Multi-network parallel acceleration system

Through network traffic monitoring module and deep clustering technology, combined with machine learning model, the shortcomings of network traffic prediction and abnormal detection in the existing technology are solved, efficient network management and failure recovery are achieved, and network stability and performance are improved.

CN119052085AActive Publication Date: 2024-11-29GUAN JULONG AUTOMATION EQUIP

Patent Information

Application Number
CN202411262510.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-10
Publication Date
2024-11-29
Estimated Expiration
2044-09-10

AI Technical Summary

Technical Problem

Existing network collaboration technologies lack flexibility and accuracy in predicting network traffic and processing delays, anomaly detection and dynamic data queue management, resulting in reduced network performance and reliability, especially in the event of network congestion or failure.

Method used

The network traffic monitoring module is used to analyze traffic trends based on the autoregressive moving average algorithm, combine deep clustering and machine learning models for abnormal detection and fault analysis, dynamic scheduling priority and resource allocation, and optimize data flow and fault recovery.

Benefits of technology

It realizes accurate prediction and abnormal identification of network behavior, improves network response speed and processing efficiency, ensures service continuity and efficiency in the event of network congestion or failure, and significantly improves the reliability and overall performance of the network.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119052085A_ABST
    Figure CN119052085A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of network collaboration, in particular to a multi-network parallel acceleration system which comprises a network flow monitoring module, an anomaly detection and response module, a data queue management module and a fault response operation module. According to the method, network flow and delay data are analyzed through an autoregression moving average algorithm, accurate prediction of future network behaviors is achieved, potential fluctuation is recognized, the prospective performance of the operation is improved through prediction, data behavior modes are deeply analyzed through trend analysis, recognition of the data behavior modes is accelerated by refining abnormal states, and the accuracy of the data behavior modes is improved. The response speed and processing efficiency of the network are improved, data queue management becomes more efficient, the priority and resource allocation of data processing are dynamically scheduled and optimized, the stability of network operation is ensured, parallel processing optimization ensures the service continuity when the network is jammed or breaks down, seamless network experience is provided for users, and the user experience is improved. The optimization of the fault response mechanism reduces the service interruption time, and improves the reliability and performance of the network.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of network collaboration technology, and in particular to a multi-network parallel acceleration system. Background Art

[0002] The field of network collaboration technology involves the interaction of multiple computing devices and network systems to improve the efficiency of data transmission and the overall performance of the system. This field pays special attention to the coordination and optimization between different network technologies, including network resource sharing, load balancing, fault recovery, and data flow priority management. Network collaboration technology implements efficient data routing strategies and communication protocols in different network environments to achieve optimal allocation and use of network resources, thereby enhancing network reliability and service quality. This technology is suitable for large-scale enterprise-level networks or cloud computing environments, as well as specific industry applications that require a high degree of network coordination.

[0003] Among them, the multi-network parallel acceleration system is a system designed to transmit data simultaneously in multiple network connections. The purpose is to accelerate data processing and transmission speeds and improve the overall performance and reliability of the network. The system achieves high-speed data transmission and processing by utilizing multiple network resources in parallel, such as different data centers or network access points. Its main uses include improving the response time of data-intensive applications, such as online video streaming, large-scale online transaction processing, and cloud services. Through multi-network parallel operation, the system can continue to maintain service stability and efficiency when the network is congested or a network fails, thereby providing users with a seamless and efficient network experience.

[0004] Although existing technologies involve the interaction of multiple computing devices and network systems, they lack sufficiently flexible and accurate tools to predict network traffic and processing delays, resulting in the inability to effectively foresee and adjust the allocation of network resources in actual operations, thereby affecting the performance and reliability of the overall network. Existing network collaboration technologies fail to accurately identify problems in real time in terms of anomaly detection, which leads to the spread of problems and increases the recovery time and cost of network systems. There is a lack of efficient dynamic data queue management strategies, which limits the flexibility and efficiency of the network in processing large-scale or high-density requests. Especially in the event of network congestion or failure, existing technologies cannot maintain the stability and efficiency of services, thereby affecting user experience and service quality. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings in the prior art and to propose a multi-network parallel acceleration system.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: A multi-network parallel acceleration system comprises:

