Optical network route intelligent planning method and system based on multi-objective optimization

Abstract

The invention relates to an optical network routing intelligent planning method and system based on multi-objective optimization. The method comprises the following steps: performing recursive partitioning on a target optical network to obtain a plurality of sub-network units and a boundary node set; inputting the plurality of sub-network units into a three-layer deep learning model for processing to obtain a routing decision matrix and a wavelength allocation vector; inputting the plurality of sub-network units and the boundary node set into a distributed controller for processing to obtain an error compensation matrix; based on the routing decision matrix and the wavelength allocation vector, performing dual strategy processing of time delay priority and hop count priority to obtain a time delay weighted evaluation function and a hop count weighted evaluation function; performing multi-target joint optimization to obtain an optimal routing scheme; and performing reconstruction cost calculation on the optical path set according to the error compensation matrix and the optimal routing scheme, and outputting a resource reconfiguration scheme. According to the implementation of the invention, the resource overhead in the network reconstruction process is reduced, and the overall operation efficiency of the network is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of optical network routing, and in particular to an optical network routing intelligent planning method and system based on multi-objective optimization. Background Art

[0002] The scale and complexity of services carried by optical networks are growing explosively, and traditional static routing planning methods are difficult to meet the needs of dynamic and changing network environments. Especially in large-scale optical networks, the diversity and time-varying nature of service requirements and the complexity of resource competition pose severe challenges to routing planning.

[0003] The current optical network routing planning methods have the following main problems: the routing decision process lacks deep perception and learning capabilities of the global network status, resulting in inefficient network resource utilization; the centralized processing of the routing control architecture leads to slow system response speed, which is difficult to adapt to the real-time scheduling needs of large-scale networks; the routing optimization goal is single, and it is impossible to take into account multiple performance indicators such as network latency, resource utilization, and service blocking rate at the same time. In addition, when dealing with dynamic network reconstruction, the existing routing planning methods often use simple path reconstruction strategies, which not only causes a large amount of resource waste, but also may cause network oscillation. Summary of the invention

[0004] The main purpose of the present invention is to provide an optical network routing intelligent planning method and system based on multi-objective optimization, which reduces the resource overhead in the network reconstruction process and improves the overall operation efficiency of the network.

[0005] To achieve the above object, the present invention provides an optical network routing intelligent planning method based on multi-objective optimization, comprising the following steps: Recursively partitioning the target optical network to obtain a plurality of sub-network units and a set of boundary nodes; Input the multiple sub-network units into a three-layer deep learning model, process the network topology feature vector, the service demand vector and the network state vector, and obtain a routing decision matrix and a wavelength allocation vector; Inputting the plurality of sub-network units and the set of boundary nodes into a distributed controller, processing status update messages, routing request messages and resource reservation messages between controllers, and obtaining an error compensation matrix; Based on the routing decision matrix and the wavelength allocation vector, the optical path set is processed by a binary search tree and an ordered search tree respectively with a dual strategy of delay priority and hop priority to obtain a delay weighted evaluation function and a hop weighted evaluation function; Constructing a target optimization function based on the delay weighted evaluation function and the hop count weighted evaluation function, performing multi-objective joint optimization, and obtaining an optimal routing solution; According to the error compensation matrix and the optimal routing solution, a reconstruction cost calculation is performed on the optical path set, and a resource reconfiguration solution is output.

[0006] The present invention also provides an optical network routing intelligent planning system based on multi-objective optimization, comprising: A partitioning module, used for recursively partitioning the target optical network to obtain a plurality of sub-network units and a set of boundary nodes; A decision module, used to input the multiple sub-network units into a three-layer deep learning model, process the network topology feature vector, the service demand vector and the network state vector, and obtain a routing decision matrix and a wavelength allocation vector; A compensation module, used for inputting the plurality of sub-network units and the boundary node set into a distributed controller, processing the state update message, the route request message and the resource reservation message between the controllers, and obtaining an error compensation matrix; A processing module, configured to perform dual strategy processing of delay priority and hop priority on the optical path set through a binary search tree and an ordered search tree based on the routing decision matrix and the wavelength allocation vector, and obtain a delay weighted evaluation function and a hop weighted evaluation function; An optimization module, used to construct a target optimization function based on the delay weighted evaluation function and the hop count weighted evaluation function, perform multi-objective joint optimization, and obtain an optimal routing solution; The output module is used to calculate the reconstruction cost of the light path set according to the error compensation matrix and the optimal routing solution, and output a resource reconfiguration solution.

[0007] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0008] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above-mentioned methods are implemented.

[0009] In summary, the technical solution provided by the present invention adopts a recursive partitioning strategy to achieve efficient decomposition of large-scale optical networks and reduce the complexity of routing calculations; the network status is perceived and learned through a three-layer deep learning model, thereby improving the intelligence level of routing decisions; a distributed control mechanism is introduced to achieve rapid response to routing requests; a dual routing strategy based on a binary search tree and an ordered search tree is designed to optimize resource utilization efficiency while ensuring delay performance; a multi-objective joint optimization method is adopted to achieve balanced optimization of multiple performance indicators such as network blocking rate, resource utilization and average delay; a dynamic reconstruction mechanism based on error compensation is proposed to reduce resource overhead in the network reconstruction process and improve the overall operation efficiency of the network. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 It is a schematic diagram of the steps of an optical network routing intelligent planning method based on multi-objective optimization in one embodiment of the present invention; Figure 2 It is a structural block diagram of an optical network routing intelligent planning system based on multi-objective optimization in one embodiment of the present invention; Figure 3 It is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0011] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0012] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0013] Reference Figure 1 This embodiment provides an optical network routing intelligent planning method based on multi-objective optimization, comprising the following steps: S1, recursively partition the target optical network to obtain multiple sub-network units and boundary node sets; Among them, the total number of nodes in the target optical network is counted to obtain the total number of nodes. The node connection relationship in the target optical network is constructed based on the total number of nodes. By analyzing the connection relationship between the nodes, a matrix reflecting the network connection status, namely the network connection matrix, is obtained. In the matrix, the value of each element is 1, indicating that there is a direct optical fiber connection between the nodes, and the value of each element is 0, indicating that there is no direct optical fiber connection between the nodes. The number of direct connections of each node in the network connection matrix is ​​counted to obtain the degree set of the node. The degree of a node reflects the number of direct connections a node has in the network and is a key indicator for measuring the importance of a node in the network. By counting the degrees of all nodes, a node degree set is obtained, which contains the degree information of each node in the network. The nodes are sorted in descending order according to the node degree set, and the nodes with larger degrees are placed in front, and the core nodes with more connections in the network are identified. The node with the largest degree is selected as the center point of the first partition. The directly connected nodes of the partition center point are included in the first sub-network unit to obtain the first sub-network range. The first sub-network unit is composed of the central node with the largest degree and all its directly connected nodes. This partitioning method ensures that nodes with higher degrees and their associated nodes in the network are managed as early as possible, thus forming a relatively compact and stable network substructure. For all remaining nodes outside the scope of the first sub-network, it is necessary to repeat the process of sorting the node degrees in descending order, and select the node with the largest degree as the center point of the next partition. Repeat similar operations to include the directly connected nodes of the central node into the corresponding sub-network units. In this way, the nodes in the network are gradually divided into multiple sub-network units, each of which has a central node and several directly connected nodes. Nodes are marked on the adjacent boundaries of multiple sub-network units. Boundary nodes are nodes that serve as connections between two or more sub-network units. They undertake communication and coordination tasks between different sub-networks, so they need to be clearly marked during the network partitioning process. By marking the adjacent boundaries of multiple sub-network units, a set of boundary nodes is obtained. S2, multiple sub-network units are input into the three-layer deep learning model, the network topology feature vector, the service demand vector and the network state vector are processed to obtain the routing decision matrix and the wavelength allocation vector; Specifically, feature extraction is performed on the node connection relationship, link capacity and physical distance in multiple sub-network units to generate a network topology feature vector. The feature vector describes the physical topology of the optical network, including the connection status between each node, the maximum capacity supported on the link and the geographical distance between nodes. Feature extraction is performed on the source-destination node pairs, bandwidth requirements and delay requirements in multiple sub-network units to obtain a service demand vector. The service demand vector reflects the specific characteristics of each traffic demand in the network, including the relationship between the source node and the destination node pair that need to transmit data, the bandwidth size of each service request, and the specific requirements for service delay. By accurately modeling the service demand, the user's requirements for transmission quality are better met during the routing process, thereby improving the quality of network services and user experience. Feature extraction is performed on the link utilization and service blocking rate of multiple sub-network units to obtain a network state vector. The network state vector is an important parameter that reflects the resource usage in the network in real time, including the current utilization of the link and the service blocking rate caused by insufficient resources or other reasons. In this way, the operation status of the network is dynamically reflected, and the current resource usage is comprehensively considered when calculating the route to improve the accuracy and practicality of the routing decision. The network topology feature vector, business demand vector and network state vector are input into the first deep learning layer to perform routing calculation within the subnetwork. The first deep learning layer contains 128 neurons, which are used to calculate the optimal path set within each subnetwork unit. Through the deep learning model, the complex relationship between network topology, business demand and network state is effectively captured, and then a routing scheme with high performance is calculated within the subnetwork range to generate a subnetwork internal routing matrix. This matrix represents the optimal data transmission path between nodes within each subnetwork unit, thereby realizing routing optimization within a single subnetwork. The subnetwork internal routing matrix and the boundary node set are input into the second deep learning layer for cross-subnetwork routing calculation. The second deep learning layer contains 256 neurons to handle cross-domain routing problems between multiple subnetworks. By combining the information of the boundary nodes with the routing matrix within the subnetwork, the deep learning model can identify the optimal path combination between different subnetworks and calculate the cross-domain routing matrix. The cross-domain routing matrix describes the optimal path combination when transmitting service requests between different subnetworks, thereby ensuring efficient use of network resources in cross-network communication, reducing latency and congestion, and improving the service quality of the overall network. The cross-domain routing matrix is ​​input into the third deep learning layer for global routing optimization. The third deep learning layer contains 512 neurons, which further optimizes the global routing performance indicators to generate the final routing decision matrix. In this process, the deep learning model comprehensively considers the global information of the network, balances the resource allocation between each sub-network, and minimizes the overall network latency and improves the utilization efficiency of the link.By optimizing the global routing performance indicators, the entire optical network is ensured to have high stability and efficiency when processing a large number of business requests, and at the same time, it can effectively reduce network congestion and resource waste caused by unreasonable routing decisions. Wavelength resource allocation calculations are performed for each optical path in the routing decision matrix. The initial wavelength allocation plan is generated based on the available wavelength status of the link. Wavelength allocation is an important part of optical network planning. The establishment of optical paths depends on the reasonable selection of available wavelengths in the link, and the wavelength continuity constraint of the optical network requires that all links on the same path use the same wavelength. After the initial wavelength allocation plan is generated, the wavelength continuity constraint is verified and conflicts are eliminated. If there is a wavelength conflict in the initial plan, the conflict is eliminated by adjusting the wavelength selection to ensure that the final wavelength allocation vector meets the wavelength continuity requirements of the optical network.

