Real-time enhancement method and device for network switching, electronic equipment and storage medium

Abstract

The invention discloses a real-time enhancement method and device for network switching, electronic equipment and a storage medium, and relates to the technical field of network traffic. According to the method, nonlinear mapping and parameter optimization are carried out on time delay data by using a vector machine model and a Bayesian statistical method to obtain a dynamic time delay interval, and joint compensation is carried out on the hysteresis of link state acquisition and the advancement of control instruction issuing according to the interval. The problems that a scheduling strategy lags behind an actual network state and instruction execution deviates from an expected effect due to the fact that a static delay compensation strategy is adopted in an existing network state acquisition and control instruction issuing method can be solved, and the technical effects of improving traffic scheduling real-time performance and accuracy, improving the network resource utilization rate and reducing the packet loss risk are achieved.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of network traffic, and particularly relates to a network exchange real-time enhancement method and device, electronic equipment and storage medium. BACKGROUND

[0002] With the deepening of digital transformation, network transmission and bearing technology as the core support of modern communication system is widely used in operator backbone network, 5G core network and industrial internet and other scenes with extremely high quality of service (QoS) requirements. In the field of traffic engineering, the existing technology usually obtains network topology information through OSPF, ISIS and other link state protocols, and distributes and controls routing strategies with the help of BGP protocol. Specifically, this technical system covers the whole process from network element state collection, interface relationship modeling to link state integration, including key links such as node state, network prefix, connection relationship, aiming to realize dynamic scheduling and optimization of network resources.

[0003] However, in the existing network state collection and control instruction issuing method, a static delay compensation strategy is directly used, which is usually set based on the average value of historical data; due to the time-varying and nonlinear characteristics of network state, the scheduling strategy generated by the controller may lag behind the actual network state, or deviate from the expected effect in the instruction execution process due to network changes, thereby affecting the real-time and accuracy of traffic scheduling, reducing network resource utilization and increasing the risk of packet loss. SUMMARY

[0004] The present disclosure provides a network exchange real-time enhancement method, device, electronic equipment and storage medium. Its main purpose is to solve the problem that the scheduling strategy generated by the controller lags behind the actual network state, or deviates from the expected effect in the instruction execution process due to network changes, thereby affecting the real-time and accuracy of traffic scheduling, reducing network resource utilization and increasing the risk of packet loss.

[0005] According to a first aspect of the present disclosure, a network exchange real-time enhancement method is provided, which comprises:

[0006] Collecting delay data of link state events based on optional transition attributes;

[0007] Inputting the delay data into a vector machine model, and performing nonlinear mapping and parameter optimization on the delay data based on the vector machine model and Bayesian statistical method to obtain a dynamic delay interval;

[0008] According to the dynamic delay interval, the hysteresis in the link state collection process and the advance in the control instruction issuing process are jointly compensated.

[0009] Optionally, the time delay data based on the optional transition attribute link state event acquisition further comprises:

[0010] Record the first time when the event actually occurs, the second time when it is transmitted to the controller, the third time when the controller parses it, and the fourth time when the controller generates the strategy;

[0011] Calculate the first time delay difference value according to the second time and the first time, the second time delay difference value according to the third time and the first time, and the third time delay difference value according to the fourth time and the second time;

[0012] The first time delay difference value, the second time delay difference value, and the third time delay difference value are divided into a hysteresis time delay of link state detection and an advance time delay of control instruction issuance according to the link direction, and feature extraction and normalization processing are performed respectively.

[0013] Optionally, in the input of the time delay data into the vector machine model, based on the vector machine model and the Bayesian statistical method, the time delay data is nonlinearly mapped and parameter optimized to obtain a dynamic time delay interval, and the method further comprises:

[0014] Jarque-Bera test is performed on the training set to screen time delay samples meeting the normal distribution characteristics;

[0015] Based on a large sample state, a Poisson distribution is used to approximately process discrete variables, and a prior distribution is combined to establish a posterior likelihood function to optimize the parameters of the support vector machine model.

[0016] Optionally, the joint compensation of the hysteresis in the link state acquisition process and the advance in the control instruction issuance process according to the dynamic time delay interval comprises:

[0017] According to the similarity of the link state acquisition path and the control instruction issuance path, the same kernel function and compensation model are used for bidirectional time delay inference;

[0018] The output results of the acquisition compensation and the issuance compensation are fused to generate a final strategy execution time offset.

[0019] Optionally, the method further comprises:

[0020] According to the network size, the compensation strategy is divided into multiple compensation levels, and different model parameters and compensation intervals are used at different levels to adapt to the real-time needs of different size networks.

[0021] According to a second aspect of the present disclosure, a network exchange real-time enhancement device is provided, comprising:

[0022] The collection unit is configured to collect time delay data of a link state event based on optional transition attributes;

[0023] The optimization unit is configured to input the time delay data into a vector machine model, perform nonlinear mapping and parameter optimization on the time delay data based on the vector machine model and a Bayesian statistical method, and obtain a dynamic time delay interval.

[0024] The compensation unit is configured to jointly compensate hysteresis in a link state collection process and advance in a control instruction issuing process according to the dynamic time delay interval.

[0025] Optionally, the collection unit is further configured to:

[0026] record a first time when an event actually occurs, a second time when the event is transmitted to a controller, a third time when the controller completes analysis, and a fourth time when the controller generates a strategy;

[0027] calculate a first time delay difference value according to the second time and the first time, a second time delay difference value according to the third time and the first time, and a third time delay difference value according to the fourth time and the second time;

[0028] divide the first time delay difference value, the second time delay difference value, and the third time delay difference value into a hysteresis time delay of link state detection and an advance time delay of control instruction issuing according to a link direction, and perform feature extraction and normalization processing respectively.

[0029] Optionally, the device further includes:

[0030] The screening unit is configured to perform Jarque-Bera test on a training set before the optimization unit inputs the time delay data into the vector machine model, performs nonlinear mapping and parameter optimization on the time delay data based on the vector machine model and the Bayesian statistical method, and obtains the dynamic time delay interval, and screen time delay samples conforming to normal distribution characteristics.

