in problems such as decreased network performance, increased latency, and low bandwidth utilization.

[0003] With the widespread application of new technologies such as the Internet of Things and cloud computing, network services are becoming more heterogeneous and diverse. Different service flows have different requirements for service quality, but existing routing methods lack the ability to differentiate service flows and cannot meet the service quality requirements of different types of service flows, especially in high-concurrency scenarios. Summary of the invention

[0004] The present application provides a method and system for adaptively adjusting network routing, thereby achieving differentiated processing of different business flows and improving the efficiency of network resource utilization.

[0005] A first aspect of the present application provides a method for adaptively adjusting network routing, the method comprising: Probe the data packet transmission and hierarchical clustering of the network interface to obtain a set of path clusters; Performing double-circular matrix coding on the path cluster set to obtain a routing path coding set; Constructing a spatiotemporal feature tensor based on the routing path encoding set, and performing tensor decomposition and feature pattern extraction to obtain a time-varying weight coefficient; Based on the spatiotemporal feature tensor and the time-varying weight coefficient, a sparse-aware proportional iterative optimization calculation is performed to obtain an optimal routing configuration solution.

[0006] A second aspect of the present application provides a network routing adaptive adjustment system, the network routing adaptive adjustment system comprising: A transmission module, used for performing detection data packet transmission and hierarchical clustering on the network interface to obtain a set of path clusters; An encoding module, used for performing double-circular matrix encoding on the path cluster set to obtain a routing path encoding set; A construction module, used to construct a spatiotemporal feature tensor based on the routing path encoding set, and perform tensor decomposition and feature pattern extraction to obtain a time-varying weight coefficient; The computing module is used to perform sparse-aware proportional iterative optimization calculation based on the spatiotemporal feature tensor and the time-varying weight coefficient to obtain an optimal routing configuration solution.

[0007] The third aspect of the present application provides an electronic device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the electronic device executes the above-mentioned adaptive adjustment method of network routing.

[0008] A fourth aspect of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned method for adaptively adjusting network routing.

[0009] Compared with the prior art, the present application has the following beneficial effects: by introducing the concept of path clusters and adopting a hierarchical clustering method, network links are dynamically grouped and managed, thereby effectively reducing the complexity of routing calculations and improving the scalability of routing algorithms; by adopting a double-circular matrix coding technique, accurate modeling of the relationships within and between path clusters is achieved, thereby enhancing the adaptability of routing schemes to changes in network topology; by combining spatiotemporal feature tensor decomposition and time-varying weight coefficients, multi-dimensional feature extraction of network states is achieved, thereby improving the accuracy of routing decisions; by using a sparse-aware proportional iterative optimization algorithm, while ensuring convergence, joint optimization of multiple performance indicators is achieved, thereby improving the comprehensive performance of routing schemes; by introducing a hybrid routing strategy based on quality of service classification, differentiated processing of different business flows is achieved, thereby improving the efficiency of network resource utilization; by using a multi-link bandwidth measurement and delay difference correction mechanism, real-time monitoring and dynamic adjustment of routing performance are achieved, thereby enhancing the robustness of routing schemes. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0011] The structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with this technology. They are not used to limit the conditions under which the present invention can be implemented, and therefore have no substantive technical significance. Any structural modification, change in proportion or adjustment of size, without affecting the effects and purposes that can be achieved by the present invention, should still fall within the scope of the technical contents disclosed by the present invention.

[0012] Figure 1 It is a flowchart of a method for adaptively adjusting network routing provided by an embodiment of the present invention; Figure 2 is a schematic block diagram of the structure of a network routing adaptive adjustment system provided by an embodiment of the present invention; Figure 3 It is a schematic block diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0013] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0014] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0015] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0016] It should be further understood that the term "and / or" used in the specification and appended claims of this application refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations. Figure 1 , an embodiment of the method for adaptively adjusting network routing in the embodiment of the present application includes: Step 100: Detect data packet transmission and hierarchical clustering on the network interface to obtain a path cluster set; It is understandable that the execution subject of the present application may be a network routing adaptive adjustment system, or a terminal or a server, which is not limited here. The present application embodiment is described by taking a server as the execution subject as an example.

[0017] Specifically, a network detection module is deployed on the network interface to monitor various characteristics of the network link. The network detection module is set up, including adding sending timestamp information and sequence number information to the detection data packet, which is used to uniquely identify the transmission process of the data packet so that the transmission status of each data packet can be accurately tracked. The detection data packet is sent periodically. In this way, the link performance in the network is continuously monitored. Each time a data packet is sent, its sending timestamp is recorded, and the receiving timestamp is recorded again when the data packet is received. Through these time information, the network link delay that the data packet passes from the sending end to the receiving end can be calculated. Based on the difference between the sending timestamp and the receiving timestamp, the obtained network link delay information can reflect the current network status and is used for subsequent path optimization and adjustment. The network link delay information is stored in the form of a matrix, where the rows of the matrix represent different network interfaces and the columns represent collection points at different times. By constructing a delay feature matrix, the network link performance characteristics can be effectively linked to the time change trend, and a first delay feature matrix is ​​obtained to record the link delay characteristics of each network interface at different time points. Since the fluctuation of network performance has a certain randomness, the first delay feature matrix is ​​subjected to data smoothing. By performing data smoothing on each column of the delay feature matrix, the influence of noise in the data can be reduced, and a second delay feature matrix is ​​obtained. Based on the second delay feature matrix, the delay feature similarity between different network links is calculated. By comparing the feature differences between different rows in the matrix, the similarity of each pair of links is calculated to obtain a similarity distance matrix, which reflects the similarity between different links. The more similar the links, the shorter their distances, while the links with greater differences are shown as longer distances. Hierarchical clustering calculation is performed based on the similarity distance matrix, and these links are classified using a hierarchical clustering algorithm. Hierarchical clustering can hierarchically divide the data in a step-by-step merging or step-by-step splitting manner to obtain clustering results at different levels. Through hierarchical clustering processing, network links with high similarity can be classified into the same cluster, and multiple initial path clusters can be obtained. The links in each initial path cluster are relatively close in features and are regarded as alternative paths in terms of performance. For each initial path cluster obtained, an identifier is assigned to it so that each path cluster can be clearly identified in subsequent operations. At the same time, a member link list is created for each initial path cluster, which contains all network links in the path cluster. The average delay characteristics of the member link list are calculated to obtain the cluster delay feature vector, which can reflect the overall performance of the entire path cluster. The identifier of the initial path cluster, the member link list, and the cluster delay feature vector are combined and stored to obtain a path cluster set, realizing group management of network links.

