Method, device and equipment for optimizing network topology structure

Abstract

The invention provides a network topology structure optimization method, device and equipment. The method comprises the following steps: acquiring a network topology structure diagram of a preset network; inputting the network topology structure diagram into a trained network topology search optimization model for network topology structure optimization adjustment to obtain a target network topology structure diagram; wherein according to a topology classifier based on a graph neural network and a network topology structure diagram of a training network, a searcher based on deep reinforcement learning is trained to obtain the network topology search optimization model; and outputting the target network topology structure diagram. According to the scheme, the automatic optimization of the network topology structure is realized, and the optimization efficiency of the network topology structure is improved.

Description

Technical Field

[0001] The present invention relates to the field of communication technology, and in particular to a method, device and equipment for optimizing a network topology structure. Background Art

[0002] With the development of communication network technology, various network services are increasingly integrated into people's daily lives. As the scale of the network continues to expand, the growing traffic demand has brought huge challenges to the capacity guarantee of the transmission network loop, and the supporting infrastructure needs to be continuously updated. In the process of supporting the update of the underlying optical fiber of the transmission network, it is necessary to plan the current complex network. Network planning is a highly complex task with diverse factors. It is necessary to fully consider the future growth trend of traffic, the overall network delay and load balancing, and effectively deal with problems such as unbalanced loop utilization. By reasonably reorganizing the load imbalance loop, the global refined structure optimization of the network can be achieved to fully reduce the cost of manpower and material resources and improve the network utilization efficiency. The existing technical solutions are based on manual adjustment based on the experience of human experts, or based on heuristic search to complete the network planning task. These methods can achieve a certain degree of optimization on small-scale network structures, but with the continuous expansion of the scale of network topology, the advantages in this aspect are not obvious, and they are not competent for the network topology optimization tasks in large-scale complex environments. Summary of the invention

[0003] The technical problem to be solved by the present invention is to provide a network topology optimization method, device and equipment to achieve automatic optimization of the network topology, get rid of the dependence on human experts, and reduce the calculation cost and complexity in the process of network performance evaluation.

[0004] In order to solve the above technical problems, an embodiment of the present invention provides a method for optimizing a network topology structure, comprising:

[0005] Get the network topology diagram of the preset network;

[0006] The network topology structure diagram is input into a trained network topology search optimization model to optimize and adjust the network topology structure to obtain a target network topology structure diagram; wherein, according to a topology classifier based on a graph neural network and the network topology structure diagram of a training network, a searcher based on deep reinforcement learning is trained to obtain the network topology search optimization model;

[0007] Output the target network topology diagram.

[0008] Optionally, according to the topological classifier of the graph neural network and the network topological structure diagram of the training network, a searcher based on deep reinforcement learning is trained to obtain the network topology search optimization model, including:

[0009] Initialize the network parameters φ(t) of the topological classifier, the network parameters θ of the searcher, the total number of episodes N for training, and the number of steps H for each episode;

[0010] The training is performed in cycles with episodes as units. In each cycle, the searcher based on deep reinforcement learning is run every H steps with steps as units, and a running result is obtained once, and N running results are obtained;

[0011] generating a set of network topologies using an actor network in the searcher according to the network topology diagram of the training network;

[0012] Classifying the network topology structures in the group of network topology structures according to the topology classifier to obtain a classification result;

[0013] A network topology search optimization model is obtained according to the N operation results and the classification results.

[0014] Optionally, classifying the network topology structures in the group of network topology structures according to the topology classifier to obtain a classification result includes:

[0015] Obtaining a value of an objective function of each network topology in the set of network topologies; wherein the objective function is determined according to a value function and a cost function of a network topology graph of the training network, the value function being used to represent the performance of the training network, and the cost function being used to represent the loss between the training network and the adjacency matrix of the training network;

[0016] The group of network topological structures are classified according to the value of the objective function to obtain a classification result.