[0007] The network traffic monitoring module uses the autoregressive moving average algorithm to analyze the data items based on the network traffic and delay data, generates a traffic trend prediction model, and predicts the traffic and delay in the future time period through the traffic trend prediction model to obtain the traffic and delay trend prediction results;

[0008] The anomaly detection and response module uses the traffic and delay trend prediction results to analyze the behavior patterns of data points, compares the degree of deviation of data points, generates preliminary anomaly detection results, and refines the anomaly points through a deep clustering method based on the preliminary anomaly detection results to obtain an abnormal state verification result;

[0009] The data queue management module evaluates the request priority according to the abnormal data points in the abnormal state verification result, performs queue reordering, optimizes the data flow through parallel processing, calculates the data flow priority, generates a priority adjustment result, and dynamically schedules the data request based on the priority adjustment result to obtain a scheduling optimization result;

[0010] The fault response operation module monitors performance and potential faults based on the scheduling optimization results, analyzes fault points using a machine learning model, generates fault point analysis results, optimizes fault recovery operations in multi-network parallel acceleration based on the fault point analysis results, and establishes multi-network fault response results.

[0011] As a further solution of the present invention, the steps of obtaining the traffic and delay trend prediction results are specifically as follows:

[0012] Based on network traffic and delay data, an autoregressive moving average algorithm is used to calculate the autoregressive coefficient and moving average coefficient using the formula:

[0013]

[0014] Get the preliminary trend analysis results, where α a represents the autoregressive coefficient, β a represents the moving average coefficient, X a Represents the previous traffic data, Z a Represents the previous delay data, Y a Represents preliminary trend analysis results;

[0015] The preliminary trend analysis results are refined using the formula:

[0016] Q a =log(γ a ·Y a +1)

[0017] Introducing logarithmic transformation parameters to generate refined trend analysis results, where γ a represents the logarithmic transformation enhancement parameter, Qa Represents the refined trend analysis results, Y a Represents preliminary trend analysis results;

[0018] The refined trend analysis results are combined with real-time network monitoring data, using the weight adjustment coefficient, using the formula:

[0019]

[0020] Get the traffic and delay trend prediction results, where θ a represents the trend adjustment coefficient, W a Represents real-time monitoring data, T a represents the traffic and delay trend prediction results, Q a Represents the refined trend analysis results.

[0021] As a further solution of the present invention, the steps of obtaining the preliminary abnormality detection results are specifically as follows:

[0022] The data point behavior pattern analysis is performed on the traffic and delay trend prediction results, and normalization is performed, through the formula:

[0023]

[0024] Generate a normalized behavior model where T a Represents the traffic and delay trend prediction results, ∈ b is a small constant added for stability calculation, U b Represents the normalized behavior pattern;

[0025] The normalized behavior model is used to perform outlier analysis, using the formula:

[0026]

[0027] Get the deviation analysis results, where V b is the deviation analysis result, It's U b The average value, U b Represents the normalized behavior pattern;

[0028] The deviation analysis result is compared with the threshold value, and the judgment threshold value is set as θ b , by comparing V b Is it greater than θ b , determine whether each data point is an anomaly and generate preliminary anomaly detection results.

[0029] As a further solution of the present invention, the step of obtaining the abnormal state verification result is specifically:

[0030] Receive data from the preliminary anomaly detection results and convert them into a format that matches the cluster analysis, using the formula:

[0031]

[0032] Generate a weighted distance matrix where x c,i represents the data points obtained from anomaly detection, μ c represents the average value of the data points, n represents the total number of data points, D c represents the weighted distance matrix;

[0033] According to the weighted distance matrix, a deep clustering algorithm is applied to classify data points, using the formula:

[0034]

[0035] Calculate the classification labels of the data points and obtain the group classification results, where G c is the group classification result, w c,i is the weight of the data point, D c represents the weighted distance matrix;

[0036] The group classification results are analyzed to check the consistency of the data points in each classification group, and the variability within the group is analyzed using multivariate statistical methods to determine the abnormal state of each group and generate abnormal state verification results.

