Routing spectrum planning method based on mixed fragment measurement and double-layer model management

Abstract

The invention relates to the technical field of communication networks, in particular to a routing spectrum planning method based on hybrid fragment measurement and double-layer model management, which comprises the following steps: acquiring topological structure data and service data of an elastic optical network; a routing spectrum planning scheme is initialized based on the service data, mixed fragment measurement is carried out on the routing spectrum planning scheme according to the topological structure data, and the mixed fragment measurement comprises local routing fragment measurement and global routing fragment measurement; a Gaussian process model is utilized to proxy global routing fragment measurement, the Gaussian process model is trained, a routing spectrum planning scheme is optimized based on the trained Gaussian process model and a double-layer model management strategy, the double-layer model management strategy comprises upper-layer management and lower-layer management, the upper-layer management selects a non-dominated solution by utilizing a reference vector, and the lower-layer management selects a non-dominated solution by utilizing a reference vector. And the lower-layer management determines a target non-dominated solution in which computing resources are concentrated in the Gaussian process model evaluation process based on the non-dominated solution.

Description

Technical Field

[0001] The present application relates to the field of communication network technology, and in particular to a routing spectrum planning method based on hybrid fragmentation metric and double-layer model management. Background Art

[0002] Routing and Spectrum Allocation (RSA) is a key factor affecting the survivability of Elastic Optical Networks (EONs). The RSA problem in EONs involves finding a path and allocating sufficient spectrum across all fiber links along the path to create a suitable end-to-end optical path for the connection. Compared to the Routing and Wavelength Assignment (RWA) algorithm used in traditional WDM (Wavelength Division Multiplexing Networks), the RSA algorithm has stricter requirements. Specifically, spectrum allocation must adhere to three fundamental constraints: spectrum continuity, spectrum consistency, and spectrum non-overlap. The RSA problem has been proven to be NP-hard (Nondeterministic Polynomial Time), meaning it's difficult to find a universal algorithm that works for all instances. This is especially true for online services, where dynamic lightpath establishment and teardown over time, coupled with the repeated release and replanning of spectrum resources, creates a significant amount of fragmentation, severely impacting network survivability.

[0003] Related technologies addressing the RSA problem primarily include precise algorithms and heuristic algorithms. While precise algorithms can guarantee the accuracy of the resulting routing spectrum planning solution, they are computationally complex, and the time required to solve the routing spectrum planning solution increases exponentially with network size, making them inadequate for large-scale dynamic business scenarios. While heuristic algorithms can obtain approximate solutions in a relatively short time, they rely solely on a single-dimensional fragmentation metric (such as local or global fragmentation), lack a comprehensive assessment of fragmentation, and are prone to falling into local optimality. Furthermore, they lack deep integration with survivability requirements and are unable to effectively address the negative impact of fragmentation accumulation on network survivability. Summary of the Invention

[0004] The present application provides a routing spectrum planning method based on hybrid fragmentation measurement and two-layer model management to solve the local optimization trap caused by single-dimensional measurement in related technologies, which leads to problems such as decreased survivability and poor sustainability of the planned routing spectrum solution.

[0005] The first aspect of the present application provides a routing spectrum planning method based on hybrid fragmentation measurement and two-layer model management, comprising the following steps: obtaining topology data and service data of an elastic optical network; initializing a routing spectrum planning scheme based on the service data, and performing hybrid fragmentation measurement on the routing spectrum planning scheme according to the topology data, wherein the hybrid fragmentation measurement includes a local routing fragmentation measurement and a global routing fragmentation measurement, the local routing fragmentation measurement includes the influence of the number of continuous idle frequency slot clusters based on the local route, and the global routing fragmentation measurement includes the influence of the number of continuous idle frequency slot clusters based on the global network at the embedded connection between the local and global networks; using a Gaussian process model to proxy the global routing fragmentation measurement, training the Gaussian process model, taking the local routing fragmentation measurement and the global fragmentation measurement as optimization targets, optimizing the routing spectrum planning scheme based on the trained Gaussian process model and the two-layer model management strategy, wherein the two-layer model management strategy includes upper-layer management and lower-layer management, the upper-layer management uses a reference vector to select a non-dominated solution, and the lower-layer management determines the target non-dominated solution on which computing resources are concentrated during the Gaussian process model evaluation process based on the non-dominated solution.

[0006] Optionally, a hybrid fragmentation measurement is performed on the routing spectrum planning scheme based on the topology data, including: determining the number of continuous idle frequency slots required for the current service based on the topology data; determining an idle frequency slot cluster consisting of any adjacent continuous idle frequency slots on the link, and a continuous idle frequency slot cluster within the path that is unoccupied on all links in the entire path domain, so as to obtain the number of continuous idle frequency slot clusters and the number of continuous idle frequency slots for the local route; calculating the routing availability based on the number of continuous idle frequency slot clusters and the number of continuous idle frequency slots, and determining the local routing fragmentation measurement result based on the routing availability and the first metric function; and determining the global routing fragmentation measurement result based on the routing availability and the second metric function.

[0007] Optionally, the first metric function is:

[0008] ΔR u,v =R u,v -R u,v (B);

[0009] Among them, (u,v) is the routing path, R u,v R is the routing availability of the routing path (u, v) before allocating service B. u ,v (B) is the routing availability of the routing path (u, v) after resource occupation according to the routing spectrum planning scheme, ΔR u,v The local routing fragmentation metric of the routing path (u, v) for the routing spectrum planning scheme of service B;

[0010] The second metric function is:

[0011]

[0012] Among them, u is the set of links at the connection between the local routing and the network, (u, v) is the routing path, R u,v R is the routing availability of the routing path (u, v) before allocating service B. u,v (B) is the routing availability of the routing path (u, v) after resource occupation according to the routing spectrum planning scheme. The global routing fragmentation metric for all routing paths in the routing spectrum planning solution for service B.

[0013] Optionally, training the Gaussian process model includes: constructing a database based on the routing spectrum planning scheme that completes the mixed fragmentation metric; and training the Gaussian process model using samples in the database, wherein each sample represents a solution and each solution is a routing spectrum planning scheme.

[0014] Optionally, a routing spectrum planning scheme is optimized based on a trained Gaussian process model and a two-layer model management strategy, including: selecting multiple samples from a database as an initial population, generating multiple candidate solutions without mixed fragmentation measurement using a target algorithm and the initial population; generating multiple reference vectors in a target space, wherein the multiple reference vectors divide the target space into different regions; calculating the vertical projection distance of each reference vector of each candidate solution, dividing the candidate solutions into regions corresponding to the reference vectors based on the vertical projection distances, and generating a candidate solution set corresponding to the reference vector based on the candidate solutions in the region to which each reference vector belongs; using a Gaussian process model to perform global routing fragmentation measurement on each candidate solution in the candidate solution set, and using a first metric function to perform local routing fragmentation measurement on each candidate solution; determining non-dominated solutions in the candidate solution set corresponding to each reference vector based on the global routing fragmentation measurement and local routing fragmentation measurement results of each candidate solution; generating a non-dominated solution set based on the non-dominated solutions corresponding to all reference vectors, and transmitting the non-dominated solution set to a lower-level management, wherein the lower-level management optimizes the non-dominated solutions in the non-dominated set based on the Gaussian process model and the second metric function.

[0015] Optionally, the non-dominated solutions in the non-dominated set are optimized based on the Gaussian process model and the second metric function, including: using the Gaussian process model to proxy the global routing fragmentation metric; using the hybrid hypervolume improvement expectation calculation formula to calculate the hypervolume improvement expectation of each non-dominated solution in the non-dominated solution set; determining the target non-dominated solution in the non-dominated set based on the hypervolume improvement expectation of the non-dominated solution, using the second metric function to perform a true evaluation of the target non-dominated solution, storing the non-dominated solution after the true evaluation in a database, and retraining the Gaussian process model using the updated database until the expensive evaluation budget is exhausted, thereby obtaining an optimized routing spectrum planning scheme.

[0016] The second aspect of the present application provides a routing spectrum planning device based on hybrid fragmentation measurement and two-layer model management, including: an acquisition module for acquiring topology data and service data of an elastic optical network; a measurement module for initializing a routing spectrum planning scheme based on the service data, and performing hybrid fragmentation measurement on the routing spectrum planning scheme according to the topology data, wherein the hybrid fragmentation measurement includes a local routing fragmentation measurement and a global routing fragmentation measurement, the local routing fragmentation measurement includes the influence of the number of continuous idle frequency slot clusters based on the local route, and the global routing fragmentation measurement includes the influence of the number of continuous idle frequency slot clusters based on the global network at the embedded connection between the local and global networks; an optimization module for using a Gaussian process model to proxy the global routing fragmentation measurement, training the Gaussian process model, taking the local routing fragmentation measurement and the global fragmentation measurement as optimization targets, and optimizing the routing spectrum planning scheme based on the trained Gaussian process model and the two-layer model management strategy, wherein the two-layer model management strategy includes upper-layer management and lower-layer management, the upper-layer management uses a reference vector to select a non-dominated solution, and the lower-layer management determines the target non-dominated solution on which computing resources are concentrated during the Gaussian process model evaluation process based on the non-dominated solution.