[0018] Step 200, double-circular matrix encoding is performed on the path cluster set to obtain a routing path encoding set; Specifically, coordinate mapping is performed on the cluster delay feature vector of each path cluster in the path cluster set, and the cluster delay feature vector is mapped to a specific coordinate system to represent the relationship between the links in the cluster. Through this mapping, a matrix representing the relationship between the links in the cluster is constructed for each path cluster, namely the first cyclic dimension matrix. The function of the first cyclic dimension matrix is ​​to describe the link structure characteristics inside the path cluster in the form of a matrix to obtain an internal correlation matrix, which can fully reflect the closeness and delay characteristics between the links in the cluster. Relationship mapping is performed on the internal correlation matrices of different path clusters in the path cluster set. There is a certain correlation between different path clusters, which is not only reflected in the mutual connection in the network structure, but also in the interactive influence on the performance. In order to capture the complex relationship between these clusters, a second cyclic dimension matrix representing the correlation between clusters is constructed to obtain an external correlation matrix. The internal correlation matrix and the external correlation matrix are subjected to tensor product operation. The two matrices are operated according to specific rules to obtain a higher-dimensional matrix, so as to capture the comprehensive characteristics between different path clusters in a wider space. On the basis of completing the tensor product operation, a double cyclic coding space is constructed to form a double cyclic matrix. The double-circular matrix can describe the global structure of the network path in a compact form by integrating the features of two circular dimensions. The double-circular matrix is ​​subjected to eigenvalue decomposition to obtain the main eigenvector of each path cluster. The main eigenvector is the main representation of the path cluster in the feature space and can reflect the global characteristics of the cluster. The main eigenvector of each path cluster is normalized to limit the value of the eigenvector to a specific range and eliminate the differences caused by different feature dimensions. The cosine distance is calculated based on the normalized eigenvector, and the path cluster similarity matrix is ​​obtained by calculating the cosine distance between different path clusters. The value of each element in the similarity matrix represents the similarity between two path clusters. The higher the value, the closer their characteristics are. A similarity merging threshold is set for the path cluster similarity matrix. By setting the threshold, the path clusters with similarity higher than the threshold are merged to form a new path cluster. The purpose of the merging operation is to reduce the number of clusters while retaining the main features in the network, making the routing path selection process more efficient. The eigenvector of the double-circular matrix is ​​recalculated for the newly formed path cluster to ensure that the merged path cluster can be accurately described. The double-circular matrix feature vector is used as the encoding of the routing path to obtain the path encoding sequence. The path encoding sequence is the result of uniquely identifying the characteristics of each path cluster, which is used for fast path search and routing decision. The path encoding sequence and its corresponding new path cluster are associated and stored to obtain the routing path encoding set.

[0019] Step 300: construct a spatiotemporal feature tensor based on the routing path code set, and perform tensor decomposition and feature pattern extraction to obtain a time-varying weight coefficient; It should be noted that the time window length for feature extraction is set. The network interface is sampled in each time window, and relevant information such as the number of data packets, the number of data bytes, and the packet loss rate are collected to obtain statistical data of network traffic, reflecting the traffic changes of the network interface in a specific time period. The network traffic statistical data is divided into path clusters according to the routing path encoding set and input into the dynamic time warping algorithm. The traffic features at different time points are aligned, and the traffic data with different time sampling features are transformed into a unified time series to eliminate the errors caused by time offset or data inconsistency. Through dynamic time warping, the regularized traffic features are obtained, making the data smoother and more unified. The performance index is calculated for each path cluster in the regularized traffic features, and the calculation results are organized in the form of a path performance index matrix. Each row in the path performance index matrix represents the performance characteristics of a path cluster, and each column is a different performance index, such as bandwidth utilization, packet loss rate, delay, etc. The path performance index matrix is ​​organized according to the three dimensions of path cluster, performance index and time series to construct a spatiotemporal feature tensor. The spatiotemporal feature tensor is decomposed. The dimension of the feature data is reduced, thereby reducing the complexity of data processing and extracting the main mode of the data. In this process, the spatiotemporal feature tensor is decomposed into a path cluster pattern tensor, a performance indicator pattern matrix, and a time pattern vector. By decomposing the tensor, the reduced-dimensional feature data is obtained. The reduced-dimensional feature data contains the pattern associations between path clusters, the relationships between different performance indicators, and the feature patterns that change over time. Based on the reduced-dimensional feature data, the feature pattern sequence is extracted. The decay coefficient of the pattern weight is set on the basis of the feature pattern sequence, and the weight of the feature pattern is gradually reduced as time goes by, reflecting the dynamic change characteristics of the network state. The decay mechanism makes the new data have a greater impact on the old pattern, thereby realizing real-time adjustment of the dynamic characteristics of the network. By setting the weight decay coefficient, the pattern weight vector is obtained, which is used to represent the importance of each feature pattern in different time periods. The pattern weight vector is weighted and combined with the path cluster pattern tensor, and the influence of the feature pattern is directly reflected on each path cluster, so that the characteristics of the path cluster can better reflect the overall state of the network. The updated path cluster contains the dynamically adjusted weight characteristics, and finally the time-varying weight coefficient of each path cluster is obtained.

[0020] Step 400: Perform sparse-aware proportional iterative optimization calculation based on the spatiotemporal feature tensor and the time-varying weight coefficient to obtain the optimal routing configuration solution.

[0021] Specifically, the objective function of the recursive least squares framework is constructed for the spatiotemporal feature tensor and the time-varying weight coefficient. In this objective function, the performance indicators of each path cluster are included in the optimization scope to form a comprehensive optimization target model. The optimization target model combines the performance characteristics of different path clusters, including performance indicators such as latency, bandwidth utilization and load balancing, so that the model can describe the overall state of the current network and determine the direction that needs to be optimized. According to the constructed optimization target model, sparse constraints are constructed for the latency, bandwidth utilization and load balancing performance indicators of the network. By setting corresponding constraint thresholds for each performance indicator, it is possible to ensure that each performance indicator is maintained within a reasonable range during the optimization process. The purpose of sparse constraints is to adjust only a few key path clusters during optimization, thereby reducing the overall changes in network configuration and maintaining the stability of the system. After completing the constraint setting, a mathematical optimization equation group is obtained, which includes the correlation and constraint conditions between each performance indicator and the path cluster. The mathematical optimization equation group is input into the sparse-aware proportional iterative algorithm to optimize the routing configuration of the current path cluster. The objective function value is calculated to obtain the initial state of optimization. The initial optimization state represents the current performance of the network and is the starting point of the entire optimization process. Based on the initial state, gradient calculation is performed to determine how to adjust the network configuration. During the gradient calculation process, the gradient direction is subjected to sparsity constraint processing. During the optimization process, only the path clusters that have a greater impact on the overall performance of the network are focused on, while the paths that do not change significantly remain unchanged. The processing of sparsity constraints can effectively reduce the complexity of the calculation and avoid excessive adjustments to the entire network. By processing the sparsity constraints on the gradient direction, the optimization direction vector is obtained, which represents the optimization direction. The optimization direction vector is adaptively calculated. The size of the step determines the adjustment amplitude of the network configuration in each iteration. The adaptive step calculation dynamically adjusts the step size according to the current network state to ensure that the adjustment process can converge quickly and will not cause network instability due to excessive adjustment. Through the adaptive step calculation, the configuration update amount is obtained, which indicates the specific adjustment amplitude of the configuration parameters of each path cluster. The configuration update amount is combined with the current routing configuration to update the configuration parameters of each path cluster and obtain the routing configuration parameters. Substitute the updated routing configuration parameters into the spatiotemporal feature tensor to recalculate the performance indicators of each path cluster. Then compare the new performance indicators with the optimization target model to obtain the convergence judgment result. The convergence judgment result is used to evaluate whether the current routing configuration has reached the optimal state or needs further adjustment. If the result of the convergence judgment indicates that the optimization process has not yet reached the convergence condition, continue to iterate according to the current state until the convergence requirements are met. When the convergence conditions are met, select the optimal configuration combination from all routing configuration parameter combinations as the final routing configuration solution.The optimal combination of routing configuration parameters represents the routing configuration method that can achieve the best performance in the current network environment. This configuration achieves the best network performance, including the lowest latency, the best bandwidth utilization, and the most reasonable load balancing.