[0031] The training unit is configured to perform approximate processing on discrete variables by using a Poisson distribution based on a large sample state, and combine a prior distribution to establish a posterior likelihood function, so as to optimize parameters of a support vector machine model.

[0032] Optionally, the compensation unit is further configured to:

[0033] according to similarity of a link state collection path and a control instruction issuing path, use the same kernel function and compensation model to perform bidirectional time delay inference;

[0034] fuse output results of collection compensation and issuing compensation to generate a final strategy execution time offset.

[0035] Optionally, the device further includes:

[0036] The division unit is configured to divide the compensation strategy into multiple compensation levels according to network scales, and adopt different model parameters and compensation intervals at different levels to adapt to real-time requirements of different scale networks.

[0037] According to a third aspect of the present disclosure, an electronic device is provided, comprising:

[0038] at least one processor; and

[0039] a memory connected to the at least one processor in communication; wherein

[0040] The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method of the first aspect.

[0041] According to a fourth aspect of the present disclosure, a non-transitory computer readable storage medium storing computer instructions is provided, wherein the computer instructions are used to make the computer perform the method of the first aspect.

[0042] According to a fifth aspect of the present disclosure, a computer program product is provided, comprising a computer program which, when executed by a processor, implements the method of the first aspect.

[0043] The network exchange real-time enhancement method, device, electronic device and storage medium provided by the present disclosure mainly include the following technical solutions: collecting time delay data of link state events based on optional transition attributes; inputting the time delay data into a vector machine model, and performing nonlinear mapping and parameter optimization on the time delay data based on the vector machine model and a Bayesian statistical method to obtain a dynamic time delay interval; and jointly compensating for the hysteresis in the link state collection process and the advance in the control instruction issuing process according to the dynamic time delay interval. Compared with related technologies, the embodiments of the present disclosure collect time delay data of link state events based on optional transition attributes, perform nonlinear mapping and parameter optimization on the time delay data by using a vector machine model and a Bayesian statistical method to obtain a dynamic time delay interval, and jointly compensate for the hysteresis in the link state collection process and the advance in the control instruction issuing process according to the interval. Therefore, the problem that the scheduling strategy lags behind the actual network state and the instruction execution deviates from the expected effect due to the use of a static time delay compensation strategy in the existing network state collection and control instruction issuing method can be solved, and the technical effects of improving the real-time and accuracy of traffic scheduling, improving the utilization rate of network resources, and reducing the risk of packet loss can be achieved.

[0044] It is to be understood that the details set forth herein do not construe a limitation on the application, but merely are intended to provide a description of certain embodiments. Furthermore, it is understood that the constructions herein illustrated can be substituted for by other BRIEF DESCRIPTION OF DRAWINGS

[0045] The accompanying drawings are included to provide a further understanding of the application, and are incorporated in and constitute a part of this specification, illustrate embodiments of the application, and together with the description serve to explain the principles of the application. In the drawings:

[0046] Figure 1 A flowchart of a network exchange real-time enhancement method provided by an embodiment of the present disclosure;

[0047] Figure 2 A structural schematic diagram of a network exchange real-time enhancement device provided by an embodiment of the present disclosure;

[0048] Figure 3 A structural schematic diagram of a network exchange real-time enhancement device provided by an embodiment of the present disclosure;

[0049] Figure 4 A schematic block diagram of an example electronic device provided by an embodiment of the present disclosure. DETAILED DESCRIPTION

[0050] Exemplary embodiments of the present disclosure are described herein with reference to the accompanying drawings, which are included to provide a further understanding of the present disclosure and are incorporated in and constitute a part of this specification. Embodiments of the present disclosure described herein are described in connection with various details, which are provided to help understand the embodiments. These details are not intended to limit the scope of the present disclosure, but merely to provide examples of the disclosure. Thus, those of ordinary skill in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Also, for clarity and conciseness, descriptions of well-known functions and constructions are omitted from the following description.

[0051] A network exchange real-time enhancement method, device, electronic device and storage medium of an embodiment of the present disclosure are described below with reference to the accompanying drawings.

[0052] Figure 1 A flowchart of a network exchange real-time enhancement method provided by an embodiment of the present disclosure.

[0053] As Figure 1 shown, the method comprises the following steps:

[0054] Step 101, collecting link state event delay data based on optional transition attribute.

[0055] Specifically, this step involves collecting link state event latency data through the optional transition attribute of the Border Gateway Protocol (BGP) protocol and recording the difference between the actual occurrence time and the controller perception time. This step is the key link in the implementation of link state collection latency compensation in this proposal, and its technical implementation is based on the extension attribute mechanism of the BGP protocol and the timestamp collection technology.

[0056] In some implementations, the BGP protocol supports multiple attribute types, among which the optional transition attribute (Optional Transitive Attributes) allows forwarding to other BGP peers without being identified by the opposite device. This proposal takes advantage of the propagation characteristics of this attribute to encapsulate the timestamp information of the link state event in the BGP Update message, inject the timestamp when the event occurs by the source network element, and extract the timestamp after the controller receives the Update message to calculate the difference between the event occurrence time and the controller perception time. This difference reflects the end-to-end transmission latency from the event occurrence to the controller receiving the event.

[0057] The specific operation includes: when the network element device (such as LSW1) detects a link state change (such as interface down, bandwidth fluctuation, etc.), it immediately generates a BGP Update message containing the event type and timestamp, and encapsulates this information in the optional transition attribute. The BGP message is forwarded hop by hop to the controller through IBGP or EBGP neighbor relationship. The controller records the reception time after receiving the Update message, and calculates the difference.

[0058] Further, the latency data collected by this step will be one of the input features for subsequent support vector machine (SVM) model training, used to establish a latency distribution model for link state events. Through large sample collection and Jarque-Bera test, the normality of the latency data can be judged, and the dynamic compensation parameter derivation can be combined with the Bayesian statistical method. This step is particularly suitable for large-scale operator networks in practical applications, such as backbone networks, metropolitan area networks, etc., where link state changes frequently and the controller needs to respond quickly to improve the real-time scheduling capability of traffic engineering.