[0014] S3, inputting multiple sub-network units and boundary node sets into a distributed controller, processing status update messages, routing request messages, and resource reservation messages between controllers, and obtaining an error compensation matrix; It should be noted that multiple sub-network units are allocated to corresponding local controller nodes, and communication links between controllers are established based on the boundary node set to obtain a distributed controller architecture. The distributed controller architecture can effectively share the management burden of each sub-network, and at the same time, realize the interconnection and coordination between different sub-networks through the communication links between the boundary nodes. The distributed control method makes network control more flexible, and can realize state sharing and mutual cooperation between different local sub-networks, so as to better cope with the dynamic changes and complex needs of the optical network. The control messages in the distributed controller are classified and counted, and these messages are divided into three categories: state update messages, routing request messages and resource reservation messages. By classifying different types of messages, the efficiency of message processing is improved, ensuring that each message can be reasonably processed according to its characteristics. A control message processing queue is established according to the classified messages to effectively manage and schedule messages from different sub-networks, avoid processing confusion caused by the mixing of message types, and improve the response speed and efficiency of the controller. The control message processing queue is synchronized to process the state update message according to the period to obtain the global network state matrix. Periodic synchronization processing can effectively cope with the dynamic changes in the optical network and ensure that the controller has a global perspective on the state of the entire network. Parse the routing request message in the global network state matrix, identify the source node and destination node pair of each service request, and generate the routing request processing sequence based on this information. The routing request processing sequence is used to describe the transmission path planning order of each service request in the network, so as to reasonably arrange the resource usage in the network and try to avoid path conflicts and resource competition. After completing the parsing of the routing request, the routing response matrix is ​​obtained. The routing response matrix is ​​the result of the response to each service request, and describes the transmission path that each service request should choose under the current network state. Based on the routing response matrix, the optical channel resources are reserved and marked to generate the resource reservation matrix. The resource reservation matrix indicates which optical channels and wavelength resources in the network have been allocated for transmitting specific service requests, and which resources are still available. By marking the optical channels with reservations, it is ensured that there are enough resources in the path selection to ensure the actual transmission of data, and avoid service congestion due to insufficient resources during the transmission process. Based on the resource reservation, the deviation between the performance parameters of the actual routing path and the expected routing performance is calculated to obtain the routing deviation matrix. The routing deviation matrix is ​​used to describe the difference between the actual routing performance and the expected performance. The accumulated errors in the route deviation matrix are quantified and counted, and the routes whose error values ​​exceed the set threshold are marked as objects to be reconstructed to obtain the reconstruction target set. The reconstruction target set includes paths that do not meet the performance standards in actual operation. These paths will cause the network performance to deteriorate due to their large deviations, so they need to be replanned. In this way, the unreasonable routes in the network can be gradually corrected to ensure that the overall operation of the network meets expectations.The error compensation matrix is ​​generated according to the reconstruction target set. The error compensation matrix is ​​used to describe the plan for adjusting and reconstructing unreasonable routes in the network, reducing errors by optimizing substandard routes and improving the overall performance of the network.

[0015] S4, based on the routing decision matrix and the wavelength allocation vector, the lightpath set is processed by a binary search tree and an ordered search tree respectively with a dual strategy of delay priority and hop priority, and a delay weighted evaluation function and a hop weighted evaluation function are obtained; Specifically, each optical path in the routing decision matrix is ​​input into a binary search tree for path analysis. The binary search tree is an efficient data structure that can quickly perform node insertion, deletion, and search operations, and is suitable for analyzing each node in the optical path. When analyzing each optical path, the physical distance of each link on the optical path is accumulated to obtain the physical delay set of the path. The physical delay set represents the delay overhead of the optical path at the physical level, which is mainly determined by the distance and speed of the optical signal propagating in the optical fiber. The number of wavelength conversions in the optical path is counted based on the wavelength allocation vector. In the optical network, due to the wavelength continuity constraint, the optical path needs to perform wavelength conversion when passing through different links, and this wavelength conversion will introduce additional delay. After counting the number of wavelength conversions, the wavelength conversion delay is superimposed with the physical delay set of the path to obtain the total delay set of the path. Based on the total delay set of the path, the delay weighting coefficient α is constructed, and the weighted result of the delay and the number of hops is used as the evaluation index to obtain the delay weighted evaluation function f(p)=Σ(ti)+α×h, where ti represents the transmission delay of the i-th link on the optical path, h represents the number of hops passed by the optical path, α is the weighting coefficient of the delay and the number of hops, and p represents the path vector of the optical path. The delay weighted evaluation function is constructed to comprehensively consider the delay and the number of hops of the path, so as to achieve a balance between minimizing the delay and reducing the number of hops when selecting the path. By adjusting the weighting coefficient α, the importance of the delay and the number of hops is adjusted according to the specific network requirements to adapt to the routing requirements in different business scenarios. The optical path set is evaluated with hop priority. Each optical path in the routing decision matrix is ​​input into the ordered search tree for path analysis. The ordered search tree is a data structure that effectively maintains the orderliness of data. By analyzing the path of the optical path and counting the number of relay nodes in the optical path, the hop number set of the path is obtained. The number of path hops is an important indicator for measuring the complexity of the optical path. The more hops, the more relay nodes there are on the path, which increases the complexity of transmission and potential failure points. The link utilization of the optical channel is calculated based on the wavelength allocation vector. Link utilization is a key parameter for measuring the efficiency of optical network resource utilization. By calculating the utilization of each link, the resource occupancy in the current network can be understood. In the path analysis process, the link utilization and the path hop set are weighted to obtain the resource occupancy set of the path. The resource occupancy set reflects the comprehensive occupancy of network resources by the optical path during transmission. By weighting the link utilization and the path hop number, the resource consumption of the optical path in the network is evaluated. The hop weighting coefficient β is constructed for the path resource occupancy set, and the weighted result of the hop number and link utilization is used as the evaluation index to obtain the hop weighted evaluation function g(p)=β×h+Σ(ui), where ui represents the resource utilization of the i-th link on the optical path, h represents the number of hops passed by the optical path, β is the weighting coefficient of the hop number and link utilization, and p represents the path vector of the optical path.The hop-count weighted evaluation function evaluates the resource occupancy of each optical path by comprehensively considering the hop count of the path and the utilization of link resources. By adjusting the weight coefficient β, it can flexibly adapt to different network optimization goals, thereby achieving the optimal configuration of optical network resources.

