Deployment method of software defined optical network controller based on machine learning

Abstract

The invention discloses a method for deploying a software defined optical network controller based on machine learning, which comprises the following steps of: traversing an SDON network topology based on a depth-first search strategy, searching all possible paths between any two nodes in all nodes in the SDON network, and counting a transmission delay value of each path; setting a delay threshold, and screening out a control path between nodes meeting a delay constraint condition; constructing a new network topological graph of the SDON network to control a path to form a new abstract link; dividing the two nodes without the direct control path into different network areas; determining a controller location for each network area; and selecting one of the controllers as a control center to complete deployment of the SDON network controller. According to the invention, the control redundancy of the SDON is reduced while the network survivability is ensured, and the reliability of the SDON is improved.

Description

Technical Field

[0001] The present invention belongs to the field of software-defined optical network controller deployment design, and specifically relates to a deployment method of a software-defined optical network controller based on machine learning. Background Art

[0002] With the large-scale deployment of 5G and B5G networks, various new applications and services, including the Internet of Things, autonomous driving, and blockchain, have developed rapidly. Correspondingly, the data traffic in the global Internet has maintained a stable and rapid growth. Optical networks have always been the basic carrier facilities of the Internet since their emergence, with their advantages such as large capacity, high transmission rate, and low loss, and have been widely used in various network application environments. However, with the exponential growth of network traffic and the continuous improvement of people's demand for network service quality, optical networks have also brought huge challenges. Faced with the growth of network bandwidth demand and the increasingly complex differences in business needs, the evolution model of optical networks that simply expand capacity can no longer meet development needs. It is necessary to introduce intelligent and reconfigurable optical networks to support services through intelligent management and control.

[0003] Since optical networks differ significantly from ordinary networks in terms of physical layer, transmission mechanism and network architecture, if software-defined networking (SDN) is used directly to manage optical networks, the SDN controller will not be able to implement dynamic configuration and optimization of the optical network, which will limit the flexibility and scalability of the optical network, resulting in longer fault recovery time, reduced network reliability and availability, and other issues. Therefore, an innovative network architecture, software-defined optical networking (SDON), is needed for optical networks. Software-defined optical networking (SDON) can solve the problems of low resource utilization and poor scalability of optical network management in ordinary software-defined networking (SDN). SDON can give full play to the advantages of high bandwidth, low latency and high reliability of optical networks, while achieving flexible resource management and efficient network operation and maintenance.

[0004] Software Defined Optical Networks (SDON) mostly use centralized controllers for network control. Once a controller has problems or fails, it is likely to cause the entire network to be paralyzed. This situation will have a significant impact on application fields that are highly dependent on network resources, such as cloud computing and big data, and may even exceed the acceptable range. At the same time, in the process of promoting large-scale deployment of SDON controllers, cost control issues have also become a key factor that cannot be ignored. How to optimize the number of deployed controllers while ensuring efficient and reliable implementation of services has always been a highly concerned issue in academia and industry, and it is also a difficulty that affects the large-scale application of SDON.

[0005] Most of the SDON controller deployment work that has been carried out focuses on aspects such as network transmission delay and control redundancy. However, in terms of SDON's most critical survivability requirements and their impact on controller deployment strategies, there are still shortcomings such as the inability to effectively reduce the probability of network failures and the large number of controllers deployed. Summary of the invention

[0006] The present invention aims to solve one of the technical problems existing in the related art at least to a certain extent.

[0007] An object of the present invention is to provide a deployment method of a software defined optical network controller based on machine learning, combining a graph convolutional network and a simulated annealing algorithm, to reduce the control redundancy of SDON while ensuring the survivability of the network and improve the reliability of SDON.