[0059] In summary, this step collects and records the link state event latency through the optional transition attribute of the BGP protocol, providing high-quality training data for the subsequent latency compensation model, thereby effectively improving the controller's perception of network state in real time. It is the basic link for this proposal to enhance the real-time performance of network switching based on high-dimensional classification algorithms.

[0060] Step 102, input the time delay data into the vector machine model, based on the vector machine model and the Bayesian statistical method, the time delay data is nonlinearly mapped and parameter optimized, and the dynamic time delay interval is obtained.

[0061] The collected time delay data is nonlinearly mapped and parameter optimized by using the support vector machine (SVM) model combined with the Bayesian statistical method, thereby generating the dynamic time delay interval for compensation. This step is the core link of realizing the adaptive compensation of the network controller on the link state collection and control instruction delay, and has significant technical innovation and practicality.

[0062] In some implementations, the SVM model maps the original time delay data to a high-dimensional feature space by introducing a kernel function to solve the problem of linearly inseparable data in the original space. In this proposal, the Laplacian kernel function is selected to control the smoothness of the feature space. This kernel function has good robustness, especially suitable for network time delay data with long-tailed distribution characteristics, and can effectively improve the generalization ability of the model.

[0063] Further, combined with the Bayesian statistical method, this proposal introduces the joint modeling of prior distribution and posterior distribution in the model training stage. By approximating the large sample state data with Poisson distribution, the likelihood function is constructed, and the prior knowledge of historical time delay data is combined to realize the dynamic optimization of model parameters. Specifically, the Markov chain Monte Carlo (MCMC) method is used for parameter sampling to obtain the posterior distribution of the model parameters, thereby improving the prediction accuracy of the model on the time delay change trend.

[0064] At the parameter index level, the kernel function width σ, the regularization parameter C, and the classification boundary tolerance ε need to be set during the model training process. Among them, σ is usually selected in the range of [0.1, 10] through cross-validation, C is optimized in the range of [10^{-3}, 10^3] through grid search, and ε is set to 10^{-4} to ensure the sensitivity of the model to nonlinear time delay changes.

[0065] In application scenarios, this step is mainly used for the network controller to dynamically predict the current delay interval based on historical time delay data after receiving link state information, thereby compensating in advance when issuing control instructions. For example, in large-scale data centers or 5G bearer networks, the network topology is complex and the business traffic fluctuates dramatically, and this method can effectively improve the real-time performance and accuracy of traffic scheduling.

[0066] Through the nonlinear mapping and the Bayesian parameter optimization, the adaptability and the prediction accuracy of the time delay compensation model are significantly improved, the problems of insufficient compensation or over-compensation of the traditional static compensation method in the time-varying network environment are solved, and the real-time performance and the stability of the network exchange are enhanced.

[0067] In step 103, according to the dynamic time delay interval, the hysteresis in the link state collection process and the advance in the control instruction issuing process are jointly compensated.

[0068] This step aims to jointly compensate the hysteresis in the link state collection process and the advance in the control instruction issuing process through modeling and analysis of the dynamic time delay interval, so as to realize accurate matching of the controller strategy and the real-time network state. This process is based on the bidirectional time delay characteristics between network state perception and control response, and uses statistical modeling and machine learning methods to dynamically predict and compensate the time delay.

[0069] Firstly, the time stamps of the transmission of link state information between network elements and controllers are recorded through the "optional transition" attribute in the BGP protocol, including the actual occurrence time T1 of the event, the transmission time T2 to the controller, the completion time T3 of the analysis of the controller, and the policy generation time T4. By calculating the time difference values such as T2-T1 and T4-T2, a time delay data set is constructed. Further, a support vector machine (SVM) model is used, combined with a Laplace kernel function, to map the original nonlinear time delay data to a high-dimensional feature space, realizing nonlinear modeling of the time delay distribution. In the model training process, Bayesian statistical methods are introduced, and parameter optimization is performed based on the prior distribution and the likelihood function, to improve the generalization ability and prediction accuracy of the model.

[0070] This step involves multiple key time delay indicators, including but not limited to link state collection delay (T2-T1), controller processing delay (T3-T2), policy generation delay (T4-T3), etc. In the model training, Jarque-Bera test is used to judge the normality of the data, to ensure the rationality of subsequent statistical modeling. In the large sample condition, the discreteness of the time delay is approximately processed using the Poisson distribution, to enhance the adaptability of the model to the time delay fluctuations. The compensation interval is usually referenced with a 95% confidence interval, and the compensation value is dynamically adjusted in combination with the historical time delay distribution.

[0071] It is suitable for traffic engineering (TE) scheduling scenarios in large-scale operator networks, especially in environments where network topology changes frequently and business traffic fluctuates significantly, such as data center interconnection and 5G bearer networks. Through the joint compensation mechanism, the controller can adjust the strategy in advance according to the predicted time delay interval, to avoid scheduling deviations caused by state lag or instruction advance.

[0072] In some embodiments, the collecting time delay data of the link state event based on the optional transition attribute further includes:

[0073] a first time instant when the event actually occurs, a second time instant when the event is transmitted to the controller, a third time instant when the controller finishes analyzing, and a fourth time instant when the controller generates a strategy;

[0074] a first time delay difference value is calculated according to the second time instant and the first time instant, a second time delay difference value is calculated according to the third time instant and the first time instant, and a third time delay difference value is calculated according to the fourth time instant and the second time instant;

[0075] the first time delay difference value, the second time delay difference value, and the third time delay difference value are divided into a hysteresis time delay of link state detection and an advance time delay of control instruction issuing according to a link direction, and feature extraction and normalization processing are respectively performed.

[0076] An event generally refers to a triggering time instant of a link state change (such as link congestion, failure, bandwidth fluctuation, etc.) in a network. To ensure high precision and consistency of time stamps, the system adopts a network time protocol (NTP) or a more advanced precision time protocol (PTP) for clock synchronization, so that the time deviation of all network elements and the controller is controlled within microseconds (μs). At the network element side, when an event occurs, a time stamp is directly triggered by a bottom layer driver or a protocol stack to avoid additional delay caused by operating system scheduling or application layer processing. The time stamp will be used as a reference time point for subsequent delay modeling and compensation calculation, for calculating the delay of each stage from event occurrence to controller perception, strategy generation, and instruction issuing.