[0016] S5, constructing a target optimization function based on the delay weighted evaluation function and the hop count weighted evaluation function, performing multi-objective joint optimization, and obtaining an optimal routing solution; Among them, based on the transmission delay ti and weighting coefficient α in the delay weighted evaluation function f(p), the number of blocked services and the total number of services in the optical network are statistically analyzed. By comparing the number of blocked services in the network with the number of all service requests, the first sub-objective function reflecting the network blocking situation is obtained. The role of the first sub-objective function is to quantify the blocking level of the optical network, help identify potential bottlenecks in the network, ensure that blocking can be effectively reduced in the selection of routing solutions, and improve the service quality and responsiveness of the network. At the same time, based on the hop count h and resource utilization ui in the hop count weighted evaluation function g(p), the resource occupancy of the optical channel is statistically analyzed. By calculating the hop count of the optical path relay node and the resource utilization of each link, the second sub-objective function reflecting the level of network resource utilization is obtained. The second sub-objective function is used to quantify the use of resources in the optical network, improve the overall performance of the network through effective management of link resources, and ensure that the optimal utilization of network resources is achieved while meeting business needs. Through these two sub-objective functions, the performance of the network in two key aspects, blocking and resource utilization, is evaluated. The first sub-objective function and the second sub-objective function are combined to construct a joint optimization objective including network blocking rate and resource utilization rate, and the target optimization function is obtained. By weighted combination of the two sub-objectives, the contradictions between different network performance objectives are flexibly balanced to ensure that the final routing solution has good performance overall. The target optimization function is input into the multi-objective genetic algorithm for optimization. The population is initialized to generate an initial solution population. The multi-objective genetic algorithm simulates the genetic and evolutionary processes in nature to find the optimal solution to the problem. The generation of the initial solution population is the starting point of this process and represents several possible routing solutions in the network. After obtaining the initial solution population, the fitness of each individual in the population is calculated. Fitness is an indicator to measure the degree of excellence of each individual under the current network conditions. Through fitness calculation, individuals with excellent performance in the population are identified. Based on the fitness calculation results, individuals are stratified using fast non-dominated sorting and congestion distance. The purpose of fast non-dominated sorting is to classify the population and place those individuals with better performance in each objective at a higher level. The crowding distance is used to measure the diversity of individuals in the same level to ensure that the population can maintain sufficient coverage of the solution space during the evolution process and avoid falling into the local optimal solution. After the stratification is completed, the stratified parent population is obtained, and the parent population contains the better routing solutions in the current network. The stratified parent population is crossover and mutation operations are performed to generate offspring individuals with new feature combinations. The crossover operation is to recombine some genes of the parent individuals to generate offspring with the characteristics of the parent, thereby introducing new solution combinations in the population. The mutation operation introduces random changes to some individuals to increase the diversity of the population and help the algorithm escape the local optimum.Through crossover and mutation, new solutions with different characteristics are continuously generated, thereby expanding the search space of solutions and improving the ability of the algorithm to find the global optimal solution. After competitive selection, the parent population and the child population are optimized. The best performing individuals are selected from the parent and child generations to ensure that the overall fitness of the next generation population can be continuously improved. In this way, the overall performance of the population is gradually optimized. The optimized population is analyzed for Pareto optimal solutions to screen out a solution set that meets the requirements of multi-objective optimization. Each solution in the Pareto optimal solution set is a routing solution that performs balanced on each objective. These solutions are not simultaneously surpassed by other solutions on any objective, so they are considered optimal. A comprehensive performance score is performed on each solution in the candidate routing solution set, and the solution with the highest score is selected as the optimal routing solution based on the scoring results. The comprehensive performance score considers the performance of each routing solution in terms of blocking rate, resource utilization, delay, hop count, etc., and selects the routing solution that best meets the overall performance requirements of the network through a comprehensive measurement of these indicators. The selected optimal routing solution is used as the final planning solution for the optical network to guide the actual network resource configuration and service scheduling.

[0017] S6, based on the error compensation matrix and the optimal routing plan, the reconstruction cost of the optical path set is calculated, and a resource reconfiguration plan is output.

[0018] Specifically, based on the reconstruction target set in the error compensation matrix, the optical paths to be reconstructed are identified to obtain a set of optical paths to be reconstructed. The error compensation matrix contains optical paths that have large differences in performance from the expected performance during actual operation. By identifying these optical paths, the parts that need to be adjusted in the network can be found. For each optical path in the set of optical paths to be reconstructed, the nodes whose link utilization exceeds the threshold η1 are marked to obtain a sequence of reconstructed nodes. The link utilization reflects the use of resources on each link. When the link utilization exceeds the set threshold, it means that the link is overloaded. These nodes need to be marked so that these nodes are given priority in the reconstruction process. Based on the sequence of reconstructed nodes, the number of wavelength conversions of each reconstructed node is counted to obtain a wavelength conversion cost matrix. In an optical network, each wavelength conversion will introduce additional delay and power loss. The number of wavelength conversions of each reconstructed node is counted to quantify the cost of wavelength conversion on the network. The end-to-end delay of each optical path in the set of optical paths to be reconstructed is checked, and those links whose end-to-end delay exceeds the threshold η2 are marked to obtain a sequence of reconstructed links. End-to-end delay is an important indicator of the quality of service of optical networks. If the delay of a certain optical path exceeds the set threshold, it means that there is a problem with the transmission performance of the optical path and it needs to be adjusted and optimized. By marking the links with delays exceeding the threshold, the parts of the network with delay bottlenecks are identified to ensure that these key links are given priority during the reconstruction process to improve the overall transmission performance of the network. The wavelength resource occupancy rate of each link is calculated based on the reconstruction link sequence to obtain the resource occupancy cost matrix. The wavelength resource occupancy rate reflects the use of wavelength resources on each link. When the occupancy rate is high, it means that the available resources of the link are small, which will lead to resource competition and service blocking. The resource occupancy cost matrix is ​​used to quantify the tension of wavelength resources on the link so as to better manage and allocate network resources during the reconstruction process. The reconstruction cost matrix is ​​constructed based on the wavelength conversion cost matrix and the resource occupancy cost matrix. The optical path reconstruction order of the optimal routing solution is optimized and sorted according to the reconstruction cost matrix to obtain the optical path reconstruction sequence. Based on the optical path reconstruction sequence, wavelength resources are reserved and optical path switching operations are performed to ensure that each optical path has sufficient resources for transmission during the reconstruction process to avoid service interruption or service quality degradation due to insufficient resources. Based on the optical path reconstruction sequence and wavelength resource reservation operations, a resource reconfiguration plan is output.

[0019] In one example, the target optical network is recursively partitioned to obtain a plurality of sub-network units and a set of boundary nodes, including: Counting the total number of nodes in the target optical network to obtain the total number of nodes; Based on the total number of nodes, a matrix is ​​constructed for the node connection relationship in the target optical network to obtain a network connection matrix, where a matrix element value of 1 indicates that there is a direct optical fiber connection between nodes, and a matrix element value of 0 indicates that there is no direct optical fiber connection between nodes; Count the number of direct connections of each node in the network connection matrix to obtain the node degree set, sort the nodes in descending order according to the node degree set, and select the node with the largest degree as the first partition center; The directly connected nodes of the partition center point are included in the first sub-network unit to obtain the first sub-network range, and the node degree descending sorting and partition center point selection operations are repeated for the remaining nodes outside the first sub-network range to obtain multiple sub-network units; Nodes are marked on adjacent boundaries of multiple sub-network units to obtain a boundary node set.

