A method for dynamically constructing a virtual network topology for SDN

Abstract

The application discloses a kind of virtual network topology dynamic construction methods for SDN, comprising: step 1, by embedded probe real-time collection multi-modal network data in teaching scene, constructs multi-modal topology perception matrix;Step 2, multi-modal topology perception matrix is input into pre-trained deep decision tree model, generates dynamic topology descriptor;Step 3, dynamic topology descriptor is converted into compatible multi-protocol standardized configuration instruction set;Step 4, according to teaching process stage feature dynamic adjustment virtual node density parameter, realize from basic single-domain topology to complex cross-domain topology gradual reconstruction;Step 5, when detecting entropy value mutation, trigger topology self-repairing mechanism, synchronously update abnormal operation mode feature library in knowledge graph.The application provides a kind of efficient, intelligent, flexible virtual network topology dynamic construction method for teaching scene by multi-link collaborative work, effectively improves teaching effect and network management efficiency.

Description

Technical Field

[0001] This invention relates to the field of SDN virtual network topology construction technology, and in particular to a method for dynamically constructing virtual network topologies for SDN. Background Technology

[0002] With the development of the internet, virtualization technology has become a research hotspot. The current internet suffers from problems such as "rigidity," significant waste of hardware resources, maintenance difficulties, and complex network equipment configuration. The concept of network virtualization has emerged to address these issues. It uses abstraction, allocation, and isolation mechanisms to build diverse virtual networks that coexist but are isolated from each other by sharing underlying physical resources. This enables the rational allocation and management of physical resources and allows for the construction of logical networks on top of existing network architectures to meet specific business needs, such as Virtual Local Area Networks (VLANs) and Virtual Private Networks (VPNs).

[0003] Software-defined networking (SDN), as a novel network architecture, separates the network control plane from the data plane. It employs a centralized controller and standard interfaces to manage network devices, constructing an open and programmable new network structure that facilitates new network architectures and service innovation. Furthermore, SDN has inherent advantages in network virtualization, simplifying network design and operation, allowing network administrators to centrally control a large number of network devices without requiring manual configuration of each device individually.

[0004] In terms of virtual network topology construction, network simulation platforms aim to provide users with a near-realistic virtual network environment for conducting network security experiments and improving security skills. However, existing technologies are insufficient in dynamically constructing and modifying virtual network topologies, failing to adequately meet users' needs for modifying and building upon existing topologies. For example, in teaching scenarios, network topologies need to be dynamically adjusted according to different teaching content and experimental requirements to simulate various complex network environments. However, existing network topology construction methods struggle to adapt to the dynamic and complex changes in teaching scenarios. Summary of the Invention

[0005] In view of this, the present invention proposes a method for dynamically constructing virtual network topology for SDN, in order to solve the problem that the dynamic construction of network topology in teaching scenarios is difficult to adapt to the dynamic and complex changes in teaching scenarios in the existing technology.

[0006] The specific technical solution of this invention is as follows:

[0007] A method for dynamically constructing virtual network topology for SDN, comprising:

[0008] Step 1: Collect multimodal network data in the teaching scenario in real time using embedded probes, and construct a multimodal topology perception matrix. The multimodal network data includes network cognitive level data stream, device command operation feature vector, and protocol interaction state matrix.

[0009] Step 2: Based on the adaptive rule engine, the dynamic reconstruction module inputs the multimodal topology perception matrix constructed in Step 1 into the pre-trained deep decision tree model to generate a dynamic topology descriptor including node correlation parameters, link time-varying weight matrix and protocol stack configuration vector.

[0010] Step 3: Through the protocol abstraction middleware of the heterogeneous controller interface adaptation layer, the dynamic topology descriptor generated in Step 2 is converted into a standardized configuration instruction set compatible with OpenFlow, Restconf and Netconf protocols;

[0011] Step 4: Based on the real-time topology calculation engine, an incremental construction algorithm with time window constraints is adopted to dynamically adjust the virtual node density parameters according to the characteristics of the teaching process stage, so as to realize the gradual reconstruction from basic single-domain topology to complex cross-domain topology.

[0012] Step 5: Construct a process visualization feedback loop. The network state complexity index is calculated in real time through the topology entropy monitoring module. When a sudden change in entropy is detected, the topology self-healing mechanism is triggered, and the abnormal operation mode feature library in the knowledge graph is updated synchronously.

[0013] Furthermore, in step 1, the network cognitive level data stream is generated by comprehensively quantifying the students' operation steps, task completion time, accuracy rate, and operation fluency indicators in the network experiment; the device instruction operation feature vector is generated by parsing the instruction types, frequencies, and abnormal operation markers received by the router and switch; and the protocol interaction state matrix is ​​generated by dynamically parsing the interaction messages between the control plane and the data plane, including protocol type, interaction frequency, and abnormal protocol behavior markers.

