Network capacity planning method and device based on Bayesian optimization, equipment and medium

Abstract

The invention provides a network capacity planning method and device based on Bayesian optimization, equipment and a medium, and aims to solve the problems of high complexity, low universality and poor generalization ability of methods provided by related technologies. The method comprises the following steps: determining a search space of resource capacity of each node and each link in a network according to service request distribution; determining a network capacity scheme of each round of iteration from the search space based on a historical evaluation point set; the prediction network capacity scheme represents that specific resource capacity is allocated to each node and each link in the network topology, and the historical evaluation point set comprises each network capacity scheme determined by historical iteration and a performance index achieved by the network capacity scheme; performing performance evaluation on the network capacity scheme to obtain a performance index; and determining the network capacity scheme corresponding to the optimal performance index in the historical iteration as the network capacity scheme corresponding to the service request distribution under the condition that the iteration end condition is satisfied.

Description

Technical Field

[0001] The present disclosure relates to the technical field of software defined networks, and in particular to a network capacity planning method, apparatus, device and medium based on Bayesian optimization. Background Art

[0002] In software-defined networking (SDN), infrastructure providers (InP) provide various physical network resources such as computing, storage, and bandwidth; Internet service providers (ISP) rent physical network resources from InP and provide users with a variety of application services. Given a physical network topology, the ISP needs to plan the capacity of the network resources rented on each node and link, that is, the network capacity plan.

[0003] In software-defined networks, user business demands are highly dynamic. When the distribution of business requests changes significantly, ISPs need to adjust network capacity plans accordingly to allow the network to adapt to changing business demands and ensure service quality.

[0004] Related technologies propose to use various machine learning methods or deep reinforcement learning training to take the performance indicators that ISPs care about as reward functions, and obtain the optimal network capacity solution by simulating the interaction between the agent and the environment. However, the methods proposed in related technologies require model training under a given network environment and service deployment solution, facing the problem of poor generalization ability, and the methods proposed in related technologies usually contain a large number of model parameters, resulting in a huge amount of calculation. Summary of the invention

[0005] In order to overcome the problems existing in the related art, the present disclosure provides a network capacity planning method, device, equipment and medium based on Bayesian optimization. The technical solution of the present disclosure is as follows: According to a first aspect of an embodiment of the present disclosure, a network capacity planning method based on Bayesian optimization is provided, comprising: Determine, according to the service request distribution, a search space for resource capacity of each node and each link in the network; the search space represents a selection range of the resource capacity; Based on the historical evaluation point set, a network capacity solution for each round of iteration is determined from the search space; the predicted network capacity solution represents: allocating specific resource capacity to each node and each link in the network topology, and the historical evaluation point set includes: each network capacity solution determined by historical iteration and the performance index achieved; Performing performance evaluation on the network capacity solution to obtain performance indicators; When the iteration end condition is met, the network capacity solution corresponding to the optimal performance indicator in the historical iteration is determined as the network capacity solution corresponding to the service request distribution.

[0006] Optionally, determining a search space for resource capacity of each node and each link in the network according to the service request distribution includes: Allocating a corresponding initial search space of resource capacity to each node and each link in the network topology; Based on the service request distribution and the network topology, determining the shortest path between the starting and ending points of each group of requests according to a target routing algorithm; Determine each of the shortest paths as a shortest path set; Determine the frequency of each link in the network topology in the shortest path set, and determine the corresponding frequency ratio; the frequency ratio represents: the frequency of the corresponding link being selected in the route used by the service request distribution; Re-allocating the search space of the corresponding optimized resource capacity for each node and each link in the network topology according to the frequency ratio; According to the optimized search space of resource capacity, the search space of resource capacity of each node and each link in the network is determined.

[0007] Optionally, based on the historical evaluation point set, determining a network capacity solution for each round of iteration from the search space includes: Determine a historical evaluation point set; any evaluation point in the historical evaluation point set represents a performance indicator under a certain network capacity solution; Determining a proxy model corresponding to this round of iteration according to the historical evaluation point set; the proxy model represents an implicit functional relationship between a network capacity solution and a performance indicator; Based on the proxy model, determining an acquisition function; the acquisition function represents: the exploration value of each network capacity solution determined from the search space based on the proxy model; We will explore the network capacity solution with the greatest value and determine it as the network capacity solution for this round of iterations.

[0008] Optionally, it also includes: Determine, according to historical changes in request distribution, a historical service request distribution corresponding to the service request distribution; Determine historical iteration data of the historical service request distribution; the historical iteration data is historical iteration data optimized for a network capacity solution; Through the transfer learning algorithm, the agent model is pre-trained before the iteration starts according to the historical iteration data.

[0009] Optionally, a set of historical evaluation points is determined, including: Based on the service deployment history, an initial evaluation point set is obtained; the initial evaluation point set at least includes evaluation points consisting of an initial network capacity solution and corresponding performance indicators; The network capacity solutions and corresponding performance indicators obtained in each round of iteration before the current round of iteration are combined to obtain various evaluation points; The evaluation point is added to the initial evaluation point set to obtain the historical evaluation point set.

[0010] Optionally, a performance evaluation is performed on the network capacity solution to obtain a performance indicator, including: Perform preset independent simulations according to the service deployment strategy actually adopted to obtain corresponding performance indicators; wherein a set of requests is generated for each simulation according to the business request distribution; Determine the mean performance index and the variance performance index corresponding to the preset number of independent simulations according to each of the performance indicators; A performance indicator of the network capacity solution is determined according to the mean performance indicator and the variance performance indicator.

[0011] Optionally, determining a network capacity solution to be explored in each iteration from the search space includes: Determine constraint conditions according to actual needs, wherein the constraint conditions represent constraint conditions of the network capacity solution; Based on the constraint condition, a network capacity solution to be explored in each round of iteration is determined from the search space.