[0020] In this example, the total number of nodes in the target optical network is counted to obtain the total number of nodes Traverse the entire optical network topology and summarize all nodes to ensure that each node is correctly identified and counted in the total Based on the total number of nodes , construct a matrix of node connection relationships in the optical network to generate a network connection matrix ,matrix is a A two-dimensional matrix that reflects the connections between nodes in the network. In this matrix, the matrix elements Representation Node and nodes There is a direct fiber optic connection between It means node and nodes For example, suppose there is a simple optical network consisting of The connection relationship between nodes is as follows: Node 1 connects Node 2 and Node 3, Node 2 connects Node 1 and Node 4, Node 3 connects Node 1 and Node 5, Node 4 connects Node 2, and Node 5 connects Node 3. Based on these connection relationships, a network connection matrix is ​​constructed. : ; In this matrix, indicates that there is a direct connection between node 1 and node 2, and It means that there is no direct connection between node 1 and node 4. The number of direct connections of each node in is counted to obtain the node degree set The degree of the node Representation Node The number of direct connections is calculated by the matrix It is obtained by summing the elements of the corresponding rows in . , whose degree is expressed as: ; In the above example, the degree of node 1 is , the degree of node 2 , the degree of node 3 , the degree of node 4 , the degree of node 5 Therefore, the node degree set for: ; According to the node degree set Sort all nodes in descending order to select the node with the largest degree as the center point of the first partition. In the above example, the degrees of nodes 1, 2, and 3 are all the maximum value 2, and any one of them is selected as the center point of the first partition. For example, suppose node 1 is selected as the center point of the first partition. After node 1 is selected as the center point of the first partition, the nodes directly connected to the partition center node 1 are included in the first sub-network unit to form the first sub-network range. The nodes directly connected to node 1 are node 2 and node 3, so the first sub-network unit contains the node set . For the remaining nodes outside the first subnetwork, i.e., nodes 4 and 5, repeat the descending order of node degree sorting and the selection of partition center points to continue to divide the remaining subnetworks. Among the remaining nodes, the degrees of nodes 4 and 5 are both 1, and one of them is selected as the next partition center point. Assuming that node 4 is selected as the second partition center point, the second subnetwork unit contains the node set , and node 5 is divided into the third sub-network unit, the set is . Nodes are marked on the adjacent boundaries of multiple sub-network units to obtain a set of boundary nodes. Boundary nodes refer to nodes that are located between different sub-networks and serve as connections. For example, node 2 belongs to both the first sub-network unit and the Belongs to the second sub-network unit , so node 2 is a boundary node. Similarly, node 3 belongs to the first sub-network unit. It also belongs to the third sub-network unit , so node 3 is also a boundary node. The set of boundary nodes is: ; Through the above steps, the target optical network is recursively partitioned, and the original large-scale network is divided into multiple sub-network units. Each sub-network unit has a relatively simple structure, which is convenient for subsequent management and optimization. At the same time, by marking adjacent boundary nodes, a basis is provided for communication and coordination between different sub-network units.

[0021] In one example, multiple sub-network units are input into a three-layer deep learning model, and the network topology feature vector, the service demand vector, and the network state vector are processed to obtain a routing decision matrix and a wavelength allocation vector, including: Feature extraction is performed on the node connection relationship, link capacity and physical distance of multiple sub-network units to obtain a network topology feature vector; feature extraction is performed on the source-destination node pairs, bandwidth requirements and delay requirements in multiple sub-network units to obtain a service demand vector; feature extraction is performed on the link utilization and service blocking rate of multiple sub-network units to obtain a network state vector; The network topology feature vector, the business demand vector and the network state vector are input into the first deep learning layer to calculate the internal routing of the sub-network. The first deep learning layer contains 128 neurons for calculating the optimal path set within the sub-network unit to obtain the internal routing matrix of the sub-network. The sub-network internal routing matrix and the boundary node set are input into the second deep learning layer for cross-sub-network routing calculation. The second deep learning layer contains 256 neurons for calculating the optimal path combination of cross-domain routing to obtain the cross-domain routing matrix. The cross-domain routing matrix is ​​input into the third deep learning layer for global routing optimization. The third deep learning layer contains 512 neurons for optimizing the global routing performance index to obtain the routing decision matrix. The wavelength resource allocation is calculated for each optical path in the routing decision matrix, an initial wavelength allocation scheme is generated according to the available wavelength status of the link, and wavelength continuity constraint verification and conflict elimination are performed based on the initial wavelength allocation scheme to obtain a wavelength allocation vector.

[0022] In this example, the physical characteristics of the optical network are analyzed. For each sub-network unit, their node connection relationship, the capacity of each link, and the physical distance are considered to form a comprehensive network topology feature vector. Let the network topology feature vector be , which contains multiple elements, such as the connection matrix between nodes , link capacity matrix And the physical distance matrix . Connection Matrix is a The matrix of Representation Node and nodes There is a direct connection between Indicates that there is no direct connection. Link capacity matrix It represents the transmission capacity of each link. Representation Node and nodes The maximum bandwidth that the link between can carry. Physical distance matrix It is used to represent the geographical distance between nodes. Representation Node and nodes For example, for a small network with four nodes, node 1 is connected to node 2 and node 3, node 2 is connected to node 4, and node 3 is connected to node 4. The link capacity and physical distance between nodes are 10 Gbps and 100 km respectively. Based on this information, the following matrix is ​​constructed: Connection Matrix : ; Link Capacity Matrix : ; Physical distance matrix : ; Network topology feature vector Represented as a set of vectors formed by concatenating these matrices: ; Feature extraction is performed on source-destination node pairs, bandwidth requirements, and delay requirements in multiple sub-network units to obtain a service demand vector S. Service demand vector Contains the service requests of each source-destination node pair in the subnetwork. For example, assuming the source-destination node pair is node 1 and node 4, the bandwidth requirement is 5 Gbps, and the latency requirement is 50 milliseconds, it is expressed in the following form: ; Among them, (1,4) represents the source-destination node pair, Indicates bandwidth requirement, equal to 5Gbps, Indicates the delay requirement, which is equal to 50 milliseconds. At the same time, feature extraction is performed on the link utilization and service blocking rate of multiple sub-network units to obtain the network state vector The network state vector is used to describe the current network resource usage and transmission performance, including the link utilization matrix and service blocking rate matrix . Link Utilization Matrix Elements Representation Node and nodes Bandwidth usage ratio on the link between them, service blocking rate matrix Elements This means that at the node and nodes The proportion of business requests blocked between . The network state vector is expressed as: ; The network topology feature vector , Business Requirements Vector and the network state vector Input to the first deep learning layer for routing calculation within the sub-network. The first deep learning layer contains 128 neurons, which are used to calculate the optimal path set within the sub-network unit. Assume that the input vector is , the weight matrix of the first layer of neurons is , the bias is , the output is expressed as: ; in, represents the activation function, such as ReLU or Sigmoid, is the feature vector obtained after processing by 128 neurons, which is used to describe the optimal path set within the sub-network. Based on this output, the routing matrix within the sub-network is obtained It describes the best transmission path between nodes in the current sub-network. and the set of boundary nodes Input to the second deep learning layer for cross-subnetwork routing calculation. The second deep learning layer contains 256 neurons, which are used to calculate the optimal path combination for cross-domain routing. Similarly, let the input vector be , the weight matrix of the second layer is , the bias is , the output is expressed as: ; This output Represents the path combination across sub-networks to form a cross-domain routing matrix , which describes the optimal path between different sub-networks. The data is input into the third deep learning layer for global routing optimization. The third layer contains 512 neurons, and the final routing decision matrix is ​​obtained by optimizing the global routing performance index. . Let the input vector be , the weight matrix is , the bias is , the output is expressed as: ; Final Output Represents the optimized global routing solution, forming a routing decision matrix , which contains the global optimal optical network routing decision. After obtaining the global routing decision matrix, the wavelength resource allocation calculation is performed for each optical path. The initial wavelength allocation scheme is generated based on the available wavelength status of the link. Assume that the available wavelength status on the link is , indicating a node and nodes The initial wavelength allocation scheme counts and matches the wavelength resources of each link to ensure that each optical path obtains sufficient wavelength resources during transmission. However, wavelength allocation in optical networks needs to satisfy the wavelength continuity constraint, that is, all links on the same path must use the same wavelength. After the initial scheme is generated, the wavelength continuity constraint is verified and conflicts are eliminated. Assuming that the optical path The set of links passed is , then the wavelength continuity of the optical path is expressed as: ; if , it means that the optical path has found a commonly available wavelength; otherwise, the wavelength needs to be adjusted until the wavelength continuity requirement is met. Through wavelength continuity constraint verification and conflict elimination, the wavelength allocation vector of each optical path is obtained. , where the elements Indicates the optical path The assigned wavelength.