[0077] The recording of the event occurrence time instant T needs to meet certain timing accuracy requirements, and generally requires that the synchronization error of the time stamp does not exceed ±10 μs to ensure the accuracy of delay calculation. In actual deployment, the time synchronization mechanism between the controller and the network element needs to comply with the NTP protocol defined in IEEE 1588v2 standard or RFC 5905. In addition, the event recording module should have high throughput capability to support processing thousands of events per second to cope with frequent link state changes in large-scale networks. In the data acquisition stage, the system pairs the event occurrence time instant T with the time stamps of subsequent stages (such as T2, T3, T4) to form time difference values (ΔT12, ΔT13, ΔT24, etc.), which are used as input features for training a support vector machine (SVM) model.

[0078] This step is widely used in large-scale operator networks, especially in complex topologies with multiple regions and multiple autonomous systems (AS). For example, in data center interconnection (DCI) or 5G bearer networks, link state changes frequently and have very high real-time requirements, so the controller needs to respond in the shortest time. By accurately recording the time T when the event occurs, the system can model the delay characteristics of different links and different time periods, so as to dynamically adjust the compensation strategy during peak hours or after network expansion, improving the accuracy and response speed of traffic scheduling.

[0079] By recording the actual time T when the event occurs, a reliable time reference is provided for subsequent delay compensation models, which is the core prerequisite for dynamic and adaptive compensation. This step effectively solves the scheduling lag or advance problem caused by fixed compensation values in traditional methods, improving the controller's perception accuracy and response efficiency to network state changes, thereby significantly enhancing the real-time and stability of network switching.

[0080] In some embodiments, before inputting the delay data into the vector machine model and performing nonlinear mapping and parameter optimization on the delay data based on the vector machine model and Bayesian statistical methods to obtain a dynamic delay interval, the method further comprises:

[0081] Jarque-Bera test is performed on the training set to screen delay samples that meet the normal distribution characteristics.

[0082] Specifically, in this proposal, the step of "performing Jarque-Bera hypothesis test on the training set to screen delay samples that meet the normal distribution characteristics" is a key preprocessing link in the model training phase, and its purpose is to improve the modeling accuracy and generalization ability of the subsequent support vector machine (SVM) model. This step is based on the Jarque-Bera (JB) test in statistics to evaluate whether the distribution characteristics of the delay samples in the training set are close to the normal distribution, thereby ensuring the statistical consistency and predictability of the model input data.

[0083] Jarque-Bera test judges whether the sample follows a normal distribution by calculating the skewness (Skewness) and kurtosis (Kurtosis) of the sample. Specifically, the calculation formula of JB statistic is:

[0084]

[0085] where n is the sample size, S is the skewness, and K is the kurtosis. In this proposal, the delay samples are derived from the link state propagation time differences recorded in the "optional transit" attribute of the BGP protocol (such as T2-T1, T3-T1, etc.). These time differences exhibit varying degrees of distribution characteristics under different network topologies and traffic loads. By setting a significance level (such as α = 0.05), if the p-value corresponding to the JB statistic is less than the threshold, the normality assumption is rejected and the sample is removed; otherwise, it is retained as valid input for subsequent modeling.

[0086] This step requires a sufficient number of samples (usually n ≥ 30) to meet the applicability conditions of the central limit theorem. At the same time, considering the mixed characteristics of continuity and discreteness of network delay, this proposal uses Poisson distribution to approximate the discrete delay samples to enhance their applicability under the assumption of normal distribution. In addition, to improve the robustness of the model, a sliding window mechanism is optionally introduced to dynamically update and real-time test the delay samples.

[0087] It is mainly used in the data preprocessing stage before the network controller collects link state and issues control instructions. By normality screening of historical delay data, abnormal or atypical samples can be effectively removed, thereby improving the prediction ability of support vector machine model in high-dimensional feature space for delay compensation.

[0088] It significantly improves the quality of model input data, reduces prediction errors caused by distribution deviation, and provides a more representative sample set for the construction of Bayesian likelihood function and the training of SVM model from a statistical point of view, thereby enhancing the real-time performance and adaptability of the entire system in dynamic network environment.

[0089] Based on large sample state, Poisson distribution is used to approximate discrete variables, and prior distribution is combined to establish posterior likelihood function to optimize the parameters of support vector machine model.

[0090] In some embodiments, based on the link state attributes extracted from the BGP protocol, a training data set containing multiple time series features is constructed. Since the collection and processing of link state involves multiple network nodes, its delay has obvious discreteness and randomness, so probability modeling of discrete variables is needed when modeling. Considering that the sample size is large (usually thousands to tens of thousands of event records), under the condition of meeting the large sample approximation, Poisson distribution is used to approximate discrete variables (such as the number of events, state update frequency, etc.), so that the originally non-continuous variables are converted into continuous form for probability derivation, facilitating subsequent statistical modeling.

[0091] Further, a Bayesian statistical framework is introduced to incorporate prior distribution to express the initial belief on the model parameters. In this proposal, the prior distribution can be set based on the statistical properties of the historical network delay data, for example, using Gamma distribution or Exponential distribution as the conjugate prior to enhance the stability of the model. Subsequently, the posterior likelihood function is constructed by combining the likelihood function of the training samples with the prior distribution through the maximum a posteriori (MAP) method, which is used to optimize the hyperparameters of the SVM model, such as the penalty coefficient C and the kernel function parameter σ.

[0092] In terms of parameter setting, the Jarque-Bera test is used to verify the normality of the data, and its statistic is where n is the sample size, S is the skewness, and K is the kurtosis. If the JB statistic is significantly greater than the critical value, the normality assumption is rejected, and the Poisson approximation is used instead. The SVM model uses the Laplace kernel function The semi-definiteness of which guarantees the convergence and stability of the model.