[0008] In order to achieve the above-mentioned object, the present invention provides a deployment method of a software-defined optical network controller based on machine learning, comprising: S100, based on the depth-first search strategy, traverse the SDON network topology to find the SDON network from any starting node To any end node All possible paths, and the transmission delay value of each path is counted ; S200, setting delay threshold ,by As a delay constraint, the path that meets the delay constraint is selected as the reachable path; when there is only one reachable path between two nodes, the reachable path is used as the control path between the two nodes; when there are multiple reachable paths between two nodes, the reachable path with the smallest delay is selected as the control path; S300: construct a new network topology diagram of the SDON network. The construction rules of the new network topology diagram include: (1) Forming a new abstract link with all control paths obtained in step S200; (2) If and If there is no reachable path between the two nodes, the two nodes are divided into nodes in different areas. and If there is a reachable path between them, the two nodes are placed in the same area, thus forming several network areas; S400, determining a controller location for each network area; S500. Select one controller from among the controllers as a control center to coordinate the work among the controllers and complete the deployment of the SDON network controller.

[0009] Furthermore, the specific steps of step S100 include: S110, abstract the SDON network into an undirected graph , is a set of nodes, representing all nodes in the network, where is the number of nodes in the network; is an edge set, representing all the edges in the network, where is the total number of edges; S120, using the stack structure to implement the traversal of the SDON network topology based on the depth-first search strategy; first select the starting node And push it into the stack, then take out the top node of the stack As the current access node, Mark it as visited and add it to the current path list, then traverse the neighbor nodes. Each unvisited neighbor node , calculated from arrive The total path delay is Push it into the stack; finally, check the algorithm termination condition. If is the target node , then record the current path and its total delay value; if If there are no unvisited neighbor nodes, backtrack: Remove from the current path list and mark as unvisited; S130, repeat the traversal process until the stack is empty, and all possible paths are searched; after the traversal is completed, we get arrive All feasible paths and their corresponding transmission delay values .

[0010] Furthermore, the method for determining the location of the controller for each network area in step S400 is: S410, obtaining structural information and node characteristics of the network topology through the GCN model, and generating an initial solution for controller deployment; S420, the initial solution generated by GCN As a heuristic solution, the simulated annealing algorithm is combined to optimize the final controller position.

[0011] Furthermore, in step S410, the structural information and node characteristics of the network topology are obtained through the GCN model to generate an initial solution for controller deployment; the specific steps are: S411, converting the controller deployment problem into an optimization problem of a network topology graph, extracting features for nodes in the graph, including node degree, node traffic load, and node clustering coefficient; S412. Build a multi-layer GCN model and update the embedding representation of each node through multi-layer graph convolution. After multi-layer GCN calculation, output a controller fitness score to indicate the possibility of each node as a controller. S413, using integer linear programming to generate optimal controller deployment solutions for multiple SDON network topologies as training sets, and training the GCN model constructed in step S412. During the training process, GCN learns the relationship between node features and graph structures, thereby predicting the optimal location for controller deployment; S414: Use the trained GCN model to predict the new network topology of the SDON network constructed in step S300, and calculate the controller fitness; select the top k nodes with the highest fitness as the initial controller deployment plan .

[0012] Further, in step S420, the initial solution generated by GCN As a heuristic solution, the simulated annealing algorithm is combined to optimize the final controller position. The specific steps are: S421, deploy the initial controller generated by GCN As the initial solution, set the initial temperature , the simulated annealing algorithm is applied for iteration. In each iteration, a node is randomly selected from the current solution and the controller deployment state of the node is reversed, that is, the node that originally deployed the controller is not deployed with the controller or the node that did not deploy the controller is deployed with the controller, and a new solution is generated. ; Check whether the new solution satisfies the coverage constraint, that is, each network node is either a controller or adjacent to at least one controller. If the coverage constraint is not met, add controllers or adjust the positions of controllers until the coverage constraint is met; S422, define the objective function of the simulated annealing algorithm to optimize the controller deployment as:

[0013] in is the total number of nodes in the network; Indicates that the node has a controller deployed. Indicates that the node has no controller deployed; is the number of uncovered nodes; is the weight coefficient; New calculation solution The objective function difference between the current solution ,like , then directly accept the new solution Otherwise, according to the probability Accept new interpretations; S423, after each iteration, the temperature is set according to the cooling coefficient Gradually reduce, that is ; When no new solution is accepted for several consecutive times, the algorithm ends and outputs the current solution as the final controller position.