[0012] According to a second aspect of an embodiment of the present disclosure, a network capacity planning device based on Bayesian optimization is provided, comprising: A search space setting module, used to determine the search space of resource capacity of each node and each link in the network according to the service request distribution; the search space represents the selection range of the resource capacity; A scheme determination module is used to determine the network capacity scheme of each iteration from the search space based on a historical evaluation point set; the predicted network capacity scheme represents: allocating specific resource capacity to each node and each link in the network topology, and the historical evaluation point set includes: each network capacity scheme determined by historical iterations and the performance indicators achieved therein; A performance evaluation module, used to perform performance evaluation on the network capacity solution to obtain performance indicators; The iteration end module is used to determine the network capacity solution corresponding to the optimal performance indicator in the historical iteration as the network capacity solution corresponding to the business request distribution when the iteration end condition is met.

[0013] According to a third aspect of an embodiment of the present disclosure, there is provided an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the computer program is executed by the processor, the steps of the network capacity planning method based on Bayesian optimization as described in the first aspect are implemented.

[0014] According to a fourth aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the network capacity planning method based on Bayesian optimization as described in the first aspect are implemented.

[0015] According to a fifth aspect of an embodiment of the present disclosure, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps of the network capacity planning method based on Bayesian optimization as described in the first aspect are implemented.

[0016] The present disclosure dynamically determines the search space of the network topology according to the distribution of business requests, and can adapt to the changes in network capacity requirements at different times and for different user needs. The present disclosure does not rely on a specific network environment and service deployment scheme for model training, but uses a Bayesian optimization algorithm to gradually approach the optimal solution through iterative search and performance evaluation, and has a stronger generalization ability. The present disclosure uses a historical evaluation point set to guide the search process, avoiding a comprehensive search in the entire search space, and can effectively reduce the amount of calculation. The present disclosure can use the performance indicators that ISPs are concerned about as optimization targets based on actual business needs and network environments, flexibly select and optimize performance indicators, and provide higher quality services. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings required for use in the description of the embodiments of the present disclosure will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0018] Figure 1 is a schematic diagram of the steps of a network capacity planning method based on Bayesian optimization shown in an embodiment of the present disclosure; Figure 2 is a schematic diagram of a test network topology shown in an embodiment of the present disclosure; Figure 3 is a schematic diagram of a Bayesian optimization convergence performance comparison diagram shown in an embodiment of the present disclosure; Figure 4is a block diagram of a network capacity planning device based on Bayesian optimization shown in an embodiment of the present disclosure; Figure 5 Schematic diagram of an electronic device according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present disclosure to clearly and completely describe the technical solutions in the embodiments of the present disclosure. Obviously, the described embodiments are part of the embodiments of the present disclosure, rather than all of the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present disclosure.

[0020] The terms "first", "second", etc. in the specification and claims of the present disclosure are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable when appropriate, so that the embodiments of the present disclosure can be implemented in an order other than those illustrated or described herein, and the objects distinguished by "first", "second", etc. are generally of the same type, and the number of objects is not limited. For example, the first object can be one or more. In addition, "and / or" in the specification and claims represents at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0021] The difference between network capacity planning and network resource allocation is that the former takes the resource capacity of each part of the network as an optimization variable, while the latter mainly optimizes the deployment plan of the service, in which the resource capacity of each node and link in the network is fixed.

[0022] To address the problem of network capacity planning, related technologies have proposed various machine learning methods to train neural network models by analyzing service request history and network resource allocation history to predict the distribution of future business requests and network resource requirements, or to use deep reinforcement learning (DRL) (Deep Reinforcement Learning) to train performance indicators that ISPs are concerned about as reward functions, and to obtain the optimal network capacity solution by simulating the interaction between the agent and the environment.

[0023] However, the methods proposed in the related art have certain problems.

[0024] Problem 1: The methods proposed by related technologies usually require model training under a given network environment and service deployment scheme, and face the problem of poor generalization ability. When the service deployment method or the performance indicators that ISPs are concerned about change, the model performance may drop significantly, resulting in low versatility of existing solutions and difficulty in adapting to different network environments and service deployment scenarios. In addition, for more complex service deployment processes, the accuracy of predicting network resource requirements using existing methods may drop, affecting the performance of network capacity planning.

[0025] Question 2: The methods proposed by related technologies usually contain a large number of model parameters, which need to be optimized during the training process, resulting in a huge amount of calculation. In actual application scenarios, ISPs may need to respond quickly to changes in business needs. The high computational complexity of existing methods may cause the model to be unable to make decisions quickly. In addition, deep learning-based models may require high-performance GPUs (Graphic Processing Units) to accelerate the training process, which not only increases hardware costs, but also limits the application of existing methods in scenarios with limited computing resources.

[0026] In order to solve the above technical problems, the present disclosure proposes a network capacity planning method based on Bayesian optimization. The method adopts a network capacity planning algorithm based on Bayesian optimization, thereby effectively solving the problems of low universality and poor generalization of the methods proposed in related technologies.

[0027] Figure 1 Schematic diagram of a method for network capacity planning based on Bayesian optimization shown in an embodiment of the present disclosure. Figure 1 As shown, the method may specifically include the following steps: Step S11: Determine the search space of the resource capacity of each node and each link in the network according to the service request distribution; the search space represents the selection range of the resource capacity.

[0028] Service request distribution refers to the frequency and pattern of various service requests received within a specific time period.

[0029] Determine the search space representation of the resource capacity of each node and each link in the network, and determine the search space corresponding to the resource capacity of each node and link. The search space represents the range of resource capacity selection, such as the minimum and maximum bandwidth on a link.

[0030] Network topology represents the connection relationship between nodes and links in the network. Network topology can be expressed as , the total number of nodes and links are V and E respectively. The search space can be expressed as , represents the upper and lower bounds of the resource capacity of each node and link in the network.

[0031] Collect the distribution data of business requests, analyze the traffic patterns and user needs in different time periods, and determine the search space of network topology based on user needs.

[0032] Step S12: Based on the historical evaluation point set, determine the network capacity plan for each round of iteration from the search space; the predicted network capacity plan represents: allocating specific resource capacity to each node and each link in the network topology, and the historical evaluation point set includes: each network capacity plan determined by historical iterations and the performance indicators achieved by it.