[0093] This step can effectively deal with the time-varying and nonlinear characteristics of link state in practical applications, especially in large-scale network topologies, and improve the prediction ability of the SVM model for network delay, providing a high-precision model basis for subsequent dynamic compensation mechanisms, thereby enhancing the real-time performance and scheduling efficiency of network switching.

[0094] In some embodiments, the joint compensation of the hysteresis in the link state collection process and the advance in the control instruction issuing process according to the dynamic delay interval includes:

[0095] According to the similarity of the link state collection path and the control instruction issuing path, the same kernel function and compensation model are used for bidirectional delay inference;

[0096] The technical implementation principle is based on modeling and analyzing the similarity of the link state collection and control instruction issuing processes in terms of path structure and delay characteristics, thereby realizing unified inference and compensation of the delays in both directions.

[0097] In some implementations, the link state collection path is the process of uploading state information from network elements (such as LSW, PE, etc.) to the controller, while the control instruction issuing path is the process of the controller sending scheduling instructions to the network elements. The two processes often have symmetry or high similarity in physical paths, especially in large-scale networks, where the communication link structure between the controller and the network elements tends to be consistent. Therefore, this proposal proposes to use the path similarity to use a unified kernel function (such as the Laplace kernel) and a support vector machine (SVM) model to jointly model and infer the delays in both directions.

[0098] The timestamp information in the link state collection and control instruction issuing process is recorded through the "optional transition" attribute in the BGP protocol, including the time when the event occurs (T1), the time when it is transmitted to the controller (T2), the time when the controller completes the analysis (T3), and the time when the control instruction reaches the controlled unit (T4). Based on these timestamps, a bidirectional delay dataset is constructed, covering the lag delay of the collection path (T2-T1, T3-T1) and the lead delay of the issuing path (T4-T2). In the model training stage, the Laplace kernel function is used to map the original nonlinear delay data to a high-dimensional feature space, the likelihood function is established through Bayesian statistical method, and the SVM model is trained and parameter optimized combined with large sample data.

[0099] The Laplace kernel function used in this application has good robustness and nonlinear mapping ability, and its kernel width parameter σ is usually optimized in the range of [0.1, 10] through cross-validation method. The division ratio of the training set and the test set of the compensation model is suggested to be 7:3 to ensure the model generalization ability. The output result of the delay compensation is presented in the form of normal distribution interval, and the confidence interval is usually set to 95% to adapt to the time-varying and random nature of network state.

[0100] In some embodiments, this application is applicable to large-scale operator networks, especially in scenarios where network topology changes frequently and traffic fluctuates dramatically, such as 5G core network, data center interconnection, edge computing network, etc. Through bidirectional delay inference, the controller can predict the execution delay of the control instruction while obtaining the link state, so as to dynamically adjust the scheduling strategy generation and improve the timeliness and accuracy of scheduling.

[0101] Further, the technical effect of this step is to significantly improve the real-time and stability of network scheduling. Through unified modeling and inference, the inconsistency problem of traditional methods caused by path differences is avoided, and the adaptability of the system to network state changes is enhanced, providing key support for realizing high-dimensional classification algorithm to enhance network switching real-time.

[0102] The output results of collection compensation and issuing compensation are fused to generate the final strategy execution time offset.

[0103] In some implementations, the collection compensation is used to correct the lag generated in the process of uploading the link state from the network element to the controller, and the delivery compensation is used to predict the advance time before the control instruction is sent from the controller to the controlled network element for execution. Both are modeled by a support vector machine (SVM) model, and the model input includes link length, network load, delay information recorded in the BGP attribute, etc., and the output is a delay compensation value. The collection compensation model output is a positive time offset (such as +50ms), indicating how far in advance the controller should predict the current state; the delivery compensation model output is a negative time offset (such as -30ms), indicating that the controller should issue the instruction in advance to offset the execution delay.

[0104] Further, the fusion process adopts methods such as weighted average or Kalman filtering, combining the confidence, historical error, and link stability of the collection and delivery compensation to generate the final policy execution time offset. For example, the collection compensation weight can be set to 0.6, and the delivery compensation weight can be set to 0.4, and the weight coefficient can be dynamically adjusted according to the volatility of the link state. This fusion mechanism needs to meet the requirements of IEEE 802.1Q standard on time synchronization and scheduling accuracy, and also meet the processing specifications of BGP protocol in RFC 4271 for attribute transmission delay.

[0105] In actual application scenarios, this step is suitable for large-scale data centers or operator backbone networks, especially in scenarios where the link state changes frequently and the control instruction response time is sensitive, such as 5G core network slice scheduling, real-time video streaming optimization, etc. Through this fusion compensation mechanism, the controller can more accurately predict the network state change trend, realize the unification of "advance execution" and "lag correction" of the policy, significantly improve the real-time performance and accuracy of traffic scheduling, reduce the risk of policy failure caused by delay, and thus enhance the overall throughput capacity and service quality of the network.

[0106] In some embodiments, the method further comprises:

[0107] According to the network size, the compensation policy is divided into multiple compensation levels, and different model parameters and compensation intervals are used at different levels to adapt to the real-time performance requirements of different size networks.

[0108] Specifically, the core of this step is to divide the compensation levels according to the network size, and use differentiated model parameters and compensation intervals at different levels to improve the compensation accuracy and real-time performance of the network controller in link state collection and control instruction delivery delay in different size networks. In some implementations, this step realizes dynamic and adaptive compensation of network delay by introducing a multi-level compensation mechanism and combining the parameter adjustment strategy of the support vector machine (SVM) model.

[0109] The network scale is usually quantified by the number of network elements managed by the controller, link density, topology complexity, and the like. For example, the network scale can be divided into three levels: low (<100 network elements), medium (100-500 network elements), and high (>500 network elements). Each level corresponds to a different compensation interval (for example, the low-level compensation interval is [0, 50 ms], the medium-level is [50 ms, 150 ms], and the high-level is [150 ms, 300 ms]), as well as different SVM model parameter configurations, including kernel function parameters (such as the bandwidth parameter σ of the Laplace kernel), regularization coefficient C, and classification boundary tolerance ε. The settings of these parameters need to be optimized according to the distribution characteristics of historical delay data to ensure the generalization ability and prediction accuracy of the model under different scale networks.