[0014] Furthermore, in step 1, let the network cognitive level data stream be D. c The device instruction operation feature vector is D0, and the protocol interaction state matrix is ​​D. p ;D c ∈R m×n Where m represents the number of students, and n represents the number of indicators measuring online cognitive level; D0∈R k k represents the number of device instruction operation features; D p ∈R l×l , l represents the number of network protocols, and matrix element D p (i, j) represents the interaction state between the i-th protocol and the j-th protocol; the multimodal topology sensing matrix M is constructed using the following formula:

[0015]

[0016] Among them, 0 a×bIt is a zero matrix with a rows and b columns. vec(·) is a vectorization operation that stacks the matrix into a vector column by column.

[0017] Furthermore, the dynamic topology descriptor generated by the deep decision tree model in step 2 includes: node correlation parameters, which are dynamically updated based on the communication dependencies between nodes to optimize node connection planning; a link time-varying weight matrix, which dynamically adjusts the weight values ​​according to the link congestion or load capacity during the teaching process to guide traffic scheduling; and a protocol stack configuration vector, which records the device protocol parameters to ensure the compatibility and communication continuity of the TCP / IP protocol stack.

[0018] Further, in step 2, the multimodal topology perception matrix M is input into the pre-trained deep decision tree model f to generate a dynamic topology descriptor T = [R0, W(t), C], where the node correlation parameter... This is the part used to generate node association parameters in a deep decision tree model, R0∈R s×s s represents the number of network nodes, R0(i,j) represents the correlation between node i and node j; the link time-varying weight matrix W(t) = f W (M, t), f W This is the part of the deep decision tree model used to generate the time-varying weight matrix of the links, where t represents time, and W(t)∈R. e×e e represents the number of network links, W(t)(i,j) represents the weight between link i and link j at time t; the protocol stack configuration vector C = f C (M), f C This is the part of the deep decision tree model used to generate protocol stack configuration vectors, C∈R q q represents the number of protocol stack configuration parameters.

[0019] Furthermore, in step 3, the protocol abstraction middleware implements a unified semantic expression for multi-protocol interfaces using a layered mapping mechanism, including: the semantic abstraction layer converts dynamic topology descriptors into protocol-independent intermediate representations; and the protocol adaptation layer maps the intermediate representations to OpenFlow flow table entries, Restconf configuration templates, or Netconf XML directives based on the target controller type.

[0020] Furthermore, the incremental construction algorithm in step 4 includes: a time window segmentation module, which divides the teaching process into basic, transitional, and advanced stage windows; a virtual node density parameter adjustment module, which sets a low density value for single-domain topology or a high density value for cross-domain topology based on stage characteristics; and a progressive reconstruction module, which updates newly added or deleted virtual nodes and links only within the current time window.

[0021] Furthermore, in step 4, the time window is set to [t1, t2]. Within this time window, the incremental construction algorithm only processes network topology information related to this time window. The virtual node density parameter ρ(t) is dynamically adjusted according to different teaching stages as follows:

[0022]

[0023] Where, ρ base It serves as the benchmark for calculating the virtual node density across all teaching stages, reflecting the topological complexity in the early stages of teaching; ΔS represents the teaching progress increment; ΔC represents the complexity of concurrent operations; and α and β are adaptive adjustment coefficients.

[0024] Furthermore, in step 5, the topology self-healing mechanism includes two modes: link redundancy rerouting and protocol stack dynamic rollback. In the link redundancy rerouting mode, when a link interruption is detected, it automatically switches to a backup link. In the protocol stack dynamic rollback mode, when the protocol configuration is abnormal, it restores the protocol parameters to the stable version.

[0025] Furthermore, in step 5, the topological entropy value E is calculated using the following formula:

[0026]

[0027] Where N is the number of network states, p i It is the probability of the i-th network state occurring; when a sudden change in topological entropy value |E(t)-E(t-Δt)|>∈ is detected, the topological self-healing mechanism is triggered, where ∈ is a preset threshold.

[0028] The beneficial effects of this invention are as follows:

[0029] 1. By leveraging embedded probes to collect multimodal network data in real time from teaching scenarios, a multimodal topology awareness matrix is ​​constructed to capture dynamic information promptly and accurately reflect the actual network operation. Integrating three types of data—network cognitive level data stream, device command operation feature vectors, and protocol interaction state matrix—provides rich data support for network topology construction, performance analysis, and fault diagnosis.

[0030] 2. The dynamic reconstruction module based on the adaptive rule engine inputs the multimodal topology awareness matrix into the pre-trained deep decision tree model to generate a dynamic topology descriptor. This descriptor includes node correlation parameters, link time-varying weight matrices, and protocol stack configuration vectors. The adaptive rule engine automatically adjusts the decision-making process to adapt the topology descriptor to new network conditions, optimize the network topology, and meet different teaching needs.

[0031] 3. The protocol abstraction middleware of the heterogeneous controller interface adaptation layer converts dynamic topology descriptors into a standardized configuration instruction set compatible with multiple protocol interfaces, enabling seamless connection and collaborative work between different protocols.

[0032] 4. The real-time topology calculation engine adopts an incremental construction algorithm with time window constraints. It dynamically adjusts the virtual node density parameters according to the characteristics of the teaching process stage, realizing the progressive reconstruction from basic single-domain topology to complex cross-domain topology. This improves topology construction efficiency, reduces computing resource consumption, fits the actual teaching, and makes the teaching process more scientific, reasonable, and efficient.