[0022] In the embodiment of the present application, by introducing the concept of path clusters and adopting a hierarchical clustering method, network links are dynamically grouped and managed, which effectively reduces the complexity of routing calculations and improves the scalability of routing algorithms; by adopting a double-loop matrix coding technique, accurate modeling of the relationships within and between path clusters is achieved, and the adaptability of the routing scheme to changes in network topology is enhanced; by combining spatiotemporal feature tensor decomposition and time-varying weight coefficients, multi-dimensional feature extraction of network states is achieved, and the accuracy of routing decisions is improved; by using a sparse-aware proportional iterative optimization algorithm, while ensuring convergence, joint optimization of multiple performance indicators is achieved, thereby improving the overall performance of the routing scheme; by introducing a hybrid routing strategy based on quality of service classification, differentiated processing of different business flows is achieved, thereby improving the efficiency of network resource utilization; by using a multi-link bandwidth measurement and delay difference correction mechanism, real-time monitoring and dynamic adjustment of routing performance are achieved, thereby enhancing the robustness of the routing scheme.

[0023] In a specific embodiment, the process of executing step 100 may specifically include the following steps: Deploy a network detection module on the network interface, set the sending timestamp information and sequence number information for the network detection module, obtain a detection data packet, and periodically send the detection data packet, record the receiving timestamp information, and calculate based on the sending timestamp information and the receiving timestamp information to obtain the network link delay information; The network link delay information is stored by rows representing network interfaces and columns representing time points to obtain a first delay feature matrix, and data smoothing is performed on the first delay feature matrix to obtain a second delay feature matrix; Calculating the delay feature similarity between network links based on the second delay feature matrix, and constructing a similarity distance matrix according to the delay feature similarity; Based on the similarity distance matrix, hierarchical clustering calculation is performed, and network link classification processing is performed to obtain initial path clusters, and identifiers are assigned to the initial path clusters. A member link list is created for each initial path cluster, and the average delay characteristics of the member link list are calculated to obtain the cluster delay feature vector. The identifier of the initial path cluster, the member link list and the cluster delay feature vector are combined and stored to obtain a path cluster set.

[0024] Specifically, a network detection module is deployed on the network interface to monitor the performance characteristics of the network link. The sending timestamp information and sequence number information of the detection data packet are set in the network detection module. The sending timestamp information is used to record the specific time when the data packet is sent from the source end, and the sequence number information is used to uniquely identify each data packet, so that different data packets can be distinguished and their transmission paths can be tracked. The detection data packet is sent periodically. Periodic transmission helps to collect the status information of the network link at different time points and obtain the time-varying characteristics of the network link. The receiving timestamp information is also recorded at the sending and receiving ends of the detection data packet. By comparing the sending timestamp with the receiving timestamp, the transmission delay of the data packet in the network link is calculated. Assume that the sending time is , receiving time is , then the delay of the network link is expressed as: ; in, Represents the delay of the network link. This calculation helps evaluate the performance of the link, especially the delay characteristics. The delay information of the network link is stored in a specific way, with rows representing different network interfaces and columns representing different time points, to construct a delay feature matrix. Assume that there are network interfaces and time point, we get a The matrix is ​​called the first delay characteristic matrix and is expressed as: ; in, Indicates The network interface is in The delay data of each time point. Since the delay of the network link is affected by many factors, such as instantaneous network congestion, the delay data will fluctuate greatly. In order to make the data smoother and more suitable for subsequent analysis, the first delay feature matrix is ​​smoothed. The data smoothing method uses the moving average method. Assuming that the sliding window length is , then the smoothed delay data is: ; By using the moving average method, the short-term fluctuations in the delay are eliminated, and a more stable feature matrix is ​​obtained, namely the second delay feature matrix, which is recorded as Based on the second delay feature matrix, the delay feature similarity between network links is calculated. The similarity is calculated using the cosine similarity method. and , and its similarity is defined as: ; The closer the similarity value is to 1, the more similar the delay characteristics of the two links are, and the closer the value is to 0, the greater the difference between the delay characteristics of the two links. By calculating the similarity between all links, a similarity matrix is ​​obtained, which is recorded as , where each element Indicates link and Link Based on the similarity matrix , construct a similarity distance matrix, and the elements in the distance matrix are defined as: ; The values ​​in the distance matrix represent the similarity differences between links. The smaller the value, the closer the characteristics between the links. , perform hierarchical clustering calculations. Group links with high similarity into one category to reduce the complexity of the network optimization process. Assuming that a bottom-up clustering method is used, by setting a distance threshold, links with distances less than the threshold are aggregated into the same cluster to obtain several initial path clusters. Assign identifiers to the initial path clusters, and create a member link list for each initial path cluster, which contains the identifiers of all links belonging to the path cluster. Calculate the average delay characteristics of the member links in each path cluster to obtain the cluster delay feature vector of the path cluster. Assume that a path cluster contains The delay characteristic vector is expressed as: ; in, Represents a path cluster The delay characteristic vector of Indicates link The characteristic vector of Represents a path cluster The member link set of the initial path cluster is stored together with the member link list and the delay feature vector of the cluster to form a path cluster set. The structure of the path cluster set is represented as containing multiple records, each of which contains the identifier of the path cluster, the member link list and the delay feature vector of the cluster.

[0025] In a specific embodiment, the process of executing step 200 may specifically include the following steps: The cluster delay feature vector of each path cluster in the path cluster set is mapped to construct a first circular dimension matrix representing the link relationship within the cluster to obtain an internal correlation matrix. The internal correlation matrices of different path clusters in the path cluster set are mapped to construct a second circular dimension matrix representing the correlation relationship between clusters to obtain an external correlation matrix. Performing tensor product operation and double-circular coding space construction on the internal correlation matrix and the external correlation matrix to obtain a double-circular matrix, and performing eigenvalue decomposition operation on the double-circular matrix to obtain the main eigenvector of each path cluster; The main feature vector of each path cluster is normalized and the cosine distance is calculated to obtain the path cluster similarity matrix, and a similarity merging threshold is set for the path cluster similarity matrix, and the path clusters with similarity higher than the similarity merging threshold are merged into a new path cluster; The double-circular matrix eigenvector is recalculated for the new path cluster, and the double-circular matrix eigenvector is used as the routing path code to obtain a path code sequence, and the path code sequence and its corresponding new path cluster are associated and stored to obtain a routing path code set.