[0017] Optionally, the measurement module is further used to: determine the number of continuous idle frequency slots required for the current business based on the topology data; determine an idle frequency slot cluster consisting of any adjacent continuous idle frequency slots on the link, and a continuous idle frequency slot cluster within the path that is unoccupied on all links in the entire path area, so as to obtain the number of continuous idle frequency slot clusters and the number of continuous idle frequency slots for the local route; calculate the route availability based on the number of continuous idle frequency slot clusters and the number of continuous idle frequency slots, and determine the local route fragmentation measurement result based on the route availability and the first measurement function; determine the global route fragmentation measurement result based on the route availability and the second measurement function.

[0018] Optionally, the first metric function is:

[0019] ΔR u,v =R u,v -R u,v (B);

[0020] Among them, (u,v) is the routing path, R u,v R is the routing availability of the routing path (u, v) before allocating service B. u ,v (B) is the routing availability of the routing path (y, v) after resource occupation according to the routing spectrum planning scheme, ΔR u,v The local routing fragmentation metric of the routing path (u, v) for the routing spectrum planning scheme of service B;

[0021] The second metric function is:

[0022]

[0023] Among them, S is the set of links at the connection between the local routing and the network, (u,v) is the routing path, R u,v R is the routing availability of the routing path (u, v) before allocating service B. u,v (B) is the routing availability of the routing path (u, v) after resource occupation according to the routing spectrum planning scheme. The global routing fragmentation metric for all routing paths in the routing spectrum planning solution for service B.

[0024] Optionally, the optimization module is further used to: construct a database based on the routing spectrum planning scheme that completes the mixed fragmentation metric; and train a Gaussian process model using samples in the database, wherein each sample represents a solution and each solution is a routing spectrum planning scheme.

[0025] Optionally, the optimization module is further used to: select multiple samples from the database as the initial population, and use the target algorithm and the initial population to generate multiple candidate solutions without mixed fragmentation measurement; generate multiple reference vectors in the target space, and the multiple reference vectors divide the target space into different areas; calculate the vertical projection distance of each reference vector of each candidate solution, divide the candidate solutions into the areas to which the corresponding reference vectors belong based on the vertical projection distance, and generate a candidate solution set corresponding to the reference vector based on the candidate solutions in the area to which each reference vector belongs; use the Gaussian process model to perform global routing fragmentation measurement on each candidate solution in the candidate solution set, and use the first metric function to perform local routing fragmentation measurement on each candidate solution; based on the global routing fragmentation measurement and local routing fragmentation measurement results of each candidate solution, determine the non-dominated solutions in the candidate solution set corresponding to each reference vector; generate a non-dominated solution set based on the non-dominated solutions corresponding to all reference vectors, and pass the non-dominated solution set to the lower-level management, wherein the lower-level management optimizes the non-dominated solutions in the non-dominated set based on the Gaussian process model and the second metric function.

[0026] Optionally, the optimization module is further used to: use a Gaussian process model to proxy the global routing fragmentation metric; use a mixed hypervolume improvement expectation calculation formula to calculate the hypervolume improvement expectation of each non-dominated solution in the non-dominated solution set; determine the target non-dominated solution in the non-dominated set based on the hypervolume improvement expectation of the non-dominated solution, use a second metric function to perform a true evaluation of the target non-dominated solution, store the non-dominated solution after the true evaluation in a database, and use the updated database to retrain the Gaussian process model until the expensive evaluation budget is exhausted to obtain an optimized routing spectrum planning scheme.

[0027] In a third aspect, an embodiment of the present application provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to execute the routing spectrum planning method based on hybrid fragmentation metric and two-layer model management as described in the above embodiment.

[0028] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program or instruction stored thereon, which is executed by a processor to perform a routing spectrum planning method based on hybrid fragmentation metric and two-layer model management as described in the above embodiment.

[0029] The fifth aspect of the present application provides a computer program product, including a computer program or instructions. When the computer program or instructions are executed, they can implement the routing spectrum planning method based on hybrid fragmentation measurement and two-layer model management as described in the above embodiment.

[0030] Therefore, this application has at least the following beneficial effects:

[0031] The embodiment of the present application can perform hybrid fragmentation measurement on multiple initial routing spectrum planning schemes, avoid the local optimization trap caused by single-dimensional measurement by comprehensively evaluating the impact of fragmentation, comprehensively evaluate the global impact of spectrum fragmentation on network resource allocation, avoid global resource fragmentation caused by local optimization, ensure the long-term effectiveness of resource allocation schemes, and use a two-layer management model to guide the optimization of routing spectrum planning schemes. The upper-layer management takes into account the diversity and convergence of routing spectrum planning schemes, and the lower-layer management realizes the concentration of computing resources on excellent routing spectrum planning schemes to improve computing efficiency and reduce computing overhead, thereby improving the survival performance of routing spectrum schemes and network resource utilization. As a result, the local optimization trap caused by single-dimensional measurement in related technologies is solved, which leads to technical problems such as decreased survival performance and poor sustainability of planned routing spectrum schemes.

[0032] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0034] Figure 1 A flowchart of a routing spectrum planning method based on hybrid fragmentation metrics and dual-layer model management according to an embodiment of the present application;

[0035] Figure 2 A schematic diagram of a hybrid fragment measurement method provided according to an embodiment of the present application;

[0036] Figure 3 A schematic diagram of dividing the target space in the upper layer of the two-layer model management strategy according to an embodiment of the present application;

[0037] Figure 4 A specific execution flow chart of a routing spectrum planning method based on hybrid fragmentation metric and dual-layer model management according to an embodiment of the present application;

[0038] Figure 5 This is an example diagram of a routing spectrum planning device based on hybrid fragmentation metrics and dual-layer model management according to an embodiment of the present application;

[0039] Figure 6 A schematic diagram of the structure of an electronic device provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0040] The following describes in detail embodiments of the present application, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0041] Before describing the solution of this application, the related technologies of this application are first explained.

[0042] RSA is an important factor affecting the survivability of elastic optical networks. The RSA problem in elastic optical networks refers to the fact that in order to establish a connection in an EON, a path must be found and sufficient spectrum must be allocated on all optical fiber links along the path to create a suitable end-to-end optical path for the connection. Compared to the RWA algorithm of traditional WDM networks, the RSA algorithm has stricter requirements. Specifically, spectrum resources must comply with three basic constraints when allocating them, including spectrum continuity, spectrum consistency, and spectrum non-overlap. The RSA problem has been proven to be an NP-hard problem, meaning it is difficult to find a general algorithm that applies to all instances. This is especially true for online services, where optical paths are established and dismantled over a period of time due to dynamic service requests. Spectrum resources are repeatedly released and replanned, resulting in a large amount of fragmentation, which seriously affects network survivability.

[0043] In recent decades, with the advancement of computer technology, the development of algorithms for solving the RSA problem can be roughly divided into two phases. Exact algorithms are commonly used to solve such problems, but the solution time of such algorithms increases exponentially with the number of activities, making them unsuitable for solving large-scale project scheduling optimization problems involving multiple activities. Heuristic algorithms, which can find optimal or near-optimal solutions in a relatively short time, have been used to solve the RSA problem. However, when applying metaheuristic algorithms to practical problems, it is necessary not only to consider the algorithm's update and iteration mechanisms but also to consider how to measure and address network fragmentation in the specific problem. Two main fragmentation mitigation mechanisms exist for such problems: maintenance-based defragmentation and preventative-based fragmentation avoidance. Maintenance-based defragmentation involves reconfiguring some requests, which is complex and time-consuming. Preventative-based fragmentation avoidance avoids the tedious reorganization and replanning of network traffic and offers strong real-time and robustness.

[0044] At the same time, when solving the RSA model, the exact algorithm is not suitable for solving large-scale optimization problems due to the excessive time overhead. In the heuristic algorithm, some metrics are used to directly plan the routing spectrum, but as the network scale increases, an approximate solution cannot be obtained. To address this problem, the present invention proposes a hybrid fragmentation metric method and designs an evolutionary algorithm assisted by a proxy model using reference vector partitioning and hypervolume improvement to solve the model, that is, to obtain an optimized routing spectrum planning solution.

