Movatterモバイル変換


[0]ホーム

URL:


CN118660014B - Dynamic load balancing method and system for optical communication device - Google Patents

Dynamic load balancing method and system for optical communication device
Download PDF

Info

Publication number
CN118660014B
CN118660014BCN202411134310.2ACN202411134310ACN118660014BCN 118660014 BCN118660014 BCN 118660014BCN 202411134310 ACN202411134310 ACN 202411134310ACN 118660014 BCN118660014 BCN 118660014B
Authority
CN
China
Prior art keywords
data
optical communication
graph
load
model
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202411134310.2A
Other languages
Chinese (zh)
Other versions
CN118660014A (en
Inventor
曲宝春
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Suzhou Aixiongsi Communication Technology Co ltd
Original Assignee
Suzhou Aixiongsi Communication Technology Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Suzhou Aixiongsi Communication Technology Co ltdfiledCriticalSuzhou Aixiongsi Communication Technology Co ltd
Priority to CN202411134310.2ApriorityCriticalpatent/CN118660014B/en
Publication of CN118660014ApublicationCriticalpatent/CN118660014A/en
Application grantedgrantedCritical
Publication of CN118660014BpublicationCriticalpatent/CN118660014B/en
Activelegal-statusCriticalCurrent
Anticipated expirationlegal-statusCritical

Links

Classifications

Landscapes

Abstract

The application provides a dynamic load balancing method and a system of optical communication devices, which relate to the field of optical communication equipment management, wherein the method comprises the steps of obtaining multi-dimensional time sequence data of each optical communication device; the method comprises the steps of extracting multi-dimensional data of each optical communication device corresponding to each historical time step in multi-dimensional time sequence data, constructing a corresponding historical graph structure according to each extracted multi-dimensional data, updating node characteristics of each graph node in the historical graph structure through a graph neural network, determining time sequence fusion characteristics based on the node characteristics of the corresponding graph nodes of each updated historical graph structure for each optical communication device, inputting the time sequence fusion characteristics into a flow prediction model to determine predicted flow load data corresponding to a future preset time period, and executing dynamic load balancing operation based on each predicted flow load data. Thus, accurate load prediction and intelligent load balancing operation under a dynamically changing network environment are realized.