[0023] In one example, multiple sub-network units and a set of boundary nodes are input into a distributed controller, and status update messages, route request messages, and resource reservation messages between controllers are processed to obtain an error compensation matrix, including: Allocating local controller nodes to multiple sub-network units, and establishing communication links between controllers based on a set of boundary nodes to obtain a distributed controller; Classify and count the control messages in the distributed controller, and establish a control message processing queue based on status update messages, routing request messages, and resource reservation messages; The control message processing queue is used to synchronously process the status update message according to the period to obtain the global network status matrix; Parse the routing request message in the global network state matrix, and generate a routing request processing sequence according to the source and destination node pairs to obtain a routing response matrix; Based on the routing response matrix, optical channel resources are reserved and marked to obtain a resource reservation matrix, and based on the resource reservation matrix, the deviation between the performance parameter of the actual routing path and the expected routing performance is calculated to obtain a routing deviation matrix; The accumulated errors in the routing deviation matrix are quantitatively counted, and the routes whose error values ​​exceed the set threshold are marked as objects to be reconstructed, and a reconstruction target set is obtained. The error compensation matrix is ​​generated according to the reconstruction target set.

[0024] In this example, local controller nodes are assigned to multiple sub-network units, and communication links between controllers are established based on the boundary node set to obtain a complete distributed controller architecture. The purpose of the distributed controller is to manage routing decisions, resource allocation, and state synchronization within each sub-network unit to ensure the coordinated operation of the entire network. Assume that each sub-network unit is ( ), each subnetwork Each node is assigned a local controller. These local controllers are responsible for monitoring and managing the nodes and links in their sub-networks. It contains nodes connecting different sub-networks. Communication links between controllers are established between these boundary nodes to transmit network status and routing information between controllers. In this way, a distributed controller architecture is formed, so that each sub-network unit is managed by a local controller, and different sub-networks communicate and collaborate through boundary nodes. After the distributed controller is established, the control messages received in the controller are classified and counted. Control messages are divided into three categories: status update messages, route request messages, and resource reservation messages. Each message type has its specific purpose and processing method. Status update messages are used to update the topological status and resource utilization of the current network, route request messages are used to request data transmission paths in the network, and resource reservation messages are used to confirm and lock the required network resources. By classifying these control messages, corresponding control message processing queues are established. , , , respectively storing status update messages, routing request messages and resource reservation messages. The establishment of the control message processing queue effectively manages the complex control information flow in the network, so that each message can be efficiently processed according to its characteristics. After the control message processing queue is established, the status update messages are synchronously processed according to the cycle to obtain the global network status matrix . Global network state matrix It is used to describe the link utilization, node status, topology and other information in the entire network. For one The matrix of Representation Node and nodes The link status between the two networks includes the current link utilization, bandwidth availability, and latency. By periodically synchronizing status update messages, the information in the global network status matrix is ​​ensured to be up to date. Parse the routing request message in the source node and generate a routing request processing sequence according to the source node and the destination node to obtain the routing response matrix . Assume that the routing request message contains the source node and destination node , and corresponding business requirements (such as bandwidth and latency requirements). By parsing these requests, a sequence containing source and destination node pairs is generated. For each source-destination node pair , based on the global network state matrix Calculate the optimal routing path and form a routing response matrix ,in Represents a slave node To Node The optimal path and its corresponding resource allocation. Based on the obtained routing response matrix , reserve the optical channel resources to obtain the resource reservation matrix . Resource reservation matrix It is a matrix used to describe the resource reservation of each link in the network, where Representation Node and nodes The wavelength and bandwidth resources that have been reserved on the link between them. The resource reservation operation ensures that each service request in the network has sufficient resource support during the transmission process, thereby avoiding service interruption due to insufficient resources during the transmission process. After generating the resource reservation matrix, the performance parameters of the actual routing path are calculated based on the matrix and compared with the expected routing performance to obtain the routing deviation matrix . Route Deviation Matrix It is used to describe the deviation between the actual routing performance (such as delay, bandwidth utilization, etc.) and the expected performance. The elements in the matrix Representation Node and nodes The actual routing performance is different from the expected one. To better manage the performance deviation in the network, the routing deviation matrix The cumulative error in the quantified statistics is counted, and the error value exceeding the set threshold is The route is marked as the object to be reconstructed, and the reconstruction target set is obtained .set up Representation Node and nodes If the routing error between , the route is marked as an object to be reconstructed and added to the reconstruction target set In this way, the paths in the network that do not meet the performance standards are identified and reconstructed to improve the overall performance of the network. Generate error compensation matrix . Error compensation matrix It is used to describe how to adjust the paths to be reconstructed to eliminate performance deviations. Representation Node and nodes The routing between the two networks needs to be adjusted, such as increasing bandwidth, reducing delay, optimizing wavelength allocation, etc. By executing the error compensation matrix, the accumulated error in the network is gradually reduced, making the network performance gradually tend to the optimal state.

[0025] In one example, based on the routing decision matrix and the wavelength allocation vector, the optical path set is processed by a binary search tree and an ordered search tree respectively with a dual strategy of delay priority and hop priority, and a delay weighted evaluation function and a hop weighted evaluation function are obtained, including: Each optical path in the routing decision matrix is ​​input into a binary search tree for path analysis, and the physical distances between nodes are accumulated to obtain a set of path physical delays; The number of wavelength conversions in the optical path is counted based on the wavelength allocation vector, and the wavelength conversion delay is superimposed on the path physical delay set to obtain the path total delay set; The delay weighting coefficient α is constructed for the total path delay set, and the weighted result of delay and hop number is used as the evaluation index to obtain the delay weighted evaluation function f(p)=Σ(ti)+α×h, where ti represents the transmission delay of the i-th link on the optical path, h represents the number of hops passed by the optical path, α is the weighting coefficient of delay and hop number, and p represents the path vector of the optical path; Each optical path in the routing decision matrix is ​​input into an ordered search tree for path analysis, the number of optical path relay nodes is counted, and a path hop count set is obtained; The link utilization of the optical channel is calculated based on the wavelength allocation vector, and the link utilization is weighted with the path hop count set to obtain the path resource occupancy set; The hop weighting coefficient β is constructed for the path resource occupancy set, and the weighted result of the hop count and the link utilization is used as the evaluation index to obtain the hop weighted evaluation function g(p)=β×h+Σ(ui), where ui represents the resource utilization of the i-th link on the optical path, h represents the number of hops passed by the optical path, β is the weighting coefficient of the hop count and the link utilization, and p represents the path vector of the optical path.