[0033] The historical evaluation point set is obtained in the initialization phase and can be used to store the network capacity solution and its corresponding performance indicators determined in each round of iteration. Specifically, several evaluation points are generated in the historical iteration, and each evaluation point is used to represent a network capacity solution that achieves a certain performance indicator. The network capacity solution and its corresponding performance indicator are taken as an evaluation point and added to the historical evaluation point set.

[0034] Performance indicators can be indicators such as latency, throughput, packet loss rate, etc. that can reflect the effectiveness of network services.

[0035] In each round of iteration, based on the historical evaluation point set, the network capacity solution obtained in the current round of iteration is determined from the search space.

[0036] The performance indicators corresponding to the evaluation points in the historical evaluation point set are indicators that the current ISP is concerned about and needs to be optimized. For example, when the indicator that the ISP is concerned about and needs to be optimized is the request rejection rate, the performance indicator may represent the number of rejected requests.

[0037] In the initialization phase of the network capacity planning algorithm, an initial network capacity plan can be set as the input of the first round of iterations, and the performance of the initial network capacity plan can be evaluated.

[0038] Step S13: Perform performance evaluation on the network capacity solution to obtain performance indicators.

[0039] After obtaining the network capacity plan in each round of iteration, the network capacity plan is simulated and tested, that is, the network configured according to the network capacity plan is used to simulate the deployment of requests, evaluate its performance under the distribution of specific business requests, and obtain the performance indicators of the current network capacity plan. The performance indicators are the indicators that the current ISP is concerned about and needs to be optimized.

[0040] Performance indicator data can be used to analyze network performance under different network capacity solutions, including service quality, resource utilization, etc.

[0041] Step S14: When the iteration end condition is met, the network capacity solution corresponding to the optimal performance indicator in the historical iteration is determined as the network capacity solution corresponding to the service request distribution.

[0042] You can set the conditions for the end of iteration. The conditions for the end of iteration can be set as reaching the preset performance index, after a certain number of iterations, or when the performance index has not improved significantly in several consecutive rounds.

[0043] At the end of each round of iteration, check whether the set iteration end condition is met. If the iteration end condition is met, end the iteration and take the current iteration as the last iteration.

[0044] In all historical iterations, the performance indicators of various network capacity solutions are compared to find the network capacity solution with the best performance. The optimal network capacity solution is determined as the final network capacity solution for business request distribution and is ready for implementation.

[0045] After the network capacity plan is implemented, network performance can be continuously monitored to ensure that it meets business needs and can be dynamically adjusted based on actual conditions.

[0046] By adopting the embodiments of the present disclosure, the network capacity solution is determined according to the service request distribution, which can ensure the best matching state between the network capacity solution and the service request distribution, thereby improving the utilization rate of network resources. By determining the network capacity solution for each round of iteration based on the historical evaluation point set, it is possible to make full use of prior information and iteratively search for a set of optimal network capacity solutions. By continuously optimizing the network capacity solution in an iterative manner, it is possible to cope with changes in the service request distribution, thereby achieving flexible adjustment according to the actual needs of the ISP, and realizing unified optimization of various heterogeneous network resources such as computing, storage, and bandwidth.

[0047] In some cases, due to the upper limit of physical resources of nodes and links and in order to accelerate the convergence of Bayesian optimization, constraints can be set and the network capacity solution can be determined based on the constraints. The following is a specific embodiment of applying the constraints.

[0048] Among them, in an optional embodiment, based on the historical evaluation point set, the network capacity solution for each round of iteration is determined from the search space, including: determining constraints according to actual needs, the constraints characterizing the constraints of the network capacity solution; based on the constraints, determining the network capacity solution that needs to be explored in each round of iteration from the search space.

[0049] Constraints can be defined based on the actual needs of the ISP In the specific implementation, the resource rental cost can usually be used as a constraint. Assume that the price of renting a computing resource per unit resource on each node is , the price of renting a unit bandwidth resource on each link is , the total cost of ISP leasing network resources is capped at , the amount of resources rented by ISP on each node and link is expressed as , then the constraint condition It can be expressed as: .

[0050] It should be noted that in addition to cost constraints, performance-related constraints can also be considered, such as request response time, service quality, etc.

[0051] Integrate the constraints into the optimization model of network capacity planning to ensure that the network capacity solution obtained in each iteration satisfies these constraints in the process of solving the optimal solution.

[0052] By adopting the embodiments of the present disclosure, the network capacity solutions to be explored are screened out from the search space based on the constraints in each round of iteration, thereby avoiding the exploration of invalid solutions and reducing unnecessary calculation and simulation time. In actual applications, service requests and resource prices may change. By setting constraints, the ISP can flexibly adjust the restrictions of the network capacity solution according to the actual situation to ensure that it can quickly respond to changes in the distribution of service requests.

[0053] In each round of iteration, for a given service request distribution, how to evaluate the performance of the network capacity solution obtained in the current iteration is a problem that needs to be solved. To address this problem, the Monte Carlo method can be used to evaluate the performance of a network capacity solution through multiple simulations. The following is a specific embodiment of the present disclosure for simulation.

[0054] In an optional embodiment, the network capacity solution is performance evaluated to obtain performance indicators, including: performing a preset number of independent simulations according to the service deployment strategy actually adopted to obtain corresponding performance indicators; generating a group of requests for each simulation according to the business request distribution; determining the mean performance indicator and variance performance indicator corresponding to the preset number of independent simulations according to each of the performance indicators; and determining the performance indicator of the network capacity solution according to the mean performance indicator and the variance performance indicator.

[0055] Determine the actual service deployment strategy, which can include network node configuration, link load balancing, etc. According to the service deployment strategy, build a simulation system that can simulate the actual network environment.

[0056] The Monte Carlo method is used to perform a preset number of simulations through the simulation system, for example, Q independent simulations. Each simulation generates a set of requests based on the business request distribution, which can be expressed as ,The request can be implemented by a random number generator.,Each request contains necessary features such as the requested service type,,request start and end points, traffic size, etc., so that it can be used in subsequent simulations.

[0057] These requests are deployed in a simulation environment. During each simulation, the performance indicators that the current ISP needs to optimize are collected, such as response time and success rate. After each simulation, the corresponding performance indicators are recorded and stored as structured data. For example, the performance indicator corresponding to the qth simulation in the nth iteration can be recorded as .