Description

Dynamic load balancing method and system for optical communication device
Technical Field
The present application relates to the field of optical communication device management technologies, and in particular, to a dynamic load balancing method and system for an optical communication device.
Background
Optical communication technology has been the heart of modern communication systems and has been under development for decades. The advantages of high transmission speed, large bandwidth and strong interference resistance are achieved, so that the method is widely applied to various communication scenes, such as metropolitan area networks, long-distance communication, data center interconnection and the like. In these application scenarios, the optical communication technology not only improves the data transmission efficiency, but also enhances the reliability and stability of the communication network.
However, with the ever-increasing network size and the ever-increasing communications demands, challenges faced by optical communications devices in actual operation are also increasing, and in particular, in current optical communications networks, load balancing is primarily dependent on static configurations and predefined load distribution policies. Although these approaches solve the load distribution problem to some extent, their drawbacks are increasingly apparent as the network size and complexity increases.
In particular, static configuration methods typically rely on an administrator to manually set load distribution parameters, which may be effective at initial deployment of the network, but may be frustrating in the face of dynamic changes in network traffic. In addition, predefined load distribution policies are typically based on analysis of historical traffic data, but because network traffic tends to exhibit large instantaneous fluctuations, it is difficult for these policies to globally schedule communication resources sufficiently efficiently, resulting in some nodes being in an overload state and other nodes being idle during certain periods of time, resulting in wasted resources.
In view of the above problems, currently, no preferred technical solution is proposed.
Disclosure of Invention
The application provides a dynamic load balancing method, a system, a storage medium, a computer program product and electronic equipment of an optical communication device, which are used for at least solving the problems that the static configuration or the predefined load distribution strategy in the prior art cannot cope with the dynamic change of network traffic, and the resource waste or overload is easy to cause.
In a first aspect, an embodiment of the present application provides a dynamic load balancing method for an optical communication device, including obtaining multidimensional time sequence data of each optical communication device in an optical communication network to be managed; the multi-dimensional time sequence data comprises a plurality of historical time steps and corresponding multi-dimensional data, wherein the multi-dimensional data comprises flow load data, equipment state data and external environment data; the traffic load data comprises a bandwidth utilization rate, a total traffic, an uploading rate, a downloading rate, a packet loss rate and a transmission delay, wherein for each historical time step, the multi-dimensional data corresponding to each optical communication device in the multi-dimensional time sequence data are extracted, a corresponding historical graph structure is built according to each extracted multi-dimensional data, then node characteristics of each graph node in the historical graph structure are updated through a graph neural network, the historical graph structure comprises a plurality of graph nodes and edge connections, the node characteristics of each graph node are respectively defined by the multi-dimensional data of the corresponding optical communication device, each edge connection is used for connecting the graph nodes corresponding to the optical communication devices with a link transmission relation, for each optical communication device, a time sequence fusion characteristic is determined based on the node characteristics of the corresponding graph nodes of each updated historical graph structure, the time sequence fusion characteristic is input into a traffic prediction model to determine predicted traffic load data corresponding to a future time period, and dynamic load balancing operation is performed based on the predicted traffic load data corresponding to each optical communication device.
In a second aspect, an embodiment of the present application provides a dynamic load balancing system for an optical communication device, including a data obtaining unit, configured to obtain multidimensional time sequence data of each optical communication device in an optical communication network to be managed; the multi-dimensional time sequence data comprises a plurality of historical time steps and corresponding multi-dimensional data, wherein the multi-dimensional data comprises flow load data, equipment state data and external environment data; the traffic load data comprises a bandwidth utilization rate, a total traffic, an uploading rate, a downloading rate, a packet loss rate and a transmission delay, a characteristic extraction unit, a traffic prediction unit and a load balancing unit, wherein the characteristic extraction unit is used for extracting multi-dimensional data corresponding to each optical communication device in the multi-dimensional time sequence data for each historical time step, constructing a corresponding historical graph structure according to each extracted multi-dimensional data, updating node characteristics of each graph node in the historical graph structure through a graph neural network, the historical graph structure comprises a plurality of graph nodes and edge connections, the node characteristics of each graph node are respectively defined by the multi-dimensional data of the corresponding optical communication device, each edge connection is used for connecting the graph node corresponding to the optical communication device with a link transmission relation, the traffic prediction unit is used for determining a time sequence fusion characteristic based on the node characteristics of the corresponding graph node of each updated historical graph structure for each optical communication device, inputting the time sequence fusion characteristic into a traffic prediction model so as to determine predicted traffic load data corresponding to a future preset time period, and the load balancing unit is used for carrying out dynamic load balancing operation on the basis of the predicted traffic load data of each optical communication device.
In a third aspect, an electronic device is provided that includes at least one processor and a memory communicatively coupled to the at least one processor, wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the steps of the dynamic load balancing method of the optical communication apparatus of any of the embodiments of the present application.
In a fourth aspect, an embodiment of the present application provides a storage medium having a computer program stored thereon, wherein the program when executed by a processor implements the steps of the dynamic load balancing method of the optical communication apparatus of any embodiment of the present application.
In a fifth aspect, embodiments of the present application provide a computer program product comprising computer programs/instructions which, when executed by a processor, implement the steps of the dynamic load balancing method of the optical communication device of any of the embodiments of the present application.
The dynamic load balancing method and the system of the optical communication device provided by the application can at least produce the following technical effects:
(1) By acquiring multi-dimensional time sequence data and constructing a history graph structure, the states and traffic load conditions of all devices in the optical communication network at different moments can be effectively captured. By constructing a history graph structure of each history time step and updating by using a graph neural network, the dynamic change trend of the network global flow is captured, and the instantaneous fluctuation of the network flow can be responded in time. The time sequence fusion characteristics of the optical communication device, which are used for fusing a plurality of historical time steps, are input into the flow prediction model, so that the future flow load condition can be accurately predicted based on historical data. Furthermore, the dynamic load balancing operation is executed based on the predicted traffic load data, so that resource allocation can be optimized in a global scope, and the conditions that part of nodes are overloaded and other nodes are idle are avoided. Therefore, the resource allocation of the optical communication network can be planned and adjusted in advance through the analysis and prediction of the future traffic load condition of the historical data, the resource utilization rate of the optical communication network can be remarkably improved, the load of each node is ensured to be in a reasonable range, and the overall performance and reliability of the communication network are improved.
(2) By introducing multi-dimensional time sequence data, traffic load data, equipment state data and external environment data are covered, and the running state of the optical communication device in different time periods can be reflected more comprehensively and accurately. By analyzing and predicting the multi-dimensional time sequence data, the flow load condition of each optical communication device in a future period can be accurately predicted, so that network resources are reasonably distributed in dynamic load balancing operation, the condition that part of nodes are overloaded and other nodes are idle is avoided, and the resource utilization rate is optimized.
(3) Through deep learning and analysis of the graph neural network on the multi-dimensional time sequence data, the change trend and the flow load characteristics of each node in the network can be accurately captured. In addition, the historical graph structure is constructed and updated by utilizing the graph neural network, the link transmission relation among all devices in the optical communication network can be comprehensively considered, the future flow load data of all nodes can be accurately predicted by extracting time sequence fusion characteristics and applying a flow prediction model, dynamic adjustment is carried out in advance, and the overload and fault risk caused by sudden flow change in the network are reduced, so that the reliability and stability of the optical communication network are enhanced.
According to the technical scheme, through introducing multidimensional time sequence data and graph neural network technology, accurate load prediction and intelligent load balancing operation under a dynamic change environment are realized, the resource utilization rate, reliability and stability of an optical communication network are remarkably improved, the manual intervention and configuration complexity are reduced, and the overall performance of the network is comprehensively optimized.
Drawings
In order to more clearly illustrate the embodiments of the present application or the technical solutions of the prior art, the following description will briefly explain the drawings used in the embodiments or the description of the prior art, and it is obvious that the drawings in the following description are some embodiments of the present application, and other drawings can be obtained according to these drawings without inventive effort for a person skilled in the art.
Fig. 1 shows a flowchart of an example of a dynamic load balancing method of an optical communication apparatus according to an embodiment of the present application;
FIG. 2 illustrates a schematic diagram of structural connections of an example of a meta learning model according to an embodiment of the present application;
FIG. 3 illustrates an example operational flow diagram for determining load optimization adjustment instructions via a load balancing model according to an example of an embodiment of the application;
FIG. 4 illustrates an operational flow diagram of an example of a second stage build operation in the construction of a meta-learning model in accordance with an embodiment of the present application;
FIG. 5 illustrates a schematic diagram of structural connections of an example of a flow prediction model according to an embodiment of the present application;
fig. 6 shows a block diagram of an example of a dynamic load balancing system of an optical communication apparatus according to an embodiment of the present application;
fig. 7 is a schematic structural diagram of an embodiment of an electronic device of the present application.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present application more apparent, the technical solutions of the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application, and it is apparent that the described embodiments are some embodiments of the present application, but not all embodiments of the present application. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.
In the technical scheme of the application, the related processes such as collection, storage, use, processing, transmission, provision, disclosure and the like of the personal information of the user accord with the regulations of related laws and regulations, and the public order is not violated.
Fig. 1 shows a flowchart of an example of a dynamic load balancing method of an optical communication apparatus according to an embodiment of the present application.
The execution main body of the method of the embodiment of the application can be any controller or processor with calculation or processing capability, and by introducing artificial intelligence technologies such as a graph neural network, a flow prediction model and the like, the dynamic load balancing operation of each device in the optical communication network can be realized, and the resource utilization rate, the stability and the reliability of the network are effectively improved.