[0026] Specifically, the delay feature vector of each path cluster is mapped into a multidimensional coordinate space to describe the characteristic relationship between the links within the path cluster. Assume that the cluster delay feature vector of each path cluster is ,in Represents a path cluster In the The delay characteristics of the feature dimensions are mapped to coordinates and placed in an n-dimensional feature space, so that each path cluster has a unique position in this space. A matrix representing the relationship between links within the cluster is constructed, namely the first cyclic dimension matrix, which describes the correlation between links within the same path cluster. , assuming it contains The link characteristics are represented by vectors as By comparing the feature vectors of the links pairwise, we construct The internal correlation matrix , where each element Indicates link and Link The correlation between them. The correlation is measured using cosine similarity: ; in, represents the dot product of two vectors, and Represents the vector norm of each. Through the calculation of cosine similarity, each value in the internal association matrix represents the similarity between the links in the path cluster. The closer the value is to 1, the more similar the link characteristics are. After obtaining the internal association matrix of each path cluster, the different path clusters in the path cluster set are mapped to construct a second circular dimension matrix that represents the association relationship between clusters, namely the external association matrix. Assume that there are path clusters, and the internal correlation matrix of each path cluster is By comparing the internal correlation matrix, we construct a The external correlation matrix , where the elements Represents a path cluster and path clusters The correlation between them is expressed as the average value of the internal correlation matrix: ; in, and Represents path clusters and path clusters The number of links in Represents a path cluster Middle Link With link The internal correlation matrix and the external correlation matrix are tensor-producted, and a double-circular encoding space is constructed to obtain a double-circular matrix. The role of the tensor product is to combine the internal and external correlation relationships in a higher-dimensional space, so that the characteristics of the path cluster and the inter-cluster characteristics are captured at the same time. Assume that the internal correlation matrix The dimension is , the external correlation matrix The dimension is , then the result of the tensor product is a vector with dimension The double-circular matrix : ; in Represents the tensor product operation, and the resulting double-circular matrix It is possible to represent the complex relationships within and between path clusters in a more comprehensive space. Perform eigenvalue decomposition to extract the main feature information of each path cluster. Eigenvalue decomposition represents the matrix as a combination of its eigenvectors and eigenvalues. Assume that the result of eigenvalue decomposition is: ; in, is a diagonal matrix with diagonal elements represents the eigenvalue of a double circulant matrix, and The column vector in represents the eigenvector of the double circulant matrix. For each path cluster, select the corresponding main eigenvector As its feature representation, these feature vectors can reflect the main characteristics of the path cluster. The main feature vector of each path cluster is normalized to eliminate the influence of the feature vector dimension. The normalization process is expressed as: ; in, represents the normalized eigenvector, Represents the norm of the feature vector. Normalization is based on the cosine distance to calculate the similarity between path clusters and define path clusters. and path clusters The similarity between them is: ; Each element of the similarity matrix represents the similarity between the path clusters. The higher the value, the closer the characteristics of the two path clusters are. Set a similarity merge threshold for the path cluster similarity matrix , the similarity is higher than the threshold The path clusters are merged to form a new path cluster set. The merging process helps to reduce the number of path clusters and simplify the management and configuration of network routing. After the merging is completed, the eigenvector of the double-circular matrix is ​​recalculated for the new path cluster in order to encode and describe it. The recalculated double-circular matrix eigenvector is used as the encoding of the routing path to obtain a path encoding sequence. The path encoding sequence is a unique identifier of the path cluster feature and is used for subsequent path search and routing configuration. In order to manage these encoding sequences, the path encoding sequence and its corresponding new path cluster are associated and stored to form a routing path encoding set. The structure of the routing path encoding set is similar to a mapping relationship, which records the detailed information of each path encoding sequence and its corresponding path cluster.

[0027] In a specific embodiment, the process of executing step 300 may specifically include the following steps: Set the feature extraction time window length, collect the number of data packets, number of data bytes and packet loss rate information of the network interface within the time window, and obtain network traffic statistics; The network traffic statistics data are divided into path clusters according to the routing path encoding set and input into the dynamic time warping algorithm to align the traffic characteristics at different time points into a unified time series to obtain the regularized traffic characteristics; The performance index of each path cluster in the regularized traffic characteristics is calculated to obtain the path performance index matrix, and the path performance index matrix is ​​organized according to the three dimensions of path cluster, performance index and time series to construct the spatiotemporal feature tensor; Decomposing the spatiotemporal feature tensor into a path cluster pattern tensor, a performance indicator pattern matrix and a time pattern vector to obtain reduced-dimensional feature data, and extracting a feature pattern sequence based on the reduced-dimensional feature data; The pattern weight attenuation coefficient is set for the feature pattern sequence to obtain the pattern weight vector, and the pattern weight vector is weighted combined with the path cluster pattern tensor and updated to the path cluster to obtain the time-varying weight coefficient.

[0028] Specifically, the length of the feature extraction time window is set. The time window is used to determine the frequency of collecting and analyzing network performance features within a specific time period. For example, if it is set to 1 minute, it means that the traffic data of the network interface is sampled every minute to obtain details reflecting the network status. In a certain time window, the number of packets is , the number of data bytes is , the packet loss rate is , this information constitutes the network traffic statistics within the time window. These data describe the traffic characteristics and network quality of the network interface at a certain moment. The collected network traffic statistics are divided according to the path clusters in the routing path encoding set, so that each path cluster has independent data input into the subsequent processing flow. In order to better compare and analyze the traffic characteristics at different time points, the dynamic time warping algorithm is used to align these traffic characteristics to a unified time series. The role of the dynamic time warping algorithm is to deal with the time differences caused by different collection time points, so that the traffic characteristics of different path clusters can be compared on the same time scale. For each path cluster, assume that its traffic characteristics at different time points are , these traffic features are aligned into a unified time series through the dynamic time warping algorithm , and obtain the regularized traffic characteristics. The performance indicators of each path cluster in the regularized traffic characteristics are calculated. The calculation of performance indicators includes multiple aspects, such as bandwidth utilization, packet loss rate, average delay, etc. Assume that for the path cluster Performance indicators include bandwidth utilization , Packet loss rate , average delay , then the performance index matrix of the path cluster is constructed at each time point, recorded as , which is of the form: ; The rows of the matrix represent different performance indicators, and the columns represent different time points. The performance indicator matrix of all path clusters is organized according to the three dimensions of path cluster, performance indicator and time series to construct a three-dimensional spatiotemporal feature tensor, denoted as , whose dimensions are ,in represents the number of path clusters, represents the number of performance indicators, Represents the number of time points. Spatiotemporal feature tensor Contains the performance characteristics of all path clusters at different time points. In order to reduce the dimension of the data and reduce the complexity of the calculation, the spatiotemporal feature tensor The decomposition process uses high-order singular value decomposition to decompose the spatiotemporal feature tensor into path cluster pattern tensors , performance indicator pattern matrix and the time pattern vector The product of . Assume that the decomposition is in the form of: ; in, Indicates that along the Tensor multiplication of patterns, is the core tensor, representing the pattern relationship between path clusters, is the performance indicator pattern matrix, which indicates the correlation between different performance indicators. is the pattern matrix of the path clusters, is a time pattern matrix, which is used to describe the characteristic patterns at different time points. By decomposing the tensor, the reduced-dimensional feature data is obtained, which can extract the most important change patterns in the network. Based on the reduced-dimensional feature data, the characteristic pattern sequence is extracted, which is used to describe the change trend of the network state over time. For example, assuming the characteristic pattern sequence is , these sequences can reflect the main change characteristics of the network at different time points. In order to optimize these characteristic patterns, the pattern weight attenuation coefficient is set based on the characteristic pattern sequence. , which is used to represent the decay of the feature mode over time. The weight decay coefficient is set to a value between 0 and 1, assuming , then the mode weight vector It is expressed as: ; The role of the weight decay coefficient is to assign different weights to features at different time points. Features at recent time points are given higher weights, while the weights of features at distant time points gradually decrease, thereby focusing on recent features. Tensor with path cluster mode Perform weighted combination operations to update the characteristics of the path cluster and obtain time-varying weight coefficients The formula for weighted combination is: ; in, Represents the weight vector Diagonalization, Indicates that a weighted operation is performed on the path cluster mode. In this way, the characteristics of the path cluster are dynamically adjusted so that the characteristics of the path cluster can be closer to the current network state, the path cluster characteristics are updated, and the time-varying weight coefficient is obtained.