[0017] Optionally, the objective function f(x)=U(x)+γCost(x,x0);

[0018] Among them, U(x) is the value function, Cost(x,x0) is the cost function, the network topology graph of the training network x is G=(V,E), V is the set of network element nodes in the training network x, E is the set of edges in the network topology graph of the training network x, and the edges are loop connections between network elements in the training network x, x0 represents the adjacency matrix of the network topology graph of the training network x, and γ is a constant parameter.

[0019] Optionally, the constraints of the objective function f(x) include:

[0020]

[0021]

[0022] M(x) = True;

[0023] Dist(e) means that the distance between two nodes in the training network does not exceed D and the nodes can be connected. Load(e) means that the bandwidth of the network loop in the training network does not exceed L. max , M(x) indicates that the network topology diagram of the training network must satisfy the legal connection relationship.

[0024] Optionally, determining a network topology search optimization model according to the N operation results and the classification result includes:

[0025] If f(x) in the classification result exceeds a preset threshold, the next training process uses a graph neural network to replace the initial objective function to obtain an objective function value and a reward function value;

[0026] According to the objective function value and the reward function value, the target graph neural network parameters φ(t) and the target deep reinforcement learning network parameters θ are corrected to obtain a network topology search optimization model.

[0027] Optionally, the network topology structure diagram of the preset network is input into a trained network topology search optimization model to optimize and adjust the network topology structure to obtain a target network topology structure diagram, including:

[0028] The network topology diagram of the preset network is input into the actor network in the searcher of the trained network topology search optimization model for operation to generate a target network topology structure.

[0029] The present invention also provides a network topology optimization device, comprising:

[0030] An acquisition module is used to acquire a network topology diagram of a preset network;

[0031] A processing module is used to input the network topology structure diagram into a trained network topology search optimization model to optimize and adjust the network topology structure to obtain a target network topology structure diagram; wherein, according to a topology classifier based on a graph neural network and the network topology structure diagram of a training network, a searcher based on deep reinforcement learning is trained to obtain the network topology search optimization model; and the target network topology structure diagram is output.

[0032] The present invention also provides a communication device, comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus;

[0033] The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform operations corresponding to the above method.

[0034] The present invention also provides a computer-readable storage medium storing instructions, which, when executed on a computer, enable the computer to execute the above method.

[0035] The above solution of the present invention includes at least the following beneficial effects:

[0036] The network topology structure diagram of the acquired preset network is input into the trained network topology search optimization model to optimize and adjust the network topology structure, so as to obtain a target network topology structure diagram; and the target network topology structure diagram is output; wherein, the network topology search optimization model is obtained by training a searcher based on deep reinforcement learning through a topology classifier based on a graph neural network and the network topology structure diagram of the training network; the automatic optimization of the network topology structure is realized, while reducing the computational cost and complexity in the process of network performance evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 is a flow chart of a method for optimizing a network topology structure provided by an embodiment of the present invention;

[0038] Figure 2 It is a module block diagram of a device for optimizing a network topology structure provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0039] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0040] like Figure 1 As shown, an embodiment of the present invention provides a method for optimizing a network topology structure, comprising:

[0041] Step 11, obtaining a network topology diagram of a preset network;

[0042] Step 12, inputting the network topology structure diagram of the preset network into the trained network topology search optimization model to optimize and adjust the network topology structure to obtain a target network topology structure diagram; wherein, according to the topology classifier based on the graph neural network and the network topology structure diagram of the training network, the searcher based on deep reinforcement learning is trained to obtain the network topology search optimization model;

[0043] Step 13: output the target network topology diagram.