[0014] Further, in step S500, the node with the smallest average delay for all controllers is selected to deploy the control center, and the specific method is as follows: Assume the location of the controller deployment node is , the deployment location of the control center meets:

[0015] in, Indicates taking the average value; Indicates the transmission delay between deployment nodes.

[0016] Furthermore, in step S500, after selecting the node with the smallest average delay for all controllers to deploy the control center, an auxiliary control center is selected to deploy the node with the second lowest average delay for all controllers as a backup of the control center.

[0017] Furthermore, the method also includes deploying optical performance monitoring and using machine learning algorithms to detect SDON network anomalies and locate fault points; the specific method is: Interact with the optical performance monitoring system through the southbound interface SBI of the SDON network controller to collect optical performance data and perform standardized processing on the data; Then, an isolation forest model is built and trained for anomaly monitoring, combining abnormal data points with network topology information to locate the location where the fault occurs; The detection results are fed back to the controller of the SDON network to trigger fault recovery operations.

[0018] On the other hand, the present invention provides a non-transitory computer-readable storage medium having computer instructions stored thereon, which enable a computer to execute the above-mentioned method for deploying a software-defined optical network controller based on machine learning.

[0019] Yet another aspect of the present invention provides an electronic device, comprising: a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus, and the processor calls logic instructions in the memory to execute the above-mentioned deployment method of the software-defined optical network controller based on machine learning.

[0020] On the other hand, the present invention provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer executes the above-mentioned method for deploying a software-defined optical network controller based on machine learning.

[0021] Beneficial effects: (1) The deployment method of the software-defined optical network controller based on machine learning of the present invention adopts a depth-first search strategy. In the software-defined optical network, the depth-first search strategy can efficiently traverse the network topology, record all possible paths and their delay values, and provide reliable support for path planning and optimization.

[0022] (2) The present invention uses a graph convolutional network (GCN) to generate an initial solution for SDON controller deployment, which can fully utilize the structural information and node characteristics of the network topology to generate a better initial solution for SDON controller deployment. The key advantage of GCN is that it can use the optimal solution of a small-scale network as training data, thereby extracting the inherent patterns and dependencies in the network topology and scaling the solution to large-scale networks.

[0023] (3) The present invention adopts a simulated annealing algorithm to determine the controller location. In this way, the controller deployment can be effectively optimized, the number of deployed controllers can be minimized, and the coverage of the network can be ensured.

[0024] (4) The present invention adopts a hierarchical controller deployment architecture, coordinates the work between the various controllers through the control center, and selects additional control centers to be deployed at auxiliary locations with lower average latency to all controllers as backups. This avoids the increased probability of network failures caused by the failure of the selected control center.

[0025] (5) The present invention uses high-performance optical performance monitoring (OPM) combined with the isolation forest algorithm to analyze the monitoring data of SDON, which can quickly identify and locate abnormal events in the optical network, thereby improving the reliability and operation and maintenance efficiency of the network. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 This is an overall flow chart of the deployment method of the software-defined optical network controller based on machine learning of the present invention.

[0027] Figure 2 A model diagram of a layered deployment architecture in an embodiment of the present invention.

[0028] Figure 3 This is the pan-European COST239 network topology diagram used in the simulation experiment in the embodiment of the present invention.

[0029] Figure 4 This is a North American NSF network topology diagram used in the simulation experiment in the embodiment of the present invention.

[0030] Figure 5 A comparison chart of the number of controllers deployed in the pan-European COST239 network topology using different algorithms.

[0031] Figure 6A comparison chart of the number of controllers deployed in the North American NSF network topology using different algorithms.

[0032] Figure 7 The graph shows the change of network failure probability in the pan-European COST239 network topology using different algorithms.

[0033] Figure 8 Figure 2 shows the change in network failure probability in the North American NSF network topology using different algorithms.

[0034] Fig. 9 The graph shows the change of network failure recovery probability in the pan-European COST239 network topology using different algorithms.