[0045] To this end, the present application provides a routing spectrum planning method based on hybrid fragmentation measurement and two-layer model management. In this method, hybrid fragmentation measurement can be performed on multiple initial routing spectrum planning schemes. By comprehensively evaluating the impact of fragmentation, the local optimization trap caused by single-dimensional measurement can be avoided. The global impact of spectrum fragmentation on network resource allocation is comprehensively evaluated to avoid global resource fragmentation caused by local optimization, thereby ensuring the long-term effectiveness of the resource allocation scheme. The two-layer management model is used to guide the optimization of the routing spectrum planning scheme. The upper-layer management takes into account the diversity and convergence of the routing spectrum planning scheme, and the lower-layer management realizes the concentration of computing resources on excellent routing spectrum planning schemes to improve computing efficiency and reduce computing overhead, thereby improving the survivability performance of the routing spectrum scheme and network resource utilization.

[0046] Specifically, Figure 1 A flowchart of a routing spectrum planning method based on hybrid fragmentation metric and dual-layer model management provided in an embodiment of the present application.

[0047] like Figure 1 As shown, the routing spectrum planning method based on hybrid fragmentation metric and two-layer model management includes the following steps:

[0048] In step S101, topology data and service data of an elastic optical network are obtained.

[0049] Among them, the topology data includes: network node set, network node connectivity relationship, link distance, link optical signal-to-noise ratio, link spectrum resource information, etc.; current business needs include: business transmission demand starting point and end point, business spectrum resource demand, etc.

[0050] In step S102, the routing spectrum planning scheme is initialized based on the business data, and the routing spectrum planning scheme is subjected to a hybrid fragmentation metric according to the topology data, wherein the hybrid fragmentation metric includes a local routing fragmentation metric and a global routing fragmentation metric, the local routing fragmentation metric includes the influence of the number of continuous idle frequency slot clusters based on the local routing, and the global routing fragmentation metric includes the influence of the number of continuous idle frequency slot clusters based on the global network at the embedded connection between the local and global networks.

[0051] Among them, local routing fragmentation refers to the degree to which spectrum resources within a single link are divided into multiple discontinuous segments, and global routing fragmentation refers to the degree to which spectrum resources are not allocated in a coordinated manner between adjacent links in the network.

[0052] It is understandable that the embodiments of the present application can initialize the routing spectrum planning scheme based on service data, and perform mixed fragmentation measurement on the routing spectrum planning scheme according to topology data, so as to subsequently optimize the quality of the routing spectrum planning scheme.

[0053] In addition, it should be noted that the multiple routing spectrum planning schemes in the embodiment of the present application can be generated by priority coding and path generation algorithm to meet multiple initial routing spectrum planning schemes. When generating multiple routing spectrum planning schemes, the constraints of the routing spectrum planning model need to be met.

[0054] After collecting the topology data of the elastic optical network, the embodiment of the present application can perform data quality inspection and preprocessing, clean and delete data for repeated disconnected links, build a network topology connectivity relationship matrix, and use priority coding for network nodes.

[0055] In an embodiment of the present application, a hybrid fragmentation measurement is performed on a routing spectrum planning scheme based on topology data, including: determining the number of continuous idle frequency slots required for the current service based on the topology data; determining an idle frequency slot cluster consisting of any adjacent continuous idle frequency slots on the link, and a continuous idle frequency slot cluster within a path that is unoccupied on all links in the entire path domain, to obtain the number of continuous idle frequency slot clusters and the number of continuous idle frequency slots for the local route; calculating the routing availability based on the number of continuous idle frequency slot clusters and the number of continuous idle frequency slots, and determining the local routing fragmentation measurement result based on the routing availability and the first metric function; and determining the global routing fragmentation measurement result based on the routing availability and the second metric function.

[0056] It can be understood that the embodiments of the present application can determine the number of continuous idle frequency slots required for the current business based on the topology data, determine the idle frequency slot clusters composed of any adjacent continuous idle frequency slots on the link, and the continuous idle frequency slot clusters in the path that are unoccupied on all links in the entire path area, so as to obtain the number of continuous idle frequency slot clusters and the number of continuous idle frequency slots for the local route, calculate the route availability based on the number of continuous idle frequency slot clusters and the number of continuous idle frequency slots, and determine the local route fragmentation measurement result based on the route availability and the first metric function, and determine the global route fragmentation measurement result based on the local route availability and the second metric function, so as to refine the fragmentation measurement dimension, and then more comprehensively evaluate the quality of the routing spectrum planning scheme from local and global perspectives.

[0057] In this embodiment of the present application, the first metric function is:

[0058] ΔR u,v =R u,v -R u,v (B);

[0059] Among them, (u,v) is the routing path, R u,v R is the routing availability of the routing path (u, v) before allocating service B. u ,v (V) is the routing availability of the routing path (u, v) after resource occupation according to the routing spectrum planning scheme, ΔR u,v The local routing fragmentation metric of the routing path (u, v) for the routing spectrum planning scheme of service B;

[0060] The second metric function is:

[0061]

[0062] Among them, S is the set of links at the connection between the local routing and the network, (u,v) is the routing path, R u,v R is the routing availability of the routing path (u, v) before allocating service B. u,v (B) is the routing availability of the routing path (u, v) after resource occupation according to the routing spectrum planning scheme. The global routing fragmentation metric for all routing paths in the routing spectrum planning solution for service B.

[0063] Specifically, this application constructs a hybrid fragmentation measurement method, including a fragmentation evaluation system for local routing resources and a fragmentation evaluation system for network-wide routing resources, specifically:

[0064] 1. Build a fragmented evaluation system for local routing resources.

[0065] A fragmentation evaluation system for local routing resources is constructed to define idle frequency slot clusters. For current service needs, assuming that the need requires n idle continuous frequency slots, any n consecutive idle frequency slots on link L can form an idle frequency slot cluster.

[0066] Define a continuous cluster of idle frequency slots within a path: If the frequency slots from i to i+n-1 on a path are considered a continuous cluster of idle frequency slots, this cluster must be unoccupied on all links along the entire path. If a large number of idle frequency slots cannot meet service requirements, it indicates a high degree of link fragmentation. The definition of an idle frequency slot cluster precisely measures the amount of idle resources available on a route to meet service needs.

[0067] Define the fragmentation index within the local path: For a reachable route of a service, assume that the number of continuous idle frequency slots on the route is I u,v , the number of continuous idle frequency slot clusters is C u,v , the service requires n frequency slots. Then the availability of the route is R u,v It can be measured as:

[0068]

[0069] If all idle frequency slots on the route (u, v) are adjacent or there is no traffic load on the link, they can form I u,v -(n-1) idle frequency slot clusters, namely C u,v The maximum value can be I u,v -(n-1), the availability of the route R u,v Equal to 1, otherwise, C u,v If it is equal to 0, it means that no continuous idle frequency slot cluster that can meet the service requirements can be found on the route, and the availability of the route is 0.

[0070] For the current service B, after the current route allocates spectrum resources to service B, the availability of the spectrum resources within it is reduced the least, that is, the availability of the route (u, v) after the spectrum is allocated R u,v The maximum value is the maximum value, so the current spectrum planning scheme is the best for this local route. Specifically, using ΔR u,v The availability of route (u, v) is reduced as shown in the following formula:

[0071] ΔR u,v =R u,v -R u,v (B).

[0072] 2. Build a fragmentation assessment system covering the entire network's routing resources to enable spectrum allocation decisions that take into account global resource utilization efficiency.

[0073] First, define the spectrum resource overlap of adjacent links: for two adjacent links, the number of overlapping idle frequency slots between the two links represents the spectrum resource overlap O(e1, e2) of the two links, as shown in the following formula:

[0074]

[0075] The impact of the current planning scheme on the global network fragmentation is defined as the change in the degree of spectrum resource fragmentation at the interface when the local route accesses the global network topology through the boundary link, as shown in the following formula:

[0076]

[0077] Among them, S is the link set at the connection between the local routing and the network, and R represents the overlap of the idle spectrum cluster of the link.

[0078] In step S103, the Gaussian process model is used as a proxy for the global routing fragmentation metric to train the Gaussian process model. The local routing fragmentation metric and the global fragmentation metric are used as optimization targets. The routing spectrum planning scheme is optimized based on the trained Gaussian process model and the two-layer model management strategy. The two-layer model management strategy includes upper-layer management and lower-layer management. The upper-layer management uses a reference vector to select a non-dominated solution. The lower-layer management determines the target non-dominated solution on which computing resources are concentrated during the Gaussian process model evaluation process based on the non-dominated solution.