[0037] As a further solution of the present invention, the step of obtaining the priority adjustment result is specifically:

[0038] According to the abnormal status verification results, a preliminary priority is assigned to each data point according to the abnormal level, using the formula:

[0039]

[0040] The priority of generating data points, where k d is the adjustment factor, S d Indicates the abnormal state of the data point, P d is the priority of the data point;

[0041] Using the priority of the data point, the data queue is initially reordered, and the priority data is processed by the function, using the formula:

[0042]

[0043] Generate a queue with adjusted priority in descending order, where R d Indicates the queues with descending priority after adjustment, P d is the priority of the data point, and sort means sorting the calculation results;

[0044] The queues arranged in descending order of priority after adjustment are applied, data is optimized by a parallel processing method, and processing resources are allocated to each queue entry to obtain a priority adjustment result.

[0045] As a further solution of the present invention, the step of obtaining the scheduling optimization result is specifically:

[0046] Based on the priority adjustment result, the data request is initially scheduled, the priority result of each data request is received, and the initial scheduling priority of each data request is calculated as a function of the data priority, weight and load status, through the formula:

[0047]

[0048] Generate preliminary scheduling priority, where P h Indicates the priority adjustment result, w h is the weight of each data request, m h is the current load state, γ h is the load regulation parameter, D h Indicates the preliminary scheduling priority;

[0049] The preliminary scheduling priority is optimized, and the data requests are sorted according to the scheduling score using the formula:

[0050]

[0051] Get the scheduling score, where l h For real-time response delay, b h is the set delay benchmark, η h is the scheduling preference parameter, c h is the additional priority factor, S h is the scheduling score, D h Indicates the preliminary scheduling priority;

[0052] According to the scheduling score, the scheduling priority of multiple data requests is analyzed, and a threshold is set to determine S h Whether the optimization conditions are met, re-evaluation and resource allocation are performed for data requests that do not meet the conditions, and the scheduling optimization result is obtained after the resource allocation is completed.

[0053] As a further solution of the present invention, the steps of obtaining the fault point analysis result are specifically as follows:

[0054] According to the scheduling optimization results, the real-time data of multiple performance indicators are monitored, and the deviation of each performance indicator is calculated after capturing the data, using the formula:

[0055]

[0056] Generate the deviation value of the monitoring indicator, where x g To monitor data values ​​in real time, μ g is the mean of previous data, σ g is the standard deviation of previous data, n g is the sample size, Δ g Indicates the deviation value of the monitoring indicator;

[0057] The deviation value of the monitoring indicator is called, combined with the load level and status data, to calculate the failure probability of each monitoring point, through the formula:

[0058]

[0059] Generate potential failure probability, where λ g is the fault sensitivity coefficient, ω g is the state influence factor, L g is the load level, F g represents the potential failure probability, Δ g Indicates the deviation value of the monitoring indicator;

[0060] Analyze the potential failure probability, compare and classify the failure type through the data of the machine learning model, determine the location of the failure point, set the judgment threshold, determine whether each monitoring point is a failure point, and generate the failure point analysis result.

[0061] As a further solution of the present invention, the step of obtaining the multi-network fault response result is specifically:

[0062] According to the fault point analysis results, determine the network paths affected in multi-network parallel acceleration, calculate the performance loss of each path in turn, and evaluate the impact on the multi-network environment using the formula:

[0063]

[0064] Get the performance loss rate of each path, where p v is the real-time data transmission rate, q v is the expected transmission rate, σ v is the path weight parameter, κ v is the redundancy coefficient of the path, L v represents the performance loss rate of each path;

[0065] According to the performance loss rate of each path, the impact of each path on the fault recovery operation is evaluated and the impact factor is calculated using the formula:

[0066]

[0067] Generate the failure recovery impact factor, where δv is the path reliability coefficient, γ v is the recovery sensitivity parameter, α v is the fault impact factor, τ v is the recovery time, I v represents the failure recovery impact factor, L v represents the performance loss rate of each path;

[0068] The fault recovery impact factor is used to set a priority threshold, sort the recovery priority of each path in turn, select the best priority path for recovery operation, arrange resource allocation and scheduling sequence, and establish a multi-network fault response result.