In some examples, it may be an optical communication load balancing platform and may be integrally configured in an electronic device, a terminal or a server by means of software, hardware or a combination of software and hardware, and the type of terminal, electronic device or server may be diversified, such as a mobile phone, a tablet computer or a desktop computer, etc.
As shown in fig. 1, in step S110, multi-dimensional time series data of each optical communication device in the optical communication network to be managed is obtained, where the multi-dimensional time series data includes a plurality of historical time steps and corresponding multi-dimensional data.
In some embodiments, the multidimensional data includes traffic load data, device status data, and external environment data. The traffic load data includes bandwidth usage, total traffic, upload rate, download rate, packet loss rate, and transmission delay. The device status data includes at least one of device failure information, computing resource consumption information, and memory resource consumption information. The external environment data includes at least one of temperature, weather forecast information, and voltage fluctuation information.
More specifically, the multi-dimensional time sequence data of each optical communication device is collected in real time by deploying a plurality of sensors and monitoring systems in the optical communication network, the flow load data is collected by the flow monitoring module, the equipment state data is collected by the equipment management system, and the external environment data is collected by the environment monitoring module. Further, all the acquired data are stored in time step units to form multi-dimensional time sequence data. Thus, by collecting and storing multi-dimensional time sequence data, a comprehensive and detailed historical data base is realized, and accurate data support is provided for subsequent flow prediction and load balancing.
It should be noted that, based on the device state data and the external environment data in the multi-dimensional data, the accuracy of the future flow prediction result and the robustness of the risk coping capability can be effectively improved. Aiming at the equipment state data, the performance degradation caused by equipment failure or equipment resource pressure can be effectively identified and positioned by monitoring the equipment failure information, the computing resource utilization rate and the storage resource utilization rate in real time, and the abnormal change of the network flow caused by abnormal working conditions can be accurately analyzed, so that the high accuracy of the flow prediction result is optimized. In addition, for external environment data, the interference of abnormal environment events on the stable operation of the equipment is rapidly captured by monitoring the environment temperature, weather forecast information and voltage fluctuation, so that the high accuracy of a flow prediction result is further optimized.
According to the embodiment, multidimensional time sequence data is introduced, the flow load data, the equipment state data and the external environment data are covered, and the running state of the optical communication device in different time periods can be reflected more comprehensively and accurately. By analyzing and predicting the multi-dimensional time sequence data, the flow load condition of each optical communication device in a future period of time can be accurately predicted.
In step S120, for each historical time step, multi-dimensional data of each optical communication device corresponding to the historical time step in the multi-dimensional time sequence data is extracted, a corresponding historical graph structure is constructed according to each extracted multi-dimensional data, and then node characteristics of each graph node in the historical graph structure are updated through the graph neural network.
Here, the history graph structure includes a plurality of graph nodes and edge connections, and node characteristics of each graph node are respectively defined by multidimensional data of the corresponding optical communication device, including traffic load, equipment status, and external environment data of each optical communication device at the history time step. Each side is connected with a graph node corresponding to the optical communication device with the link transmission relation, and the side connection in the graph structure is constructed according to the actual link topology.
Here, the types of the graph neural network (Graph Neural Network, GNN) may be diversified, such as GAT (Graph Attention Network, graph annotation network) or GCN (Graph Convolutional Network, graph convolution network), and the like. More preferably, the graph neural network employs a depth map convolution model, more details of which will be developed below in connection with other examples.
In some embodiments, the history graph structure is input into a trained graph neural network, the characteristics of each node are updated through multi-layer propagation, and the characteristics of each node are updated through a message passing mechanism so that the interaction and the association between each device in the network can be reflected.
By constructing and updating the history graph structure of each history time step, the embodiment can capture the topological relation and the data transmission characteristic of the optical communication device at different time steps, and can comprehensively capture the link relation between the devices. The node characteristics are updated by using the graph neural network, the mutual influence among all devices is fully considered, the hidden relation in the multidimensional data can be fully mined, and the accuracy and the expressive force of characteristic representation are improved.
In step S130, for each optical communication apparatus, a time-series fusion feature is determined based on the node features of the corresponding graph nodes of each updated history graph structure, and the time-series fusion feature is input to the traffic prediction model to determine predicted traffic load data corresponding to a preset time period in the future.
In some embodiments, for each optical communication device, a corresponding graph node is extracted from each updated history graph structure, and node features of each graph node and a corresponding time step are spliced and fused, so that corresponding time sequence fusion features are obtained. Here, the time sequence fusion feature comprehensively considers multidimensional data, time sequence change and network global relation, so that future traffic load trend of the device can be reflected more accurately. And then, inputting the time sequence fusion characteristics into a pre-trained flow prediction model to predict flow load data in a preset time period in the future, wherein the flow prediction model can adopt various time sequence models, such as LSTM, GRU and the like, and the time sequence fusion characteristics are extracted through the time sequence model, so that the time dynamic change of the node characteristics can be effectively integrated, and the timeliness and the accuracy of a flow prediction result are improved. In some examples of embodiments of the application, the traffic prediction model may employ an LSTM model based on a multi-head attention mechanism, and more details will be developed below in connection with other examples.
In step S140, a dynamic load balancing operation is performed based on the predicted traffic load data corresponding to each optical communication apparatus.
It should be noted that various non-limiting dynamic load balancing operations may be used herein, for example, an optimization algorithm (such as linear programming, genetic algorithm, deep reinforcement learning, etc.) may be used to make dynamic load balancing decisions based on the predicted traffic load data, to determine an optimal resource allocation scheme for each optical communication device. Therefore, according to the optimization decision result, specific load balancing operation is executed by adjusting the load distribution parameters of each optical communication device in real time, so that the balance of network flow distribution is ensured, and the utilization efficiency of network resources can be improved.
With respect to implementation details of the above step S140, in some embodiments, the predicted traffic load data corresponding to each optical communication device is input to the load balancing model to determine the load optimization adjustment instruction for at least one target optical communication device. Here, the load balancing model employs a meta-learning model and the load optimization adjustment instructions include at least one of adjusting traffic routing, enabling standby nodes, dynamically adjusting bandwidth allocation, and adjusting transmission paths.
It should be noted that the core of the Meta-Learning Model (Meta-Learning Model) is to train on a plurality of tasks, so that the Model can adapt to new tasks quickly. In the embodiment of the application, the meta-learning model is utilized to implement dynamic load balancing operation, so that the adaptability and generalization capability for a plurality of tasks can be realized. More specifically, the plurality of tasks are defined as load balancing requirements under different network load scenarios or network load types, respectively. Accordingly, the data sample set of the meta-learning model includes a plurality of data subsets, each data subset having a respective load type. Specifically, the plurality of data subsets includes a peak load data subset, a valley load data subset, a regular load data subset, and a gradual load data subset. Peak load scenarios may represent scenarios of high bandwidth usage, high latency, rising packet loss rates, e.g., where the network is carrying a large amount of data traffic during peak hours (e.g., the morning and evening peaks of weekdays), requiring fast response and load distribution. The off-peak load scenario may represent a low bandwidth usage, low latency scenario, such as where the network is less traffic during off-peak periods (e.g., late night or holidays), but still requires that the device operate efficiently. A conventional load scenario may represent a scenario where traffic is smooth and bandwidth demands are common, such as network traffic on a common workday. The gradual load scenario may represent a gradual change in flow, which may be a gradual increase or decrease, such as a change in flow before the end of a shift peak or workday.
It should be appreciated that the requirements for load balancing control are different for different load scenarios, in order to enable adaptive regulation for different load scenarios. Here, the meta-learning model is adopted to realize that good performance can be generated on a plurality of tasks, and fine adjustment is performed on each task to generate an optimal load distribution strategy, so that scene self-adaptive dynamic load balancing is realized. Thus, by training on multiple scene data sets (e.g., peak load, valley load, regular load, and gradual load), the meta-learning model can quickly adapt to new tasks and new environments, ensuring efficient execution of load balancing policies. In addition, based on the ability of multi-task learning, the model can predict and cope with various complex network environments and emergencies, and the overall adaptability and the robustness of the network are improved.
FIG. 2 illustrates a schematic diagram of structural connections of an example of a meta learning model according to an embodiment of the present application.
As shown in fig. 2, meta-learning model 200 includes random forest classifier 210, meta-learner 220, and a plurality of base model modules (231, 233..23n), each having a respective load type, e.g., base model module 231 for processing predicted peak load scenarios, base model module 233 for processing predicted valley load scenarios, and so forth.
FIG. 3 illustrates an example operational flow diagram for determining load optimization adjustment instructions via a load balancing model according to an example of an embodiment of the application.
As shown in fig. 3, in step S310, a target load type matching each of the input predicted traffic load data is determined by a random forest classifier.
In some embodiments, the random forest classifier performs feature extraction and analysis on the predicted traffic load data in each optical communication device in the input network through the combination of a plurality of decision trees, and determines a target load type matched with the input data from peak load type, valley load type, regular load type and gradual load type.
The random forest classifier can effectively improve the accuracy of load type classification, reduce the situation of misclassification, rapidly process input data and timely determine the target load type by integrating the judgment results of a plurality of decision trees.
In step S320, each predicted traffic load data is processed according to a target base model module that matches the target load type among the plurality of base model modules, so as to obtain a corresponding load optimization adjustment instruction.
According to the embodiment, the predicted traffic load data of each optical communication device in the network is analyzed through the random forest classifier, so that the corresponding load type is obtained, and the matched basic model module is called to output the corresponding load optimization adjustment instruction, so that the load optimization adjustment strategy is ensured to be suitable for the current network load condition or type, and the optimization effect of the network global load balance is improved. In addition, along with the real-time change of the network flow, the system can rapidly identify the load type and adjust the load distribution strategy, and the dynamic response capability of the network balance control is improved.
In some examples of embodiments of the application, the construction of the meta-learning model includes a continuous first-stage construction operation for constructing a random forest classifier and a second-stage construction operation for updating the respective base model modules.