[0079] Among them, the Gaussian process model is a probabilistic model that uses historical data to learn the relationship between routing spectrum planning schemes and the global fragmentation impact measurement for rapid prediction; the optimization goal is to minimize local routing fragmentation and global routing fragmentation.

[0080] Since the computational complexity of the local routing fragmentation metric does not increase with the network scale, the routing spectrum allocation quality can be directly calculated and evaluated in practice, and different routing schemes require retraining of the proxy model and cannot inherit historical parameters. Therefore, the use of real evaluation of the local routing fragmentation degree can speed up the measurement speed and avoid errors caused by multiple training models due to routing changes. The global routing fragmentation metric evaluation is expensive and still requires the introduction of a proxy model to replace the real evaluation. Therefore, the embodiment of the present application can use a Gaussian process model to represent the global routing fragmentation metric, train the Gaussian process model, and use the local routing fragmentation metric and the global fragmentation metric as optimization targets. The routing spectrum planning scheme is optimized based on the trained Gaussian process model and the two-layer model management strategy. The routing spectrum planning scheme is taken into account through upper-layer management, and the computing resources are concentrated on the excellent routing spectrum planning scheme through lower-layer management to improve computing efficiency. The speed and survival performance of the routing spectrum planning scheme optimization are balanced through the two-layer model management strategy and the hybrid fragmentation metric.

[0081] In an embodiment of the present application, the Gaussian process model is trained, including: constructing a database based on the routing spectrum planning scheme that completes the mixed fragmentation metric; using samples in the database to train the Gaussian process model, wherein each sample represents a solution, and each solution is a routing spectrum planning scheme.

[0082] It is understandable that, in the embodiment of the present application, a database can be constructed based on the evaluated routing spectrum planning solutions, and the Gaussian process model can be trained using samples in the database.

[0083] Specifically, the embodiment of the present application trains a Gaussian process model through the evaluated routing spectrum planning scheme (hereinafter referred to as the scheme) and its fragmentation measurement results to predict the global fragmentation measurement of the new scheme, so as to avoid time-consuming global fragmentation calculation each time the scheme is evaluated, thereby reducing computational complexity.

[0084] In an embodiment of the present application, a routing spectrum planning scheme is optimized based on a trained Gaussian process model and a two-layer model management strategy, including: selecting multiple samples from a database as an initial population, generating multiple candidate solutions without mixed fragmentation measurement using a target algorithm and the initial population; generating multiple reference vectors in a target space, wherein the multiple reference vectors divide the target space into different regions; calculating the vertical projection distance of each reference vector of each candidate solution, dividing the candidate solutions into regions corresponding to the corresponding reference vectors based on the vertical projection distances, and generating a candidate solution set corresponding to the reference vector based on the candidate solutions in the region corresponding to each reference vector; performing a global routing fragmentation measurement on each candidate solution in the candidate solution set using a Gaussian process model, and performing a local routing fragmentation measurement on each candidate solution using a first metric function; determining non-dominated solutions in the candidate solution set corresponding to each reference vector based on the global routing fragmentation measurement and the local routing fragmentation measurement results of each candidate solution; generating a non-dominated solution set based on the non-dominated solutions corresponding to all reference vectors, and transmitting the non-dominated solution set to a lower-level management, wherein the lower-level management optimizes the non-dominated solutions in the non-dominated set based on the Gaussian process model and the second metric function.

[0085] The target algorithm can be the NSGA-II genetic algorithm; the reference vector is a set of direction vectors emitted from the origin of the target space, which is used to divide the target space into different regions; the vertical projection distance is the vertical distance from the candidate solution to the reference vector in the target space, which is used to determine the matching degree between the routing spectrum planning scheme and the reference vector.

[0086] It can be understood that the embodiments of the present application can screen high-quality solutions from the database as initial solutions, and use genetic algorithms to generate new candidate solutions, generate uniformly distributed reference vectors in the target space, divide the target space into multiple regions, calculate the distance between the candidate solutions and each reference vector, and assign them to the nearest region to form a candidate set, ensuring that the optimization process covers all directions of the target space and maintains the diversity of the solution set. The candidate solutions are processed in different regions to reduce the computational complexity. In each reference vector region, non-dominated solutions are screened out, that is, there is no other solution that is better than all the optimization objectives of this solution at the same time, and the set of non-dominated solutions is further optimized. High-quality solutions are quickly screened through non-dominated sorting, redundant calculations are reduced, and the focus is on the Pareto front solution to improve the quality of the solution set.

[0087] In an embodiment of the present application, non-dominated solutions in a non-dominated set are optimized based on a Gaussian process model and a second metric function, including: using a Gaussian process model to proxy a global routing fragmentation metric; using a hybrid hypervolume improvement expectation calculation formula to calculate the hypervolume improvement expectation of each non-dominated solution in the non-dominated solution set; determining a target non-dominated solution in the non-dominated set based on the hypervolume improvement expectation of the non-dominated solution, using the second metric function to perform a true evaluation of the target non-dominated solution, storing the non-dominated solution after the true evaluation in a database, and retraining the Gaussian process model using the updated database until the expensive evaluation budget is exhausted, thereby obtaining an optimized routing spectrum planning scheme.

[0088] Among them, the expected hypervolume improvement is the expectation of the increase in the area dominated by the solution set in the target space after the new solution is added to the solution set.

[0089] It can be understood that the embodiments of the present application can use the Gaussian process proxy model to proxy the global routing fragmentation metric, and evaluate the potential of each candidate solution to contribute to the hypervolume of the current solution set, select the solution with a high expected value of hypervolume improvement and use the second metric function for real evaluation, update the solution set, and balance exploration (discovering new areas) and utilization (optimizing known areas) by prioritizing the evaluation of the most promising solutions, and reduce the number of real evaluations, so as to concentrate the expensive evaluation budget on high-value solutions.

[0090] Specifically, the present invention proposes an intelligent optimization algorithm based on a hybrid fragmentation measurement method and a two-tier model management strategy assisted by a proxy model. This algorithm is based on a local-global network fragmentation measurement mechanism, provides a network routing priority encoding mechanism, and designs an intelligent optimization algorithm assisted by a proxy model. Specifically, the algorithm includes the following steps:

[0091] 1. Define the preventive routing spectrum planning problem model.

[0092] 1. Minimize service blocking rate.

[0093] In the preventive routing spectrum planning problem model, the most common optimization goal is to minimize the service blocking rate. This is because in real-world scenarios, the most direct correlation between network survivability and the successful establishment of dynamic service connections is: the lower the service blocking rate, the more reasonable the routing spectrum planning scheme in the network, and the stronger the network survivability. The goal is to calculate a reasonable routing spectrum scheme for the service under the condition that the model constraints are met due to the frequent establishment and removal of dynamic services in the network, thereby minimizing the degree of network fragmentation and maintaining network survivability. The model formula based on the service blocking rate is as follows:

[0094] Min(BR);

[0095]

[0096] Where R represents the total number of dynamic RSA requests currently processed by the network; Represents the number of blocked requests.

[0097] 2. Maximize network spectrum utilization.

[0098] Network spectrum utilization refers to the process of concentrating spectrum resources as much as possible when planning routing spectrum for dynamic services. This not only reserves more idle resources for subsequent services, but also enables as few links as possible to save energy.

[0099] This application defines spectrum utilization as the ratio of the total link resource units occupied in each service allocation process to the total link resource units as the spectrum utilization of the network, denoted as SU, which can be calculated by the following method:

[0100]

[0101] Among them, FS occupied Refers to the total number of frequency slot resources currently occupied by all services, FS useful Refers to the total number of available frequency slots on links through which services pass.

[0102] Therefore, the optimization objectives of the preventive network survivability spectrum planning problem are: Min(BR) and Max(SU).

[0103] 2. Solving the model.

[0104] The embodiment of the present application proposes a network fragmentation degree evaluation and planning scheme based on the local-global fragmentation metric of spectrum continuity, and uses a two-layer agent model management strategy to guide the evolutionary algorithm to select high-quality solutions (the solution is the routing spectrum planning scheme).

[0105] When the network considers dynamic traffic, the request, configuration, and release of optical paths are random. When one or more idle time slots on different links on the optical path are different, misaligned idle time slots will appear. When one or more idle time slots are not adjacent to each other, discontinuous idle time slots will be created in the spectrum domain. These misaligned and discontinuous idle time slots are spectrum fragments, which are very likely to not meet future optical path requests, increasing the probability of network service blocking. As the scale of decision variables in the optimization problem grows, it is difficult for the evolutionary optimization algorithm to traverse the huge search space within a reasonable resource range, and the performance may drop sharply. In this application, the following improvements are made to the intelligent optimization algorithm:

[0106] 1. Hybrid fragmentation measurement method.