[0044] In this embodiment, the network topology structure diagram of the preset network can be expressed as G1=(V1, E1), wherein V1 represents the set of all network element nodes in the network topology structure of the preset network, and E1 represents the set of all edges in the network topology structure of the preset network, which can be understood as a loop connection between network element nodes in a physical sense;

[0045] The network topology search optimization model described here may include a topology classifier based on a graph neural network and a searcher based on deep reinforcement learning; wherein the topology classifier is used to classify the network topology structure generated during the training process, and is used to optimize the computational cost and computational complexity in the process of evaluating the performance of the network topology structure; the searcher is used to repeatedly train based on the classified network topology structure, and then determine the relevant parameters of the network topology search optimization model;

[0046] The network topology search optimization model may also include a feasibility determiner of complex network topology links, which may determine the feasibility of the network connection relationship of the initially input preset network topology structure and the newly generated network topology structure based on the constraints of the network topology structure, so as to ensure the accuracy of subsequent searcher training and the accuracy of optimizing the topology structure of the preset network;

[0047] By optimizing the topology of the preset network through the searcher in the trained network topology search optimization model, automatic search and optimization of the network topology structure is realized, the accuracy and efficiency of the network topology structure adjustment and optimization are improved, and excessive reliance on artificial expert experience and the low efficiency of heuristic search are avoided; at the same time, the searcher based on deep reinforcement learning adjusts the network topology structure, making full use of the globality of the network topology structure and improving the performance of the optimized network topology structure.

[0048] In an optional embodiment of the present invention, according to a topological classifier based on a graph neural network and a network topological structure diagram of a training network, training a searcher based on deep reinforcement learning to obtain the network topology search optimization model may include:

[0049] Step 201, initializing the network parameter φ(t) of the topology classifier, the network parameter θ of the searcher, the total number of episodes N for training, and the number of steps H for each episode;

[0050] Step 202, perform cyclic training with episodes as units, and in each cycle, run the searcher based on deep reinforcement learning every H steps with steps as units, obtain a running result, and obtain N running results;

[0051] Step 203, generating a set of network topologies using the actor network in the searcher according to the network topology diagram of the training network;

[0052] Step 204, classifying the network topology structures in the group of network topology structures according to the topology classifier to obtain a classification result;

[0053] Step 205: Determine a network topology search optimization model according to the N operation results and the classification results.

[0054] In this embodiment, the network parameter φ(t) of the topology classifier, the network parameter θ of the searcher, the total number of training episodes N of the network topology structure of the training network, and the number of steps H of each episode can all be set according to the actual operation of the model and the performance of the network topology structure of the training network; the setting of the initial related parameters can also include a performance threshold Q of the network topology structure;

[0055] Before the topology structure of the training network is input into the searcher for operation, the feasibility determiner can also be used to judge the feasibility of the topology structure of the training network based on the loop connection of the network topology structure of the training network, the compliance of the network topology structure and the network element, and the performance threshold Q. After meeting the feasibility requirements, the searcher in the network topology search optimization model is trained based on the network topology structure of the training network and the deep reinforcement learning model;

[0056] Specifically, during training: the network topology of the training network is input into a searcher based on deep reinforcement learning, and cyclic training is performed in the searcher in units of scenes. Here, the deep reinforcement learning algorithm can be the A2C reinforcement learning algorithm in the existing scheme; in each cycle, the searcher based on deep reinforcement learning is run every H steps in units of steps, and the network topology of the training network is used as the state of reinforcement learning, and the changes between network element nodes are used as actions to obtain a running result. After N scenes of cyclic running (after N cyclic running), N running results are obtained; wherein, each run corresponds to a set of running results including: network topology state elements, changes between network element nodes as action elements, and value reward elements of the topology structure, and the corresponding elements of the running results are cached; here, the value reward of the topology structure can be obtained by calculating the objective function corresponding to the topology structure, and can also be obtained by approximate calculation by fitting the objective function of the GCN graph convolutional neural network algorithm;

[0057] Further, the network topology of the training network can be adjusted by using the actor network in the searcher and some elements in the cached N sets of operation results, and a new set of network topology structures is generated by interacting with the environment through some elements in the cached N sets of operation results. It should be noted that the set of network topology structures actually includes N sets of new network topology structures.