[0033] 5. A process visualization feedback loop is constructed. The network state complexity index is calculated in real time through the topology entropy monitoring module. When the entropy value changes abruptly, a network anomaly is identified, and the topology self-healing mechanism is automatically activated. It has dual modes of link redundancy rerouting and dynamic protocol stack rollback, which can quickly repair the network. At the same time, the abnormal operation mode is updated to the knowledge graph feature base to provide a reference for subsequent fault diagnosis and self-healing, improve network reliability and stability, and ensure the smooth progress of teaching. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1 This is a flowchart illustrating the SDN-oriented method for dynamically constructing virtual network topology according to the present invention.

[0036] Figure 2 This is a schematic diagram of the structure of the SDN-oriented dynamic virtual network topology construction method of the present invention. Detailed Implementation

[0037] To make the technical problems to be solved, the technical solutions, and the beneficial effects of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the present invention.

[0038] This invention proposes a method for dynamically constructing virtual network topologies for SDN, such as... Figure 1-2 As shown, the method includes:

[0039] Step 1: Real-time acquisition of multimodal network data in the teaching scenario using embedded probes to construct a multimodal topology awareness matrix. Utilizing embedded probes widely distributed across key network nodes, three types of multimodal network data are acquired in real-time within the teaching scenario: network cognitive level data stream, device command operation feature vectors, and protocol interaction state matrix.

[0040] The teaching environment is constantly changing, with students' actions and network status evolving due to factors such as the teaching process, learning needs, and external interference. Real-time acquisition of multimodal data can capture these changes promptly, enabling the constructed topology to dynamically adapt to changes in the teaching environment and better reflect the actual network operation. For example, when students perform new operations in network experiments or encounter network failures, the topology-aware matrix can quickly update the data, providing network administrators with the latest network status information so that appropriate measures can be taken in a timely manner.

[0041] The specific implementation of embedded probes can be flexibly chosen from hardware devices or software modules based on the network architecture and monitoring needs of the teaching scenario. At the hardware level, embedded probes can be designed as miniaturized network monitoring devices, integrating sensor modules and embedded processors. They connect to key network nodes via physical interfaces (such as Ethernet interfaces, USB interfaces, or wireless communication modules), for example, deployed on the management ports of routers and switches, or the serial interfaces of experimental terminals. These hardware probes possess real-time data capture capabilities, accurately acquiring low-level signals from the characteristic vectors of device command operations. For example, by listening to console command streams or parsing device firmware logs, parameters such as command type, execution frequency, and response time can be extracted. Their advantage lies in their direct access to hardware-level operations, enabling them to capture the physical details of student device operations in teaching experiments, such as the input timing of configuration commands and device restart cycles, providing a high-precision data source for device status analysis.

[0042] The implementation of software modules focuses on protocol parsing and behavioral analysis, typically embedded in network operating systems, experimental platform virtual machines, or teaching management software in the form of lightweight agents or virtual probes. These probes dynamically construct protocol interaction state matrices by calling system APIs or intercepting network stack packets. For example, in a virtualized teaching environment, software probes can be deployed at the Hypervisor layer to parse the protocol header information of inter-virtual machine communication in real time and generate a dynamic view of protocol interactions by combining traffic statistics. Simultaneously, software probes can integrate the collection of network cognitive level data streams, recording student interface interaction events through integration with online experimental platforms, and transforming behavioral data into cognitive ability assessment indicators using knowledge graph mapping technology.

[0043] The data stream on students' network cognition reflects their understanding and use of the internet. It is an important indicator for measuring students' absorption of network knowledge and their practical skills. By analyzing data from various aspects such as students' operational steps in network experiments, their answers to questions, their grasp of network concepts, and their proficiency in operating network devices, their network cognition level can be quantified. For example, their network cognition ability can be comprehensively evaluated based on indicators such as the time taken, accuracy, and fluency of their operation in completing specific tasks in network experiments.

[0044] Device command operation feature vectors record various commands and their characteristics received by network devices (such as routers and switches). By analyzing these feature vectors, we can clearly understand the device's operating mode, such as whether the device is in normal operation or has undergone special configuration; we can grasp the device's usage frequency and determine whether the device is overloaded or idle; and we can promptly detect any abnormal operations, such as unauthorized command input or frequent erroneous commands. This helps to analyze the device's operating status and user operating habits, providing precise guidance for device maintenance and management.

[0045] The protocol interaction state matrix describes the interactions between different network protocols. By dynamically parsing the interaction messages generated between the control plane and the data plane, it can intuitively display the interaction flow and state of protocols in the network. Specifically, the control plane is responsible for policy decisions (such as the OpenFlow controller issuing flow table rules), while the data plane is responsible for executing forwarding actions (such as switches processing data packets). The protocol messages continuously exchanged between the two (such as LLDP discovery messages, flow table update requests / responses, etc.) carry key information about the network's operational status. By monitoring the protocol interaction state in real time, abnormal protocol behaviors in the network, such as protocol conflicts and illegal protocol interactions, can be detected in a timely manner. For example, when data transmission chaos or incorrect protocol message format is detected between different protocols, the problem can be quickly located, ensuring the stable operation of the network and avoiding network failures and data loss caused by protocol issues.