[0069] Compared with the prior art, the advantages and positive effects of the present invention are:

[0070] In the present invention, the network traffic and delay data are analyzed by implementing the autoregressive moving average algorithm, which allows accurate prediction of network behavior in future time periods, identifies trends and potential fluctuations, and improves the accuracy of predictions and the foresight of operations. The traffic and delay trend results are used to deeply analyze the behavior patterns of data points, and through comparison and clustering refinement, the identification and verification of abnormal states are accelerated, the response speed and processing efficiency of the network are improved, and the rapid response to abnormal states makes the management of data queues more efficient. Through dynamic scheduling, the priority and resource allocation of data processing are optimized to achieve high efficiency of data transmission and stability of network operations. This parallel processing optimization of data flows ensures service continuity and efficiency when network congestion or failure occurs, providing users with a seamless network experience, and optimizing the fault response mechanism. By accurately analyzing the fault point and quickly establishing a recovery strategy, the reliability and overall performance of the network are significantly improved, and the potential service interruption time is reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 is a system flow chart of the present invention;

[0072] Figure 2 It is a flow chart of the flow and delay trend prediction results in the present invention;

[0073] Figure 3 It is a flow chart of the preliminary anomaly detection results in the present invention;

[0074] Figure 4 It is a flow chart of abnormal state verification results in the present invention;

[0075] Figure 5 A flow chart of the priority adjustment result in the present invention;

[0076] Figure 6 It is a flow chart of the scheduling optimization result in the present invention;

[0077] Figure 7 It is a flow chart of the fault point analysis results in the present invention;

[0078] Figure 8 It is a flow chart of the multi-network failure response results in the present invention. DETAILED DESCRIPTION

[0079] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0080] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0081] See also Figure 1 , a multi-network parallel acceleration system comprises:

[0082] The network traffic monitoring module uses the autoregressive moving average algorithm to analyze the data items based on the network traffic and delay data, generates a traffic trend prediction model, and uses the traffic trend prediction model to predict the traffic and delay in the future time period to obtain the traffic and delay trend prediction results;

[0083] The anomaly detection and response module uses the traffic and delay trend prediction results to analyze the behavior patterns of data points, compare the degree of deviation of data points, and generate preliminary anomaly detection results. Based on the preliminary anomaly detection results, the anomaly points are refined through deep clustering methods to obtain abnormal status verification results;

[0084] The data queue management module evaluates the request priority and reorders the queue according to the abnormal data points in the abnormal state verification results, optimizes the data flow through parallel processing, calculates the data flow priority, generates the priority adjustment result, and dynamically schedules the data request based on the priority adjustment result to obtain the scheduling optimization result;

[0085] The fault response operation module monitors performance and potential faults based on the scheduling optimization results, uses machine learning models to analyze fault points, generates fault point analysis results, optimizes fault recovery operations in multi-network parallel acceleration based on the fault point analysis results, and establishes multi-network fault response results.

[0086] The traffic and delay trend prediction results include future traffic patterns, delay pattern changes, and key time nodes. The preliminary anomaly detection results include the preliminary identified abnormal nodes, anomaly intensity, and abnormal time window. The abnormal status verification results include the verified key abnormal nodes, anomaly level, and affected network scope. The priority adjustment results include the reordered request sequence, the optimized priority list, and the priority change log. The scheduling optimization results include the scheduling algorithm efficiency, comparison before and after scheduling, and optimized queue length. The fault point analysis results include the fault point location, fault level, and service scope affected by the fault. The multi-network fault response results include the recovery measures taken, the network status after recovery, and the update of preventive measures.

[0087] See also Figure 2 ,The specific steps for obtaining the traffic and delay trend prediction results are:

[0088] Based on network traffic and delay data, an autoregressive moving average algorithm is used to calculate the autoregressive coefficient and moving average coefficient using the formula:

[0089]

[0090] Get the preliminary trend analysis results, where α a represents the autoregressive coefficient, β a represents the moving average coefficient, X a Represents the previous traffic data, Z a Represents the previous delay data, Y a Represents preliminary trend analysis results;

[0091] The preliminary trend analysis results are refined using the formula:

[0092] Q a =log(γ a ·Y a +1)

[0093] Introducing logarithmic transformation parameters to generate refined trend analysis results, where γ a represents the logarithmic transformation enhancement parameter, Q a Represents the refined trend analysis results, Y a Represents preliminary trend analysis results;

[0094] Combine the refined trend analysis results with real-time network monitoring data and use the weight adjustment coefficient using the formula:

[0095]

[0096] Get the traffic and delay trend prediction results, where θ a represents the trend adjustment coefficient, W a Represents real-time monitoring data, T a represents the traffic and delay trend prediction results, Q a Represents the refined trend analysis results.