[0058] Further, the newly generated network topology structure in the set of network topology structures is classified according to the topology classifier to obtain a classification result; the classification result may include but is not limited to: the strength of the performance of the newly generated network topology structure, etc.; the separation processing here can be regarded as screening the newly generated network topology structure in the set of network topology structures;

[0059] By classifying the newly generated set of network topology structures through the topology classifier, the original method of network performance evaluation calculation can be replaced, and the computational cost and complexity in the network performance evaluation process can be optimized; then, based on the classification result and the N operation results, the network topology search optimization model is determined, and the network topology structure of the preset network is subsequently input into the model to realize automatic optimization of the network topology structure of the preset network, thereby improving the efficiency and accuracy of network topology structure optimization.

[0060] In an optional embodiment of the present invention, the above step 204 may include:

[0061] Step 2041, obtaining a value of an objective function of each network topology structure in the set of network topologies; the objective function is determined according to a value function and a cost function of the network topology graph of the training network, the value function is used to represent the performance of the training network, and the cost function is used to represent the loss between the training network and the adjacency matrix of the training network;

[0062] Step 2042: classify the group of network topology structures according to the value of the objective function to obtain a classification result.

[0063] Furthermore, the objective function f(x)=U(x)+γCost(x,x0);

[0064] In the above formula, U(x) represents the value function, Cost(x,x0) represents the cost function, and the network topology structure graph of the training network x can be expressed as G=(V,E), wherein V is a set of network element nodes in the training network x, E is a set of edges in the network topology structure graph of the training network x, and the edges represent loop connections between network elements in the training network x, x0 represents the adjacency matrix of the network topology structure graph of the training network x, and γ represents a constant parameter, and γ<0, which is used to adjust the relationship between the value function U(x) and the cost function Cost(x,x0);

[0065] In this embodiment, the objective function of each network topology structure is determined according to the value function corresponding to the network performance of each network topology structure in the newly generated set of network topology structures and the cost function corresponding to the loss between each network topology structure and its connection matrix;

[0066] Further, based on the value of the objective function of each network topology structure, the network topology structures in the group of network topology structures are classified and processed; the network topology structures corresponding to the values ​​of the objective function greater than the performance threshold Q are stored in the set of candidate topology structures, and the network topology structures corresponding to the values ​​of the objective function less than or equal to the performance threshold Q can be directly deleted, and the network topology search optimization model is subsequently trained and determined based on the network topology structures in the set of candidate topology structures, which can ensure the accuracy of the model determination, and at the same time, the newly generated network topology structures are classified and processed by the topology classifier, which accelerates the process of model training optimization, thereby improving the efficiency of network topology structure optimization;

[0067] Preferably, before classifying the network topology structures in the group of network topology structures, the feasibility of the network topology structures in the group of network topology structures may be judged by the constraint conditions of the objective function corresponding to each network topology structure;

[0068] Furthermore, the constraints of the objective function f(x) include:

[0069]

[0070]

[0071] M(x) = True;

[0072] Among them, Dist(e)<D is the distance constraint between network element nodes in the network topology structure, indicating that the distance between two nodes in the training network does not exceed the preset distance D. The nodes can be connected, and Load(e)<L max is the network loop utility constraint of the network topology structure, indicating that the bandwidth of the network loop in the training network does not exceed the preset maximum bandwidth L max , M(x) = True is the legality constraint of the network topology (actually the set of legality constraints in), indicating that the network topology diagram of the training network must satisfy the legal connection relationship, such as the number of network element nodes, the order of loop connections, etc.;

[0073] The feasibility of the network topology structure in the group of network topology structures is judged by the constraint conditions of the objective function, and it is determined that the newly generated network topology structure that meets the constraint conditions is feasible. Subsequently, the network topology structure based on the feasibility is classified and processed, which further accelerates the model training and optimization efficiency.

[0074] In an optional embodiment of the present invention, the above step 205 may include:

[0075] Step 2051, if f(x) in the classification result exceeds a preset threshold, the next training process uses a graph neural network to replace the initial objective function to obtain an objective function value and a reward function value;

[0076] Step 2052: According to the objective function value and the reward function value, the target graph neural network parameter φ(t) and the target deep reinforcement learning network parameter θ are modified to obtain a network topology search optimization model.