[0046] The collected network cognitive level data stream, device command operation feature vectors, and protocol interaction state matrix are deeply integrated to construct a multimodal topology perception matrix. This matrix comprehensively reflects the overall state of the network in the teaching scenario, including user behavior, device operation status, and protocol interaction state, providing rich data support for subsequent network topology construction, performance analysis, and fault diagnosis.

[0047] The following example illustrates the construction process of a multimodal topology sensing matrix: Let the network cognitive level data flow be D. c The device instruction operation feature vector is D0, and the protocol interaction state matrix is ​​D. p D c ∈R m×n Where m represents the number of students, and n represents the number of metrics measuring online cognitive level. For example, n metrics may include time taken to complete a specific task, accuracy, and operational fluency. D0∈R k , k represents the number of device instruction operation characteristics, such as instruction type, execution frequency, response time, etc. D p ∈R l×l , l represents the number of network protocols, and matrix element D p(i, j) represents the interaction state between the i-th protocol and the j-th protocol. The multimodal topology sensing matrix M can then be constructed using the following formula:

[0048]

[0049] Among them, 0 a×b It is a zero matrix with a rows and b columns. vec(·) is a vectorization operation that stacks the matrix into a vector column by column.

[0050] Step 2: The dynamic reconstruction module based on the adaptive rule engine inputs the constructed multimodal topology awareness matrix into the pre-trained deep decision tree model to generate dynamic topology descriptors, including node correlation parameters, link time-varying weight matrices, and protocol stack configuration vectors. These descriptors can accurately describe the dynamic characteristics of the network topology, providing a basis for decision-making in subsequent topology construction and configuration.

[0051] The dynamic reconstruction module based on the adaptive rule engine first inputs the constructed multimodal topology awareness matrix into a pre-trained deep decision tree model. This deep decision tree model, trained on a large amount of historical network data, possesses powerful pattern recognition and decision-making capabilities, accurately learning and establishing the mapping relationship between network state and topology. Simultaneously, the module's adaptive rule engine continuously monitors various changes in network state in real time, such as fluctuations in network traffic and changes in link performance, preparing for potential subsequent decision adjustments.

[0052] The deep decision tree model generates a dynamic topology descriptor that includes the following key parameters based on the data features in the multimodal topology perception matrix:

[0053] The node correlation parameter measures the degree of connection between nodes in a network, directly reflecting the dependencies between nodes during communication. Taking a data center network as an example, some server nodes will have relatively high correlation parameter values ​​due to frequent data interaction, clearly indicating a strong communication dependency between these nodes. This parameter provides insight into the communication patterns between nodes in the network, offering a strong basis for rationally planning node connections during subsequent topology construction, ensuring efficient and smooth network communication.

[0054] Considering that network link status and performance dynamically change over time during actual operation, especially considering potential traffic surges during teaching processes, the time-varying link weight matrix dynamically adjusts based on these real-time characteristics. It dynamically adjusts the weight of each link according to its real-time status and performance changes. The weight reflects characteristics such as the link's "importance" or "load capacity." For example, when a link becomes congested, its weight decreases accordingly, indicating to subsequent traffic scheduling mechanisms that this link is not suitable for handling excessive traffic in its current state; conversely, when a link is lightly loaded and performs well, its weight increases, so that subsequent traffic scheduling prioritizes its use. The time-varying link weight matrix more realistically reflects the actual network operation, providing crucial information for topology optimization and enabling more rational allocation and utilization of network resources.

[0055] Protocol stack configuration vectors record the configuration information of different protocol stacks in a network. In network communication, various protocol stacks need to be correctly configured to ensure the normal operation of network protocols and effective interaction between different devices. In a network containing various network devices (such as routers, switches, servers, etc.), different devices may use different protocol stacks. Protocol stack configuration vectors record detailed configuration parameters for the protocol stack on each device, such as IP addresses, subnet masks, gateway settings in the TCP / IP protocol stack, and specific parameters for each protocol layer. This ensures the correct operation of network protocols and smooth interaction between different devices, providing a solid guarantee for the stability and functionality of the network topology and preventing network failures or communication interruptions caused by protocol configuration errors.

[0056] Node correlation parameters, link time-varying weight matrices, and protocol stack configuration vectors together constitute the dynamic topology descriptor. These parameters comprehensively and dynamically describe the characteristics of the network topology, providing a solid basis for subsequent topology configuration and adjustments. For example, during network topology optimization, the connection methods between nodes can be adjusted based on node correlation parameters to make the network structure more rational and improve network communication efficiency; traffic can be rationally allocated based on the link time-varying weight matrix to avoid link congestion and ensure efficient network operation; and the protocol stack configuration vector can be referenced to ensure the normal operation of network protocols, guarantee the integrity of network functions, and enable the network to stably support various service requirements.