[0035] Fig.10 The graph shows the change in network failure recovery probability in the North American NSF network topology using different algorithms.

[0036] Fig.11 This is a diagram showing the optical network traffic prediction performance in an embodiment of the present invention. DETAILED DESCRIPTION

[0037] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings in the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments, and they should not be understood as limitations on the present invention. 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. In the description of the present invention, it should be understood that the terms used are only for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0038] Combine the following Figure 1-Figure 11 The present invention describes a deployment method of a software-defined optical network controller based on machine learning.

[0039] Embodiment 1: This embodiment provides a method for deploying a software-defined optical network controller based on machine learning.

[0040] For software-defined optical networks, the main factor affecting the survivability of optical networks is the security of optical fiber links. When a link fails, data is easily lost. Therefore, in SDON networks, multiple controllers need to be deployed according to actual network transmission requirements. This requires a balance between improving the survivability of optical networks and reducing deployment costs. This embodiment provides an improved deployment method for software-defined optical network controllers. For large-scale network structures, a layered deployment architecture is adopted. The specific deployment method is as follows: Figure 1 As shown, including: S100, based on the depth-first search strategy (DFS), traverse the SDON network topology to find all possible paths between any two nodes in the SDON network, and count the transmission delay value of each path. The specific steps include: S110, abstract the SDON network into an undirected graph , is a set of nodes, representing all nodes in the network, where is the number of nodes in the network; is an edge set, representing all the edges in the network, where is the total number of edges; S120, using the stack structure to implement the traversal of the SDON network topology based on the depth-first search strategy; first select the starting node And push it into the stack, then take out the top node of the stack As the current access node, Mark it as visited and add it to the current path list, then traverse the neighbor nodes. Each unvisited neighbor node , calculated from arrive The total path delay is Push it into the stack; finally, check the algorithm termination condition. If is the target node , then record the current path and its total delay value; if If there are no unvisited neighbor nodes, backtrack: Remove from the current path list and mark as unvisited; S130, repeat the traversal process until the stack is empty, and all possible paths are searched; after the traversal is completed, we get arrive All feasible paths and their corresponding transmission delay values .

[0041] S200, set the delay threshold and filter the control path between nodes. The specific method is: Setting latency thresholds , get the node from the path search algorithm and All possible paths between and the corresponding transmission delay values .if , the corresponding path Add the path to the valid path set as a reachable path; otherwise, Add to the unreachable path set. When there is only one reachable path between two nodes, this reachable path is used as the control path between the two nodes; when there are multiple reachable paths between two nodes, the reachable path with the smallest delay is selected as the control path.

[0042] S300, construct a new network topology diagram of the SDON network.

[0043] The new network topology construction rules include: (1) All control paths obtained in step S200 form a new abstract link; in the new network topology, nodes and The abstract link between represents a arrive The path that is reachable and has the lowest delay.

[0044] (2) If and If there is no reachable path between them (i.e., the path does not satisfy the delay constraint), it means that the two nodes belong to different network areas, and the two nodes are divided into nodes in different areas; if and If there is a reachable path between the two nodes, the two nodes are placed in the same area, thus forming several network areas.

[0045] A simplified regional topology can be constructed using the above rules, where each region represents a subnet with internal connectivity, and the connections between regions are represented by abstract links that are reachable communication paths.

[0046] S400: Determine the controller location for each network area. This includes: S410, obtain the structural information and node features of the network topology through the graph convolutional network (GCN) model, and generate the initial solution for controller deployment. The specific steps are: S411, converting the controller deployment problem into an optimization problem of a network topology graph, extracting features for nodes in the graph, including node degree, node traffic load, and node clustering coefficient; S412. Build a multi-layer GCN model and update the embedding representation of each node through multi-layer graph convolution. After multi-layer GCN calculation, output a controller fitness score to indicate the possibility of each node as a controller. S413, using integer linear programming to generate optimal controller deployment solutions for multiple SDON network topologies as training sets, and training the GCN model constructed in step S412. During the training process, GCN learns the relationship between node features and graph structures, thereby predicting the optimal location for controller deployment; S414: Use the trained GCN model to predict the new network topology of the SDON network constructed in step S300, and calculate the controller fitness; select the top k nodes with the highest fitness as the initial controller deployment plan , where the number of k is determined by the number of nodes in the SDON network topology.