For specific implementation details of the first stage build operation, in some embodiments, the data set is divided into a training set and a testing set. A random forest classifier model is trained using a training dataset, specifically, a plurality of samples are randomly extracted from the training dataset with a put-back ground, and a plurality of subsets are generated. For each subset, a decision tree is built separately, specifically, feature subsets are randomly selected and features are recursively partitioned until a stop condition (e.g., maximum depth or minimum number of samples) is met. Further, all decision trees are combined into a random forest. After the independent training is finished, the random forest classifier is applied to actual flow prediction data, corresponding load types (such as peak load, valley load, normal load and gradual load) are identified, and the load types corresponding to future flow data can be accurately predicted.
For specific implementation details of the second stage construction operation, according to the output result (load type) of the classifier, corresponding basic model modules are selected and trained, so that each basic model module is optimized for a specific load type, and dynamic adjustment of a load balancing strategy is achieved.
In the embodiment of the application, the meta-learning model is trained and constructed in stages, and the meta-learning device and the basic model module are trained after the training and construction of the random forest classifier are finished. Therefore, different load types can be accurately identified through advanced training and optimization of the classifier, an accurate basis is provided for selection of the basic model module, and the basic model module is dynamically selected according to the output result of the classifier, so that the system can flexibly cope with different load conditions, and the overall adaptability of the system is improved. In addition, the first stage only needs to pay attention to the training of the classifier, and the second stage focuses on the optimization of the basic model module, so that the tasks of each stage are more definite and concentrated, the coupling between the classifier and the basic model module is reduced through staged construction, and the implementation difficulty and the training complexity are reduced.
FIG. 4 illustrates an operational flow diagram of an example of a second stage build operation in the construction of a meta-learning model in accordance with an embodiment of the present application.
As shown in fig. 4, in step S410, model parameters of each basic model module are initialized based on the shared parameters of the meta learner.
Here, by sharing parametersAnd initializing each basic model module, realizing common optimization on a plurality of tasks, providing good initial parameters, and enabling the basic model to be capable of rapidly adapting to different load types.
In step S420, for each subset of data, a plurality of gradient descent optimizations are performed on a base model module having a matching load type using the subset of data to update model parameters of the base model module.
More specifically, each base model moduleOptimizing on corresponding load type data subset, and updating model parameters thereof. By way of example, the corresponding basic model modules are respectively trained and optimized through peak load data subsets, valley load data subsets and the like, and parameters of the basic model modules are updated so that the parameters can generate good performance on the data subsets of the respective tasks.
Here, through the initialization setting of the sharing parameters, the basic model module has good adaptability in the initial stage, and can be quickly adapted to different load types. In addition, the shared parameters are optimized together on a plurality of tasks, so that the quality of initial parameters of the model is improved, and the basic model can reach an optimal state in fewer iterations.
In step S430, meta-losses for all the base model modules are calculated, and in case it is detected that the meta-losses do not satisfy the meta-loss convergence condition, the shared parameters are updated based on the calculated meta-losses, and model parameters of the respective base model modules are reset using the updated shared parameters to iteratively update the model parameters of the respective base model modules.
In some embodiments, after training of each base Model module is completed, model parameters and losses of each base Model module are passed to a meta-learner update module, the shared parameters are updated by a meta-Learning algorithm (e.g., model-Agnostic Meta-Learning, MAML), and the optimized meta-learner parameters are used for the next round of base Model training, such that the shared parametersUpdating is performed through gradient descent of the meta-loss until the meta-loss convergence condition is satisfied, for example, the optimization amplitude of the meta-loss corresponding to two iterations is smaller than a threshold value.
Here, through the calculation of the meta-loss and gradient descent optimization, the global performance of the model over multiple tasks is ensured to be optimal. In addition, the optimization of each basic model module on each load type obviously improves the accuracy and effect of the load balancing strategy.
In some examples of embodiments of the present application, the loss of all base model modules may be comprehensively considered by means of weighted summation, resulting in meta-loss. More specifically, the meta-loss for all base model modules is calculated by:
Formula (1);
Formula (2);
Formula (3);
In the formula,Representing the meta-loss for all of the underlying model modules,Representing the total number of basic model modules contained in the meta-learning model; Represent the firstThe type of the individual loads to be applied,Representation ofThe weight of the corresponding weight is set to be equal to the weight,Representation ofA loss function of the corresponding basic model module; Is atThe corresponding data subset is trained to form a basic model module, and the parameters are as follows;Representation ofThe corresponding load type task is lost; Representation ofThe total number of samples in the corresponding data subset; Representation ofThe corresponding data subset is the firstA data sample and a corresponding tag; representing a loss function for a single sample; Representation ofGradient norms of the loss function of the corresponding base model module,The sum of the gradient norms of the loss functions for all base model modules is represented as a normalization factor.
In the meta-loss function used in the embodiment of the application, the model performance is optimized from the global perspective by comprehensively considering the losses of all the basic model modules. Furthermore, the weight of the individual tasks or base model modulesThe dynamic adjustment can ensure that the influence of different tasks is balanced in the optimization process, and the overall optimization effect is improved. More specifically, the weightThe model is dynamically adjusted according to the gradient of each task, so that the model can pay attention to important tasks in a self-adaptive manner in the training process, and the performance of the model under different load types is improved.
In some examples of embodiments of the application, the base model module employs a reinforcement learning model. At this point, the loss function of the reinforcement learning model may be a corresponding policy gradient loss function (Policy Gradient Loss), a function for optimizing parameters of the policy network to enable actions taken to maximize the expected rewards for supporting iterative computation of element losses.
More specifically, the state of the reinforcement learning model is defined by the predicted traffic load data of each optical communication device in the optical communication network and the overall load balancing index, overall network delay, and overall bandwidth utilization of the optical communication network, the actions of the reinforcement learning model are defined by the load optimization adjustment instructions for one or more optical communication devices, and the rewards of the reinforcement learning model are defined by the overall load balancing index, overall network delay, and the degree of optimization of the overall bandwidth utilization of the optical communication network after the actions are performed.
According to the embodiment, the load balancing regulation and control scheme is integrated into the reinforcement learning model, and through interactive learning with the environment, the load balancing strategy can be adjusted in real time, the load balancing strategy is suitable for the network load condition which changes rapidly, and the dynamic response capability of the system is improved. Through continuous interactive learning and optimization, the basic model module can show good robustness in the face of different load changes and network conditions, and through selection and execution of an optimal strategy, the bandwidth utilization rate can be improved, network delay is reduced, resource allocation is optimized, and overall network performance is improved. For different load types, each corresponding basic model module can adaptively learn an optimal strategy to ensure optimization of system performance under various load conditions. Therefore, through detailed state, action and rewarding design, the self-adaptive reinforcement learning model is adopted as a basic model module, and the dynamic load balancing of the high-efficiency and stable optical communication device can be realized through real-time learning and dynamic optimization.
In some examples of embodiments of the application, the reward function of the reinforcement learning model is:
formula (4);
formula (5);
formula (6);
formula (7);
formula (8);
Formula (9);
Formula (10);
Formula (11);
In the formula,Representing the previous stateAfter the action is executed, the new state is reachedThe generated rewards; State before execution for actionThe corresponding overall load balancing index is used for balancing the load,New state after execution of actionIs a global load balancing index; State before execution for actionIs used to determine the overall network delay of the (c) network,New state after execution of actionIs a global network delay of (1); State before execution for actionIs used for the overall bandwidth utilization of the system,New state after execution of actionIs a whole bandwidth utilization of the system; Is in a previous stateAfter the action is executed, the new state is reachedDynamic weight coefficients of (2); As an initial weight to be used,In order to adjust the coefficient of the coefficient,Respectively an overall load balance index reference value, a network delay reference value and a bandwidth utilization ratio reference value; Indicating the total number of optical communication devices in the optical communication network,Is in state ofTime NoThe traffic load of the individual optical communication devices,Is in state ofAverage flow load at time: Is in state ofTime NoNetwork delay of the individual optical communication devices; Is in state ofTime NoBandwidth utilization of individual optical communication devices.
In the reward function provided in this embodiment, the overall load balance index, network delay and improvement degree of bandwidth utilization rate are comprehensively considered in the reward function, so that the load balance strategy can more effectively and comprehensively optimize each network performance index, and excessive fluctuation of the reward value can be avoided. The reward function reflects the system state change after the action is executed in real time, guides the reinforcement learning model to learn the optimal strategy, improves the dynamic response capability of the system, enables the system to adapt to the load change quickly, adjusts the load balancing strategy in time, and avoids network congestion and resource waste. In addition, through the design of each dynamic weight coefficient, the reward function can adaptively adjust the optimization target, so that the system can be excellent under different load conditions, for example, under high load conditions, the system can pay more attention to the improvement of the load balance index and the network delay, and under low load conditions, the system can pay more attention to the improvement of the bandwidth utilization rate, and the dynamic load balance capacity of the network system is optimized.
In some examples of embodiments of the present application, the graph neural network employs a depth map convolution model that includes a plurality of map convolution layers to update node characteristics of each map node in the history map structure by:
formula (12);
Formula (13);
In the formula,Represent the firstThe layer graph convolves the node feature matrix of the layer,The activation function is represented as a function of the activation,Representing graph nodesIs a neighbor map node set; AndRespectively, are graph nodesAndDegree of (3); AndRespectively the firstThe layer diagram is rolled up and layered with weights and biases; Represent the firstLayer graph nodeIs characterized by the node characteristics of (a),Representing the mean aggregation function.
In the embodiment of the application, the multidimensional data are fused through the graph convolution layer, so that the working load condition of each optical communication device in the optical communication network can be comprehensively reflected, and each updated node characteristic not only contains own load information, but also fuses the characteristics of neighbor nodes. Specifically, the graph volume stacking layer can effectively aggregate the information of the neighbor nodes through the mean value aggregation function, so that the characteristics of each node are not only dependent on own data, but also influenced by the neighbor nodes, the relevance among the nodes in the network can be better reflected, the load balancing strategy has higher integrity and global performance by considering the load condition of the neighbor nodes, the local optimal phenomenon is avoided, the load balancing is carried out through the global view, and the utilization efficiency of network resources is improved.