[0057] The adaptive rule engine plays a crucial role in automatically adjusting the decision-making process throughout the entire process. When it detects a change in network status, such as a sudden increase in network traffic, it automatically triggers the decision adjustment mechanism. At this time, it adjusts the time-varying weight matrix of links in real time, allowing the weights to more accurately reflect the current link status. Through this dynamic adjustment, the generated topology descriptor can better adapt to new network conditions, thereby optimizing the network topology and ensuring that the network maintains good performance under different teaching needs and network conditions, meeting the requirements of various practical application scenarios.

[0058] The following embodiment illustrates the generation process of dynamic topology descriptors: A multimodal topology-aware matrix M is input into a pre-trained deep decision tree model f to generate a dynamic topology descriptor T = [R0, W(t), C], where the node correlation parameter... This is the part used to generate node association parameters in a deep decision tree model, R0∈R s ×s s represents the number of network nodes, R0(i,j) represents the correlation between node i and node j; the link time-varying weight matrix W(t) = f W (M, t), f W This is the part of the deep decision tree model used to generate the time-varying weight matrix of the links, where t represents time, and W(t)∈R. e×e e represents the number of network links, W(t)(i,j) represents the weight between link i and link j at time t; the protocol stack configuration vector C = f C (M), f C This is the part of the deep decision tree model used to generate protocol stack configuration vectors, C∈R q q represents the number of protocol stack configuration parameters.

[0059] Step 3 involves using the protocol abstraction middleware of the heterogeneous controller interface adaptation layer to convert the dynamic topology descriptor into a standardized configuration instruction set compatible with multiple protocol interfaces such as OpenFlow / Restconf / Netconf, thus adapting to different types of controllers. The protocol abstraction middleware employs a layered mapping mechanism to achieve a unified semantic expression for multiple protocol interfaces.

[0060] The role of the heterogeneous controller interface adaptation layer is to ensure that generated configuration commands can be correctly executed on different controllers and network devices, achieving seamless integration and collaborative operation between different protocols. Since different network controllers in an SDN network may support different protocol interfaces, such as OpenFlow, Restconf, and Netconf, and protocol abstraction middleware can shield these differences, it is introduced to convert and adapt dynamic topology descriptors, transforming them into a standardized set of configuration commands compatible with multiple protocol interfaces. In this way, regardless of the type of network controller used, these configuration commands can be easily received and processed, enabling dynamic adjustment of the virtual network topology.

[0061] The protocol abstraction middleware constructs a syntax-semantic decoupling architecture through a layered mapping mechanism, achieving a unified expression of multi-protocol interfaces. Its semantic abstraction layer extracts topology intent based on the YANG model to generate intermediate instructions, shielding protocol differences. The protocol adaptation layer dynamically compiles the intermediate instructions into polymorphic configuration streams according to the message format and interaction paradigm of the target protocol. This architecture, through a two-level mapping table, a semantic logic table, and a protocol syntax template library, achieves efficient conversion from topology descriptors to OpenFlow flow tables / Restconf RPC / Netconf configuration blocks, ensuring consistent northbound semantic expression while maintaining southbound interface diversity.

[0062] Step 4: Based on the real-time topology calculation engine, an incremental construction algorithm with time window constraints is adopted to dynamically adjust the virtual node density parameters according to the characteristics of the teaching process stages, thereby achieving a progressive reconstruction from basic single-domain topology to complex cross-domain topology. The virtual node density parameters exhibit a non-linear relationship with the concurrent operation complexity of the teaching scenario in advanced stages of teaching.

[0063] Specifically, the real-time topology calculation engine precisely divides the timeline into multiple time windows of varying lengths, each with its own specific time span and processing tasks. Within each independent time window, the incremental construction algorithm focuses only on processing and updating network topology information closely related to the current time window. For example, it only processes network connections that have changed, newly added or deleted virtual nodes, etc., within the current time window, while ignoring stable topology information already processed in other time windows. This refined processing method cleverly avoids frequent and time-consuming full reconstruction operations of the entire large and complex network topology, greatly improving the efficiency of topology construction and significantly shortening the time required for topology updates, enabling the network topology to quickly respond to changes in the teaching environment. The incremental construction algorithm only updates the changed parts, avoiding the reconstruction of the entire topology, thus improving the efficiency of topology construction and reducing the consumption of computing resources.

[0064] The incremental construction algorithm also deeply integrates time factors and dynamic changes in network status. By continuously monitoring the progress of the teaching process and real-time fluctuations in network status, such as increases or decreases in network traffic and fluctuations in node performance, it can more accurately and flexibly adjust the virtual node density parameters based on the unique characteristics exhibited at different stages of the teaching process, such as the pace of teaching and shifts in teaching focus. The virtual node density parameter, as a key quantitative indicator measuring the density of virtual node distribution in the virtual network topology, directly and profoundly determines the complexity of the topology. It is not fixed but dynamically and finely adjusted according to the stage characteristics of the teaching process.