[0058] The performance indicators corresponding to all independent simulations are statistically analyzed to calculate the mean performance indicator and the variance performance indicator. The mean performance indicator reflects the performance of the network capacity solution under average conditions, while the variance performance indicator reveals the fluctuation of the performance indicator. For the nth iteration, the mean performance indicator can be expressed as , the variance performance index can be expressed as .

[0059] Based on the mean performance index and variance performance index, a comprehensive performance evaluation of the network capacity solution is performed. Specifically, the evaluation can be performed using the following formula:

[0060] in, is an adjustable weight parameter. This adjustment balances the impact of mean and variance in the final performance evaluation.

[0061] The meaning is: the network capacity solution obtained in this round of iteration Performance under .

[0062] In getting Performance under After that, it is taken as a tuple and added to the evaluation point set to construct a new evaluation point set to provide data support for the next round of iteration.

[0063] By adopting the embodiments of the present disclosure, through multiple independent simulations based on the service deployment strategy actually adopted, the performance of the network capacity solution in actual applications can be more realistically reflected, and the reliability and accuracy of the evaluation results can be improved. A set of requests is generated for each simulation based on the business request distribution, so that the evaluation process can take into account the randomness and uncertainty of the business requests, and can better simulate the request situation in the real environment, thereby improving the effectiveness of the performance evaluation. In addition, by simulating the service deployment process to evaluate network performance, it is not limited by the specific algorithm of service deployment and can be used in a variety of network environments and service deployment scenarios. By presetting multiple independent simulations, the performance of the network capacity solution in different situations can be captured, the randomness of a single simulation result can be reduced, and the robustness of the evaluation results can be enhanced.

[0064] Among them, in an optional embodiment, based on the historical evaluation point set, the network capacity plan for each round of iteration is determined from the search space, including: determining the historical evaluation point set; any evaluation point in the historical evaluation point set represents the performance index under a certain network capacity plan; determining the proxy model corresponding to this round of iteration according to the historical evaluation point set; the proxy model represents the implicit functional relationship between the network capacity plan and the performance index; based on the proxy model, determining the acquisition function; the acquisition function represents: based on the proxy model, the exploration value of each network capacity plan determined from the search space; the network capacity plan with the largest exploration value is determined as the network capacity plan for this round of iteration.

[0065] The network capacity solutions and corresponding performance indicators obtained from historical iterations are collected, and each historical capacity solution and its corresponding performance indicator are regarded as an evaluation point.

[0066] Select a suitable proxy model based on the historical evaluation point set, and use the data from the historical evaluation point set to train the proxy model. Specifically, the proxy model can accurately reflect the implicit functional relationship between the network capacity solution and the performance index by fitting the model parameters. By fitting the existing evaluation point set, the performance of the unevaluated solution can be predicted.

[0067] According to the characteristics of the proxy model and the characteristics of the search space, a suitable acquisition function is designed. The acquisition function is used to evaluate the exploration value of each network capacity solution determined from the search space under the proxy model. The acquisition function can help the network capacity planning algorithm based on Bayesian optimization decide which solution should be evaluated next.

[0068] In each round of iteration, the network capacity solution with the highest exploration value is found by maximizing the acquisition function. The network capacity solution with the highest exploration value represents the network capacity solution that is most likely to bring performance improvement under the current proxy model. The network capacity solution with the highest exploration value is taken as the solution for this round of iteration.

[0069] With the embodiments of the present disclosure, the proxy model is used to approximate the functional relationship between complex network capacity solutions and performance indicators. The proxy model can quickly predict the performance indicators of different network capacity solutions without directly performing actual evaluation, reducing the consumption of computing resources and time and improving iteration efficiency. The design of the acquisition function enables the algorithm to select those solutions that are most likely to bring performance improvements under the current proxy model, thereby effectively exploring the search space.

[0070] Among them, in an optional embodiment, determining a historical evaluation point set includes: obtaining an initial evaluation point set based on the service deployment history; the initial evaluation point set at least includes evaluation points consisting of an initial network capacity plan and corresponding performance indicators; combining the network capacity plans and corresponding performance indicators obtained in each round of iteration before the current round of iteration to obtain each evaluation point; adding the evaluation point to the initial evaluation point set to obtain the historical evaluation point set.

[0071] Historical data can be collected from the ISP's service deployment system, including detailed records of each service deployment, including network capacity plans, start and end points of service requests, deployment time, performance indicators, etc. Performance indicators can include service quality, request acceptance rate, total deployment revenue, etc.

[0072] The performance indicators that need to be paid attention to can be determined based on the actual needs of the current ISP. They can be performance indicators related to service quality, such as latency, bandwidth utilization, packet loss rate, etc., performance indicators related to request acceptance rate, such as the ratio of the number of successfully processed requests to the total number of requests, or performance indicators related to total deployment revenue, such as the difference between the cost of ISP leasing resources and the revenue from successful business deployment.

[0073] The collected historical data is organized into an evaluation point set. Each evaluation point in the evaluation point set can be represented as a tuple, which contains the network capacity plan and the corresponding performance indicator. The evaluation point can be represented as , The meaning is that in a certain network capacity solution Performance indicators under .

[0074] In the initial stage before Bayesian optimization, a set of network capacity solutions can be set up based on historical data, business request distribution and network requirements, and used as the initial solution as the input for the first round of Bayesian optimization iteration. The performance indicators of the initial solution are then obtained, and the initial solution and the corresponding performance indicators are used as an evaluation point and added to the initial evaluation point set.

[0075] In addition, in each round of iteration, the network capacity plan corresponding to each round of iteration and the performance indicators corresponding to the plan will be obtained. According to the network capacity plan and the corresponding performance indicators obtained in each round of iteration, new evaluation points are formed and added to the initial evaluation point set, thereby continuously enriching and improving the evaluation point set.