In some examples of embodiments of the application, the flow prediction model employs an LSTM model based on a multi-head attention mechanism. Fig. 5 shows a schematic structural connection diagram of an example of a flow prediction model according to an embodiment of the present application.
As shown in fig. 5, the traffic prediction model 500 includes an input layer 510, an LSTM layer 520, a multi-headed attention layer 530, and an output layer 540.
The input layer 510 is for receiving the timing fusion feature.
Here, the amount of information included in the timing fusion feature is huge, which can reflect the historical traffic feature, the equipment status feature, the external environment feature, and the corresponding timing feature change information of the optical communication device, and the features of other optical communication devices related to the network, and can comprehensively reflect the network status.
The LSTM layer 520 is configured to process the timing fusion feature, capture the dependency between time steps:
Formula (14);
In the formula,Representing the fusion characteristics of the corresponding time steps in the time sequence fusion characteristics,Is a time stepIs used to determine the hidden state of the (c),Is a time stepIs a hidden state of (c).
The LSTM layer can capture the dependency relationship in the long-time sequence, ensure the effective extraction and utilization of the time sequence characteristics, and improve the prediction capability of future flow trend.
The multi-headed attention layer 530 is used to capture timing characteristics from different angles and levels:
Formula (15);
formula (16);
A compound of formula (17;
Formula (18);
formula (19);
Formula (20);
In the formula,A multi-head attention feature output for a multi-head attention layer; AndRespectively represent the firstAnd (b)The output characteristics of the individual attention heads,Representing the total number of attention heads,The characteristic stitching operation is represented as such,Representing a multi-headed attention mechanism function; The function of the attention mechanism is represented,An output weight matrix for the multi-head attention; Is the firstThe query matrix weights of the individual attention headers,Is the firstThe key matrix weights of the individual attention headers,Is the firstA value matrix weight for each attention header; AndRepresenting a query matrix, a key matrix and a value matrix respectively,A dimension represented as a key vector; representing the softmax activation function.
By a multi-head attention mechanism, time sequence characteristics can be captured from different angles and levels, and the expression capability of the model is enhanced. Therefore, the multi-head attention mechanism enables the model to understand and process time sequence data from different perspectives, the model can capture the characteristics which have important influence on the prediction result more finely, and the comprehensiveness and the prediction reliability of the characteristic extraction are improved.
The output layer 540 is configured to map the output of the multi-head attention mechanism through the full connection layer, so as to obtain final predicted traffic load data:
;
In the formula,Is a time stepIs used for the flow prediction result of the (a),AndThe weights and offsets of the full connection layer respectively,Represents the leak ReLU activation function.
In addition, in order to realize the processing of non-negative data by combining the characteristics of flow prediction, a leakage ReLU activation function is used, and the leakage ReLU activation function allows a small negative slope, so that the problem that the output of the traditional ReLU is zero in a negative area is relieved, neurons can still keep active when the negative value is input, the stability of a model is improved, the problem of dead neurons can be effectively relieved, and the stability of the model is improved. In addition, because the neurons cannot be completely inactivated due to negative input, the model can be better adapted to different input data, and the overall robustness is improved. On the other hand, leakyReLU activation functions introduce nonlinear characteristics, so that the model can better fit complex data distribution, the model can more accurately predict complex flow load conditions, and a load balancing strategy is optimized. Therefore, accurate flow prediction is realized, the system can quickly respond to network load change, and the system is ensured to keep high-efficiency running under different load conditions by dynamically adjusting load balancing strategies, so that the overall network performance is improved.
It should be noted that, for simplicity of description, the foregoing method embodiments are all illustrated as a series of acts combined, but it should be understood and appreciated by those skilled in the art that the present application is not limited by the order of acts, as some steps may be performed in other orders or concurrently in accordance with the present application. Further, those skilled in the art will also appreciate that the embodiments described in the specification are all preferred embodiments, and that the acts and modules referred to are not necessarily required for the present application. In the foregoing embodiments, the descriptions of the embodiments are emphasized, and for parts of one embodiment that are not described in detail, reference may be made to related descriptions of other embodiments.
Fig. 6 shows a block diagram of an example of a dynamic load balancing system of an optical communication apparatus according to an embodiment of the present application.
As shown in fig. 6, the dynamic load balancing system 600 of the optical communication apparatus includes a data acquisition unit 610, a feature extraction unit 620, a traffic prediction unit 630, and a load balancing unit 640.
The data obtaining unit 610 is configured to obtain multi-dimensional time sequence data of each optical communication device in the optical communication network to be managed, where the multi-dimensional time sequence data includes a plurality of historical time steps and corresponding multi-dimensional data, the multi-dimensional data includes traffic load data, device state data and external environment data, and the traffic load data includes bandwidth usage rate, total traffic, upload rate, download rate, packet loss rate and transmission delay.
The feature extraction unit 620 is configured to extract, for each of the historical time steps, multi-dimensional data of each optical communication device corresponding to the historical time step in the multi-dimensional time sequence data, construct a corresponding historical graph structure according to each of the extracted multi-dimensional data, and then update node features of each graph node in the historical graph structure through a graph neural network, where the historical graph structure includes a plurality of graph nodes and edge connections, and node features of each graph node are respectively defined by the multi-dimensional data of the corresponding optical communication device, and each edge connection is used to connect the graph nodes corresponding to the optical communication devices having a link transmission relationship.
The traffic prediction unit 630 is configured to determine, for each optical communication device, a time-series fusion feature based on node features of corresponding graph nodes of each updated historical graph structure, and input the time-series fusion feature to a traffic prediction model to determine predicted traffic load data corresponding to a future preset time period.
The load balancing unit 640 is configured to perform a dynamic load balancing operation based on the predicted traffic load data corresponding to each of the optical communication devices.
In some embodiments, embodiments of the present application provide a non-transitory computer readable storage medium having stored therein one or more programs including execution instructions that are readable and executable by an electronic device (including, but not limited to, a computer, a server, or a network device, etc.) for performing the steps of the dynamic load balancing method of any of the above-described optical communication apparatuses of the present application.
In some embodiments, embodiments of the present application also provide a computer program product comprising a computer program stored on a non-volatile computer readable storage medium, the computer program comprising program instructions which, when executed by a computer, cause the computer to perform the steps of the dynamic load balancing method of any of the above-mentioned optical communication apparatuses.
In some embodiments, the present application further provides an electronic device comprising at least one processor, and a memory communicatively coupled to the at least one processor, wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the steps of a dynamic load balancing method of an optical communication apparatus.
Fig. 7 is a schematic hardware structure of an electronic device for performing a dynamic load balancing method of an optical communication apparatus according to another embodiment of the present application, where, as shown in fig. 7, the device includes:
one or more processors 710, and a memory 720, one processor 710 being illustrated in fig. 7.
The apparatus for performing the dynamic load balancing method of the optical communication device may further include an input device 730 and an output device 740.
Processor 710, memory 720, input device 730, and output device 740 may be connected by a bus or other means, for example in fig. 7.
The memory 720 is used as a non-volatile computer readable storage medium for storing a non-volatile software program, a non-volatile computer executable program, and modules, such as program instructions/modules corresponding to the dynamic load balancing method of the optical communication device in the embodiment of the present application. The processor 710 executes various functional applications of the server and data processing by running non-volatile software programs, instructions and modules stored in the memory 720, i.e., implements the dynamic load balancing method of the optical communication device of the above-described method embodiment.
The memory 720 may include a storage program area that may store an operating system, application programs required for at least one function, and a storage data area that may store data created according to the use of the electronic device, etc. In addition, memory 720 may include high-speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, memory 720 may optionally include memory located remotely from processor 710, which may be connected to the electronic device via a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.
The input device 730 may receive input digital or character information and generate signals related to user settings and function control of the electronic device. The output device 740 may include a display device such as a display screen.
The one or more modules are stored in the memory 720 that, when executed by the one or more processors 710, perform the dynamic load balancing method of the optical communication device in any of the method embodiments described above.
The product can execute the method provided by the embodiment of the application, and has the corresponding functional modules and beneficial effects of the execution method. Technical details not described in detail in this embodiment may be found in the methods provided in the embodiments of the present application.
The electronic device of the embodiments of the present application exists in a variety of forms including, but not limited to:
(1) Mobile communication devices, which are characterized by mobile communication functionality and are aimed at providing voice, data communication. Such terminals include smart phones, multimedia phones, functional phones, low-end phones, and the like.
(2) Ultra mobile personal computer equipment, which belongs to the category of personal computers, has the functions of calculation and processing and generally has the characteristic of mobile internet surfing. Such terminals include PDA, MID, and UMPC devices, etc.
(3) Portable entertainment devices such devices can display and play multimedia content. The device comprises an audio player, a video player, a palm game machine, an electronic book, an intelligent toy and a portable vehicle navigation device.
(4) Other on-board electronic devices with data interaction functions, such as on-board devices mounted on vehicles.
The apparatus embodiments described above are merely illustrative, wherein the elements illustrated as separate elements may or may not be physically separate, and the elements shown as elements may or may not be physical elements, may be located in one place, or may be distributed over a plurality of network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
From the above description of embodiments, it will be apparent to those skilled in the art that the embodiments may be implemented by means of software plus a general purpose hardware platform, or may be implemented by hardware. Based on such understanding, the foregoing technical solution may be embodied essentially or in a part contributing to the related art in the form of a software product, which may be stored in a computer readable storage medium, such as ROM/RAM, a magnetic disk, an optical disk, etc., including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform the method described in the respective embodiments or some parts of the embodiments.
It should be noted that the above-mentioned embodiments are merely for illustrating the technical solution of the present application, and not for limiting the same, and although the present application has been described in detail with reference to the above-mentioned embodiments, it should be understood by those skilled in the art that the technical solution described in the above-mentioned embodiments may be modified or some technical features may be equivalently replaced, and these modifications or substitutions do not make the essence of the corresponding technical solution deviate from the spirit and scope of the technical solution of the embodiments of the present application.