[0065] The required complexity of network topology varies significantly across different teaching stages. In the initial stages of teaching basic network knowledge, students are just beginning to learn about networks and lack an intuitive understanding of complex network structures and concepts. At this time, the real-time topology calculation engine, based on the characteristics of this teaching stage, sets the virtual node density parameter to a low value. With this parameter setting, the constructed single-domain topology is simple and clear, with a relatively small number of nodes and links, and the connections between nodes are clear and easy to understand. This provides students with a clear and easy-to-understand starting point for network learning, enabling them to quickly grasp basic network knowledge, establish a basic cognitive framework of networks, and lay a solid foundation for subsequent in-depth learning.

[0066] As the teaching progresses, a transition phase begins. At this stage, students have mastered basic network knowledge, gained a preliminary understanding of networks, and are eager to explore more advanced network knowledge and skills. The real-time topology calculation engine continuously monitors the characteristics of each teaching stage, closely integrating with students' learning progress and actual needs. It accurately assesses students' learning status and needs by collecting data on classroom performance, homework completion, and experimental results. Based on this, it dynamically adjusts the virtual node density parameters, gradually increasing the number and distribution of virtual nodes as the teaching progresses. New nodes and links are cleverly added to the basic single-domain topology, gradually increasing the topology's complexity. These new nodes and links guide students to gradually master deeper network knowledge and skills, improving their network cognition and operational abilities, allowing them to continuously consolidate and deepen their knowledge through practice.

[0067] As teaching progresses to advanced levels, students need to engage with and master knowledge of complex network environments to better adapt to the network demands of future work. Complex cross-domain topologies simulate real-world complex network environments such as large enterprise networks and internet data centers. These environments include multiple network domains, each with specific functions and security policies, widespread node distribution, and intricate link relationships. The real-time topology calculation engine adjusts the virtual node density parameter to a high value based on the characteristics of the current teaching stage. By significantly increasing the number of nodes and links in the topology, a carefully constructed network topology structure with multiple domains, widespread node distribution, and complex link relationships is created. This complex cross-domain topology provides students with a practical scenario that closely resembles a real-world network environment, allowing them to experience various problems and challenges in complex network environments, such as network congestion, security vulnerabilities, and routing. By solving these problems, students can effectively improve their ability to solve complex network problems, cultivate their innovative thinking and practical skills, and lay a solid foundation for their career development.

[0068] As the teaching process progresses and its complexity increases, the adjustment of virtual node density parameters does not follow a simple linear growth pattern. Linear growth means that the number of nodes increases at a fixed ratio or magnitude as the teaching stage progresses, but this simple adjustment method cannot meet the diverse needs of actual teaching. In fact, the adjustment of virtual node density parameters is based on actual teaching needs and scenario characteristics, allowing for more flexible and precise adjustments. For example, in the basic network knowledge explanation stage, only a small number of nodes may be needed to build a simple topology. These nodes are sufficient to demonstrate the basic concepts and principles of the network, such as clients, servers, and routers. However, in the complex network application practice stage, as the teaching complexity increases, the number of nodes will increase significantly, but the magnitude and manner of the increase are not linear. When dealing with network security practices, a large number of simulated attack nodes and defense nodes may be suddenly added; when studying network performance optimization, a large number of nodes will be added to simulate different traffic loads and device performance. This non-linear adjustment method is determined according to specific teaching objectives and content to better adapt to the teaching objectives at different stages, ensuring that students can obtain the most suitable learning experience at each stage, making the teaching process more scientific, reasonable, and efficient.

[0069] The following is an example illustrating the process of incrementally reconstructing the virtual network topology: Let the time window be [t1, t2]. Within this time window, the incremental construction algorithm only processes network topology information relevant to that time window. Depending on the teaching stage, such as the basic stage, transition stage, or advanced stage, the virtual node density parameter ρ(t) is dynamically adjusted according to the following rules:

[0070]

[0071] Where, ρ base This serves as the benchmark for calculating virtual node density across all teaching stages, reflecting the topological complexity at the beginning of the teaching process. ΔS represents the teaching progress increment. ΔC represents the complexity of concurrent operations. α and β are adaptive adjustment coefficients.

[0072] Step 5: Construct a process visualization feedback loop. The network state complexity index is calculated in real time through the topology entropy monitoring module. When a sudden change in entropy is detected, the topology self-healing mechanism is triggered, and the abnormal operation mode feature library in the knowledge graph is updated synchronously.

[0073] Topological entropy is an indicator that reflects the complexity and disorder of a network structure. It can intuitively reflect the network's operating state; the higher the entropy value, the more complex and disordered the network state. For example, in a simple star network, because the node connections are relatively simple and the data flow is relatively clear, its topological entropy value may be at a low level. However, in a complex mesh network, there are multiple connection paths between nodes, and the data transmission paths are diverse and complex, so the topological entropy value will be relatively high.

[0074] The anomaly detection process relies heavily on data provided by the topology entropy monitoring module. Its judgment is based on continuous monitoring of changes in topology entropy. When a significant change in entropy that deviates from the normal fluctuation range within a short period is detected—a sudden entropy change—it is determined that a network anomaly may have occurred. Such a "sudden change" can be caused by a variety of factors. For example, equipment failure is a common cause; when a router or switch experiences a hardware problem, it may lead to data transmission interruptions or errors, thus causing a sudden change in topology entropy. Link interruptions are also significant; accidental disconnection of network cables or damage to fiber optic lines can disrupt normal network connectivity, causing changes in entropy. Furthermore, network attacks can also cause anomalies; malicious software may generate a large amount of abnormal traffic in the network, causing network chaos and resulting in a sudden change in topology entropy.