[0107] A fragmentation evaluation system for local routing resources is constructed, and an idle frequency slot cluster is defined. For the current service demand, assuming that the demand requires n idle continuous frequency slots, then any n consecutive adjacent idle frequency slots on link L can form an idle frequency slot cluster.

[0108] Define a continuous cluster of idle frequency slots within a path: If the frequency slots from i to i+n-1 on a path are considered a continuous cluster of idle frequency slots, this cluster must be unoccupied on all links along the entire path. If a large number of idle frequency slots cannot meet service requirements, it indicates a high degree of link fragmentation. The definition of an idle frequency slot cluster precisely measures the amount of idle resources available on a route to meet service needs.

[0109] Define the fragmentation index within the local path: For a reachable route of a service, assume that the number of continuous idle frequency slots on the route is I u,v , the number of continuous idle frequency slot clusters is C u,v , the service requires n frequency slots. Then the availability of the route is R u,v It can be measured as:

[0110]

[0111] If all idle frequency slots on the route (u, v) are adjacent or there is no traffic load on the link, they can form I u,v -(n-1) idle frequency slot clusters, namely C u,v The maximum value can be I u,v -(n-1), the availability of the route R u,v Equal to 1, otherwise, C u,v If it is equal to 0, it means that no continuous idle frequency slot cluster that can meet the service requirements can be found on the route, and the availability of the route is 0.

[0112] For the current service B, after the current route allocates spectrum resources to service B, the availability of the spectrum resources within it is reduced the least, that is, the availability of the route (u, v) after the spectrum is allocated R u,v Maximum, then for this local route, the current spectrum planning scheme is the best. Figure 2 Specifically, using ΔR u,v The availability of route (u, v) is reduced as shown in the following formula:

[0113] ΔR u,v =R u,v -R u,v (B).

[0114] Build a fragmentation evaluation system covering the entire network's routing resources to make spectrum allocation decisions that take into account global resource utilization efficiency.

[0115] First, define the spectrum resource overlap of adjacent links: for two adjacent links, the number of overlapping idle frequency slots between the two links represents the spectrum resource overlap O(e1, e2) of the two links, as shown in the following formula:

[0116]

[0117] The impact of the current planning scheme on the global network fragmentation is defined as the change in the degree of spectrum resource fragmentation at the interface when the local route accesses the global network topology through the boundary link, as shown in the following formula:

[0118]

[0119] Among them, S is the link set at the connection between the local routing and the network, and R represents the overlap of the idle spectrum cluster of the link. Figure 2 As shown, Figure 2 Schematic diagram of the hybrid fragmentation measurement method.

[0120] 2. Model management strategy based on target space partitioning.

[0121] To address the algorithm's challenge of balancing exploration and exploitation, the model management problem is transformed into a two-level optimization problem. This problem is clearly divided into two levels: the upper level and the lower level. The upper level program in the entire model management system undertakes the important task of balancing the diversity of the decision space and the convergence of the target value.

[0122] This application proposes a candidate solution generation strategy based on reference vectors, which selects high-quality unevaluated solutions by balancing population diversity and convergence. To control the computational complexity, a uniformly distributed reference vector equal to the population size (N) is generated in the target space in advance, and the nearest neighbor matching mechanism is used to associate the unevaluated offspring: the vertical projection distance from the offspring to the reference vector is calculated to perform region division, and the offspring is assigned to the reference vector region corresponding to the minimum projection distance (similar to the NSGA-II mechanism), where the target space is divided as follows: Figure 3 This method effectively improves the convergence efficiency of the population to the Pareto front while maintaining the breadth of the solution target space distribution, and achieves efficient search under diversity constraints.

[0123] For each reference vector, the candidate solution set is first screened for non-dominated solutions. If a reference vector has more than one non-dominated solution, the maximum crowding distance between these solutions and the archived samples is calculated, and the individual with the largest distance is selected as a surviving candidate. If only a single non-dominated solution exists for the corresponding reference vector, it is directly included in the surviving offspring set. Through this screening strategy, all surviving individuals are selected according to the principle of "non-dominated priority and distribution optimization." This mechanism dynamically balances the convergence and distribution of solutions, preventing excessive concentration of high-quality solutions in local areas while ensuring the population's exploration capability in the target space. The algorithm's screening mechanism consists of two core steps: first, a pre-screening based on local dominance relationships is performed, selecting individuals with non-dominated advantages by analyzing the dominance status of offspring in adjacent areas of the target space. Second, a distribution density control strategy is introduced, prioritizing individuals with the largest distance from the archived sample space among the non-dominated individuals to avoid solution redundancy in local areas. Overall, the reference vector-guided distance metric ensures a uniform distribution of the offspring population in the target space, while the non-dominated sorting mechanism drives the population to converge efficiently toward the Pareto front. The synergistic effect of the two enables the algorithm to dynamically maintain the balance between the diversity and convergence of the solution set during the evolution process.

[0124] 3. Design of acquisition function based on cheap and expensive hybrid objectives.

[0125] The computational complexity of local routing fragmentation metrics does not increase dramatically with network scale, and in practice, spectrum allocation quality can be directly evaluated. Different routing schemes require retraining proxy models, which cannot inherit historical parameters. Therefore, using a true assessment of local routing fragmentation can speed up the measurement and avoid errors caused by multiple model training due to routing changes. However, global routing fragmentation assessment is expensive and still requires the use of a proxy model instead of a true assessment.

[0126] Based on the aforementioned cheap-expensive mixing characteristics of the objective function, a single-surrogate model acquisition function is designed. A Gaussian process is used as the surrogate model to provide a predictive distribution given new input data. The Gaussian process defines a prior for the function f(x)~GP(m(x),k(x,x′)). The Gaussian process (GP) is completely determined by its mean function m(x) and its semi-positive covariance function k(x,x′). The predictive distribution of new data X = [x1,…,xN] has a zero mean and can be calculated as follows:

[0127]

[0128] The probability density function of the above Gaussian process function is defined as follows:

[0129] φ j [y j ]:=φ j [y j ;μ j (x),σ j (x)];

[0130] Introducing the vector y:=(y1,y2), where y1 and y2 represent the potential observation value and actual observation value of the objective function f1 (i.e., the first metric function) and f2 (i.e., the second metric function), the expected improvement index of cheap hypervolume can be calculated as follows:

[0131]

[0132] Where A is the screened unevaluated solution and φ(y1) is the probability density function at y1.

[0133] In general, the complete execution process of the routing spectrum planning solution optimization method of the embodiment of the present application is as follows: Figure 4 Shown, including:

[0134] 1. Collect topological structure data and service demand data of the elastic optical network, including network node sets, network node connectivity, link distance, link optical signal-to-noise ratio, link spectrum resource information, service transmission starting and end points, service spectrum resource requirements, etc., and perform data quality inspection and preprocessing.

[0135] 2. Data cleaning and deletion of duplicate disconnected links are performed to construct a network topology connectivity matrix. Priority coding and Latin hypercube sampling are used to initialize the routing spectrum planning scheme for network nodes. The hybrid fragmentation metric method is used to evaluate the initialized routing spectrum planning scheme.

[0136] Among them, the hybrid fragmentation metric is used for the initial routing spectrum planning scheme, including the impact of the number of continuous idle frequency slot clusters based on local routing on ΔRu,v , and the effect of the number of clusters of continuous idle frequency slots on the global network at the connection between local and global networks

[0137] 3. Only the global fragmentation metric is trained on the Gaussian process model to evaluate the agent, and the offspring are generated using NSGA-II. The non-dominated solutions are divided according to the reference vector of the target space, and the non-dominated solution with the largest crowding distance on each vector is calculated and passed to the lower-level strategy.

[0138] Here, the reference vector of the partitioned target space is used to generate unevaluated solutions that balance diversity and convergence:

[0139] The Das-Dennis method is used to divide the reference vector in the target space, and the Gaussian process model is trained using the evaluated planning scheme to proxy the global network fragmentation impact measurement, and the NSGA-II is used to generate offspring, and the local routing fragmentation metric ΔR is used. u,v The offspring are evaluated using the Gaussian process model, and the offspring are divided into reference vectors using the nearest distance projection method. The non-dominated solutions on each reference vector are screened. If the number of non-dominated solutions is greater than one, the non-dominated solution with the largest distance from the evaluated solution set is selected. The screened solution is then passed to the lower-level strategy after a crossover mutation operation with the non-dominated solutions in the evaluated solutions.