Claims (8)

Translated fromChinese
1.一种光通信装置的动态负载均衡方法,包括:1. A dynamic load balancing method for an optical communication device, comprising:获取待管理的光通信网络中各个光通信装置的多维度时序数据;所述多维度时序数据包含多个历史时间步与相应的多维度数据,所述多维度数据包含流量负荷数据、设备状态数据和外部环境数据;所述流量负荷数据包含带宽使用率、总流量、上传速率、下载速率、丢包率和传输时延;Acquire multi-dimensional time series data of each optical communication device in the optical communication network to be managed; the multi-dimensional time series data includes multiple historical time steps and corresponding multi-dimensional data, the multi-dimensional data includes traffic load data, device status data and external environment data; the traffic load data includes bandwidth utilization, total traffic, upload rate, download rate, packet loss rate and transmission delay;针对各个所述历史时间步,提取所述多维度时序数据中对应所述历史时间步各个光通信装置的多维度数据,并根据所提取的各个所述多维度数据构建相应的历史图结构,然后通过图神经网络更新所述历史图结构中各个图节点的节点特征;所述历史图结构包含多个图节点和边连接,每一所述图节点的节点特征分别由相应光通信装置的多维度数据而定义,每一所述边连接用于连接具有链路传输关系的光通信装置所对应的图节点;For each of the historical time steps, extract the multi-dimensional data of each optical communication device corresponding to the historical time step in the multi-dimensional time series data, and construct a corresponding historical graph structure according to each of the extracted multi-dimensional data, and then update the node features of each graph node in the historical graph structure through a graph neural network; the historical graph structure includes a plurality of graph nodes and edge connections, the node features of each of the graph nodes are respectively defined by the multi-dimensional data of the corresponding optical communication device, and each of the edge connections is used to connect the graph nodes corresponding to the optical communication devices having a link transmission relationship;针对各个光通信装置,基于各个经更新的历史图结构的相应图节点的节点特征确定时序融合特征,并将该时序融合特征输入至流量预测模型,以确定对应未来的预设时间段的预测流量负荷数据;For each optical communication device, determining a time series fusion feature based on the node features of the corresponding graph nodes of each updated historical graph structure, and inputting the time series fusion feature into a traffic prediction model to determine predicted traffic load data corresponding to a preset time period in the future;基于各个所述光通信装置所对应的预测流量负荷数据,执行动态负载均衡操作;Performing a dynamic load balancing operation based on the predicted traffic load data corresponding to each of the optical communication devices;其中,所述图神经网络采用深度图卷积模型,所述深度图卷积模型包含多个图卷积层,以通过以下方式更新历史图结构中各个图节点的节点特征:The graph neural network adopts a deep graph convolution model, which includes multiple graph convolution layers to update the node features of each graph node in the historical graph structure in the following way:式中,表示第层图卷积层的节点特征矩阵,表示激活函数,表示图节点的邻居图节点集合;分别为图节点的度;分别为第层图卷积层的权重和偏置;表示第层图节点的节点特征,表示均值聚合函数;In the formula, Indicates The node feature matrix of the graph convolution layer, represents the activation function, Representing a graph node The set of neighbor graph nodes of ; and Graph nodes and degree; and Respectively The weights and biases of the graph convolutional layers; Indicates Layer Graph Node The node characteristics of represents the mean aggregation function;其中,所述流量预测模型采用基于多头注意力机制的LSTM模型,其包含输入层、LSTM层、多头注意力层和输出层;The traffic prediction model adopts an LSTM model based on a multi-head attention mechanism, which includes an input layer, an LSTM layer, a multi-head attention layer and an output layer;所述输入层用于接收时序融合特征;The input layer is used to receive time series fusion features;所述LSTM层用于处理所述时序融合特征,捕捉时间步之间的依赖关系:The LSTM layer is used to process the time series fusion features and capture the dependencies between time steps:式中,表示时序融合特征中对应时间步的融合特征,为时间步的隐藏状态,为时间步的隐藏状态;In the formula, Indicates the corresponding time step in the time series fusion feature The fusion characteristics of is the time step The hidden state of is the time step The hidden state of所述多头注意力层用于从不同的角度和层次捕捉时序特征:The multi-head attention layer is used to capture temporal features from different angles and levels:式中,为多头注意力层输出的多头注意力特征;分别表示第个和第个注意力头的输出特征,表示注意力头的总数,表示特征拼接操作,表示多头注意力机制函数;表示注意力机制函数,为多头注意力的输出权重矩阵;为第个注意力头的查询矩阵权重,为第个注意力头的键矩阵权重,为第个注意力头的值矩阵权重;分别表示查询矩阵、键矩阵和值矩阵,表示为键向量的维度;表示softmax激活函数;In the formula, The multi-head attention features output by the multi-head attention layer; and Respectively represent and The output features of the attention head are represents the total number of attention heads, represents the feature concatenation operation, Represents the multi-head attention mechanism function; represents the attention mechanism function, is the output weight matrix of multi-head attention; For the The query matrix weights of the attention heads, For the The key matrix weights of the attention heads, For the The value matrix weights of the attention heads; , and denote the query matrix, key matrix and value matrix respectively, Represented as the dimension of the key vector; represents the softmax activation function;所述输出层用于通过全连接层将多头注意力机制的输出进行映射,得到最终的预测流量负荷数据:The output layer is used to map the output of the multi-head attention mechanism through the fully connected layer to obtain the final predicted traffic load data:式中,为时间步的流量预测结果,分别为全连接层的权重和偏置,表示Leaky ReLU激活函数。In the formula, is the time step The traffic prediction results are and are the weights and biases of the fully connected layer, Represents the Leaky ReLU activation function.2.根据权利要求1所述的方法,其中,所述基于各个所述光通信装置所对应的预测流量负荷数据,执行动态负载均衡操作,包括:2. The method according to claim 1, wherein the performing of a dynamic load balancing operation based on the predicted traffic load data corresponding to each of the optical communication devices comprises:将各个所述光通信装置所对应的预测流量负荷数据输入至负载均衡模型,以确定针对至少一个目标光通信装置的负载优化调整指令;所述负载均衡模型采用元学习模型,所述负载优化调整指令包含以下中的至少一者:调整流量路由、启用备用节点、动态调整带宽分配和调整传输路径;所述元学习模型的数据样本集包含多个数据子集,每一所述数据子集分别具有相应的负荷类型;所述多个数据子集包含高峰负载数据子集、低谷负载数据子集、常规负载数据子集和渐变负载数据子集。The predicted traffic load data corresponding to each of the optical communication devices is input into a load balancing model to determine a load optimization adjustment instruction for at least one target optical communication device; the load balancing model adopts a meta-learning model, and the load optimization adjustment instruction includes at least one of the following: adjusting traffic routing, enabling spare nodes, dynamically adjusting bandwidth allocation, and adjusting transmission paths; the data sample set of the meta-learning model includes multiple data subsets, each of the data subsets has a corresponding load type; the multiple data subsets include a peak load data subset, a valley load data subset, a normal load data subset, and a gradual load data subset.3.根据权利要求2所述的方法,其中所述元学习模型包含随机森林分类器、元学习器和多个基础模型模块,每一基础模型模块分别具有相应的负荷类型;3. The method according to claim 2, wherein the meta-learning model comprises a random forest classifier, a meta-learner and a plurality of basic model modules, each of which has a corresponding load type;所述将各个所述光通信装置所对应的预测流量负荷数据输入至负载均衡模型,以确定针对至少一个目标光通信装置的负载优化调整指令,包括:The step of inputting the predicted traffic load data corresponding to each of the optical communication devices into a load balancing model to determine a load optimization adjustment instruction for at least one target optical communication device includes:通过所述随机森林分类器,确定与所输入的各个预测流量负荷数据相匹配的目标负荷类型;Determining, by means of the random forest classifier, a target load type matching each input predicted flow load data;根据所述多个基础模型模块中与所述目标负荷类型相匹配的目标基础模型模块处理各个所述预测流量负荷数据,以得到相应的负载优化调整指令。Each of the predicted traffic load data is processed according to a target basic model module that matches the target load type among the multiple basic model modules to obtain a corresponding load optimization adjustment instruction.4.根据权利要求3所述的方法,其中,针对所述元学习模型的构建包含连续的第一阶段构建操作和第二阶段构建操作,所述第一阶段构建操作用于构建所述随机森林分类器,所述第二阶段构建操作用于更新各个基础模型模块;4. The method according to claim 3, wherein the construction of the meta-learning model comprises a continuous first-stage construction operation and a second-stage construction operation, wherein the first-stage construction operation is used to construct the random forest classifier, and the second-stage construction operation is used to update each basic model module;其中,所述第二阶段构建操作,包括:The second stage construction operation includes:基于元学习器的共享参数,初始化设置各个基础模型模块的模型参数;Based on the shared parameters of the meta-learner, the model parameters of each basic model module are initialized and set;针对各个所述数据子集,利用该数据子集对具有相匹配的负荷类型的基础模型模块进行多次梯度下降优化,以更新该基础模型模块的模型参数;For each of the data subsets, using the data subset to perform multiple gradient descent optimizations on a basic model module with a matching load type, so as to update model parameters of the basic model module;计算针对所有基础模型模块的元损失,并在检测到所述元损失不满足元损失收敛条件的情况下,基于所计算的元损失更新所述共享参数,并利用经更新的共享参数重新设置各个基础模型模块的模型参数,以迭代更新各个基础模型模块的模型参数。The meta-loss for all base model modules is calculated, and when it is detected that the meta-loss does not satisfy the meta-loss convergence condition, the shared parameters are updated based on the calculated meta-loss, and the model parameters of each base model module are reset using the updated shared parameters to iteratively update the model parameters of each base model module.5.