[0075] When the topology entropy monitoring module detects a sudden change in entropy value and confirms a network anomaly, the topology self-healing mechanism automatically activates to attempt to repair the network topology and restore it to a normal state. This mechanism features dual modes: link redundancy rerouting and dynamic protocol stack rollback, to handle different types of network anomalies. When a major link experiences a sudden entropy change due to a fault, the system can quickly and automatically select other available redundant links to reroute data traffic. For example, core switches are typically connected by multiple fiber optic links. When one link fails, the system will switch data traffic to other normal links within a very short time, ensuring continued data transmission and avoiding network interruptions caused by link failures, thus guaranteeing business continuity. If the network anomaly is caused by configuration errors or software vulnerabilities in the network protocol stack, the dynamic protocol stack rollback mode will automatically activate. For example, after upgrading network device protocol software, compatibility issues may cause network anomalies. In this case, this function will automatically roll back the protocol stack to a previously stable operating state, eliminating anomalies caused by protocol problems, allowing the network to quickly return to normal function, and reducing network downtime caused by protocol failures.

[0076] While triggering the topology self-healing mechanism, the system records the detected abnormal operation patterns in detail and updates them synchronously to the abnormal operation pattern feature library in the knowledge graph. The knowledge graph, as a powerful structured model specifically designed to represent and organize knowledge, stores a wealth of feature information about normal network operation and abnormal situations. By continuously accumulating and updating the abnormal operation pattern feature library, the system can continuously accumulate experience with network faults and abnormal operations. In subsequent network operations, when encountering similar problems, the system can rely on the relevant feature information stored in the knowledge graph to more quickly identify and handle abnormal situations, providing a strong reference for fault diagnosis and self-healing, thereby further improving network reliability and stability and ensuring the smooth progress of teaching.

[0077] The following example illustrates the process of topological entropy monitoring and self-healing: The topological entropy value E can be calculated using the following formula:

[0078]

[0079] Where N is the number of network states, p i It represents the probability of the i-th network state occurring. When a sudden change in topological entropy is detected, i.e., |E(t)-E(t-Δt)|>∈ (∈ is a preset threshold), the topological self-healing mechanism is triggered.

[0080] The beneficial effects of this invention are as follows:

[0081] 1. By leveraging embedded probes to collect multimodal network data in real time from teaching scenarios, a multimodal topology awareness matrix is ​​constructed to capture dynamic information promptly and accurately reflect the actual network operation. Integrating three types of data—network cognitive level data stream, device command operation feature vectors, and protocol interaction state matrix—provides rich data support for network topology construction, performance analysis, and fault diagnosis.

[0082] 2. The dynamic reconstruction module based on the adaptive rule engine inputs the multimodal topology awareness matrix into the pre-trained deep decision tree model to generate a dynamic topology descriptor. This descriptor includes node correlation parameters, link time-varying weight matrices, and protocol stack configuration vectors. The adaptive rule engine automatically adjusts the decision-making process to adapt the topology descriptor to new network conditions, optimize the network topology, and meet different teaching needs.

[0083] 3. The protocol abstraction middleware of the heterogeneous controller interface adaptation layer converts dynamic topology descriptors into a standardized configuration instruction set compatible with multiple protocol interfaces, enabling seamless connection and collaborative work between different protocols.

[0084] 4. The real-time topology calculation engine adopts an incremental construction algorithm with time window constraints. It dynamically adjusts the virtual node density parameters according to the characteristics of the teaching process stage, realizing the progressive reconstruction from basic single-domain topology to complex cross-domain topology. This improves topology construction efficiency, reduces computing resource consumption, fits the actual teaching, and makes the teaching process more scientific, reasonable, and efficient.

[0085] 5. A process visualization feedback loop is constructed. The network state complexity index is calculated in real time through the topology entropy monitoring module. When the entropy value changes abruptly, a network anomaly is identified, and the topology self-healing mechanism is automatically activated. It has dual modes of link redundancy rerouting and dynamic protocol stack rollback, which can quickly repair the network. At the same time, the abnormal operation mode is updated to the knowledge graph feature base to provide a reference for subsequent fault diagnosis and self-healing, improve network reliability and stability, and ensure the smooth progress of teaching.