[0140] 4. Using the constructed Gaussian process model as a proxy for the global fragmentation metric, the hypervolume improvement expectation of the incoming unevaluated solutions is calculated according to the designed cheap and expensive hybrid hypervolume improvement expectation calculation formula, and the solutions are screened for real evaluation.

[0141] The constructed Gaussian process model is coupled with the hypervolume improvement in multi-objective optimization, and the candidate solutions are screened based on the probability distribution of the Gaussian process model and the hypervolume improvement index:

[0142]

[0143] In the above formula, z * The reference is dominated by any non-dominated solution. According to the above hypervolume calculation formula, the hypervolume improvement index can be obtained Gaussian process can obtain the prior probability distribution of decision variables φ j :φ j [y j ]=φ j [y j ;μ j (x),σ j (x)], the expected value of the hypervolume improvement of the candidate solution y can be obtained according to the probability distribution density function of the candidate solution y:

[0144]

[0145] The above calculation method only performs proxy evaluation on the Gaussian process model designed by the global fragmentation metric method, avoiding the superposition error of multiple proxy models.

[0146] The following describes the routing spectrum planning scheme optimization method of the present application through a specific embodiment.

[0147] This example uses two representative data sets of real-world services and optical network topologies provided by a telecommunications company for testing. Dataset 1 contains 1,200 nodes and 5,361 bidirectional fiber links, while Dataset 2 contains 149 nodes and 885 bidirectional fiber links. These provide realistic scenarios for testing the algorithm's ability to solve preventive routing spectrum planning scenarios.

[0148] The test set mainly contains two tables, namely Table 1 Network Topology Table and Table 2 Business Requirements Table.

[0149] Table 1

[0150] OMS omsId remoteOmsId src snk distance ots osnr colors OMS 4526 4537 0 1 100 1 0.000776 :0-960

[0151] Table 2

[0152] src snk sourceOtu targetOtu m_width sourceDimColors targetDimColors 812 803 1702 1262 24 :0-24:960-964 :0-24:960-964

[0153] Among them, OMS in Table 1 is a CSV file containing optical network path (OMS, Optical Multiplex Section) information. It is used to describe the logical paths in the optical network and their attributes (cost, distance, spectrum, etc.), supporting path calculation, resource allocation, and network optimization (such as selecting low-cost or specific spectrum paths).

[0154] OMS: Identifies the record type as an optical network path (fixed value). omsId: Unique identifier of the path. remoteOmsId: Identifier of the remote path (a symmetric record for bidirectional paths). src: The starting node ID of the path. snk: The ending node ID of the path. cost: The cost of the path. distance: The distance of the path. ots: The optical transmission segment identifier (fixed value 1). osnr: The optical signal-to-noise ratio (fixed value 0.000776). slice: The slice capacity (fixed value 6250). colors: Wavelength or spectrum allocation information (format: 0-960, some parts are empty).

[0155] Data characteristics: Each path has forward and reverse records (for example, omsId=4526 and remoteOmsId=4537 correspond to another omsId=4537 and remoteOmsId=4526).

[0156] Spectrum allocation: The colors field is mostly :0-960, indicating standard spectrum allocation. A few records contain other values ​​(such as :96-864, :0-964), indicating that special spectrum ranges or resources are occupied.

[0157] In the service information table in Table 2, Index: service number; src: source node ID; snk: target node ID; sourceOtu: source OTU (Optical Transport Unit) ID; targetOtu: target OTU ID; m_width: bandwidth, all values ​​are 24; bandType: band type; sourceDimColors: source node spectrum range; targetDimColors: target node spectrum range. This table primarily describes the complete service set, its corresponding source and target nodes, and the corresponding spectrum ranges.

[0158] Based on the above data, the specific implementation steps of this application include:

[0159] 1. Define a preventive routing spectrum planning model.

[0160] The service blocking rate and spectrum utilization of the model are calculated according to the formulas described above, and they will be used as the objective functions of the model in the following solution steps.

[0161] 2. Solving the model.

[0162] The network fragmentation degree is evaluated based on the local-global fragmentation metric of spectrum continuity, and the two-layer agent model management strategy is used to guide the evolutionary algorithm to select high-quality solutions. Figure 1 This is the flow chart of the algorithm. The specific implementation steps are as follows:

[0163] (1) Chromosome generation strategy.

[0164] Priority coding is used as the coding method. A priority is set for each node in the network. The topological relationship of the network is maintained in the connectivity matrix. The decision variable dimension is the number of network nodes. The planned path is determined based on the priority coding and the connectivity matrix.

[0165] (2) Initial solution evaluation.

[0166] Latin hypercube sampling is used to evaluate the local fragmentation metric ΔR of the initial sample in the solution space using real functions f1 (the change in the degree of local fragmentation influence, i.e., the first metric function) and f2 (the change in the degree of global fragmentation influence, i.e., the second metric function). u,v and global fragmentation metrics Save the samples to Archive. In the algorithm, Archive is used to store all the individual solutions that have been fully evaluated with real functions. These serve as training samples for building the Gaussian process model (surrogate model). Select N (population size) sample points from the initial solution to form the initial population.

[0167] (3) Construction of Gaussian process model (agent model).

[0168] The Gaussian process model is constructed using the decision variables of all solutions in Archive and the f2 objective function value. The Gaussian process defines the prior of the function f(x)~GP(m(x),k(x,x′)). With the help of the Gaussian process model, the new data point x can be * Make predictions and estimate the mean μ1 and variance σ1 associated with f2 2 .

[0169] (4) Model management based on target space partitioning method.

[0170] First, the Das-Dennis method is used to generate N (population size) reference vectors that are uniformly distributed in the target space as the basis for space division. The coordinates of each vector are calculated as follows.

[0171]

[0172] The NSGA-II algorithm is used to generate unevaluated candidate solutions. The perpendicular projection distance of each unevaluated candidate solution to each reference vector is calculated, and the candidate solution is then divided into regions corresponding to the reference vectors with the minimum projection distance. All candidate solutions contained within the region of the reference vector form the candidate solution set corresponding to that reference vector. For all candidate solutions, the local fragmentation metric is evaluated using the real function f1, and the global fragmentation metric is evaluated using the resulting Gaussian process model. The objective function value for each individual solution is then used to determine the non-dominance relationship between individuals.

[0173] For each reference vector, the candidate solution set corresponding to the non-dominated solution is first screened out. If there are more than one non-dominated solutions associated with a reference vector, the maximum crowding distance between these solutions and the archived samples is further calculated, and the individual with the largest distance value is selected as the surviving candidate solution; if there is only a single non-dominated solution corresponding to the reference vector, it is directly included in the surviving offspring set. Finally, the predicted non-dominated solution set X is obtained. ND-pre , passing this part of the unevaluated solution to the lower-level acquisition function.

[0174] (5) Population individual iteration and evaluation.

[0175] 1) Population iteration. ND-pre Perform crossover and mutation operations with the non-dominated solutions in the parent generation to generate unevaluated offspring Pun-evaluated .

[0176] 2) P un-evaluated For each individual y, the real function f1 is used to evaluate the local routing fragmentation impact ΔR of the planning scheme u,v , according to the super volume calculation formula is as follows:

[0177]

[0178] In the above formula, z * The reference is dominated by any non-dominated solution, and the hypervolume improvement function is defined as:

[0179]

[0180] According to the Gaussian process model, The probability density function is as follows

[0181] φ j [y j ]=φ j [y j ;μ j (x),σ j (x)];

[0182] The expected hypervolume improvement value CHEVI(y) of the individual y is calculated as follows

[0183]

[0184] In the above formula, A is the unevaluated non-dominated solution obtained by the upper management strategy screening. All solutions with expected hypervolume improvement value CHEVI(y) greater than 0 are screened out for real evaluation.

[0185] (6) Gaussian process model (proxy model) update.

[0186] For all true evaluated solutions, their decision variables and two objective function values ​​are stored in Archive, and all solutions in Archive are used to train and update the Gaussian process model.

[0187] (7) Repeat (4) until the expensive evaluation budget is exhausted.

[0188] Based on the above operations, we tested and solved the aforementioned dataset. The algorithm parameters were set as follows: the initial population size was set to 50, the population size N was set to 30, the number of target space reference vectors was the same as the population size N, 30, and the cost evaluation budget was set to 500. The crossover probability in the upper-layer strategy was set to 0.8, and the mutation probability was set to 0.05. Through this implementation, we obtained the service blocking rate and spectrum utilization under different load conditions, as shown in the following table. Table 3 shows the results for service set 1 and network 1, and Table 4 shows the results for service set 2 and network 2.