[0047] S420, the initial solution generated by GCN As a heuristic solution, the simulated annealing algorithm is combined to optimize the final controller position. When the simulated annealing algorithm optimizes the controller deployment strategy, the algorithm goal is to minimize the number of controllers and ensure the coverage of the network. The specific steps are: S421, deploy the initial controller generated by GCN As the initial solution, set the initial temperature , the simulated annealing algorithm is applied for iteration. In each iteration, a node is randomly selected from the current solution and the controller deployment state of the node is reversed, that is, the node that originally deployed the controller is not deployed with the controller or the node that did not deploy the controller is deployed with the controller, and a new solution is generated. ; Check whether the new solution satisfies the coverage constraint, that is, each network node is either a controller or adjacent to at least one controller. If the coverage constraint is not met, add controllers or adjust the positions of controllers until the coverage constraint is met; S422, define the objective function of the simulated annealing algorithm to optimize the controller deployment as:

[0048] in is the total number of nodes in the network; Indicates that the node has a controller deployed. Indicates that the node has no controller deployed; is the number of uncovered nodes; is the weight coefficient; New calculation solution The objective function difference between the current solution ,like , then directly accept the new solution Otherwise, according to the probability Accept new interpretations; S423, after each iteration, the temperature is set according to the cooling coefficient Gradually reduce, that is ; When no new solution is accepted for several consecutive times, the algorithm ends and outputs the current solution as the final controller position.

[0049] S500: Select the node with the smallest average delay to all controllers to deploy the control center. The specific method is: Assume the location of the controller deployment node is , the deployment location of the control center meets:

[0050] in, Indicates taking the average value; Indicates the transmission delay between deployment nodes.

[0051] S600. To avoid an increased probability of network failure due to failure of the selected control center, an additional control center is deployed at an auxiliary location with a lower average delay for all controllers as a backup.

[0052] So far, a controller hierarchical deployment architecture for large-scale network structures has been built. Figure 2 As shown in the figure, in this architecture, the controller is responsible for controlling the OpenFlow switch, the control center is responsible for coordinating the work of the controller, and a backup control center is deployed to ensure the stability of the network.

[0053] S700, deploy optical performance monitoring, use machine learning algorithms to detect SDON network anomalies and locate fault points. The specific methods are: Interact with the optical performance monitoring system through the southbound interface SBI of the SDON network controller to collect optical performance data and perform standardized processing on the data; Then, an isolation forest model is built and trained for anomaly monitoring, combining abnormal data points with network topology information to locate the location where the fault occurs; The detection results are fed back to the controller of the SDON network to trigger fault recovery operations.

[0054] The present embodiment is verified by simulation experiments below.

[0055] The deployment method of the software defined optical network controller based on machine learning (hereinafter referred to as USCD method) of this embodiment is applied to Figure 3 The pan-European COST239 network topology (11 nodes, 26 edges) and Figure 4 The simulation environment of the North American NSF network topology (14 nodes, 21 edges) is shown. In order to verify the performance of the control plane, the control plane survivability is tested using the control plane failure rate and the probability of failure of the optical fiber per 100 kilometers. The failure probability is defined according to the following equation.

[0056] Definition 1: Probability of fault alarm occurrence The probability of failure of the working route of the control channel is expressed as:

[0057] Definition 2: Failure recovery probability The probability of fast recovery of the control channel by relying on the control channel protection route is expressed as:

[0058] in, It indicates the probability of failure of optical fiber per 100 km; Indicates OpenFlow switch nodes; Indicates controller nodes; Indicates the working route length of the control channel. Indicates the length of the protection route.