[0026] In this example, for each light path in the routing decision matrix, a binary search tree is input for path analysis, and the physical distances between nodes are accumulated to obtain the physical delay set of each path. Suppose the light path in the routing decision matrix is , the set of nodes on the path is , binary search tree is used to quickly find the distance information between nodes and accumulate the physical distance between nodes Get the physical delay of the path ,Right now: ; in, Representation Node and nodes By making the optical path Input the binary search tree to efficiently query and accumulate the distance between nodes to obtain the physical delay set of each optical path. ,in Indicates the optical path After obtaining the physical delay of the path, the number of wavelength conversions in the optical path is counted based on the wavelength allocation vector. Wavelength conversion in optical networks is a very important process. When an optical path passes through different links, if the same wavelength cannot be maintained, wavelength conversion is required. The wavelength conversion operation usually brings additional delay overhead. The wavelength allocation vector is , the number of wavelength conversions is , the delay introduced by wavelength conversion is Therefore, the total delay of the path is By adding the physical delay and the wavelength conversion delay, we get: ; Through this step, the total delay set of all optical paths is obtained. . Weight the total delay set of the path, and use the delay and hop count as comprehensive evaluation indicators. Construct the delay weighting coefficient , and weighted combination of delay and hop count. The number of hops passed is , the first The transmission delay of the link is , the delay weighted evaluation function is expressed as: ; in, Indicates the optical path The comprehensive evaluation value of is the weighted coefficient of delay and hop count, which is used to adjust the importance of delay and hop count in routing selection. The value of , depending on the application scenario, focuses on minimizing the delay or reducing the number of hops to achieve different optimization goals. Each light path in the routing decision matrix is ​​input into the ordered search tree for path analysis to count the number of relay nodes in the light path and obtain the set of hops of the path. The ordered search tree is a data structure that efficiently sorts and searches for node information. The number of relay nodes in is By inputting the ordered search tree, the number of hops of each light path is quickly counted to obtain the hop count set of the path ,in Indicates the optical path After obtaining the path hop count set, the link utilization of the optical channel is calculated based on the wavelength allocation vector. Suppose the link utilization is , which indicates the resource occupancy of the i-th link on the optical path. The higher the utilization rate, the more scarce the resources of the link. , calculate the sum of resource utilization of all links on it, and get the resource occupancy set of the path ,in Indicates the optical path The total resource utilization of the path resource occupancy set is weighted to construct a hop-weighted evaluation function. The hop-weighted coefficient is introduced , and weightedly combine the number of hops and link utilization. The number of hops is , the link utilization is , the hop-count weighted evaluation function is expressed as: ; in, Indicates the optical path The resource occupancy evaluation value of is the weighted coefficient of hop count and link utilization, which is used to adjust the importance of hop count and resource utilization in path selection. The value of can be used to flexibly balance resource usage and hop count during path selection to improve network resource utilization efficiency.

[0027] In one example, a target optimization function is constructed based on a delay weighted evaluation function and a hop weighted evaluation function, and a multi-objective joint optimization is performed to obtain an optimal routing solution, including: Based on the transmission delay ti and the weighting coefficient α in the delay weighted evaluation function f(p), the number of blocked services and the total number of services in the optical network are statistically analyzed to obtain the first sub-objective function reflecting the network blocking situation; Based on the hop count h and resource utilization ui in the hop count weighted evaluation function g(p), the resource occupancy of the optical channel is statistically analyzed to obtain the second sub-objective function reflecting the level of network resource utilization; Performing a combined operation on the first sub-objective function and the second sub-objective function, constructing a joint optimization objective including a network blocking rate and a resource utilization rate, and obtaining an objective optimization function; Input the target optimization function into the multi-objective genetic algorithm, initialize the population, obtain the initial solution population, calculate the fitness of each individual in the initial solution population, stratify the individuals based on fast non-dominated sorting and crowding distance, and obtain the stratified parent population; Perform crossover and mutation operations on the stratified parent population to generate offspring individuals with new feature combinations, and conduct competitive selection between the parent population and the offspring population to obtain the optimized population; The Pareto optimal solution analysis is performed on the optimized population to screen out the solution set that meets the multi-objective optimization requirements, obtain the candidate routing solution set, and perform a comprehensive performance score on each solution in the candidate routing solution set, and select the solution with the highest score as the optimal routing solution.

[0028] In this example, the delay weighted evaluation function ,in Indicates the first The transmission delay of the link, Indicates the number of hops the light path has passed through. is the weighted coefficient of delay and hop count. In order to understand the congestion of the network, based on the transmission delay and weighted coefficient, the number of blocked services and the total number of services in the optical network are statistically analyzed to obtain the first sub-objective function . Assume the total number of services in the optical network is , the number of blocked services is , the network blocking rate is expressed as: ; The network blocking rate reflects the situation where service needs cannot be met due to high latency or insufficient resources. By combining the statistics of the latency weighted evaluation function, we can calculate which services are blocked due to large transmission latency and obtain the blocking rate of the entire network. . Hop-based weighted evaluation function Perform statistical analysis on the resource usage of the optical channel. Hop-count weighted evaluation function ,in is the weighted coefficient of hop count and link utilization, h represents the number of hops the optical path passes through, Indicates the first By analyzing all the optical paths in the optical network and counting the resource occupancy level of the optical channel, the second sub-objective function reflecting the network resource utilization is obtained. . Let the total resources of all links in the network be , the occupied resources are , the network resource utilization is expressed as: ; Network resource utilization It reflects the overall usage of each link resource in the optical network and is an important measure of network efficiency and resource management. Through statistics based on hop count and link utilization, the resource occupancy in the network is obtained. and the second sub-objective function Perform combinatorial operations to construct a joint optimization objective that includes network blocking rate and resource utilization, and obtain the target optimization function The objective optimization function is expressed as: ; in, and are the weight coefficients of network blocking rate and resource utilization, respectively, which are used to adjust the relative importance of these two sub-goals in joint optimization. and The value of can be used to flexibly optimize the performance of the optical network according to different business requirements and network conditions. Input the multi-objective genetic algorithm for optimization. The multi-objective genetic algorithm is an optimization algorithm that simulates the biological evolution process, initializes the population, and generates the initial solution population. , each individual represents a possible routing solution in the optical network. Let the population size be , each individual ( ) represents a routing scheme. The fitness of each individual in the initial solution population is calculated, and the fitness value is used to measure the pros and cons of the individual in the current environment. The fitness value is calculated based on the target optimization function , the goal is to minimize the blocking rate and resource occupancy level, so as to obtain the optimal routing solution. After calculating the fitness of the initial population, the individuals in the population are sorted. The individuals are stratified using the fast non-dominated sorting and crowding distance methods. Fast non-dominated sorting is used to divide individuals in the population into different levels, and individuals in higher levels are those that perform better on all objectives. The crowding distance is used to measure the diversity between individuals in the same level to ensure that the population can maintain sufficient coverage of the solution space during the evolution process. Through the stratification method, the stratified parent population is obtained , where individuals at higher levels will be preferentially selected for subsequent evolutionary operations. The stratified parent population is subjected to crossover and mutation operations to generate offspring individuals with new feature combinations. The crossover operation is to combine some genes of two parent individuals to generate new individuals with parental characteristics. Let the parent individual be and , crossover operation generates offspring individuals The process is expressed as: ; The mutation operation introduces random changes to the offspring individuals to increase the diversity of the population and prevent the algorithm from falling into the local optimum. The mutation operation is expressed as: ; Through crossover and mutation, new individuals with new features are generated, thereby expanding the search space for solutions and improving the algorithm's ability to find the global optimal solution. After the generation of offspring individuals, the parent population and the offspring population are competitively selected to obtain the optimized population. The best performing individuals are selected from the parent and offspring generations to ensure that the overall fitness of the next generation population can be continuously improved. Through this selection mechanism, the quality of the population is gradually improved, making the individuals in the population closer and closer to the optimal solution. Pareto optimal solution analysis is performed to screen out the solution set that meets the requirements of multi-objective optimization. Each solution in the Pareto optimal solution set is a routing solution that performs balanced on all objectives. These solutions are not simultaneously surpassed by other solutions on any one objective, so they are considered optimal. Let the Pareto optimal solution set be , and then each solution in the solution set is scored for comprehensive performance. The comprehensive performance score considers the performance of each solution in terms of blocking rate, resource utilization, delay, number of hops, etc. The solution with the highest score is selected as the final optimal routing solution. .

[0029] In one example, based on the error compensation matrix and the optimal routing solution, the reconstruction cost of the optical path set is calculated, and a resource reconfiguration solution is output, including: The optical path to be reconstructed is identified based on the reconstruction target set in the error compensation matrix to obtain the optical path set to be reconstructed; The nodes whose link utilization rate of each lightpath in the lightpath set to be reconstructed exceeds the threshold η1 are marked to obtain a reconstructed node sequence; Based on the reconstruction node sequence, the wavelength conversion times of each reconstruction node are counted to obtain the wavelength conversion cost matrix; Mark the link whose end-to-end delay of each lightpath in the lightpath set to be reconstructed exceeds the threshold η2, and obtain a reconstructed link sequence; Based on the reconstructed link sequence, the wavelength resource occupancy rate of each link is calculated to obtain a resource occupancy cost matrix, and a reconstruction cost matrix is ​​constructed based on the wavelength conversion cost matrix and the resource occupancy cost matrix; The optical path reconstruction order of the optimal routing solution is optimized and sorted according to the reconstruction cost matrix to obtain the optical path reconstruction sequence, and wavelength resource reservation and optical path switching operations are performed based on the optical path reconstruction sequence to output a resource reconfiguration solution.