[0029] In a specific embodiment, the process of executing step 400 may specifically include the following steps: The objective function of the recursive least squares framework is constructed for the spatiotemporal feature tensor and the time-varying weight coefficients, and the performance indicators of each path cluster are included in the optimization range to obtain the optimization target model; According to the optimization target model, sparse constraints are constructed for network delay, bandwidth utilization and load balancing performance indicators, and constraint thresholds for each indicator are set to obtain a set of mathematical optimization equations. The mathematical optimization equations are input into the sparse-aware proportional iterative algorithm, the objective function value is calculated for the routing configuration of the current path cluster, the initial optimization state is obtained, and the gradient is calculated based on the initial optimization state, and the sparsity constraint is processed on the gradient direction to obtain the optimization direction vector; Adaptively calculate the step size of the optimization direction vector to obtain the configuration update amount, and combine the configuration update amount with the current routing configuration to update the configuration parameters of each path cluster to obtain the routing configuration parameters; The routing configuration parameters are substituted into the spatiotemporal feature tensor to calculate the performance indicators of each path cluster, which are compared with the optimization target model to obtain the convergence judgment result. The optimal combination of routing configuration parameters is selected according to the convergence judgment result, and the optimal combination of routing configuration parameters is used as the optimal routing configuration solution.

[0030] Specifically, the objective function of the recursive least squares framework is constructed for the spatiotemporal feature tensor and the time-varying weight coefficient. Based on, combined with the time-varying weight coefficient Optimize each path cluster of the network. Assume that the dimension of the spatiotemporal feature tensor is ,in represents the number of path clusters, represents the number of performance indicators, represents the length of the time series. For each path cluster, the performance indicators include network delay, bandwidth utilization, and load balancing. Assume that The path cluster is Performance vector at each time point The recursive least squares method optimizes the performance of the path cluster by minimizing the prediction error. Therefore, when constructing the objective function, an error term is defined, which is recorded as , which is represented as a path cluster At the point in time The difference between the actual performance and the predicted performance. Assume that the predicted performance is , then the error term is expressed as: ; In order to minimize the overall error, the objective function of the recursive least squares method is defined as the sum of squared errors of all path clusters over the entire time series. It is expressed as: ; in, It is Path clusters at time points The time-varying weight coefficient of Represents the square of the error. The minimization process of the objective function updates the weights recursively to continuously approach the optimal state. Sparse constraints are constructed for the performance indicators of the network. For each performance indicator of the path cluster, including network delay, bandwidth utilization, and load balancing, corresponding constraint thresholds are set for each indicator to ensure the rationality of each indicator. The threshold is set to and , then the constraints are added to the optimization model to form a set of mathematical optimization equations. The purpose of sparse constraints is to adjust only the path clusters that significantly affect the objective function during the optimization process, thereby avoiding excessive disturbances to the entire network. The mathematical optimization equations are expressed as: ; The above mathematical optimization equations are input into the sparse-aware proportional iterative algorithm to optimize the routing configuration of the current path cluster. In the initial stage of the iterative algorithm, the objective function value is calculated to obtain the initial optimization state , which represents the performance of the network under the initial configuration. Based on the initial state, gradient calculation is performed to determine the optimization direction. Assume that the gradient of the objective function with respect to the path cluster configuration parameters is , then the optimization direction vector is expressed as In order to ensure sparsity, the gradient direction is processed with sparsity constraints, and only the direction with a greater impact on the objective function is selected for optimization to obtain the sparse optimization direction vector , expressed as: ; After the optimization direction is determined, the optimization direction vector is adaptively calculated to ensure that it can be updated with the optimal step size in each iteration, thereby ensuring convergence speed and stability. Assume that the step size is , then the line search algorithm is used to determine The optimal value of , so that the value of the objective function in the current direction is minimized. Configure the update amount It is expressed as: ; Configure the update amount With the current path cluster configuration parameters Combine to get the updated configuration parameters : ; The updated configuration parameters represent the result of adjusting the routing configuration of each path cluster, that is, the routing configuration parameters. Substitute the spatiotemporal feature tensor In order to calculate the performance index of each path cluster under the new configuration. Assume that the performance index of the path cluster under the new configuration is , then compare these performance indicators with the optimization target model to obtain the convergence judgment result. If the objective function value under the new configuration meets the convergence condition, it means that the optimal configuration has been found and the iteration is terminated; otherwise, continue to the next round of iterative update. The convergence judgment condition is measured by the change in the objective function value. Set a convergence threshold ,if , it is considered to have converged and the current routing configuration parameter combination is selected As the optimal routing configuration solution; otherwise, continue to iterate until the convergence condition is met. Through the optimization method combining recursive least squares method with sparse-aware proportional iterative algorithm, the routing configuration of the network path cluster is effectively optimized to ensure that the performance indicators of each path cluster are kept within a reasonable range and the overall performance of the network reaches the optimal state.

[0031] In a specific embodiment, the process of executing the above-mentioned method for adaptively adjusting network routing may specifically include the following steps: Extract and classify the quality of service features of network service traffic to obtain service flow classification data, set the quality of service priority threshold based on the service flow classification data, allocate path clusters to service flows above the quality of service priority threshold, and obtain service flow priority tags; Match the service flow priority tag with the optimal routing configuration solution, assign an accurate routing path to the high-priority service flow of each path cluster, obtain a directional forwarding routing sub-table, and perform multi-path distribution processing on the remaining low-priority service flows in the path cluster based on the flooding strategy, set the load balancing weight, and obtain a flooding forwarding routing sub-table; Integrate the routing entries of the directional forwarding routing sub-table and the flooding forwarding routing sub-table according to the path cluster to obtain a hybrid routing policy table, and allocate routing paths to the service flow classification data according to the hybrid routing policy table to obtain an initial routing table entry; The service quality index of each path cluster in the initial routing table entry is detected. When the index is lower than the preset threshold, the routing policy adjustment mechanism is triggered to obtain the update policy instruction, and the initial routing table entry is dynamically updated according to the update policy instruction to obtain the target routing table entry; Perform multi-link bandwidth measurement and delay difference correction on the target routing table entry, dynamically adjust the data forwarding ratio, and feed back the effect data to the detection data packet of the network interface for transmission processing.