[0059] The USCD method proposed in this embodiment takes the survivability of the SDON control plane as a constraint, and the survivability of the control plane can be converted into the control link length as a constraint. and the maximum network failure probability provided by the user The longest control link length As a constraint condition of the USCD method proposed in this embodiment.

[0060] like Figure 5 and Figure 6 As shown, the comparison of the number of controllers deployed in the pan-European COST239 network topology and the North American NSF network topology is shown in the figure using the DCD algorithm, the SCD algorithm and the USCD method proposed in this embodiment under the constraint of the same longest control link length (under the acceptable failure probability of the user). From the simulation results, it can be seen that the number of controllers deployed decreases as the user's demand for network survivability decreases, and the number of controllers deployed by the USCD method proposed in this embodiment is better than that of the DCD algorithm and the SCD algorithm in different parameters and different network topologies. It shows that the USCD method of this embodiment can effectively reduce the number of controllers deployed and reduce control redundancy.

[0061] like Figure 7 and Figure 8 As shown in FIG. 1 , when the same number of controllers are deployed, the network failure probability changes in the pan-European COST239 network topology and the North American NSF network topology are respectively obtained by applying the C-MPC algorithm, the SCD algorithm, and the USCD method proposed in this embodiment. Figure 7 It can be seen that the three algorithms are basically equal in terms of network reliability. This is because in the COST239 network topology, there are fewer nodes, more edges, and the distance between nodes is shorter, so the advantages and disadvantages of the algorithms cannot be clearly distinguished. In the NSF network with low connectivity and long link length, Figure 8As shown, the USCD method proposed in this embodiment is obviously better than the other two algorithms, greatly reducing the probability of optical network failure, and making the optical network have good performance.

[0062] like Fig. 9 and Fig.10 As shown in the figure, after the fault alarm occurs in the SDON network, the MCC algorithm, the SVVR algorithm and the USCD method proposed in this embodiment are respectively applied in the pan-European COST239 network topology and the North American NSF network topology, and the network fault recovery probability change diagram is realized through the control plane. It can be clearly seen that the fault recovery probability of the USCD method proposed in this embodiment in the COST239 and NSF network topologies is better than that of the other two deployment models. At the same time, because the network node connectivity in the COST239 network is greater than that in the NSF network, the controller can start the recovery mechanism faster and has a better protection effect on the network, so the USCD method proposed in this embodiment has greater benefits in the COST239 network.

[0063] The method of this embodiment can be applied to optical network traffic prediction. It has been verified that when it is applied to optical network traffic prediction, its performance is as follows: Fig.11 shown. Fig.11 In the figure, the blue solid line represents the simulated optical network traffic model, and the red dotted line is the traffic prediction result of the present invention. By comparison, it can be seen that the simulated optical network traffic model shows multiple obvious peak fluctuations, indicating that the network traffic has significant burstiness and instability. However, the prediction results of the USCD method proposed in this embodiment are highly consistent with the actual situation. Even in the extreme case of a sudden increase in traffic, it can still accurately track the traffic change trend and show excellent prediction performance. During the entire prediction period, the prediction curve and the actual traffic curve maintain a good fit, which fully proves the effectiveness and reliability of the present invention in optical network traffic prediction.

[0064] Embodiment 2: This embodiment provides a non-transitory computer-readable storage medium on which computer instructions are stored. The computer instructions enable a computer to execute a method for deploying a software-defined optical network controller based on machine learning. The method includes the following steps: S100, based on a depth-first search strategy, traverse the SDON network topology to find all possible paths between any two nodes in all nodes in the SDON network, and count the transmission delay value of each path; S200, setting a delay threshold, screening out paths that meet the delay constraint condition as reachable paths; when there is only one reachable path between two nodes, the reachable path is used as the control path between the two nodes; when there are multiple reachable paths between the two nodes, the reachable path with the smallest delay is selected as the control path; S300, construct a new network topology diagram of the SDON network, form a new abstract link with the control path; divide two nodes without direct control paths into different network areas; S400, determine the controller location for each network area; and select one from each controller as a control center to complete the deployment of the SDON network controller.