[0076] By using the embodiments of the present disclosure, by using the service deployment historical data, the initialization evaluation point set can provide the model with a real reference based on past performance, which helps to improve the accuracy of the agent model in the initial stage and reduce the time and resource consumption of model training. The initialization evaluation point set can provide a better starting point for Bayesian optimization, thereby accelerating the convergence process and reducing the number of iterations required to reach the optimal solution. Adding the evaluation points generated during the iteration process to the initial evaluation point set can build a more comprehensive and rich historical evaluation point set, and because the evaluation point set is dynamically updated, the algorithm can continue to learn and improve its strategy.

[0077]

[0078] Table 1 Table 1 shows the steps of a network capacity planning algorithm based on Bayesian optimization shown in an embodiment of the present disclosure. Referring to Table 1, steps 1-3 are the initialization phase, steps 4-14 are the iterative optimization phase, including multiple rounds of iterations, each round of iteration includes an evaluation phase of steps 8-11, and the evaluation phase includes multiple independent simulations.

[0079] Step 1: Define the search space and constraints .

[0080] Step 2: Initialize network capacity plan .

[0081] set up Set a set of network capacity plans for the resource rental amount of the ISP on each node and link as an initial solution.

[0082] Step 3: Initialize the evaluation point set .

[0083] Initialize the evaluation point set based on service deployment history , where each evaluation point can be expressed as , that is, in a certain network capacity solution Performance indicators under ,This performance indicator is determined according to the actual needs of the ISP and can be set as service quality, request acceptance rate, total deployment revenue, etc.

[0084] Step 4-14: After initialization, the iterative process of Bayesian optimization begins. Step 4-14 is executed once in each iteration.

[0085] Step 5: According to Building a proxy model .

[0086] In each iteration, a proxy model is constructed based on the existing set of evaluation points. , surrogate model for estimating network capacity solutions and performance indicators The implicit functional relationship between them.

[0087] Step 6: Construct a sampling function based on the surrogate model .

[0088] Compute the acquisition function based on the proxy model , the acquisition function can use models such as expected improvement or probability improvement to measure the search space The exploration value of various network capacity solutions in .

[0089] Step 7: Calculate the current network capacity solution: .

[0090] By maximizing the acquisition function, we can get the network capacity solution that needs to be selected in this round of iteration .

[0091] Steps 8-11: Perform Q independent simulations based on the service deployment strategy actually adopted by the ISP to evaluate the performance of the network capacity solution obtained in step 7.

[0092] Step 9: Distribution based on business requests Generate a set of requests .

[0093] Step 10: Simulate deployment , get the network performance under the current network capacity solution .

[0094] Step 12: Evaluate the current network capacity solution Overall performance under .

[0095] In order to improve the evaluation accuracy, the mean of Q simulation results is considered at the same time and variance Determine the current network capacity plan The overall performance under.

[0096] Specifically expressed as a network capacity solution Performance under ,in is an adjustable weight parameter.

[0097] Step 13: Expand the evaluation point set: .

[0098] The evaluation points obtained in this round of iteration Added to the set of evaluation points for updating the proxy model and acquisition function in the next iteration.

[0099] Step 15: After the iteration process is completed, select the optimal network capacity solution from the iteration history .

[0100] When using Bayesian optimization for network capacity planning, another key issue is how to improve the convergence speed of the network capacity planning algorithm. This is because in each round of iteration, the service deployment process needs to be simulated multiple times, and complex service deployment scenarios may cause each round of iteration to take a long time, generally in minutes. Therefore, in order to improve the efficiency of Bayesian optimization, it is necessary to minimize the number of iterations required for the algorithm to converge. The present disclosure provides two methods for accelerating convergence to optimize the network capacity planning algorithm, namely Reduced Search Space (RSS) and Transfer Learning (TL). Each method of accelerating convergence can make full use of various types of prior information to guide the convergence of Bayesian optimization.

[0101] Method 1: Narrowing the search space Among them, in an optional embodiment, the search space of resource capacity of each node and each link in the network is determined according to the service request distribution, including: allocating a corresponding initial search space of resource capacity to each node and each link in the network topology; based on the service request distribution and the network topology, according to the target routing algorithm, determining the shortest path between the starting and ending points of each group of requests; determining each of the shortest paths as a shortest path set; determining the frequency of each link in the network topology in the shortest path set, and determining the corresponding frequency ratio; the frequency ratio represents: the frequency of the corresponding link being selected in the route used by the service request distribution; through the frequency ratio, reallocating the corresponding optimized search space of resource capacity to each node and each link in the network topology; according to the optimized search space of resource capacity, determining the search space of resource capacity of each node and each link in the network.

[0102] In the initial stage of Bayesian optimization, we can first make a rough estimate of the network resource demand, and then narrow the search space based on the understanding of the traffic demand characteristics in different time periods and locations. This can be done in the following way.

[0103] An initial resource capacity search space is allocated for each link and each node in the network topology. The initial resource capacity search space can be determined based on the basic design parameters of the network and the preset capacity range, for example, according to the bandwidth, delay, load and other characteristics of the link.

[0104] Based on the business request distribution and network topology, a target routing algorithm is selected to calculate the shortest path between the starting and ending points of each group of requests. The target routing algorithm can be a routing calculation algorithm such as the Dijkstra algorithm.

[0105] All the calculated shortest paths are aggregated into a shortest path set. The shortest path set contains all possible service traffic paths in the network under the service request distribution. The shortest path set can be expressed as .

[0106] Traverse the shortest path set, count the frequency of each link in the network topology appearing in the shortest path set, and determine the frequency ratio by the ratio of the link frequency to the sum of all link frequencies. This frequency ratio reflects the frequency of the link being selected in the route used for business request distribution.

[0107] Analyze the statistical results to identify links with higher frequencies and links with lower frequencies. Links with higher frequencies can be understood as hot links. Hot links usually show high traffic loads and are frequently used.

[0108] According to the frequencies of different links being selected under the service request distribution, the search space of resource capacity is reallocated for each link and each node, so as to achieve the reduction of the search space.