[0030] In this example, the light paths to be reconstructed are identified based on the reconstruction target set in the error compensation matrix to obtain a set of light paths to be reconstructed. It describes the optical paths that need to be reconstructed due to performance deviations caused by various factors during network operation. Let the reconstruction target set be , contains all the light paths that need to be reconstructed. By traversing the error compensation matrix, the light paths whose performance does not meet expectations are identified and added to the light path set to be reconstructed For example, when the delay deviation and resource utilization deviation of a light path exceed a certain threshold, the light path is considered to need to be reconstructed. After identifying the set of light paths to be reconstructed, the link utilization of each of these light paths exceeds the threshold. The nodes are marked to obtain the reconstructed node sequence. Link utilization Representation Node and nodes The resource occupancy of the link between them. When the link utilization exceeds the set threshold When , it means that the link is in an overloaded state and needs to be adjusted first. Suppose the set of optical paths to be reconstructed is , for each light path , traverse all the links it passes through and mark the link utilization Nodes, get the reconstructed node sequence ,in Represents all nodes that need to be reconstructed. Based on the reconstructed node sequence , count the number of wavelength conversions of each reconstruction node to obtain the wavelength conversion cost matrix Representation Node The number of wavelength conversions performed in the optical path. Wavelength conversion is a resource-consuming operation that introduces additional delay and power loss. The wavelength conversion cost is ,in is the cost of each wavelength conversion, usually given in the form of delay or power loss. Therefore, the wavelength conversion cost matrix Represented as a A matrix where each element Representation Node and nodes At the same time, for each optical path in the set of optical paths to be reconstructed, the end-to-end delay exceeds the threshold The links are marked to obtain the reconstructed link sequence. It is an important indicator to measure the performance of optical path transmission. When the delay exceeds the set threshold When , it indicates that the transmission performance of the optical path is poor and needs to be optimized. , traverse all its links and mark those with end-to-end delay The link sequence of reconstructed links is obtained ,in Indicates all links that need to be reconstructed. Based on the reconstructed link sequence , calculate the wavelength resource occupancy rate of each link and obtain the resource occupancy cost matrix Wavelength resource occupancy rate Representation Node and nodes The wavelength resource occupancy of the links between them and the resource occupancy cost of each link are calculated ,in represents the cost of unit wavelength resources. Therefore, the resource occupation cost matrix A matrix where each element Indicates link and Link The total cost of resource occupation between. Based on the wavelength conversion cost matrix Construct reconstruction cost matrix The reconstruction cost matrix is ​​used to comprehensively consider the cost of wavelength conversion and resource occupation to measure the overall cost of the reconstruction operation. Let the elements in the reconstruction cost matrix be Indicates link and Link The reconstruction cost is expressed as: ; in, and are the weight coefficients of wavelength conversion cost and resource occupation cost, respectively, which are used to adjust the relative importance of these two costs in the reconstruction decision. and The value of , according to the specific network requirements, focuses on different reconstruction optimization goals. According to the reconstruction cost matrix The order of light path reconstruction of the optimal routing solution is optimized and sorted to obtain a light path reconstruction sequence. Represents the reconstruction priority of all lightpaths to be reconstructed. By sorting the costs in the reconstruction cost matrix, lightpaths with lower costs are selected for reconstruction first, thereby reducing the overall reconstruction cost. Suppose each lightpath in the lightpath reconstruction sequence is , the reconstruction order is obtained by sorting the reconstruction costs: ; Based on the optical path reconstruction sequence, wavelength resource reservation and optical path switching operations are performed to output a resource reconfiguration plan. Wavelength resource reservation means that before the optical path reconstruction, sufficient wavelength resources are first reserved for each optical path to ensure that there is no service interruption due to insufficient resources during the reconstruction process. The optical path switching operation is to adjust the transmission path of the optical signal during the reconstruction process to achieve switching from the original path to the new path. Resource reconfiguration plan This paper describes how to reallocate wavelength resources and adjust optical paths in optical networks to achieve optimal network performance.

[0031] Reference Figure 2 This embodiment provides an optical network routing intelligent planning system based on multi-objective optimization, including: Partitioning module 1, used for recursively partitioning the target optical network to obtain a plurality of sub-network units and a set of boundary nodes; Decision module 2 is used to input multiple sub-network units into a three-layer deep learning model, process the network topology feature vector, the service demand vector and the network state vector, and obtain a routing decision matrix and a wavelength allocation vector; The compensation module 3 is used to input a plurality of sub-network units and a set of boundary nodes into a distributed controller, process the state update message, the routing request message and the resource reservation message between the controllers, and obtain an error compensation matrix; Processing module 4 is used to perform dual strategy processing of delay priority and hop priority on the optical path set through a binary search tree and an ordered search tree based on the routing decision matrix and the wavelength allocation vector, and obtain a delay weighted evaluation function and a hop weighted evaluation function; Optimization module 5, used to construct a target optimization function based on the delay weighted evaluation function and the hop weighted evaluation function, perform multi-objective joint optimization, and obtain the optimal routing solution; The output module 6 is used to calculate the reconstruction cost of the optical path set according to the error compensation matrix and the optimal routing solution, and output the resource reconfiguration solution.

[0032] In this embodiment, for the specific implementation of each unit in the above system embodiment, please refer to the above method embodiment, which will not be repeated here.

[0033] Reference Figure 3In an embodiment of the present invention, a computer device is also provided. The computer device may be a server, and its internal structure may be as follows: Figure 3 As shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0034] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0035] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0036] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided by the present invention and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.

[0037] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, system, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, system, article or method. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the presence of other identical elements in the process, system, article or method including the element.

[0038] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. An optical network routing intelligent planning method based on multi-objective optimization, characterized in that: The following steps are involved: Recursively partitioning the target optical network to obtain a plurality of sub-network units and a set of boundary nodes; Input the multiple sub-network units into a three-layer deep learning model, process the network topology feature vector, the service demand vector and the network state vector, and obtain a routing decision matrix and a wavelength allocation vector; Inputting the plurality of sub-network units and the set of boundary nodes into a distributed controller, processing status update messages, routing request messages and resource reservation messages between controllers, and obtaining an error compensation matrix; Based on the routing decision matrix and the wavelength allocation vector, the optical path set is processed by a binary search tree and an ordered search tree respectively with a dual strategy of delay priority and hop priority to obtain a delay weighted evaluation function and a hop weighted evaluation function; Constructing a target optimization function based on the delay weighted evaluation function and the hop count weighted evaluation function, performing multi-objective joint optimization, and obtaining an optimal routing solution; According to the error compensation matrix and the optimal routing solution, a reconstruction cost calculation is performed on the optical path set, and a resource reconfiguration solution is output.

2. The optical network routing intelligent planning method based on multi-objective optimization according to claim 1 is characterized in that: The target optical network is recursively partitioned to obtain a plurality of sub-network units and a set of boundary nodes, including: Counting the total number of nodes in the target optical network to obtain the total number of nodes; Based on the total number of nodes, a matrix is ​​constructed for the node connection relationship in the target optical network to obtain a network connection matrix, wherein a matrix element value of 1 indicates that there is a direct optical fiber connection between the nodes, and a matrix element value of 0 indicates that there is no direct optical fiber connection between the nodes; Counting the number of direct connections of each node in the network connection matrix to obtain a node degree set, and sorting the nodes in descending order according to the node degree set, and selecting the node with the largest degree as the first partition center point; Incorporate directly connected nodes of the partition center into the first sub-network unit to obtain a first sub-network range, and repeat the node degree descending order sorting and partition center point selection operations for the remaining nodes outside the first sub-network range to obtain multiple sub-network units; Node marking is performed on adjacent boundaries of the plurality of sub-network units to obtain a boundary node set.