[0065] Embodiment 3: This embodiment provides an electronic device, which may include: a processor, a communications interface, a memory, and a communication bus, wherein the processor, the communications interface, and the memory communicate with each other via the communication bus. The processor may call logic instructions in the memory to execute a deployment method of a software-defined optical network controller based on machine learning, the method comprising the following steps: S100, based on a depth-first search strategy, traverse the SDON network topology to find all possible paths between any two nodes in all nodes in the SDON network, and count the transmission delay value of each path; S200, setting a delay threshold, screening out paths that meet the delay constraint condition as reachable paths; when there is only one reachable path between two nodes, the reachable path is used as the control path between the two nodes; when there are multiple reachable paths between the two nodes, the reachable path with the smallest delay is selected as the control path; S300, construct a new network topology diagram of the SDON network, form a new abstract link with the control path; divide two nodes without direct control paths into different network areas; S400, determine the controller location for each network area; and select one from each controller as a control center to complete the deployment of the SDON network controller.

[0066] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program code.

[0067] Embodiment 4: This embodiment provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute a deployment method of a software-defined optical network controller based on machine learning, and the method includes the following steps: S100, based on a depth-first search strategy, traverse the SDON network topology to find all possible paths between any two nodes in all nodes in the SDON network, and count the transmission delay value of each path; S200, setting a delay threshold, screening out paths that meet the delay constraint condition as reachable paths; when there is only one reachable path between two nodes, the reachable path is used as the control path between the two nodes; when there are multiple reachable paths between the two nodes, the reachable path with the smallest delay is selected as the control path; S300, construct a new network topology diagram of the SDON network, form a new abstract link with the control path; divide two nodes without direct control paths into different network areas; S400, determine the controller location for each network area; and select one from each controller as a control center to complete the deployment of the SDON network controller.

[0068] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0069] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0070] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention 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 invention.

Claims

1. A method for deploying a software-defined optical network controller based on machine learning, characterized in that: include: S100, based on the depth-first search strategy, traverse the SDON network topology to find the SDON network from any starting node To any end node All possible paths, and the transmission delay value of each path is counted ; S200, setting delay threshold ,by As a delay constraint, the path that meets the delay constraint is selected as the reachable path; when there is only one reachable path between two nodes, the reachable path is used as the control path between the two nodes; when there are multiple reachable paths between two nodes, the reachable path with the smallest delay is selected as the control path; S300: construct a new network topology diagram of the SDON network. The construction rules of the new network topology diagram include: (1) Forming a new abstract link with all control paths obtained in step S200; (2) If and If there is no reachable path between the two nodes, the two nodes are divided into nodes in different areas. and If there is a reachable path between them, the two nodes are placed in the same area, thus forming several network areas; S400, determining a controller location for each network area; S500. Select one controller from among the controllers as a control center to coordinate the work among the controllers and complete the deployment of the SDON network controller.

2. The method for deploying a software-defined optical network controller based on machine learning according to claim 1, characterized in that: The specific steps of step S100 include: S110, abstract the SDON network into an undirected graph , is a set of nodes, representing all nodes in the network, where is the number of nodes in the network; is an edge set, representing all the edges in the network, where is the total number of edges; S120, using the stack structure to implement the traversal of the SDON network topology based on the depth-first search strategy; first select the starting node And push it into the stack, then take out the top node of the stack As the current access node, Mark it as visited and add it to the current path list, then traverse the neighbor nodes. Each unvisited neighbor node , calculated from arrive The total path delay is Push it into the stack; finally, check the algorithm termination condition. If is the target node , then record the current path and its total delay value; if If there are no unvisited neighbor nodes, backtrack: Remove from the current path list and mark as unvisited; S130, repeat the traversal process until the stack is empty, and all possible paths are searched; after the traversal is completed, we get arrive All feasible paths and their corresponding transmission delay values .

3. The method for deploying a software-defined optical network controller based on machine learning according to claim 1, characterized in that: The method for determining the controller location for each network area in step S400 is: S410, obtaining structural information and node characteristics of the network topology through the GCN model, and generating an initial solution for controller deployment; S420, the initial solution generated by GCN As a heuristic solution, the simulated annealing algorithm is combined to optimize the final controller position.