[0109] Through the frequency ratio, the search space of the corresponding optimized resource capacity is reallocated for each node and each link in the network topology. For example, a larger search space can be allocated to a link with a relatively high frequency and its corresponding node, which can specifically expand the capacity selection range of the link and can also include increasing the upper limit of the search space; a smaller search space can be allocated to a link with a relatively low frequency and its corresponding node, which can specifically narrow the capacity selection range of the link and can also include lowering the upper limit of the search space.

[0110] According to the initial search space and the optimized search space, a search space of resource capacity of each node and each link in the network corresponding to the network topology is determined. In a subsequent Bayesian optimization process, a network capacity solution is determined from the search space.

[0111] Among them, after analyzing and determining the hot links and low-frequency links, the initial capacity value can be assigned to each link according to the frequency ratio of each link, and input into the first round of Bayesian optimization as the initial network capacity plan. Compared with the initial network capacity plan determined by the initial search space, the initial network capacity plan determined after narrowing the search space can provide a better starting point for Bayesian optimization.

[0112] By adopting the embodiments of the present disclosure, by analyzing the distribution of service requests and setting a larger search space for high-frequency links, higher service demands under the service request distribution can be met, and the search space can be reduced for low-frequency links, which can effectively reduce unnecessary search and calculation burdens, thereby effectively utilizing prior information about the network and service demands, making the search more accurate and efficient. Since the search space has been optimized according to the service request distribution and network topology, Bayesian optimization can perform iterative search in a smaller search space, which not only reduces the number of iterations, but also improves the efficiency of each iteration.

[0113] Method 2: Transfer Learning In order to improve the convergence speed of Bayesian, transfer learning (TL) can also be used to improve the convergence speed. The following is a specific example of using transfer learning to speed up the convergence speed of Bayesian.

[0114] Among them, in an optional embodiment, it also includes: determining the historical business request distribution corresponding to the business request distribution according to the periodic changes in the request distribution; determining the historical business request distribution and optimizing the historical iteration data for the network capacity solution; and pre-training the proxy model before the iteration starts according to the historical iteration data through a transfer learning algorithm.

[0115] Since the distribution of business requests often changes periodically in actual business application scenarios, the optimization history can be used as prior information to improve the efficiency of Bayesian optimization.

[0116] Real-time monitoring of business request data, including key information such as request time, type, and quantity. The business request data obtained through monitoring can be used to identify historical changes in the distribution of business requests using statistical methods, such as time series analysis or machine learning algorithms.

[0117] According to the cycle phase of the current request distribution, a historical service request distribution matching the current cycle phase is selected from the historical data. The historical service request distribution can be verified to determine whether the historical service request distribution is comparable to the request distribution currently targeted for iteration.

[0118] The historical iteration data may be stored in the historical database, so that the historical iteration data of the network capacity solution corresponding to the selected historical service request distribution may be found in the historical database. The historical iteration data may include information such as the network capacity solution for each iteration, the corresponding performance indicators, and the proxy model.

[0119] You can choose a suitable transfer learning algorithm based on the characteristics and goals of the historical iterative data, such as feature-based transfer, instance-based transfer, or model-based transfer.

[0120] The data of historical optimization tasks is used as input to construct a training set. The selected transfer learning algorithm is used to train the proxy model to capture the relationship between the business request distribution and the network capacity plan in the historical optimization tasks, so as to transfer the knowledge and experience in the historical iteration data to the currently running Bayesian optimization-based capacity planning algorithm, and train a proxy model that matches the current business request distribution.

[0121] In a specific implementation, for a service request distribution , assuming that there are K iterations of Bayesian optimization before, before building the proxy model for the business request distribution, that is, before running the network capacity planning algorithm, As input, use transfer learning algorithms such as RGPE (Relational Gaussian Process Expertise), SGPR (Sparse Gaussian Process Regression), and TransBO (Transfer Bayesian Optimization) to pre-train the proxy model .

[0122] By adopting the embodiments of the present disclosure, by determining the proxy model corresponding to this round of iteration based on the evaluation point set and using the transfer learning algorithm to integrate the prior information in the historical data, the convergence speed of Bayesian optimization in dealing with periodically changing business request distribution can be significantly improved, which not only improves the efficiency of network capacity planning, but also reduces the consumption of computing resources.

[0123] In order to verify the effects of the two methods for accelerating convergence provided by the present disclosure, the present disclosure can be verified by simulation experiments. The simulation experiments respectively test the network capacity planning algorithm OSS without the optimization of method one and method two, the network capacity planning algorithm RSS only with the optimization of method one, and the network capacity planning algorithm TL only with the optimization of method two.

[0124] Figure 2FIG. 1 is a schematic diagram of a test network topology shown in an embodiment of the present disclosure. Figure 2 As shown, the test network topology This is the network topology used for simulation experiment verification. Test network topology There are 15 nodes and 30 links in total. Figure 2 The larger nodes have computing power and can deploy services; the smaller nodes have no computing power and can only forward service data.

[0125] Based on the test network topology , now given a business request distribution , in business request distribution The possible starting and ending points of the request are Figure 2 The ISP plans to deploy 200 service requests in the network, of which 150 requests will have their start and end points randomly selected from the above nodes with rectangular frames, and the other 50 requests will have their start and end points randomly selected from the entire network.

[0126] Since network bandwidth resources are limited, when the price of renting a unit of bandwidth resources on each link is 1, the constraint condition is set as the total cost of renting network resources. .

[0127] The optimization goal is to minimize the request rejection rate, that is, the performance indicator The number of rejected requests out of 200 requests.

[0128] In order to verify the effectiveness of the method 1 for accelerating convergence provided by the present disclosure. Figure 2 Given a network topology and a given business request distribution, use different search spaces to run the network capacity planning algorithm. The initial search space and the optimized search space are shown in the following table:

[0129] Table 2: Initial search space

[0130] Table 3: Optimized search space Among them, the initial bandwidth composition of each link in Table 2 is the initial network capacity solution determined in the initial search space, and the initial bandwidth composition of each link in Table 3 is the initial network capacity solution determined in the optimized search space. Compared with the initial network capacity solution determined in the initial search space, the initial network capacity solution determined in the optimized search space can provide a better initial solution for Bayesian optimization.