[0189] Table 3

[0190] Erlangs Spectrum utilization Service blocking rate 100 0.015604 0.02 300 0.015287 0.05 500 0.020312 0.078 1000 0.027054 0.062222 1500 0.035025 0.08 1800 0.041021 0.102222

[0191] Table 4

[0192] Erlangs Spectrum utilization Service blocking rate 100 0.021776 0.03 300 0.023138 0.04 500 0.025178 0.138 1000 0.029606 0.129 1500 0.033899 0.271333 2000 0.036941 0.3195 2500 0.039073 0.3596

[0193] According to the routing spectrum planning method based on hybrid fragmentation measurement and two-layer model management proposed in the embodiment of the present application, hybrid fragmentation measurement can be performed on multiple initial routing spectrum planning schemes. By comprehensively evaluating the impact of fragmentation, the local optimization trap caused by single-dimensional measurement can be avoided. The global impact of spectrum fragmentation on network resource allocation can be comprehensively evaluated to avoid global resource fragmentation caused by local optimization, thereby ensuring the long-term effectiveness of the resource allocation scheme. In addition, a two-layer management model is used to guide the optimization of the routing spectrum planning scheme. The upper-layer management takes into account the diversity and convergence of the routing spectrum planning scheme, and the lower-layer management realizes the concentration of computing resources on excellent routing spectrum planning schemes to improve computing efficiency and reduce computing overhead, thereby improving the survivability performance of the routing spectrum scheme and network resource utilization.

[0194] Next, a routing spectrum planning device based on hybrid fragmentation metrics and dual-layer model management proposed in accordance with an embodiment of the present application will be described with reference to the accompanying drawings.

[0195] Figure 5 4 is a block diagram of a routing spectrum planning device based on hybrid fragmentation metric and dual-layer model management according to an embodiment of the present application.

[0196] like Figure 5 As shown, the route spectrum planning device 10 based on hybrid fragmentation metric and dual-layer model management includes: an acquisition module 100 , a metric module 200 and an optimization module 300 .

[0197] Among them, the acquisition module 100 is used to obtain the topology data and service data of the elastic optical network; the measurement module 200 is used to initialize the routing spectrum planning scheme based on the service data, and perform hybrid fragmentation measurement on the routing spectrum planning scheme according to the topology data, wherein the hybrid fragmentation measurement includes local routing fragmentation measurement and global routing fragmentation measurement, the local routing fragmentation measurement includes the influence of the number of continuous idle frequency slot clusters based on local routing, and the global routing fragmentation measurement includes the influence of the number of continuous idle frequency slot clusters based on the global network at the embedded connection between the local and global networks; the optimization module 300 is used to use the Gaussian process model to proxy the global routing fragmentation measurement, train the Gaussian process model, take the local routing fragmentation measurement and the global fragmentation measurement as the optimization targets, and optimize the routing spectrum planning scheme based on the trained Gaussian process model and the two-layer model management strategy, wherein the two-layer model management strategy includes upper-layer management and lower-layer management, the upper-layer management uses the reference vector to select the non-dominated solution, and the lower-layer management determines the target non-dominated solution on which the computing resources are concentrated during the Gaussian process model evaluation process based on the non-dominated solution.

[0198] In an embodiment of the present application, the measurement module 200 is further used to: determine the number of continuous idle frequency slots required for the current service based on the topology data; determine an idle frequency slot cluster consisting of any adjacent continuous idle frequency slots on the link, and a continuous idle frequency slot cluster within the path that is unoccupied on all links in the entire path area, so as to obtain the number of continuous idle frequency slot clusters and the number of continuous idle frequency slots for the local route; calculate the route availability based on the number of continuous idle frequency slot clusters and the number of continuous idle frequency slots, and determine the local route fragmentation measurement result based on the route availability and the first measurement function; and determine the global route fragmentation measurement result based on the route availability and the second measurement function.

[0199] In this embodiment of the present application, the first metric function is:

[0200] ΔR u,v =R u,v -R u,v (B);

[0201] Among them, (u,v) is the routing path, R u,v R is the routing availability of the routing path (u, v) before allocating service B. u ,v (B) is the routing availability of the routing path (u, v) after resource occupation according to the routing spectrum planning scheme, ΔR u,v The local routing fragmentation metric of the routing path (u, v) for the routing spectrum planning scheme of service B;

[0202] The second metric function is:

[0203]

[0204] Among them, S is the set of links at the connection between the local routing and the network, (u,v) is the routing path, R u,v R is the routing availability of the routing path (u, v) before allocating service B. u,v (B) is the routing availability of the routing path (u, v) after resource occupation according to the routing spectrum planning scheme. The global routing fragmentation metric for all routing paths in the routing spectrum planning solution for service B.

[0205] In an embodiment of the present application, the optimization module 300 is further used to: construct a database based on the routing spectrum planning scheme that completes the mixed fragmentation metric; use samples in the database to train a Gaussian process model, where each sample represents a solution, and each solution is a routing spectrum planning scheme.

[0206] In an embodiment of the present application, the optimization module 300 is further used to: select multiple samples from a database as an initial population, and use the target algorithm and the initial population to generate multiple candidate solutions that are not subjected to mixed fragmentation measurement; generate multiple reference vectors in the target space, and the multiple reference vectors divide the target space into different regions; calculate the vertical projection distance of each reference vector of each candidate solution, divide the candidate solutions into the regions to which the corresponding reference vectors belong based on the vertical projection distance, and generate a candidate solution set corresponding to the reference vector based on the candidate solutions in the region to which each reference vector belongs; use a Gaussian process model to perform global routing fragmentation measurement on each candidate solution in the candidate solution set, and use a first metric function to perform local routing fragmentation measurement on each candidate solution; based on the global routing fragmentation measurement and local routing fragmentation measurement results of each candidate solution, determine the non-dominated solutions in the candidate solution set corresponding to each reference vector; generate a non-dominated solution set based on the non-dominated solutions corresponding to all reference vectors, and pass the non-dominated solution set to the lower-level management, wherein the lower-level management optimizes the non-dominated solutions in the non-dominated set based on the Gaussian process model and the second metric function.

[0207] In an embodiment of the present application, the optimization module 300 is further used to: use a Gaussian process model to proxy the global routing fragmentation metric; use a hybrid hypervolume improvement expectation calculation formula to calculate the hypervolume improvement expectation of each non-dominated solution in the non-dominated solution set; determine the target non-dominated solution in the non-dominated set based on the hypervolume improvement expectation of the non-dominated solution, use a second metric function to perform a true evaluation of the target non-dominated solution, store the non-dominated solution after the true evaluation in a database, and use the updated database to retrain the Gaussian process model until the expensive evaluation budget is exhausted to obtain an optimized routing spectrum planning scheme.

[0208] It should be noted that the above explanation of the embodiment of the routing spectrum planning method based on hybrid fragmentation metric and dual-layer model management is also applicable to the routing spectrum planning device based on hybrid fragmentation metric and dual-layer model management in this embodiment, and will not be repeated here.

[0209] According to the routing spectrum planning device based on hybrid fragmentation measurement and two-layer model management proposed in the embodiment of the present application, hybrid fragmentation measurement can be performed on multiple initial routing spectrum planning schemes. By comprehensively evaluating the impact of fragmentation, the local optimization trap caused by single-dimensional measurement can be avoided. The global impact of spectrum fragmentation on network resource allocation can be comprehensively evaluated to avoid global resource fragmentation caused by local optimization, thereby ensuring the long-term effectiveness of the resource allocation scheme. In addition, a two-layer management model is used to guide the optimization of the routing spectrum planning scheme. The upper-layer management takes into account the diversity and convergence of the routing spectrum planning scheme, and the lower-layer management realizes the concentration of computing resources on excellent routing spectrum planning schemes to improve computing efficiency and reduce computing overhead, thereby improving the survivability performance of the routing spectrum scheme and network resource utilization.

[0210] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. The electronic device may include:

[0211] A memory 601 , a processor 602 , and a computer program stored in the memory 601 and executable on the processor 602 .

[0212] When the processor 602 executes the program, the routing spectrum planning solution optimization method provided in the above embodiment is implemented.

[0213] Furthermore, the electronic device further includes:

[0214] The communication interface 603 is used for communication between the memory 601 and the processor 602 .

[0215] The memory 601 is used to store computer programs that can be run on the processor 602 .