根据权利要求4所述的方法,其中,所述计算针对所有基础模型模块的元损失,包括:5. The method according to claim 4, wherein the calculating the meta-loss for all base model modules comprises:式中,表示针对所有基础模型模块的元损失,表示元学习模型所包含的基础模型模块的总数;表示第个负荷类型,表示所对应的权重,表示所对应的基础模型模块的损失函数;是在所对应的数据子集训练后的基础模型模块,参数为表示所对应的负荷类型任务损失;表示所对应的数据子集中样本的总数;表示所对应的数据子集中的第个数据样本和对应的标签;表示单个样本的损失函数;表示所对应的基础模型模块的损失函数的梯度范数,为归一化因子,表示对所有基础模型模块的损失函数的梯度范数之和。In the formula, represents the meta-loss for all base model modules, Represents the total number of basic model modules contained in the meta-learning model; Indicates load type, express The corresponding weight, express The loss function of the corresponding basic model module; is The basic model module after training the corresponding data subset has the following parameters: ; express The corresponding load type task loss; express The total number of samples in the corresponding data subset; express The corresponding data subset data samples and corresponding labels; Represents the loss function of a single sample; express The gradient norm of the loss function of the corresponding basic model module, is a normalization factor, which represents the sum of the gradient norms of the loss functions of all basic model modules.6.根据权利要求4所述的方法,其中,所述基础模型模块采用强化学习模型;6. The method according to claim 4, wherein the basic model module adopts a reinforcement learning model;所述强化学习模型的状态是由光通信网络中各个光通信装置的预测流量负荷数据及光通信网络的整体负荷均衡指标、整体网络延迟和整体带宽利用率而定义的,所述强化学习模型的动作是由关于一个或多个光通信装置的负载优化调整指令而定义的,所述强化学习模型的奖励是由动作执行后光通信网络的整体负荷均衡指标、整体网络延迟和整体带宽利用率的优化程度而定义的。The state of the reinforcement learning model is defined by the predicted traffic load data of each optical communication device in the optical communication network and the overall load balancing index, overall network delay and overall bandwidth utilization of the optical communication network. The action of the reinforcement learning model is defined by the load optimization adjustment instructions for one or more optical communication devices. The reward of the reinforcement learning model is defined by the degree of optimization of the overall load balancing index, overall network delay and overall bandwidth utilization of the optical communication network after the action is executed.7.根据权利要求6所述的方法,其中,所述强化学习模型的奖励函数为:7. The method according to claim 6, wherein the reward function of the reinforcement learning model is:式中,表示从前状态动作执行后到新状态所产生的奖励;为动作执行前状态对应的整体负荷均衡指标,为动作执行后新状态的整体负荷均衡指标;为动作执行前状态的整体网络延迟,为动作执行后新状态的整体网络延迟;为动作执行前状态的整体带宽利用率,为动作执行后新状态的整体带宽利用率;为从前状态动作执行后到新状态的动态权重系数;为初始权重,为调整系数,分别为整体负荷均衡指标参考值、网络延迟参考值和带宽利用率参考值;表示光通信网络中光通信装置的总数,为状态时第个光通信装置的流量负荷,为状态时的平均流量负荷:为状态时第个光通信装置的网络延迟;为状态时第个光通信装置的带宽利用率。In the formula, Indicates the previous state After the action is executed, the new state is reached the rewards generated; The state before the action is executed The corresponding overall load balancing index, The new state after the action is executed Overall load balancing indicator; The state before the action is executed The overall network latency, The new state after the action is executed Overall network latency The state before the action is executed The overall bandwidth utilization of The new state after the action is executed Overall bandwidth utilization; For the previous state After the action is executed, the new state is reached Dynamic weight coefficient of is the initial weight, is the adjustment factor, They are the overall load balancing index reference value, network delay reference value and bandwidth utilization reference value respectively; represents the total number of optical communication devices in the optical communication network, Status Time The traffic load of an optical communication device, Status Average traffic load at: Status Time Network latency of optical communication devices; Status Time The bandwidth utilization of an optical communication device.8.一种光通信装置的动态负载均衡系统,包括:8. A dynamic load balancing system for an optical communication device, comprising:数据获取单元,用于获取待管理的光通信网络中各个光通信装置的多维度时序数据;所述多维度时序数据包含多个历史时间步与相应的多维度数据,所述多维度数据包含流量负荷数据、设备状态数据和外部环境数据;所述流量负荷数据包含带宽使用率、总流量、上传速率、下载速率、丢包率和传输时延;A data acquisition unit, used to acquire multi-dimensional time series data of each optical communication device in the optical communication network to be managed; the multi-dimensional time series data includes multiple historical time steps and corresponding multi-dimensional data, the multi-dimensional data includes traffic load data, device status data and external environment data; the traffic load data includes bandwidth utilization, total traffic, upload rate, download rate, packet loss rate and transmission delay;特征提取单元,用于针对各个所述历史时间步,提取所述多维度时序数据中对应所述历史时间步各个光通信装置的多维度数据,并根据所提取的各个所述多维度数据构建相应的历史图结构,然后通过图神经网络更新所述历史图结构中各个图节点的节点特征;所述历史图结构包含多个图节点和边连接,每一所述图节点的节点特征分别由相应光通信装置的多维度数据而定义,每一所述边连接用于连接具有链路传输关系的光通信装置所对应的图节点;A feature extraction unit is used to extract, for each of the historical time steps, the multi-dimensional data of each optical communication device corresponding to the historical time step in the multi-dimensional time series data, and construct a corresponding historical graph structure according to each of the extracted multi-dimensional data, and then update the node features of each graph node in the historical graph structure through a graph neural network; the historical graph structure includes a plurality of graph nodes and edge connections, the node features of each of the graph nodes are respectively defined by the multi-dimensional data of the corresponding optical communication device, and each of the edge connections is used to connect the graph nodes corresponding to the optical communication devices having a link transmission relationship;流量预测单元,用于针对各个光通信装置,基于各个经更新的历史图结构的相应图节点的节点特征确定时序融合特征,并将该时序融合特征输入至流量预测模型,以确定对应未来的预设时间段的预测流量负荷数据;A traffic prediction unit, for determining, for each optical communication device, a time series fusion feature based on the node features of the corresponding graph nodes of each updated historical graph structure, and inputting the time series fusion feature into a traffic prediction model to determine predicted traffic load data corresponding to a preset time period in the future;负载均衡单元,用于基于各个所述光通信装置所对应的预测流量负荷数据,执行动态负载均衡操作;A load balancing unit, configured to perform a dynamic load balancing operation based on the predicted traffic load data corresponding to each of the optical communication devices;其中,所述图神经网络采用深度图卷积模型,所述深度图卷积模型包含多个图卷积层,以通过以下方式更新历史图结构中各个图节点的节点特征:The graph neural network adopts a deep graph convolution model, which includes multiple graph convolution layers to update the node features of each graph node in the historical graph structure in the following way:式中,表示第层图卷积层的节点特征矩阵,表示激活函数,表示图节点的邻居图节点集合;分别为图节点的度;分别为第层图卷积层的权重和偏置;表示第层图节点的节点特征,表示均值聚合函数;In the formula, Indicates The node feature matrix of the graph convolution layer, represents the activation function, Representing a graph node The set of neighbor graph nodes of ; and Graph nodes and degree; and Respectively The weights and biases of the graph convolutional layers; Indicates Layer Graph Node The node characteristics of represents the mean aggregation function;其中,所述流量预测模型采用基于多头注意力机制的LSTM模型,其包含输入层、LSTM层、多头注意力层和输出层;The traffic prediction model adopts an LSTM model based on a multi-head attention mechanism, which includes an input layer, an LSTM layer, a multi-head attention layer and an output layer;所述输入层用于接收时序融合特征;The input layer is used to receive time series fusion features;所述LSTM层用于处理所述时序融合特征,捕捉时间步之间的依赖关系:The LSTM layer is used to process the time series fusion features and capture the dependencies between time steps:式中,表示时序融合特征中对应时间步的融合特征,为时间步的隐藏状态,为时间步的隐藏状态;In the formula, Indicates the corresponding time step in the time series fusion feature The fusion characteristics of is the time step The hidden state of is the time step The hidden state of所述多头注意力层用于从不同的角度和层次捕捉时序特征:The multi-head attention layer is used to capture temporal features from different angles and levels:式中,为多头注意力层输出的多头注意力特征;分别表示第个和第个注意力头的输出特征,表示注意力头的总数,表示特征拼接操作,表示多头注意力机制函数;表示注意力机制函数,为多头注意力的输出权重矩阵;为第个注意力头的查询矩阵权重,为第个注意力头的键矩阵权重,为第个注意力头的值矩阵权重;分别表示查询矩阵、键矩阵和值矩阵,表示为键向量的维度;表示softmax激活函数;In the formula, The multi-head attention features output by the multi-head attention layer; and Respectively represent and The output features of the attention head are represents the total number of attention heads, represents the feature concatenation operation, Represents the multi-head attention mechanism function; represents the attention mechanism function, is the output weight matrix of multi-head attention; For the The query matrix weights of the attention heads, For the The key matrix weights of the attention heads, For the The value matrix weights of the attention heads; , and denote the query matrix, key matrix and value matrix respectively, Represented as the dimension of the key vector; represents the softmax activation function;所述输出层用于通过全连接层将多头注意力机制的输出进行映射,得到最终的预测流量负荷数据:The output layer is used to map the output of the multi-head attention mechanism through the fully connected layer to obtain the final predicted traffic load data:式中,为时间步的流量预测结果,分别为全连接层的权重和偏置,表示Leaky ReLU激活函数。In the formula, is the time step The traffic prediction results are and are the weights and biases of the fully connected layer, Represents the Leaky ReLU activation function.
CN202411134310.2A2024-08-192024-08-19Dynamic load balancing method and system for optical communication deviceActiveCN118660014B (en)