[0110] The compensation level division is based on the Jarque-Bera test to determine the normality of the data. If the data is not normally distributed, a Poisson distribution is used for approximation, and the model is trained based on the Bayesian likelihood function. The compensation interval is statistically modeled based on the delay samples recorded by the "optional transition" attribute in the BGP protocol, and the reasonable compensation range is determined by calculating the 95% confidence interval or the quantile interval (such as 25%-75%). In addition, the update frequency of the compensation model should also be adjusted according to the network scale, for example, the high-level network can be set to update the model parameters every 5 minutes to adapt to more frequent topology changes.

[0111] This step is suitable for large-scale data center networks, 5G bearer networks, cross-domain transmission networks, and other scenarios with high real-time requirements. For example, in the 5G core network, the controller needs to complete path adjustment within a millisecond-level response time. If a unified compensation strategy is used, it will lead to insufficient or excessive compensation, affecting the scheduling effect. Through the hierarchical compensation mechanism, the controller can automatically select the optimal compensation model and parameters according to the current network scale it manages, thereby improving the timeliness and accuracy of the scheduling instructions.

[0112] The technical effect of this step is to significantly improve the delay compensation ability of the network controller under different scale networks, avoiding the compensation deviation caused by the change of the network state in the traditional static compensation strategy. Through the multi-level compensation mechanism, the controller can more accurately predict the lag of the link state and the advance of the control instruction, thereby achieving better path selection and resource allocation in traffic scheduling and improving the throughput capacity and quality of service (QoS) of the overall network.

[0113] The real-time enhancement method for network switching of the embodiments of the present application divides the compensation levels according to the network scale and uses differentiated model parameters and compensation intervals, further improving the adaptability and refinement of the compensation strategy, so that the network controller can achieve better delay compensation effect in different scale networks, thereby significantly enhancing the switching real-time and scheduling accuracy in large-scale network environments.

[0114] The following describes a network switching real-time enhancement method provided by the application in an embodiment.

[0115] After the link state database is established by the OSPF protocol and the ISIS protocol, the database is delivered to the controller by the BGP protocol. The BGP protocol is a path vector routing protocol, and supports multiple protocol versions, including IPV4 unicast, IPV4 multicast, VPNv4 and VPNv6, etc. A BGP neighbor relationship is established with a peripheral network element. If both belong to one autonomous system, an IBGP neighbor relationship is established, and if they belong to different autonomous systems, an EBGP neighbor relationship is established.

[0116] In the BGP protocol, there are a large number of attributes for carrying information required when a policy is used. Please refer to Table 1, which is a BGP protocol attribute table provided by the embodiment of the application, as shown in Table 1:

[0117] Table 1

[0118]

[0119] Here, the "optional transition" attribute is used to record the time delay generated in the information transmission process.

[0120] After the information carried by the attribute is obtained, a data set can be established, including:

[0121] Training data set: used for training model parameters

[0122] Test data set: used for verifying model parameters

[0123] According to the network size, the compensation scale of the controller can be divided into several levels. Generally, the larger the network size, the more compensation levels are required, and the greater the differentiation degree is. For a medium-sized network, it is generally divided into three levels of high, medium and low.

[0124] In order to achieve better results, the normality of the data needs to be first tested by Jarque-Bera hypothesis. The selected statistic is the commonly used JB statistic.

[0125]

[0126] Considering the relationship between continuous variables and discrete variables in actual application, Poisson distribution is used here to approximate in a large sample state, so that the posterior distribution can be naturally introduced, and then combined with the known prior distribution, the Bayesian statistical method is adopted. Specifically, a likelihood function can be established for the latest sample.

[0127]

[0128] Then the original linearly inseparable nonlinear samples are mapped to a higher dimensional feature space by using support vector machines, so that the samples are linearly separable in this feature space. In order to find this hyperplane, the key is to find a suitable kernel function κ(x t ,x t ) such that the kernel matrix K is always semi-positive definite for any data D = {x y ,x2,…,x m}

[0129]

[0130] Here, the Laplace kernel is selected as the kernel function

[0131]

[0132] Thus, a feasible solution can be obtained for the originally linearly inseparable problem in a low-dimensional space.

[0133] The specific execution steps are as follows:

[0134] 1. Collect the link state carried by the BGP attribute;

[0135] 2. Form a training set and a test set;

[0136] 3. Perform Jarque-Bera hypothesis testing on the training set;

[0137] 4. Approximate the normal distribution by the Poisson distribution based on a large sample;

[0138] 5. Introduce Bayesian statistics to establish a likelihood function;

[0139] 6. Train the support vector machine model;

[0140] 7. Use the support vector machine model on the test set;

[0141] 8. Measure the bias of the model according to the results;

[0142] 9. Correct the model according to the bias;

[0143] 10. Deploy the corrected model.

[0144] Compensation for link state collection delay

[0145] During the link state collection process, delay inevitably occurs, that is, there is a time difference from the occurrence of an event to the perception of the controller. Please refer to Table 2, which is a link state collection delay table provided by an embodiment of the present application, as shown in Table 2:

[0146] Table 2

[0147]

[0148] Due to the different network states, the transmission delay is a random variable, which will appear in an interval, sometimes it may be large, sometimes it may be small, so the compensation also needs to adapt to the randomness accordingly, and cannot adopt a fixed compensation value

[0149] In order to deal with this feature, the present application adopts the method of large sample to smooth the randomness, and establishes the training set required by the model, and through the obtained model, the reasonable interval of delay compensation is dynamically calculated

[0150] The specific execution steps are as follows:

[0151] 1. Record the time when the link event occurs;

[0152] 2. Calculate the time difference of transmission to the controller;

[0153] 4. Calculate the time difference of the controller analysis completion;

[0154] 5. Calculate the time difference of the controller generating strategy;

[0155] 6. Call the support vector machine model for inference;

[0156] 7. Compensate for the link delay.