[0032] Specifically, data is collected for each service flow in the network in order to extract its service quality characteristics. Service quality characteristics (QoS characteristics) include latency, bandwidth requirements, packet loss rate, jitter, etc. Each service flow is represented by a vector ,in Indicates business flow In the By extracting the feature vectors of all service flows, we can get the feature matrix , which is of the form: ; in, Indicates the number of business flows, Indicates the number of QoS features. By analyzing and classifying these features, the service flows are divided into different categories to form service flow classification data. For example, a clustering-based algorithm (such as K-means) is used to classify these service flows and group service flows with similar features into one category. Based on the classification data of service flows, a service quality priority threshold is set. Set a service quality priority threshold , for each business flow If some key QoS indicators (such as bandwidth or delay) in its feature vector are higher than the threshold , it is considered that the service flow needs a higher priority to ensure its service quality. For high-priority service flows, they are assigned to specific path clusters to provide more stable and reliable transmission paths for these service flows. For the priority of each service flow, a marker is used To indicate that Indicates high priority. Indicates low priority. Match the priority tag of the service flow with the optimal routing configuration scheme to allocate accurate routing paths to the high-priority service flows of each path cluster. Assume that the optimal routing configuration scheme contains the path cluster , for each path cluster , assign its high priority service flow to the optimal link, and obtain the directional forwarding routing sub-table Each entry in the directional forwarding routing subtable represents the mapping relationship between the service flow and the precise routing path, ensuring that the high-priority service flow can obtain the best transmission path to meet its service quality requirements such as latency and bandwidth. For the remaining low-priority service flows in the path cluster, multi-path distribution processing based on the flooding strategy is adopted. The flooding strategy achieves load balancing and improves network reliability by distributing traffic to multiple different paths. For each low-priority service flow, set the load balancing weight on different paths. Assuming that the path cluster Contains the path , and their load balancing weights are , then the distribution ratio of data packets of low-priority service flows on these paths is expressed as: ; in, Indicates the path The distribution ratio, is the sum of the weights of all paths. In this way, we get the flood forwarding routing subtable , where each entry represents the distribution ratio of low-priority service flows on different paths. The directional forwarding routing subtable and the flooding forwarding routing subtable are integrated according to the path cluster to obtain the hybrid routing strategy table Comprehensively consider the precise forwarding requirements of high-priority business flows and the multi-path distribution requirements of low-priority business flows, so as to achieve efficient resource utilization and service quality assurance in the network. According to the hybrid routing strategy table, the classified data of the business flow is specifically allocated to the routing path to obtain the initial routing table entry During network operation, in order to ensure the stability of network performance and the guarantee of service quality, the service quality indicators of each path cluster in the initial routing table are tested in real time. , regularly monitor its bandwidth utilization, delay, packet loss rate and other QoS indicators. Assume that these indicators are , , When the service quality index of a path cluster is detected to be lower than the preset threshold , , When the routing policy adjustment mechanism is triggered, an instruction to update the policy is generated. The purpose of the update policy instruction is to dynamically adjust the current routing configuration to restore the service quality indicator to the expected range. Perform dynamic updates to obtain new target routing table entries During the update process, the routing paths in some path clusters are reselected to meet the new service quality requirements, thereby ensuring that the business flows in the network can continue to receive stable services. Multi-link bandwidth measurement and delay difference correction are performed on the target routing table entries to optimize the data forwarding ratio. The purpose of multi-link bandwidth measurement is to obtain the real-time bandwidth of each link. The bandwidth is , by measuring the bandwidth of all links, we can get the link bandwidth matrix The delay difference correction is to balance the delay between different links. The delay is , then calculate the average delay of all paths , and correct the delay of each path. The corrected delay is: ; Through bandwidth measurement and delay correction, the data forwarding ratio in the target routing table is dynamically adjusted to make the network load more balanced and the transmission more stable. The adjusted data forwarding ratio is applied to the target routing table to achieve optimal traffic distribution. After completing the update and adjustment of the routing table, the effect data is fed back to the detection module of the network interface, and the detection module will update the parameter settings of the detection data packet based on the feedback data. For example, the feedback data includes the real-time bandwidth of the link, delay changes and other information, which are used for the next round of detection to improve the detection accuracy and adaptive adjustment capabilities of the network.

[0033] In a specific embodiment, the execution step performs multi-link bandwidth measurement and delay difference correction on the target routing table entry, dynamically adjusts the data forwarding ratio, and feeds back the effect data to the detection data packet of the network interface for transmission processing, which may specifically include the following steps: Initiate bidirectional bandwidth measurement for the network links of each path cluster in the target routing table entry to obtain link performance measurement results, and perform feature matching on the network links of each path cluster based on the link performance measurement results to obtain link matching features; The link matching features are input into the delay difference correction algorithm based on time series, and the delay data of the links in each path cluster are dynamically corrected to obtain the link delay compensation value. The network links of each path cluster in the target routing table entry are then delayed and corrected according to the link delay compensation value to obtain the corrected routing table entry. Calculate the data forwarding ratio for each path cluster in the correction routing table entry to obtain a forwarding ratio allocation plan, and deploy the forwarding ratio allocation plan to the target routing table entry of each path cluster, perform hierarchical data traffic forwarding control, and obtain routing execution effect data; The routing execution effect data is evaluated to obtain performance feedback data, which is divided into path clusters and written into the detection data packet of the network interface as new detection parameters, and the next round of bandwidth measurement is started according to the preset period.