[0131] In order to verify the effectiveness of the second method for accelerating convergence provided by the present disclosure, 10 Bayesian optimization iteration histories are used as input to pre-train the proxy model, and the initial search space is used for Bayesian optimization.

[0132] OSS, RSS, and TL were used to conduct 30 independent experiments, each with 200 iterations. The convergence of Bayesian optimization is shown in Figure 3 shown.

[0133] Figure 3 1 is a schematic diagram of a Bayesian optimization convergence performance comparison diagram shown in an embodiment of the present disclosure. Figure 3 As shown in the figure, the convergence curve composed of dots is the average convergence of OSS in 30 independent experiments, the convergence curve composed of triangles is the average convergence of RSS in 30 independent experiments, and the convergence curve composed of rectangles is the average convergence of TL in 30 independent experiments. The corresponding colored range represents the maximum fluctuation range of the convergence curve in 30 experiments.

[0134] Comparing the convergence of the three groups of experiments, OSS, RSS, and TL, the convergence speed of RSS and TL is significantly faster than that of OSS, and the fluctuation of the convergence curve is significantly smaller than that of OSS, and OSS failed to reach convergence within 200 iterations. Comparing RSS and TL, it can be found that RSS reaches convergence faster than TL because it can determine a better initial network capacity solution while narrowing the search space.

[0135] based on Figure 3 The simulation results of OSS, RSS, and TL are shown. The method 1 and method 2 provided by the present disclosure to improve the convergence speed can effectively accelerate the convergence of the network capacity planning algorithm, thereby improving the real-time performance of the algorithm and helping the ISP to quickly plan new network capacity when the service request changes.

[0136] Based on the same technical concept, the present disclosure provides a network capacity planning device based on Bayesian optimization. Figure 4 1 is a block diagram of a network capacity planning device based on Bayesian optimization shown in an embodiment of the present disclosure. Figure 4 As shown, the device comprises: A search space setting module 410 is used to determine the search space of the resource capacity of each node and each link in the network according to the service request distribution; the search space represents the selection range of the resource capacity; A solution determination module 420 is used to determine the network capacity solution of each iteration from the search space based on the historical evaluation point set; the predicted network capacity solution represents: allocating specific resource capacity to each node and each link in the network topology, and the historical evaluation point set includes: each network capacity solution determined by historical iteration and the performance index achieved; A performance evaluation module 430 is used to perform performance evaluation on the network capacity solution to obtain performance indicators; The iteration ending module 440 is used to determine the network capacity solution corresponding to the optimal performance indicator in the historical iteration as the network capacity solution corresponding to the service request distribution when the iteration ending condition is met.

[0137] Optionally, the range determination module is specifically configured to execute: Allocating a corresponding initial search space of resource capacity to each node and each link in the network topology; Based on the service request distribution and the network topology, determining the shortest path between the starting and ending points of each group of requests according to a target routing algorithm; Determine each of the shortest paths as a shortest path set; Determine the frequency of each link in the network topology in the shortest path set, and determine the corresponding frequency ratio; the frequency ratio represents: the frequency of the corresponding link being selected in the route used by the service request distribution; Re-allocating the search space of the corresponding optimized resource capacity for each node and each link in the network topology according to the frequency ratio; According to the optimized search space of resource capacity, the search space of resource capacity of each node and each link in the network is determined.

[0138] Optionally, the solution determination module is specifically used to execute: Determine a historical evaluation point set; any evaluation point in the historical evaluation point set represents a performance indicator under a certain network capacity solution; Determining a proxy model corresponding to this round of iteration according to the historical evaluation point set; the proxy model represents an implicit functional relationship between a network capacity solution and a performance indicator; Based on the proxy model, determining an acquisition function; the acquisition function represents: the exploration value of each network capacity solution determined from the search space based on the proxy model; We will explore the network capacity solution with the greatest value and determine it as the network capacity solution for this round of iterations.

[0139] Optionally, the device further comprises: A historical business request distribution determination module, used to determine the historical business request distribution corresponding to the business request distribution according to the historical changes in the request distribution; A historical iteration data determination module, used to determine the historical iteration data of the historical service request distribution; the historical iteration data is the historical iteration data optimized for the network capacity solution; The pre-training module is used to pre-train the agent model before the iteration starts according to the historical iteration data by using a transfer learning algorithm.

[0140] Optionally, the solution determination module is specifically used to execute: Based on the service deployment history, an initial evaluation point set is obtained; the initial evaluation point set at least includes evaluation points consisting of an initial network capacity solution and corresponding performance indicators; The network capacity solutions and corresponding performance indicators obtained in each round of iteration before the current round of iteration are combined to obtain various evaluation points; The evaluation point is added to the initial evaluation point set to obtain the historical evaluation point set.

[0141] Optionally, the performance evaluation module is specifically used to execute: The network capacity solution is evaluated for performance to obtain performance indicators, including: Perform preset independent simulations according to the service deployment strategy actually adopted to obtain corresponding performance indicators; wherein a set of requests is generated for each simulation according to the business request distribution; Determine the mean performance index and the variance performance index corresponding to the preset number of independent simulations according to each of the performance indicators; A performance indicator of the network capacity solution is determined according to the mean performance indicator and the variance performance indicator.

[0142] Optionally, the solution determination module is specifically used to execute: Determine constraint conditions according to actual needs, wherein the constraint conditions represent constraint conditions of the network capacity solution; Based on the constraint condition, a network capacity solution to be explored in each round of iteration is determined from the search space.

[0143] It should be noted that the device embodiment is similar to the method embodiment, so the description is relatively simple, and the relevant parts can be referred to the method embodiment.

[0144] The present disclosure also provides an electronic device, referring to Figure 5 , Figure 5 Schematic diagram of an electronic device according to an embodiment of the present disclosure. Figure 5As shown, the electronic device 500 includes: a memory 510 and a processor 520. The memory 510 and the processor 520 are connected via a bus communication. A computer program is stored in the memory 510. The computer program can be run on the processor 520 to implement the steps in the network capacity planning method based on Bayesian optimization disclosed in the embodiment of the present disclosure.