[0216] The memory 601 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0217] If the memory 601, processor 602, and communication interface 603 are implemented independently, the communication interface 603, memory 601, and processor 602 can be connected to each other via a bus and communicate with each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 6 Only one thick line is used in the diagram, but this does not mean that there is only one bus or one type of bus.

[0218] Optionally, in a specific implementation, if the memory 601, the processor 602 and the communication interface 603 are integrated on a chip, the memory 601, the processor 602 and the communication interface 603 can communicate with each other through an internal interface.

[0219] The processor 602 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application.

[0220] An embodiment of the present application further provides a computer-readable storage medium having a computer program or instruction stored thereon. When the computer program or instruction is executed by a processor, the routing spectrum planning method based on hybrid fragmentation metric and dual-layer model management is implemented as described above.

[0221] An embodiment of the present application further provides a computer program product, including a computer program or instructions, which, when executed, implements the above-mentioned routing spectrum planning method based on hybrid fragmentation metric and dual-layer model management.

[0222] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or N embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0223] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of technical features indicated. Thus, a feature specified as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of this application, "N" means at least two, for example, two, three, etc., unless otherwise specifically defined.

[0224] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or N executable instructions for implementing a custom logical function or process step, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may be performed in a different order than shown or discussed, including performing functions in a substantially simultaneous manner or in a reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present application pertain.

[0225] It should be understood that various parts of the present application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiment, the N steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, it can be implemented using any one or a combination of the following technologies known in the art: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application-specific integrated circuit having a suitable combination of logic gate circuits, a programmable gate array, a field programmable gate array, etc.

[0226] Those skilled in the art will understand that all or part of the steps in the method of the above embodiment can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiment.

Claims

1. A routing spectrum planning method based on hybrid fragmentation metric and two-layer model management, characterized in that: The following steps are involved: Obtain topology data and service data of the elastic optical network; Initializing a routing spectrum planning scheme based on the service data, and performing a hybrid fragmentation metric on the routing spectrum planning scheme according to the topology data, wherein the hybrid fragmentation metric includes a local routing fragmentation metric and a global routing fragmentation metric, the local routing fragmentation metric includes an impact of a number of continuous idle frequency slot clusters based on local routing, and the global routing fragmentation metric includes an impact of a number of continuous idle frequency slot clusters based on a global network at an embedded connection between a local and global network; A Gaussian process model is used as a proxy for the global routing fragmentation metric, and the Gaussian process model is trained. The local routing fragmentation metric and the global fragmentation metric are used as optimization targets. The routing spectrum planning scheme is optimized based on the trained Gaussian process model and a two-layer model management strategy, wherein the two-layer model management strategy includes upper-layer management and lower-layer management, the upper-layer management uses a reference vector to select a non-dominated solution, and the lower-layer management determines the target non-dominated solution on which computing resources are concentrated during the Gaussian process model evaluation process based on the non-dominated solution.

2. The routing spectrum planning method based on hybrid fragmentation metric and two-layer model management according to claim 1, characterized in that: The performing hybrid fragmentation measurement on the routing spectrum planning scheme according to the topology data includes: Determining the number of consecutive idle frequency slots required for the current service based on the topology data; Determine an idle frequency slot cluster consisting of any adjacent continuous idle frequency slots on the link, and a continuous idle frequency slot cluster within the path that is unoccupied on all links in the entire path area, to obtain the number of continuous idle frequency slot clusters and the number of continuous idle frequency slots of the local route; Calculating a routing availability according to the number of continuous idle frequency slot clusters and the number of continuous idle frequency slots, and determining the local routing fragmentation measurement result based on the routing availability and a first metric function; The global routing fragmentation metric result is determined based on the routing availability and a second metric function.

3. The routing spectrum planning method based on hybrid fragmentation metric and two-layer model management according to claim 2, characterized in that: The first metric function is: ΔR u,v =R u,v -R u,v (B); Among them, (u,v) is the routing path, R u,v R is the routing availability of the routing path (u, v) before allocating service B. u,v (B) is the routing availability of the routing path (u, v) after resource occupation according to the routing spectrum planning scheme, ΔR u,v The local routing fragmentation metric of the routing path (U, v) for the routing spectrum planning scheme of service B; The second metric function is: Among them, S is the set of links at the connection between the local routing and the network, (u,v) is the routing path, R u,v R is the routing availability of the routing path (u, v) before allocating service B. u,v (B) is the routing availability of the routing path (u, v) after resource occupation according to the routing spectrum planning scheme. The global routing fragmentation metric for all routing paths in the routing spectrum planning solution for service B.

4. The routing spectrum planning method based on hybrid fragmentation metric and two-layer model management according to claim 1, characterized in that: The training of the Gaussian process model includes: Building a database based on the routing spectrum planning scheme that completes the hybrid fragmentation metric; The Gaussian process model is trained using samples in the database, wherein each sample represents a solution, and each solution is a routing spectrum planning scheme.

5. The routing spectrum planning method based on hybrid fragmentation metric and two-layer model management according to claim 2, characterized in that: The routing spectrum planning solution optimized based on the trained Gaussian process model and the two-layer model management strategy includes: Selecting a plurality of samples from the database as an initial population, and generating a plurality of candidate solutions without mixed fragment measurement using a target algorithm and the initial population; generating a plurality of reference vectors in a target space, wherein the plurality of reference vectors divide the target space into different regions; Calculating the vertical projection distance of each reference vector of each candidate solution, dividing the candidate solutions into regions corresponding to the reference vectors based on the vertical projection distances, and generating a candidate solution set corresponding to the reference vector based on the candidate solutions in the region to which each reference vector belongs; Performing a global routing fragmentation metric on each candidate solution in the candidate solution set using the Gaussian process model, and performing a local routing fragmentation metric on each candidate solution using the first metric function; Determining a non-dominated solution in the candidate solution set corresponding to each reference vector based on the global routing fragmentation metric and the local routing fragmentation metric results of each candidate solution; A non-dominated solution set is generated based on the non-dominated solutions corresponding to all reference vectors, and the non-dominated solution set is passed to the lower-level management, wherein the lower-level management optimizes the non-dominated solutions in the non-dominated set based on the Gaussian process model and the second metric function.

6. The routing spectrum planning method based on hybrid fragmentation metric and two-layer model management according to claim 5, characterized in that: Optimizing the non-dominated solutions in the non-dominated set based on the Gaussian process model and the second metric function includes: Utilizing the Gaussian process model to proxy the global routing fragmentation metric; Calculating the expected super volume improvement of each non-dominated solution in the non-dominated solution set using a hybrid super volume improvement expectation calculation formula; A target non-dominated solution in the non-dominated set is determined based on the expected supervolume improvement of the non-dominated solution, the target non-dominated solution is truly evaluated using the second metric function, the non-dominated solution after the true evaluation is stored in the database, and the Gaussian process model is retrained using the updated database until the expensive evaluation budget is exhausted, thereby obtaining an optimized routing spectrum planning scheme.

7. A routing spectrum planning device based on hybrid fragmentation metric and two-layer model management, characterized in that: include: An acquisition module, used to acquire topology data and service data of the elastic optical network; a metric module, configured to initialize a routing spectrum planning scheme based on the service data, and perform a hybrid fragmentation metric on the routing spectrum planning scheme according to the topology data, wherein the hybrid fragmentation metric includes a local routing fragmentation metric and a global routing fragmentation metric, the local routing fragmentation metric includes an impact of a number of continuous idle frequency slot clusters based on local routing, and the global routing fragmentation metric includes an impact of a number of continuous idle frequency slot clusters based on a global network at an embedded connection between a local and global network; An optimization module is used to use a Gaussian process model to proxy the global routing fragmentation metric, train the Gaussian process model, use the local routing fragmentation metric and the global fragmentation metric as optimization targets, and optimize the routing spectrum planning scheme based on the trained Gaussian process model and a two-layer model management strategy, wherein the two-layer model management strategy includes upper-layer management and lower-layer management, the upper-layer management uses a reference vector to select a non-dominated solution, and the lower-layer management determines the target non-dominated solution on which computing resources are concentrated during the Gaussian process model evaluation process based on the non-dominated solution.

8. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the routing spectrum planning method based on hybrid fragmentation metric and dual-layer model management according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program or instruction stored thereon, characterized in that: The computer program or instruction is executed by a processor to implement the routing spectrum planning method based on hybrid fragmentation metric and dual-layer model management according to any one of claims 1 to 6.

10. A computer program product comprising a computer program or instructions, characterized in that When the computer program or instruction is executed, the routing spectrum planning method based on hybrid fragmentation metric and dual-layer model management is implemented.

Images