[0034] Specifically, bidirectional bandwidth measurement is started for the network links of each path cluster in the target routing table entry to obtain the current performance status of the network links, especially the bandwidth in the uplink and downlink directions. Assume that there are multiple path clusters in the network , each path cluster contains multiple links. For each link, the time window length of bandwidth measurement is set to , by using the time window Send and receive test data packets within the time, record their transmission time, and obtain the link bandwidth measurement result. Assume that the path cluster The link in , then the bandwidth measurement result of each link is expressed as ,in Indicates link The average bandwidth in the time window. Through bidirectional bandwidth measurement, the link bandwidth characteristics in each path cluster are obtained to form a link performance measurement result matrix. The link performance measurement results are matched with features to select the best path according to the link characteristics. For each path cluster , match the link performance measurement results with the historical performance characteristics, and use the feature matching algorithm to determine whether the current link performance status meets expectations. The result of feature matching is represented by the link matching feature vector express, ,in Indicates link The matching characteristics of the links are calculated, including the matching degree of bandwidth utilization, packet loss rate, delay and other indicators. The link matching characteristics are input into the delay difference correction algorithm based on time series to dynamically adjust the delay characteristics of each link to achieve delay balance between links. The link delay is The goal of delay difference correction is to adjust the delay of each link to a close level to reduce the delay difference between links. By performing time series analysis on historical delay data, the delay compensation value of each link is obtained. ,in Indicates link The delay compensation value of the corrected link delay It is expressed as: ; in, Represents the corrected delay value. Through dynamic correction, the delay of each link is balanced to obtain the corrected routing table entry. The data forwarding ratio is calculated for the corrected routing table entry to achieve efficient data traffic scheduling. In each path cluster, the data forwarding ratio is determined based on the bandwidth and delay characteristics of the link. Assume that the path cluster The link bandwidth in is , the delay is , then the forwarding weight of the link Determined by weighting the inverse of bandwidth and delay: ; in, Indicates link The forwarding weight, Indicates the link bandwidth, Represents the corrected delay. Through normalization, the forwarding ratio of each link is obtained: ; in, Indicates link The data forwarding ratio is allocated to the forwarding ratio. Deploy to a path cluster In the target routing table entry, the hierarchical forwarding control of data traffic between each link is realized. Through the above operations, after executing the hierarchical data traffic forwarding control, the routing execution effect data is obtained, including the bandwidth utilization, delay, packet loss rate, etc. of each link. Assume that the effect data is ,in Indicates link The performance of the current configuration is evaluated to determine whether the current routing configuration has achieved the expected performance target. The performance evaluation process is based on a set of preset performance indicator thresholds. Assume that the bandwidth utilization threshold is The delay threshold is , the packet loss rate threshold is For each link , if its performance indicators meet the following conditions: ; If the link performance meets the standard, the link is considered to have met the standard. Otherwise, the configuration needs to be further adjusted. The performance feedback data obtained by evaluation is divided into path clusters, and these feedback data are written into the detection data packet of the network interface as new detection parameters for use in the next round of measurement. For example, the feedback data includes information such as the real-time bandwidth change and delay change of the link. This information will be used to dynamically adjust the measurement parameters in the next round of bandwidth measurement, thereby improving the measurement accuracy and adaptability to network changes. The next round of bandwidth measurement is started according to the preset period, and a new cycle of bandwidth measurement, feature matching, delay correction, forwarding ratio calculation and performance evaluation is entered. Periodic measurement and optimization can continuously adjust the configuration parameters of the network to adapt to the changing network conditions, thereby ensuring that the network is always in the optimal operating state.

[0035] The above describes the adaptive adjustment method of the network routing in the embodiment of the present application. The following describes the adaptive adjustment system 10 of the network routing in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of the adaptive adjustment system 10 of network routing includes: The transmission module 11 is used to perform detection data packet transmission and hierarchical clustering on the network interface to obtain a path cluster set; The encoding module 12 is used to perform double-circular matrix encoding on the path cluster set to obtain a routing path encoding set; A construction module 13 is used to construct a spatiotemporal feature tensor based on the routing path encoding set, and perform tensor decomposition and feature pattern extraction to obtain a time-varying weight coefficient; The calculation module 14 is used to perform sparse-aware proportional iterative optimization calculation based on the spatiotemporal feature tensor and the time-varying weight coefficient to obtain the optimal routing configuration solution.

[0036] Through the collaborative cooperation of the above components, by introducing the concept of path clusters and adopting a hierarchical clustering method, network links are dynamically grouped and managed, effectively reducing the complexity of routing calculations and improving the scalability of routing algorithms; by adopting double-circular matrix coding technology, accurate modeling of the relationship between and within path clusters is achieved, enhancing the adaptability of routing schemes to changes in network topology; by combining spatiotemporal feature tensor decomposition and time-varying weight coefficients, multi-dimensional feature extraction of network status is achieved, improving the accuracy of routing decisions; by using a sparse-aware proportional iterative optimization algorithm, while ensuring convergence, joint optimization of multiple performance indicators is achieved, improving the comprehensive performance of the routing scheme; by introducing a hybrid routing strategy based on quality of service classification, differentiated processing of different business flows is achieved, improving the efficiency of network resource utilization; by using a multi-link bandwidth measurement and delay difference correction mechanism, real-time monitoring and dynamic adjustment of routing performance are achieved, enhancing the robustness of the routing scheme.

[0037] See also Figure 3 , Figure 3 This is a schematic block diagram of the structure of an electronic device 300 provided in an embodiment of the present application. The electronic device 300 includes a processor 301 and a memory 302. The processor 301 and the memory 302 are connected via a system bus 303, wherein the memory 302 may include a non-volatile storage medium and an internal memory.

[0038] The non-volatile storage medium can store a computer program. The computer program includes program instructions, and when the program instructions are executed by the processor 301, the processor 301 can execute any of the above-mentioned adaptive adjustment methods for network routing.

[0039] The processor 301 is used to provide computing and control capabilities to support the operation of the entire electronic device 300 .

[0040] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor 301, the processor 301 can execute any of the above-mentioned methods for adaptively adjusting network routing.

[0041] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a partial structure related to the present application scheme, and does not constitute a limitation on the electronic device 300 involved in the present application scheme. The specific electronic device 300 may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0042] It should be understood that the processor 301 may be a central processing unit (CPU), and the processor 301 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0043] It should be noted that those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the electronic device 300 described above can refer to the corresponding process of the aforementioned adaptive adjustment method of network routing, which will not be repeated here.

[0044] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by one or more processors, the one or more processors implement the adaptive adjustment method of network routing provided in the embodiment of the present application.

[0045] The computer-readable storage medium may be an internal storage unit of the electronic device 300 in the aforementioned embodiment, such as a hard disk or memory of the electronic device 300. The computer-readable storage medium may also be an external storage device of the electronic device 300, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped with the electronic device 300.

[0046] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0047] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or 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 an electronic device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0048] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for adaptively adjusting network routing, characterized in that: The method comprises: Probe the data packet transmission and hierarchical clustering of the network interface to obtain a set of path clusters; Performing double-circular matrix coding on the path cluster set to obtain a routing path coding set; Constructing a spatiotemporal feature tensor based on the routing path encoding set, and performing tensor decomposition and feature pattern extraction to obtain a time-varying weight coefficient; Based on the spatiotemporal feature tensor and the time-varying weight coefficient, a sparse-aware proportional iterative optimization calculation is performed to obtain an optimal routing configuration solution.

2. The method for adaptively adjusting network routing according to claim 1, characterized in that: The detecting data packet transmission and hierarchical clustering of the network interface to obtain a path cluster set includes: Deploy a network detection module on the network interface, set sending timestamp information and sequence number information for the network detection module, obtain a detection data packet, and periodically send the detection data packet, record the receiving timestamp information, and calculate based on the sending timestamp information and the receiving timestamp information to obtain network link delay information; The network link delay information is stored by rows representing network interfaces and by columns representing time points to obtain a first delay feature matrix, and data smoothing is performed on the first delay feature matrix to obtain a second delay feature matrix; Calculating the delay feature similarity between network links based on the second delay feature matrix, and constructing a similarity distance matrix according to the delay feature similarity; Performing hierarchical clustering calculation based on the similarity distance matrix and executing network link classification processing to obtain initial path clusters, assigning identifiers to the initial path clusters, creating a member link list for each initial path cluster, calculating the average delay characteristics of the member link list, and obtaining a cluster delay feature vector; The identifier of the initial path cluster, the member link list and the cluster delay feature vector are combined and stored to obtain a path cluster set.