[0077] In this embodiment, when the value of the objective function f(x) of the network topology structure in the classification result is greater than a preset threshold, the objective function of the network topology structure initially input is replaced with the objective function greater than the preset threshold in the next training process, and the above steps are repeated until the number of loops N ends, and the final objective function is obtained, and then the replaced objective function value is calculated, and the reward function value of the objective function can be obtained by fitting the objective function through the GCN algorithm.

[0078] Furthermore, based on the objective function determined after the end of the cycle and the reward function value obtained based on the objective function fitting, the initial parameters φ(t) of the target graph neural network and the initial parameters θ of the target deep reinforcement learning network are corrected to obtain the target parameters φ(t) and the target parameters θ, and then the network topology search optimization model can be obtained to facilitate the subsequent optimization processing of the network model of the preset network.

[0079] In an optional embodiment of the present invention, the above step 12 may include:

[0080] Step 121, the network topology diagram of the preset network is input into the actor network in the searcher of the trained network topology search optimization model for operation to generate a target network topology structure.

[0081] In this embodiment, the network topology structure diagram of the preset network is input into the trained network topology search optimization model, and based on the actor network in the searcher of the trained network topology search optimization model and the objective function of the network topology structure of the preset network, the network topology structure is adjusted to obtain the target network topology structure of the network topology structure of the preset network;

[0082] In the above-mentioned embodiment of the present invention, based on the topological classifier of the graph neural network and the network topological structure diagram of the training network, the searcher based on deep reinforcement learning is trained to obtain a network topology search optimization model; then, the network topology structure of the preset network is optimized according to the trained network topology search optimization model, thereby avoiding excessive reliance on artificial expert experience and the defects of low search optimization efficiency, and realizing automatic search optimization of the network topology structure; the network search and optimization are implemented by the deep reinforcement learning algorithm, which makes full use of the globality of the network topology structure and improves the network topology structure optimization performance.

[0083] like Figure 2 As shown, an embodiment of the present invention further provides a network topology optimization device 20, comprising:

[0084] An acquisition module 21 is used to acquire a network topology diagram of a preset network;

[0085] The processing module 22 is used to input the network topology structure diagram of the preset network into the trained network topology search optimization model to optimize and adjust the network topology structure to obtain a target network topology structure diagram; wherein, according to the topology classifier based on the graph neural network and the network topology structure diagram of the training network, the searcher based on deep reinforcement learning is trained to obtain the network topology search optimization model; and the target network topology structure diagram is output.

[0086] Optionally, according to the topological classifier of the graph neural network and the network topological structure diagram of the training network, a searcher based on deep reinforcement learning is trained to obtain the network topology search optimization model, including:

[0087] Initialize the network parameters φ(t) of the topological classifier, the network parameters θ of the searcher, the total number of episodes N for training, and the number of steps H for each episode;

[0088] The training is performed in cycles with episodes as units. In each cycle, the searcher based on deep reinforcement learning is run every H steps with steps as units, and a running result is obtained once, and N running results are obtained;

[0089] generating a set of network topologies using an actor network in the searcher according to the network topology diagram of the training network;

[0090] Classifying the network topology structures in the group of network topology structures according to the topology classifier to obtain a classification result;

[0091] A network topology search optimization model is obtained according to the N operation results and the classification results.

[0092] Optionally, classifying the network topology structures in the group of network topology structures according to the topology classifier to obtain a classification result includes:

[0093] Obtaining a value of an objective function of each network topology in the set of network topologies; wherein the objective function is determined according to a value function and a cost function of a network topology graph of the training network, the value function being used to represent the performance of the training network, and the cost function being used to represent the loss between the training network and the adjacency matrix of the training network;

[0094] The group of network topological structures are classified according to the value of the objective function to obtain a classification result.