[0157] Compensation of control instruction issuing delay

[0158] In the control instruction issuing, the delay phenomenon encountered is similar to the link state collection, but not completely the same. In terms of timing, as follows:

[0159] (1) Link state collection: the link state event occurs, and after a certain random delay, it reaches the controller

[0160] (2) Control instruction issuing: control instruction generation and sending, after a certain random delay, reaches the controlled unit

[0161] It can be seen that the compensation of control instruction issuing delay and the collection of link state have opposite characteristics, one is lagging behind and one is ahead of time.

[0162] At the same time, the delay path passed by the control instruction issuing is very similar to the link state monitoring, so the calculation result has a certain degree of reference significance. Please refer to Table 3, which is a link state monitoring table provided by the embodiment of the present application, as shown in Table 3:

[0163] Table 3

[0164]

[0165] Corresponding to the real-time enhancement method of network exchange described above, the application also provides a real-time enhancement device of network exchange. Since the device embodiment of the application corresponds to the method embodiment described above, for the details not disclosed in the device embodiment, please refer to the method embodiment described above, which will not be described in detail in the application.

[0166] Figure 2 A structural schematic diagram of a real-time enhancement device of network exchange provided by an embodiment of the present disclosure is shown in Figure 2 , which includes:

[0167] The acquisition unit 21 is configured to acquire time delay data of a link state event based on optional transition attributes;

[0168] The optimization unit 22 is configured to input the time delay data into a vector machine model, and perform nonlinear mapping and parameter optimization on the time delay data based on the vector machine model and a Bayesian statistical method, to obtain a dynamic time delay interval.

[0169] The compensation unit 23 is configured to jointly compensate for hysteresis in a link state acquisition process and advance in a control instruction issuing process according to the dynamic time delay interval.

[0170] Further, in a possible implementation manner of an embodiment of the present disclosure, as shown in Figure 3 , the acquisition unit 21 is further configured to:

[0171] record a first time when an event actually occurs, a second time when the event is transmitted to a controller, a third time when the controller completes analysis, and a fourth time when the controller generates a strategy;

[0172] calculate a first time delay difference value according to the second time and the first time, a second time delay difference value according to the third time and the first time, and a third time delay difference value according to the fourth time and the second time;

[0173] divide the first time delay difference value, the second time delay difference value, and the third time delay difference value into a hysteresis time delay of link state detection and an advance time delay of control instruction issuing according to a link direction, and perform feature extraction and normalization processing respectively.

[0174] Further, in a possible implementation manner of an embodiment of the present disclosure, the device further includes:

[0175] The screening unit 24 is configured to perform Jarque-Bera test on a training set before the optimization unit 22 inputs the time delay data into the vector machine model, and performs nonlinear mapping and parameter optimization on the time delay data based on the vector machine model and the Bayesian statistical method, to obtain a dynamic time delay interval, and screen time delay samples conforming to normal distribution characteristics.

[0176] The training unit 25 is configured to approximate a discrete variable by using a Poisson distribution based on a large sample state, and establish a posterior likelihood function by combining a prior distribution, so as to optimize parameters of a support vector machine model.

[0177] Further, in a possible implementation of the embodiment of the present disclosure, the compensation unit 23 is further configured to:

[0178] According to the similarity between the link state collection path and the control instruction issuing path, the same kernel function and compensation model are used for bidirectional time delay inference.

[0179] The output results of the collection compensation and the issuing compensation are fused to generate a final policy execution time offset.

[0180] Further, in a possible implementation of the embodiment of the present disclosure, as shown in Figure 3 The apparatus further includes:

[0181] The division unit 26 is configured to divide the compensation policy into multiple compensation levels according to the network scale, and use different model parameters and compensation intervals at different levels to adapt to the real-time requirements of different scale networks.

[0182] It should be noted that the foregoing explanation and description of the method embodiments are also applicable to the apparatus of the embodiment of the present disclosure, and the principle is the same, which is not limited in the embodiment of the present disclosure.

[0183] According to the embodiments of the present disclosure, the present disclosure further provides an electronic device, a readable storage medium and a computer program product.

[0184] Figure 4 A schematic block diagram of an example electronic device 300 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptops, desktops, tablets, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular telephones, smartphones, wearable devices, and other similar computing devices. The components shown here, their connections and relationships, and their functions, are meant to be examples only, and are not intended to limit the implementations of the present disclosure described and / or claimed in this document.

[0185] As Figure 4As shown, the device 300 includes a computing unit 301 that can perform various appropriate actions and processes in accordance with a computer program stored in a ROM (Read-Only Memory) 302 or a computer program loaded into a RAM (Random Access Memory) 303 from a storage unit 308. In the RAM 303, various programs and data required for the operation of the device 300 can also be stored. The computing unit 301, the ROM 302, and the RAM 303 are connected to each other through a bus 304. An I / O (Input / Output) interface 305 is also connected to the bus 304.

[0186] A plurality of components in the device 300 are connected to the I / O interface 305, including: an input unit 306, such as a keyboard, a mouse, and the like; an output unit 307, such as various types of displays, speakers, and the like; a storage unit 308, such as a magnetic disk, an optical disk, and the like; and a communication unit 309, such as a network card, a modem, a wireless communication transceiver, and the like. The communication unit 309 allows the device 300 to exchange information / data with other devices through a computer network, such as the Internet, and / or various telecommunication networks.

[0187] The computing unit 301 can be various general-purpose and / or special-purpose processing components having processing and computing capabilities. Some examples of the computing unit 301 include, but are not limited to, a CPU (Central Processing Unit), a GPU (Graphic Processing Unit), various special-purpose AI (Artificial Intelligence) computing chips, various computing units running machine learning model algorithms, a DSP (Digital Signal Processor), and any appropriate processor, controller, microcontroller, and the like. The computing unit 301 performs various methods and processes described above, such as the real-time enhancement method for network exchanges. For example, in some embodiments, the real-time enhancement method for network exchanges can be implemented as a computer software program that is tangibly embodied in a machine-readable medium, such as the storage unit 308. In some embodiments, part or all of the computer program can be loaded and / or installed on the device 300 via the ROM 302 and / or the communication unit 309. When the computer program is loaded into the RAM 303 and executed by the computing unit 301, one or more steps of the methods described above can be performed. Alternatively, in other embodiments, the computing unit 301 can be configured to perform the aforementioned real-time enhancement method for network exchanges by any other appropriate means, such as by means of firmware.