Priority Applications (1)

Application NumberPriority DateFiling DateTitle
CN202411134310.2ACN118660014B (en)2024-08-192024-08-19Dynamic load balancing method and system for optical communication device

Applications Claiming Priority (1)

Application NumberPriority DateFiling DateTitle
CN202411134310.2ACN118660014B (en)2024-08-192024-08-19Dynamic load balancing method and system for optical communication device

Publications (2)

Publication NumberPublication Date
CN118660014A CN118660014A (en)2024-09-17
CN118660014Btrue CN118660014B (en)2024-12-13

Family

ID=92702550

Family Applications (1)

Application NumberTitlePriority DateFiling Date
CN202411134310.2AActiveCN118660014B (en)2024-08-192024-08-19Dynamic load balancing method and system for optical communication device

Country Status (1)

CountryLink
CN (1)CN118660014B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication numberPriority datePublication dateAssigneeTitle
CN119179552A (en)*2024-11-222024-12-24国网浙江省电力有限公司衢州供电公司Self-adaptive container expansion method, system, equipment and medium
CN119544603A (en)*2024-11-282025-02-28广东电网有限责任公司 A method, device, terminal equipment and computer-readable storage medium for repairing power distribution communication network faults
CN119966996B (en)*2025-04-112025-07-11北京睿智聚合科技有限公司Industrial load control method based on cloud edge cooperative technology

Citations (2)

* Cited by examiner, † Cited by third party
Publication numberPriority datePublication dateAssigneeTitle
CN117479231A (en)*2023-10-202024-01-30北京大学武汉人工智能研究院Load balancing optimization method and device based on flow prediction and reinforcement learning
CN117914794A (en)*2024-01-182024-04-19杭州师范大学 An adaptive load balancing method and system based on temporal graph convolutional network

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication numberPriority datePublication dateAssigneeTitle
US10116521B2 (en)*2015-10-152018-10-30Citrix Systems, Inc.Systems and methods for determining network configurations using historical real-time network metrics data
US11134016B2 (en)*2018-10-262021-09-28Hughes Network Systems, LlcMonitoring a communication network
CN112200373A (en)*2020-10-152021-01-08中国科学院深圳先进技术研究院Training method and training device for load prediction model, storage medium and equipment
CN113572697B (en)*2021-07-202023-09-22电子科技大学Load balancing method based on graph convolution neural network and deep reinforcement learning
CN118118400A (en)*2024-01-302024-05-31北京邮电大学 A method and device for adaptive protection path planning in metropolitan area transmission network

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication numberPriority datePublication dateAssigneeTitle
CN117479231A (en)*2023-10-202024-01-30北京大学武汉人工智能研究院Load balancing optimization method and device based on flow prediction and reinforcement learning
CN117914794A (en)*2024-01-182024-04-19杭州师范大学 An adaptive load balancing method and system based on temporal graph convolutional network

Also Published As

Publication numberPublication date
CN118660014A (en)2024-09-17

Similar Documents

PublicationPublication DateTitle
CN118660014B (en)Dynamic load balancing method and system for optical communication device
CN109818786B (en)Method for optimally selecting distributed multi-resource combined path capable of sensing application of cloud data center
CN118312861A (en)File export method and system based on AI cloud computing
CN111669291A (en) Deployment method of virtualized network service function chain based on deep reinforcement learning
CN119211585B (en)Private domain live broadcast content distribution and visitor interaction method and system based on edge calculation
CN110471621B (en) An edge collaborative storage method for real-time data processing applications
CN117669741A (en) Dynamic collaborative reasoning method for UAV cluster size model based on genetic algorithm
CN118713059A (en) Power grid modeling simulation method, device and computer equipment integrating multi-source data
CN115562746A (en) A MEC associated task offloading method with limited computing resources
CN117939518A (en) Mobile network usage data prediction method and device suitable for base station
CN118636864A (en) Energy management optimization method, device and storage medium for extended-range vehicles
CN117081163A (en)Method and device for generating source-load coupling wind-light output scene
CN119094425B (en)Cloud edge computing power network route optimization method
CN112257977B (en)Logistics project construction period optimization method and system with resource limitation under fuzzy man-hour
US12380359B2 (en)Constraint sampling reinforcement learning for recommendation systems
CN118524001A (en)Method and device for scheduling computing network resources
CN118504943A (en)Multi-star task scheduling method, system, electronic equipment and storage medium
KR20080087571A (en) Context prediction system and method
CN114819096A (en)Model training method and device, electronic equipment and storage medium
CN119718855B (en)Parameter determination method and device, storage medium and electronic equipment
CN120499018B (en)Multi-domain Internet of things task allocation optimization model training method, device, equipment and medium
CN119829297B (en) Edge-end dynamic collaborative task batch processing method and system based on reinforcement learning
CN120234014B (en)Model deployment method and device combined with hardware deployment
CN120335890A (en) Service placement and task offloading method based on tree-structured genetic programming hyper-heuristic algorithm in cloud-edge collaborative environment
CN120528813A (en) A business-driven multi-dimensional resource collaborative adaptation method for power information communication networks

Legal Events

DateCodeTitleDescription
PB01Publication
PB01Publication
SE01Entry into force of request for substantive examination
SE01Entry into force of request for substantive examination
GR01Patent grant
GR01Patent grant

[8]ページ先頭

©2009-2025 Movatter.jp