[0095] Optionally, the objective function f(x)=U(x)+γCost(x,x0);

[0096] Among them, U(x) is the value function, Cost(x,x0) is the cost function, the network topology graph of the training network x is G=(V,E), V is the set of network element nodes in the training network x, E is the set of edges in the network topology graph of the training network x, and the edges are loop connections between network elements in the training network x, x0 is the adjacency matrix of the network topology graph of the training network x, and γ is a constant parameter.

[0097] Optionally, the constraints of the objective function f(x) include:

[0098]

[0099]

[0100] M(x) = True;

[0101] Dist(e) means that the distance between two nodes in the training network does not exceed D and the nodes can be connected. Load(e) means that the bandwidth of the network loop in the training network does not exceed L. max , M(x) indicates that the network topology diagram of the training network must satisfy the legal connection relationship.

[0102] Optionally, determining a network topology search optimization model according to the N operation results and the classification result includes:

[0103] If f(x) in the classification result exceeds a preset threshold, the next training process uses a graph neural network to replace the initial objective function to obtain an objective function value and a reward function value;

[0104] According to the objective function value and the reward function value, the target graph neural network parameters φ(t) and the target deep reinforcement learning network parameters θ are corrected to obtain a network topology search optimization model.

[0105] Optionally, the processing module 22 inputs the network topology structure diagram of the preset network into a trained network topology search optimization model to optimize and adjust the network topology structure, and when obtaining the target network topology structure diagram, is specifically used for:

[0106] The network topology diagram of the preset network is input into the actor network in the searcher of the trained network topology search optimization model for operation to generate a target network topology structure.

[0107] It should be noted that the device is a device corresponding to the above-mentioned network topology optimization method, and all implementation methods in the above-mentioned method embodiments are applicable to the embodiments of the device and can achieve the same technical effects.

[0108] The embodiment of the present invention further provides a communication device, comprising: a processor and a memory storing a computer program, wherein when the computer program is executed by the processor, the method described above is executed. All implementations in the above method embodiment are applicable to this embodiment and can achieve the same technical effect.

[0109] The embodiment of the present invention further provides a computer-readable storage medium, including instructions, which, when executed on a computer, enable the computer to execute the method described above. All implementations in the above method embodiment are applicable to this embodiment and can achieve the same technical effect.

[0110] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0111] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0112] In the embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0113] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0114] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0115] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, ROM, RAM, magnetic disks, or optical disks.

[0116] In addition, it should be noted that in the apparatus and method of the present invention, it is obvious that each component or each step can be decomposed and / or recombined. These decompositions and / or recombinations should be regarded as equivalent schemes of the present invention. Moreover, the steps of performing the above-mentioned series of processing can naturally be performed in chronological order according to the order of description, but it is not necessary to perform them in chronological order, and some steps can be performed in parallel or independently of each other. For those of ordinary skill in the art, it is understood that all or any steps or components of the method and apparatus of the present invention can be implemented in any computing device (including processors, storage media, etc.) or a network of computing devices in hardware, firmware, software or a combination thereof, which can be achieved by those of ordinary skill in the art using their basic programming skills after reading the description of the present invention.

[0117] Therefore, the purpose of the present invention can also be achieved by running a program or a group of programs on any computing device. The computing device can be a well-known general device. Therefore, the purpose of the present invention can also be achieved by simply providing a program product containing a program code that implements the method or device. That is to say, such a program product also constitutes the present invention, and the storage medium storing such a program product also constitutes the present invention. Obviously, the storage medium can be any well-known storage medium or any storage medium developed in the future. It should also be pointed out that in the device and method of the present invention, it is obvious that each component or each step can be decomposed and / or recombined. These decompositions and / or recombinations should be regarded as equivalent schemes of the present invention. In addition, the steps of performing the above-mentioned series of processing can naturally be performed in chronological order according to the order of description, but it is not necessary to perform them in chronological order. Some steps can be performed in parallel or independently of each other.