[0188] Various implementations of the systems and techniques described above can be realized in digital electronic circuitry, integrated circuitry, a Field Programmable Gate Array (FPGA), an Application-Specific Integrated Circuit (ASIC), an Application Specific Standard Product (ASSP), a System on a Chip (SOC), a Complex Programmable Logic Device (CPLD), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include implementation in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

[0189] Program code for carrying out methods of the present disclosure can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general or special purpose computer, such that the program code, when executed by the processor or controller, causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can execute entirely on a machine, partly on the machine, as a stand-alone software package, partly on the machine and partly on a remote machine or entirely on the remote machine or server.

[0190] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, RAM, ROM, EPROM (Electrically Programmable Read-Only Memory) or flash memory, optical fiber, CD-ROM (Compact Disc Read-Only Memory), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0191] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (Cathode-Ray Tube) or LCD (Liquid Crystal Display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0192] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include LANs (Local Area Networks), WANs (Wide Area Networks), the Internet, and blockchain networks.

[0193] The computer system can include clients and servers. This relationship can be between a client and a server that are typically remote from each other and typically interact through a communication network. The relationship between client and server exists by virtue of computer programs running on the respective computer systems and having a client-server relationship to each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system, to solve the defects of large management difficulty and weak business scalability in traditional physical host and VPS (Virtual Private Server, or VPS for short) services. The server can also be a server of a distributed system, or a server combined with a blockchain.

[0194] It should be noted that artificial intelligence is a discipline that studies enabling computers to simulate some thinking processes and intelligent behaviors of humans (such as learning, reasoning, thinking, planning, etc.), both hardware and software technologies. Artificial intelligence hardware technology generally includes technologies such as sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing, etc.; artificial intelligence software technology mainly includes computer vision technology, speech recognition technology, natural language processing technology, and machine learning / deep learning, big data processing technology, knowledge graph technology, etc. several major directions.

[0195] It should be understood that the various forms of the flow shown above can be used to reorder, add or delete steps. For example, each step described in the present disclosure can be executed in parallel, sequentially or in a different order, as long as the desired results of the technical solutions disclosed in the present disclosure can be achieved, which is not limited herein.

[0196] The above detailed description does not constitute a limitation on the scope of protection of the present disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent replacements and improvements made within the spirit and principles of the present disclosure shall be included in the scope of protection of the present disclosure.

Claims

1. A method for enhancing the real-time performance of network switching, characterized in that, include: Delay data of link status events is collected based on optional transition attributes; The time delay data is input into a vector machine model. Based on the vector machine model and Bayesian statistical methods, the time delay data is nonlinearly mapped and the parameters are optimized to obtain a dynamic time delay interval. Based on the dynamic delay interval, joint compensation is performed to address the lag in the link status acquisition process and the lead in the control command issuance process.

2. The method according to claim 1, characterized in that, The latency data for collecting link status events based on optional transition attributes also includes: Record the first moment when the event actually occurs, the second moment when it is transmitted to the controller, the third moment when the controller completes parsing, and the fourth moment when the controller generates the strategy; Calculate the first time delay difference based on the second time point and the first time point, calculate the second time delay difference based on the third time point and the first time point, and calculate the third time delay difference based on the fourth time point and the second time point; The first delay difference, the second delay difference, and the third delay difference are divided into the lag delay of link status detection and the lead delay of control command issuance according to the link direction, and feature extraction and normalization are performed respectively.

3. The method according to claim 1, characterized in that, Before inputting the time delay data into the vector machine model, and performing nonlinear mapping and parameter optimization on the time delay data based on the vector machine model and Bayesian statistical methods to obtain the dynamic time delay interval, the method further includes: The training set was subjected to the Jarque-Bera test to select time delay samples that conform to the characteristics of a normal distribution; Based on a large sample state, the Poisson distribution is used to approximate discrete variables, and the posterior likelihood function is established in combination with the prior distribution to optimize the parameters of the support vector machine model.

4. The method according to claim 1, characterized in that, The joint compensation for the lag in the link status acquisition process and the lead in the control command issuance process based on the dynamic delay interval includes: Based on the similarity between the link status acquisition path and the control command issuance path, the same kernel function and compensation model are used for bidirectional delay inference; The output results of the collected compensation and the distributed compensation are merged to generate the final policy execution time offset.

5. The method according to claim 1, characterized in that, The method further includes: The compensation strategy is divided into multiple compensation levels based on the network size, and different model parameters and compensation ranges are used under different levels to adapt to the real-time requirements of networks of different sizes.

6. A real-time performance enhancement device for network switching, characterized in that, include: The acquisition unit is used to acquire latency data of link status events based on optional transition attributes; An optimization unit is used to input the time delay data into a vector machine model, and perform nonlinear mapping and parameter optimization on the time delay data based on the vector machine model and Bayesian statistical methods to obtain a dynamic time delay interval. The compensation unit is used to jointly compensate for the lag in the link status acquisition process and the lead in the control command issuance process based on the dynamic delay interval.

7. The apparatus according to claim 6, characterized in that, The acquisition unit is also used for: Record the first moment when the event actually occurs, the second moment when it is transmitted to the controller, the third moment when the controller completes parsing, and the fourth moment when the controller generates the strategy; Calculate the first time delay difference based on the second time point and the first time point, calculate the second time delay difference based on the third time point and the first time point, and calculate the third time delay difference based on the fourth time point and the second time point; The first delay difference, the second delay difference, and the third delay difference are divided into the lag delay of link status detection and the lead delay of control command issuance according to the link direction, and feature extraction and normalization are performed respectively.

8. An electronic device, characterized in that, include: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-5.

9. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-5.

10. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method according to any one of claims 1-5.

Images