4. The method for deploying a software-defined optical network controller based on machine learning according to claim 3, characterized in that: In step S410, the structural information and node characteristics of the network topology are obtained through the GCN model to generate an initial solution for controller deployment; The specific steps are: S411, converting the controller deployment problem into an optimization problem of a network topology graph, extracting features for nodes in the graph, including node degree, node traffic load, and node clustering coefficient; S412. Build a multi-layer GCN model and update the embedding representation of each node through multi-layer graph convolution. After multi-layer GCN calculation, output a controller fitness score to indicate the possibility of each node as a controller. S413, using integer linear programming to generate optimal controller deployment solutions for multiple SDON network topologies as training sets, and training the GCN model constructed in step S412. During the training process, GCN learns the relationship between node features and graph structures, thereby predicting the optimal location for controller deployment; S414: Use the trained GCN model to predict the new network topology of the SDON network constructed in step S300, and calculate the controller fitness; select the top k nodes with the highest fitness as the initial controller deployment plan .

5. The method for deploying a software-defined optical network controller based on machine learning according to claim 4, characterized in that: In step S420, the initial solution generated by GCN As a heuristic solution, the simulated annealing algorithm is combined to optimize the final controller position. The specific steps are: S421, deploy the initial controller generated by GCN As the initial solution, set the initial temperature , the simulated annealing algorithm is applied for iteration. In each iteration, a node is randomly selected from the current solution and the controller deployment state of the node is reversed, that is, the node that originally deployed the controller is not deployed with the controller or the node that did not deploy the controller is deployed with the controller to generate a new solution. ; Check whether the new solution satisfies the coverage constraint, that is, each network node is either a controller or adjacent to at least one controller. If the coverage constraint is not met, add controllers or adjust the position of controllers until the coverage constraint is met; S422, define the objective function of the simulated annealing algorithm to optimize the controller deployment as: ； in is the total number of nodes in the network; Indicates that the node has a controller deployed. Indicates that the node has no controller deployed; is the number of uncovered nodes; is the weight coefficient; New calculation solution The objective function difference between the current solution ,like , then directly accept the new solution Otherwise, according to the probability Accept new interpretations; S423, after each iteration, the temperature is set according to the cooling coefficient Gradually reduce, that is ; When no new solution is accepted for several consecutive times, the algorithm ends and outputs the current solution as the final controller position.

6. The method for deploying a software-defined optical network controller based on machine learning according to claim 1, characterized in that: In step S500, the control center is deployed by selecting the node with the smallest average delay for all controllers. The specific method is: Assume the location of the controller deployment node is , the deployment location of the control center meets: ； in, Indicates taking the average value; Indicates the transmission delay between deployment nodes.

7. The method for deploying a software-defined optical network controller based on machine learning according to claim 6, characterized in that: In step S500, after the control center is deployed on the node with the smallest average delay among all controllers, an auxiliary control center is deployed on the node with the second lowest average delay among all controllers as a backup of the control center.

8. The method for deploying a software-defined optical network controller based on machine learning according to claim 6, characterized in that: The method also includes deploying optical performance monitoring and using machine learning algorithms to detect SDON network anomalies and locate fault points; the specific method is: Interact with the optical performance monitoring system through the southbound interface SBI of the SDON network controller to collect optical performance data and perform standardized processing on the data; Then, an isolation forest model is built and trained for anomaly monitoring, combining abnormal data points with network topology information to locate the location where the fault occurs; The detection results are fed back to the controller of the SDON network to trigger fault recovery operations.

9. A non-transitory computer-readable storage medium, characterized in that: Computer instructions are stored thereon, and the computer instructions enable the computer to execute the method for deploying a software-defined optical network controller based on machine learning as described in any one of claims 1-8.

10. A computer program product, characterized in that The computer program product includes a computer program stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer executes the method for deploying a software-defined optical network controller based on machine learning as described in any one of claims 1 to 8.

Images