[0118] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for optimizing a network topology structure, characterized in that: include: Get the network topology diagram of the preset network; The network topology structure diagram is input into a trained network topology search optimization model to optimize and adjust the network topology structure to obtain a target network topology structure diagram; wherein, according to a topology classifier based on a graph neural network and the network topology structure diagram of a training network, a searcher based on deep reinforcement learning is trained to obtain the network topology search optimization model; Output the target network topology diagram.

2. The method for optimizing network topology according to claim 1, characterized in that: According to the topological classifier of the graph neural network and the network topological structure diagram of the training network, the searcher based on deep reinforcement learning is trained to obtain the network topology search optimization model, including: Initialize the network parameters φ(t) of the topological classifier, the network parameters θ of the searcher, the total number of episodes N for training, and the number of steps H for each episode; The training is performed in cycles with episodes as units. In each cycle, the searcher based on deep reinforcement learning is run every H steps with steps as units, and a running result is obtained once, and N running results are obtained; generating a set of network topologies using an actor network in the searcher according to the network topology diagram of the training network; Classifying the network topology structures in the group of network topology structures according to the topology classifier to obtain a classification result; A network topology search optimization model is obtained according to the N operation results and the classification results.

3. The method for optimizing network topology according to claim 2, characterized in that: Classifying the network topology structures in the group of network topology structures according to the topology classifier to obtain classification results includes: Obtaining a value of an objective function of each network topology in the set of network topologies; wherein the objective function is determined according to a value function and a cost function of a network topology graph of the training network, the value function being used to represent the performance of the training network, and the cost function being used to represent the loss between the training network and the adjacency matrix of the training network; The group of network topological structures are classified according to the value of the objective function to obtain a classification result.

4. The method for optimizing network topology according to claim 3, characterized in that: The objective function f(x)=U(x)+γCost(x,x0); Among them, U(x) is the value function, Cost(x,x0) is the cost function, the network topology graph of the training network x is G=(V,E), V is the set of network element nodes in the training network x, E is the set of edges in the network topology graph of the training network x, and the edges are loop connections between network elements in the training network x, x0 is the adjacency matrix of the network topology graph of the training network x, and γ is a constant parameter.

5. The method for optimizing network topology according to claim 4, characterized in that: The constraints of the objective function f(x) include: M(x) = True; Dist(e) means that the distance between two nodes in the training network does not exceed D and the nodes can be connected. Load(e) means that the bandwidth of the network loop in the training network does not exceed L. max , M(x) indicates that the network topology diagram of the training network must satisfy the legal connection relationship.

6. The method for optimizing network topology according to claim 3, characterized in that: Determining a network topology search optimization model according to the N operation results and the classification result includes: If f(x) in the classification result exceeds a preset threshold, the next training process uses a graph neural network to replace the initial objective function to obtain an objective function value and a reward function value; According to the objective function value and the reward function value, the target graph neural network parameters φ(t) and the target deep reinforcement learning network parameters θ are corrected to obtain a network topology search optimization model.

7. The method for optimizing network topology according to claim 1, characterized in that: The network topology structure diagram of the preset network is input into the trained network topology search optimization model to optimize and adjust the network topology structure to obtain the target network topology structure diagram, including: The network topology diagram of the preset network is input into the actor network in the searcher of the trained network topology search optimization model for operation to generate a target network topology structure.

8. A device for optimizing a network topology structure, characterized in that: include: An acquisition module is used to acquire a network topology diagram of a preset network; A processing module is used to input the network topology structure diagram into a trained network topology search optimization model to optimize and adjust the network topology structure to obtain a target network topology structure diagram; wherein, according to a topology classifier based on a graph neural network and the network topology structure diagram of a training network, a searcher based on deep reinforcement learning is trained to obtain the network topology search optimization model; and the target network topology structure diagram is output.

9. A communication device, characterized in that: include: A processor and a memory storing a computer program, wherein when the computer program is executed by the processor, the method according to any one of claims 1 to 7 is performed.

10. A computer-readable storage medium, characterized in that: Instructions are stored, and when the instructions are executed on a computer, the computer is caused to execute the method according to any one of claims 1 to 7.

Images