[0086] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for dynamically constructing virtual network topology for SDN, characterized in that, include: Step 1: Collect multimodal network data in the teaching scenario in real time using an embedded probe, and construct a multimodal topology perception matrix. The multimodal network data includes network cognitive level data stream, device command operation feature vector, and protocol interaction state matrix. In step 1, let the network cognitive level data stream be... The device instruction operation feature vector is The protocol interaction state matrix is ; , where m represents the number of students and n represents the number of indicators for measuring online cognitive level; k represents the number of device instruction operation features; 'l' represents the number of network protocols, and the matrix elements are... , representing the interaction state between the i-th protocol and the j-th protocol; the multimodal topology sensing matrix M is constructed using the following formula: in, It is a zero matrix with rows of a and columns of b. It is a vectorization operation that stacks the matrix column by column into a vector; Step 2: The dynamic reconstruction module based on the adaptive rule engine inputs the multimodal topology perception matrix constructed in Step 1 into the pre-trained decision tree-based machine learning model to generate a dynamic topology descriptor including node correlation parameters, link time-varying weight matrix and protocol stack configuration vector. In step 2, the multimodal topology awareness matrix M is input into the pre-trained decision tree-based machine learning model f to generate a dynamic topology descriptor. Among them, the node correlation parameter , This is the part used to generate node association parameters in a decision tree-based machine learning model. , Indicates the number of network nodes. Represents the correlation between node i and node j; link time-varying weight matrix , This is the part used to generate the time-varying weight matrix of the link in a decision tree-based machine learning model, where t represents time. 'e' represents the number of network links. Represents the weight between link i and link j at time t; Protocol stack configuration vector , It is the part of a decision tree-based machine learning model used to generate protocol stack configuration vectors. q represents the number of protocol stack configuration parameters; Step 3: Through the protocol abstraction middleware of the heterogeneous controller interface adaptation layer, the dynamic topology descriptor generated in Step 2 is converted into a standardized configuration instruction set compatible with OpenFlow, Restconf and Netconf protocols; Step 4: Based on the real-time topology calculation engine, an incremental construction algorithm with time window constraints is adopted to dynamically adjust the virtual node density parameters according to the characteristics of the teaching process stage, so as to realize the gradual reconstruction from basic single-domain topology to complex cross-domain topology. In step 4, the time window is set as follows: Within this time window, the incremental construction algorithm only processes network topology information relevant to that time window; depending on the teaching stage, the virtual node density parameter is adjusted. The following dynamic adjustments will be made: in, It serves as the benchmark for calculating virtual node density across all teaching stages, reflecting the topological complexity in the early stages of teaching. To increase the teaching progress; For the complexity of concurrent operations; , This is the adaptive adjustment coefficient; Step 5: Construct a process visualization feedback loop. Calculate the network state complexity index in real time through the topology entropy monitoring module. When a sudden change in entropy is detected, trigger the topology self-healing mechanism and synchronously update the abnormal operation mode feature library in the knowledge graph. In step 5, the topological entropy value E is calculated using the following formula: Where N is the number of network states. It represents the probability of the i-th network state occurring; when a sudden change in topological entropy is detected... At that time, the topology self-healing mechanism is triggered, in which It is a preset threshold.

2. The method for dynamically constructing virtual network topology for SDN as described in claim 1, characterized in that, In step 1, the network cognitive level data stream is generated by comprehensively quantifying the student's operation steps, task completion time, accuracy rate, and operation fluency indicators in the network experiment; the device instruction operation feature vector is generated by parsing the instruction type, frequency, and abnormal operation markers received by the router and switch; the protocol interaction state matrix is ​​generated by dynamically parsing the interaction messages between the control plane and the data plane, including protocol type, interaction frequency, and abnormal protocol behavior markers.

3. The method for dynamically constructing virtual network topology for SDN as described in claim 1, characterized in that, The dynamic topology descriptor generated by the decision tree-based machine learning model in step 2 includes: a node correlation parameter, which is dynamically updated based on the communication dependencies between nodes to optimize node connection planning; a link time-varying weight matrix, which dynamically adjusts the weight values ​​according to the link congestion or load capacity during the teaching process to guide traffic scheduling; and a protocol stack configuration vector, which records the device protocol parameters to ensure the compatibility and communication continuity of the TCP / IP protocol stack.

4. The method for dynamically constructing virtual network topology for SDN as described in claim 1, characterized in that, In step 3, the protocol abstraction middleware implements a unified semantic expression for multi-protocol interfaces using a layered mapping mechanism, including: the semantic abstraction layer converts dynamic topology descriptors into protocol-independent intermediate representations; and the protocol adaptation layer maps the intermediate representations to OpenFlow flow table entries, Restconf configuration templates, or Netconf XML directives based on the target controller type.

5. The method for dynamically constructing virtual network topology for SDN as described in claim 1, characterized in that, The incremental construction algorithm in step 4 includes: a time window segmentation module, which divides the teaching process into basic, transitional, and advanced stage windows; a virtual node density parameter adjustment module, which sets a low density value for single-domain topology or a high density value for cross-domain topology based on stage characteristics; and a progressive reconstruction module, which updates newly added or deleted virtual nodes and links only within the current time window.

6. The method for dynamically constructing virtual network topology for SDN as described in claim 1, characterized in that, In step 5, the topology self-healing mechanism includes two modes: link redundancy rerouting and protocol stack dynamic rollback. In the link redundancy rerouting mode, when a link interruption is detected, the system automatically switches to a backup link. In the protocol stack dynamic rollback mode, when the protocol configuration is abnormal, the system restores the protocol parameters to the stable version.

Images