3. The optical network routing intelligent planning method based on multi-objective optimization according to claim 2 is characterized in that: The step of inputting the plurality of sub-network units into a three-layer deep learning model, processing a network topology feature vector, a service demand vector, and a network state vector to obtain a routing decision matrix and a wavelength allocation vector includes: Extracting features of node connection relationships, link capacities, and physical distances of the multiple sub-network units to obtain a network topology feature vector; extracting features of source-destination node pairs, bandwidth requirements, and latency requirements in the multiple sub-network units to obtain a service demand vector; extracting features of link utilization and service blocking rates of the multiple sub-network units to obtain a network state vector; Input the network topology feature vector, the service demand vector and the network state vector into a first deep learning layer to perform sub-network internal routing calculation, wherein the first deep learning layer includes 128 neurons for calculating the optimal path set within the sub-network unit to obtain a sub-network internal routing matrix; Inputting the sub-network internal routing matrix and the boundary node set into the second deep learning layer to perform cross-sub-network routing calculation, wherein the second deep learning layer includes 256 neurons for calculating the optimal path combination of cross-domain routing to obtain a cross-domain routing matrix; Inputting the cross-domain routing matrix into a third deep learning layer for global routing optimization, wherein the third deep learning layer comprises 512 neurons for optimizing global routing performance indicators, and obtaining a routing decision matrix; A wavelength resource allocation calculation is performed on each optical path in the routing decision matrix, an initial wavelength allocation scheme is generated according to the available wavelength status of the link, and wavelength continuity constraint verification and conflict elimination are performed based on the initial wavelength allocation scheme to obtain a wavelength allocation vector.

4. The optical network routing intelligent planning method based on multi-objective optimization according to claim 3 is characterized in that: The step of inputting the plurality of sub-network units and the boundary node set into a distributed controller, processing the status update message, the routing request message and the resource reservation message between the controllers, and obtaining an error compensation matrix includes: Allocating local controller nodes to the multiple sub-network units, and establishing communication links between controllers based on the boundary node set to obtain a distributed controller; Classify and count the control messages in the distributed controller, and establish a control message processing queue according to the status update message, the routing request message and the resource reservation message; The control message processing queue synchronously processes the status update message according to a period to obtain a global network status matrix; Parsing the routing request message in the global network state matrix, and generating a routing request processing sequence according to source and destination node pairs to obtain a routing response matrix; Based on the routing response matrix, optical channel resources are reserved and marked to obtain a resource reservation matrix, and based on the resource reservation matrix, deviations between performance parameters of actual routing paths and expected routing performance are calculated to obtain a routing deviation matrix; The accumulated errors in the routing deviation matrix are quantitatively counted, and the routes whose error values ​​exceed a set threshold are marked as objects to be reconstructed, so as to obtain a reconstruction target set, and an error compensation matrix is ​​generated according to the reconstruction target set.

5. The optical network routing intelligent planning method based on multi-objective optimization according to claim 4 is characterized in that: Based on the routing decision matrix and the wavelength allocation vector, the optical path set is processed by a binary search tree and an ordered search tree respectively with a dual strategy of delay priority and hop priority to obtain a delay weighted evaluation function and a hop weighted evaluation function, including: Input each optical path in the routing decision matrix into a binary search tree for path analysis, accumulate the physical distances between nodes, and obtain a path physical delay set; Counting the number of wavelength conversions in the optical path based on the wavelength allocation vector, and superimposing the wavelength conversion delay with the path physical delay set to obtain a path total delay set; A delay weighting coefficient α is constructed for the total delay set of the path, and the weighted result of the delay and the number of hops is used as an evaluation index to obtain a delay weighted evaluation function f(p)=Σ(ti)+α×h, where ti represents the transmission delay of the i-th link on the optical path, h represents the number of hops passed by the optical path, α is the weighting coefficient of the delay and the number of hops, and p represents the path vector of the optical path; Input each optical path in the routing decision matrix into an ordered search tree for path analysis, count the number of optical path relay nodes, and obtain a path hop count set; Calculating the link utilization of the optical channel based on the wavelength allocation vector, and weighting the link utilization and the path hop count set to obtain a path resource occupancy set; A hop weighting coefficient β is constructed for the path resource occupancy set, and the weighted result of the hop count and the link utilization is used as an evaluation index to obtain a hop weighted evaluation function g(p)=β×h+Σ(ui), wherein ui represents the resource utilization of the i-th link on the optical path, h represents the number of hops passed by the optical path, β is the weighting coefficient of the hop count and the link utilization, and p represents the path vector of the optical path.

6. The optical network routing intelligent planning method based on multi-objective optimization according to claim 5 is characterized in that: The objective optimization function is constructed based on the delay weighted evaluation function and the hop weighted evaluation function, and multi-objective joint optimization is performed to obtain the optimal routing solution, including: Based on the transmission delay ti and the weighting coefficient α in the delay weighted evaluation function f(p), statistical analysis is performed on the number of blocked services and the total number of services in the optical network to obtain a first sub-objective function reflecting the network blocking situation; Based on the hop count h and resource utilization rate ui in the hop count weighted evaluation function g(p), statistical analysis is performed on the resource occupancy of the optical channel to obtain a second sub-objective function reflecting the level of network resource utilization; Performing a combined operation on the first sub-objective function and the second sub-objective function to construct a joint optimization objective including a network blocking rate and a resource utilization rate, and obtaining an objective optimization function; Inputting the objective optimization function into a multi-objective genetic algorithm, initializing the population to obtain an initial solution population, calculating the fitness of each individual in the initial solution population, stratifying the individuals based on fast non-dominated sorting and crowding distance, and obtaining a stratified parent population; Performing crossover and mutation operations on the stratified parent population to generate offspring individuals with a new feature combination, and performing competitive selection on the parent population and the offspring population to obtain an optimized population; A Pareto optimal solution analysis is performed on the optimized population to screen out a solution set that meets the multi-objective optimization requirements, obtain a set of candidate routing solutions, and perform a comprehensive performance score on each solution in the set of candidate routing solutions, and select the solution with the highest score as the optimal routing solution.

7. The optical network routing intelligent planning method based on multi-objective optimization according to claim 6 is characterized in that: The step of calculating the reconstruction cost of the optical path set according to the error compensation matrix and the optimal routing solution and outputting a resource reconfiguration solution comprises: Identifying the light paths to be reconstructed based on the reconstruction target set in the error compensation matrix to obtain a light path set to be reconstructed; Marking nodes whose link utilization rate of each lightpath in the lightpath set to be reconstructed exceeds a threshold value η1 to obtain a reconstructed node sequence; Based on the reconstruction node sequence, the number of wavelength conversions of each reconstruction node is counted to obtain a wavelength conversion cost matrix; Marking the links whose end-to-end delay of each optical path in the set of optical paths to be reconstructed exceeds a threshold value η2 to obtain a reconstructed link sequence; Based on the reconstructed link sequence, the wavelength resource occupancy rate of each link is calculated to obtain a resource occupancy cost matrix, and a reconstruction cost matrix is ​​constructed based on the wavelength conversion cost matrix and the resource occupancy cost matrix; The optical path reconstruction order of the optimal routing solution is optimized and sorted according to the reconstruction cost matrix to obtain an optical path reconstruction sequence, and wavelength resource reservation and optical path switching operations are performed based on the optical path reconstruction sequence to output a resource reconfiguration solution.

8. An optical network routing intelligent planning system based on multi-objective optimization, characterized in that: For implementing the steps of the method according to any one of claims 1 to 7, the system comprises: A partitioning module, used for recursively partitioning the target optical network to obtain a plurality of sub-network units and a set of boundary nodes; A decision module, used to input the multiple sub-network units into a three-layer deep learning model, process the network topology feature vector, the service demand vector and the network state vector, and obtain a routing decision matrix and a wavelength allocation vector; A compensation module, used for inputting the plurality of sub-network units and the boundary node set into a distributed controller, processing the state update message, the route request message and the resource reservation message between the controllers, and obtaining an error compensation matrix; A processing module, configured to perform dual strategy processing of delay priority and hop priority on the optical path set through a binary search tree and an ordered search tree based on the routing decision matrix and the wavelength allocation vector, and obtain a delay weighted evaluation function and a hop weighted evaluation function; An optimization module, used to construct a target optimization function based on the delay weighted evaluation function and the hop count weighted evaluation function, perform multi-objective joint optimization, and obtain an optimal routing solution; The output module is used to calculate the reconstruction cost of the light path set according to the error compensation matrix and the optimal routing solution, and output a resource reconfiguration solution.

9. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images