3. The method for adaptively adjusting network routing according to claim 2, characterized in that: The step of performing double-circular matrix coding on the path cluster set to obtain a routing path coding set includes: Coordinate mapping is performed on the cluster delay feature vector of each path cluster in the path cluster set, a first circular dimension matrix representing the link relationship within the cluster is constructed, and an internal correlation matrix is ​​obtained; and relationship mapping is performed on the internal correlation matrices of different path clusters in the path cluster set, a second circular dimension matrix representing the correlation relationship between clusters is constructed, and an external correlation matrix is ​​obtained; Performing a tensor product operation and a double-circular coding space construction on the internal correlation matrix and the external correlation matrix to obtain a double-circular matrix, and performing an eigenvalue decomposition operation on the double-circular matrix to obtain a main eigenvector of each path cluster; Normalizing and calculating the cosine distance of the main feature vector of each path cluster to obtain a path cluster similarity matrix, setting a similarity merging threshold for the path cluster similarity matrix, and merging path clusters with similarities higher than the similarity merging threshold into a new path cluster; The double-circular matrix feature vector is recalculated for the new path cluster, and the double-circular matrix feature vector is used as a routing path code to obtain a path code sequence, and the path code sequence and its corresponding new path cluster are associated and stored to obtain a routing path code set.

4. The method for adaptively adjusting network routing according to claim 3, characterized in that: The step of constructing a spatiotemporal feature tensor based on the routing path encoding set, performing tensor decomposition and feature pattern extraction to obtain a time-varying weight coefficient includes: Setting the feature extraction time window length, collecting the number of data packets, the number of data bytes and the packet loss rate information of the network interface within the time window, and obtaining network traffic statistics; Input the network traffic statistics data into a dynamic time warping algorithm according to the path cluster division of the routing path coding set, align the traffic characteristics at different time points into a unified time series, and obtain the regularized traffic characteristics; Calculating the performance index of each path cluster in the regularized traffic feature to obtain a path performance index matrix, and organizing the path performance index matrix according to three dimensions of path cluster, performance index and time series to construct a spatiotemporal feature tensor; Decomposing the spatiotemporal feature tensor into a path cluster pattern tensor, a performance indicator pattern matrix and a time pattern vector to obtain reduced-dimensional feature data, and extracting a feature pattern sequence based on the reduced-dimensional feature data; A pattern weight attenuation coefficient is set for the characteristic pattern sequence to obtain a pattern weight vector, and a weighted combination operation is performed on the pattern weight vector and the path cluster pattern tensor, and the path cluster is updated to obtain a time-varying weight coefficient.

5. The method for adaptively adjusting network routing according to claim 4, characterized in that: The sparse-aware proportional iterative optimization calculation is performed based on the spatiotemporal feature tensor and the time-varying weight coefficient to obtain the optimal routing configuration solution, including: Constructing an objective function of a recursive least squares framework for the spatiotemporal feature tensor and the time-varying weight coefficient, incorporating the performance index of each path cluster into the optimization range, and obtaining an optimization target model; According to the optimization target model, sparse constraints are constructed for network delay, bandwidth utilization and load balancing performance indicators, constraint thresholds of each indicator are set, and a mathematical optimization equation group is obtained; Input the mathematical optimization equation group into a sparse-aware proportional iterative algorithm, calculate the objective function value for the routing configuration of the current path cluster, obtain an initial optimization state, perform gradient calculation based on the initial optimization state, perform sparsity constraint processing on the gradient direction, and obtain an optimization direction vector; Performing adaptive step calculation on the optimization direction vector to obtain a configuration update amount, and combining the configuration update amount with the current routing configuration to update the configuration parameters of each path cluster to obtain routing configuration parameters; Substituting the routing configuration parameters into the spatiotemporal feature tensor to calculate the performance index of each path cluster, comparing with the optimization target model, obtaining a convergence judgment result, and selecting the optimal routing configuration parameter combination according to the convergence judgment result, and taking the optimal routing configuration parameter combination as the optimal routing configuration solution.

6. The method for adaptively adjusting network routing according to claim 5, characterized in that: The method for adaptively adjusting the network routing also includes: Extracting and classifying service quality features of network service traffic to obtain service flow classification data, setting a service quality priority threshold based on the service flow classification data, allocating a path cluster to service flows above the service quality priority threshold, and obtaining a service flow priority tag; Match the service flow priority tag with the optimal routing configuration scheme, assign an accurate routing path to the high-priority service flow of each path cluster, obtain a directional forwarding routing sub-table, and perform multi-path distribution processing on the remaining low-priority service flows in the path cluster based on a flooding strategy, set a load balancing weight, and obtain a flooding forwarding routing sub-table; Integrate the routing entries of the directional forwarding routing subtable and the flooding forwarding routing subtable according to the path cluster to obtain a hybrid routing policy table, and allocate routing paths to the service flow classification data according to the hybrid routing policy table to obtain an initial routing table entry; The service quality indicator of each path cluster in the initial routing table entry is detected, and when the indicator is lower than a preset threshold, a routing policy adjustment mechanism is triggered to obtain an update policy instruction, and the initial routing table entry is dynamically updated according to the update policy instruction to obtain a target routing table entry; The target routing table entry is subjected to multi-link bandwidth measurement and delay difference correction, the data forwarding ratio is dynamically adjusted, and the effect data is fed back to the detection data packet of the network interface for transmission processing.

7. The method for adaptively adjusting network routing according to claim 6, characterized in that: The method of measuring the multi-link bandwidth and correcting the delay difference of the target routing table entry, dynamically adjusting the data forwarding ratio, and feeding back the effect data to the detection data packet of the network interface for transmission processing includes: Initiate bidirectional bandwidth measurement on the network links of each path cluster in the target routing table entry to obtain link performance measurement results, and perform feature matching on the network links of each path cluster based on the link performance measurement results to obtain link matching features; The link matching feature is input into a delay difference correction algorithm based on a time series, the delay data of the links in each path cluster are dynamically corrected to obtain a link delay compensation value, and the network link of each path cluster in the target routing table entry is delayed and corrected according to the link delay compensation value to obtain a corrected routing table entry; Calculating the data forwarding ratio for each path cluster in the corrected routing table entry to obtain a forwarding ratio allocation scheme, and deploying the forwarding ratio allocation scheme to the target routing table entry of each path cluster to perform hierarchical data traffic forwarding control to obtain routing execution effect data; The routing execution effect data is subjected to performance evaluation to obtain performance feedback data, and the performance feedback data is divided into path clusters and written into the detection data packet of the network interface as new detection parameters, and the next round of bandwidth measurement is started according to a preset period.

8. A network routing adaptive adjustment system, characterized in that: A method for adaptively adjusting network routing according to any one of claims 1 to 7, the system comprising: A transmission module, used for performing detection data packet transmission and hierarchical clustering on the network interface to obtain a set of path clusters; An encoding module, used for performing double-circular matrix encoding on the path cluster set to obtain a routing path encoding set; A construction module, used to construct a spatiotemporal feature tensor based on the routing path encoding set, and perform tensor decomposition and feature pattern extraction to obtain a time-varying weight coefficient; The computing module is used to perform sparse-aware proportional iterative optimization calculation based on the spatiotemporal feature tensor and the time-varying weight coefficient to obtain an optimal routing configuration solution.

9. An electronic device, characterized in that: The electronic device comprises: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory to enable the electronic device to execute the adaptive adjustment method of network routing as described in any one of claims 1-7.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the method for adaptively adjusting network routing according to any one of claims 1 to 7 is implemented.

Images