[0145] The embodiment of the present disclosure further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps in the network capacity planning method based on Bayesian optimization disclosed in the embodiment of the present disclosure are implemented.

[0146] The embodiment of the present disclosure further provides a computer program product, including a computer program. When the computer program is executed by a processor, the steps in the network capacity planning method based on Bayesian optimization disclosed in the embodiment of the present disclosure are implemented.

[0147] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0148] It should be understood by those skilled in the art that the embodiments of the present disclosure may be provided as methods, devices or computer program products. Therefore, the embodiments of the present disclosure may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the embodiments of the present disclosure may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0149] The embodiments of the present disclosure are described with reference to the flowcharts and / or block diagrams of the methods, apparatuses, electronic devices, and computer program products according to the embodiments of the present disclosure. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0150] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0151] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device so that a series of operating steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable terminal device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0152] Although some embodiments of the present disclosure have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiment and all changes and modifications that fall within the scope of the present disclosure.

[0153] The above is a detailed introduction to a network capacity planning method, device, equipment and medium based on Bayesian optimization provided by the present disclosure. Specific examples are used in this article to illustrate the principles and implementation methods of the present disclosure. The description of the above embodiments is only used to help understand the method of the present disclosure and its core idea. At the same time, for general technical personnel in this field, according to the ideas of the present disclosure, there will be changes in the specific implementation methods and application scopes. In summary, the content of this specification should not be understood as a limitation on the present disclosure.

Claims

1. A network capacity planning method based on Bayesian optimization, characterized in that: include: Determine the search space of resource capacity of each node and each link in the network according to the distribution of service requests; The search space represents a selection range of the resource capacity; Determining a network capacity solution for each round of iteration from the search space based on a set of historical evaluation points; The network capacity solution characterizes: allocating specific resource capacity to each node and each link in the network topology, and the historical evaluation point set includes: each network capacity solution determined by historical iteration and the performance index achieved by it; Performing performance evaluation on the network capacity solution to obtain performance indicators; When the iteration end condition is met, the network capacity solution corresponding to the optimal performance indicator in the historical iteration is determined as the network capacity solution corresponding to the service request distribution.

2. The method according to claim 1, characterized in that: According to the service request distribution, the search space for resource capacity of each node and each link in the network is determined, including: Allocating a corresponding initial search space of resource capacity to each node and each link in the network topology; Based on the service request distribution and the network topology, determining the shortest path between the starting and ending points of each group of requests according to a target routing algorithm; Determine each of the shortest paths as a shortest path set; Determine the frequency of each link in the network topology in the shortest path set, and determine the corresponding frequency ratio; the frequency ratio represents: the frequency of the corresponding link being selected in the route used by the service request distribution; Re-allocating the search space of the corresponding optimized resource capacity for each node and each link in the network topology according to the frequency ratio; According to the optimized search space of resource capacity, the search space of resource capacity of each node and each link in the network is determined.

3. The method according to claim 1, characterized in that Based on the historical evaluation point set, a network capacity solution for each round of iteration is determined from the search space, including: Determine a historical evaluation point set; any evaluation point in the historical evaluation point set represents a performance indicator under a certain network capacity solution; Determining a proxy model corresponding to this round of iteration according to the historical evaluation point set; the proxy model represents an implicit functional relationship between a network capacity solution and a performance indicator; Based on the proxy model, determining an acquisition function; the acquisition function represents: the exploration value of each network capacity solution determined from the search space based on the proxy model; We will explore the network capacity solution with the greatest value and determine it as the network capacity solution for this round of iterations.

4. The method according to claim 3, characterized in that Also includes: Determine, according to historical changes in request distribution, a historical service request distribution corresponding to the service request distribution; Determine historical iteration data of the historical service request distribution; the historical iteration data is historical iteration data optimized for a network capacity solution; Through the transfer learning algorithm, the agent model is pre-trained before the iteration starts according to the historical iteration data.

5. The method according to claim 3, characterized in that: Determine the set of historical assessment points, including: Based on the service deployment history, an initial evaluation point set is obtained; the initial evaluation point set at least includes evaluation points consisting of an initial network capacity solution and corresponding performance indicators; The network capacity solutions and corresponding performance indicators obtained in each round of iteration before the current round of iteration are combined to obtain various evaluation points; The evaluation point is added to the initial evaluation point set to obtain the historical evaluation point set.

6. The method according to claim 1, characterized in that The network capacity solution is evaluated for performance to obtain performance indicators, including: Perform preset independent simulations according to the service deployment strategy actually adopted to obtain corresponding performance indicators; wherein a set of requests is generated for each simulation according to the business request distribution; Determine the mean performance index and the variance performance index corresponding to the preset number of independent simulations according to each of the performance indicators; A performance indicator of the network capacity solution is determined according to the mean performance indicator and the variance performance indicator.

7. The method according to any one of claims 1 to 6, characterized in that: Based on the historical evaluation point set, a network capacity solution for each round of iteration is determined from the search space, including: Determine constraint conditions according to actual needs, wherein the constraint conditions represent constraint conditions of the network capacity solution; Based on the constraint condition, a network capacity solution to be explored in each round of iteration is determined from the search space.

8. A network capacity planning device based on Bayesian optimization, characterized in that: include: A search space setting module, used to determine the search space of resource capacity of each node and each link in the network according to the distribution of service requests; The search space represents a selection range of the resource capacity; A solution determination module, used to determine the network capacity solution for each round of iteration from the search space based on a historical evaluation point set; The predicted network capacity solution represents: allocating specific resource capacity to each node and each link in the network topology, and the historical evaluation point set includes: each network capacity solution determined by historical iterations and the performance indicators achieved therein; A performance evaluation module, used to perform performance evaluation on the network capacity solution to obtain performance indicators; The iteration end module is used to determine the network capacity solution corresponding to the optimal performance indicator in the historical iteration as the network capacity solution corresponding to the business request distribution when the iteration end condition is met.

9. 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 when the computer program is executed by the processor, the steps of the network capacity planning method based on Bayesian optimization are implemented as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the network capacity planning method based on Bayesian optimization are implemented as described in any one of claims 1 to 7.

Images