[0097] Preliminary trend analysis result formula:

[0098]

[0099] α a : Autoregressive coefficient, which represents the influence of past traffic data on current prediction. This coefficient can be obtained through historical data analysis, such as estimating the parameters of the autoregressive model through the least squares method;

[0100] β a : Moving average coefficient, which represents the influence of previous delayed data on the current prediction, and can be obtained through data fitting;

[0101] X a : Past traffic data, such as the average traffic in the previous hour, is set to 500MB;

[0102] Z a : Past delay data, such as the average delay of the previous hour, is set to 200ms;

[0103] Assume that by fitting the data, we get α a = 0.7 and β a =0.3, using actual data X a =500MB and Z a = 200ms to calculate:

[0104]

[0105] This result Y a =175060 represents the weighted historical impact score calculated by autoregression and moving average, which can be used for further trend analysis.

[0106] The refined trend analysis result formula is:

[0107] Q a =log(γ a ·Y a +1)

[0108] γ a : Logarithmic transformation enhancement parameter, used to adjust Y aThe influence of is assumed to be set to 0.05 through the variability analysis of the data;

[0109] Y a : The result in the previous step, that is, 175060;

[0110] Use Y a =175060 and γ a =0.05 to calculate:

[0111] Q a =log(0.05·175060+1)

[0112] Q a =log(8753.1)=3.942

[0113] This result Q a =3.942 represents the trend analysis score after logarithmic transformation, which further refines the impact analysis results and provides an intermediate step for the final prediction.

[0114] Traffic and delay trend prediction result formula:

[0115]

[0116] Q a : The result obtained from the previous step, which is 3.942;

[0117] θ a : Trend adjustment coefficient, used to weigh the impact of real-time data, is set to 0.1 based on system performance and response time considerations;

[0118] W a : Real-time monitoring data, such as the average traffic in the current hour, set to 520MB;

[0119] Using Q a =3.942,θ a = 0.1 and W a =520 to calculate:

[0120]

[0121] This result T a =61.391 represents the final traffic and delay trend prediction result, which combines the impact of historical trends and real-time data.

[0122] See also Figure 3 ,The specific steps for obtaining the preliminary anomaly detection results are:

[0123] The data point behavior pattern analysis is performed on the traffic and delay trend prediction results, and normalization is performed using the formula:

[0124]

[0125] Generate a normalized behavior model where T a Represents the traffic and delay trend prediction results, ∈ b is a small constant added for stability calculation, U b Represents the normalized behavior pattern;

[0126] Use the normalized behavior pattern to perform outlier analysis, using the formula:

[0127]

[0128] Get the deviation analysis results, where V b is the deviation analysis result, It's U b The average value, U b Represents the normalized behavior pattern;

[0129] The deviation analysis results are compared with the threshold value, and the judgment threshold is set as θ b , by comparing V b Is it greater than θ b , determine whether each data point is an anomaly and generate preliminary anomaly detection results.

[0130] The formula of the normalized behavior model is:

[0131]

[0132] T a : Traffic and delay trend prediction results, assuming a series of data obtained through network monitoring, such as {100, 150, 200, 250, 300};

[0133] min(T a ) and max(T a ): T a The minimum and maximum values ​​in the data set, here are 100 and 300 respectively;

[0134] ∈ b : A small constant, set to 0.01, to avoid the situation where the denominator is zero;

[0135] Consider the specific value T b =150, calculate:

[0136]

[0137] This result U b =0.25 represents the normalized behavior mode value, which reflects the Tb Position relative to the observation range provides a standardized basis for comparison.

[0138] Deviation analysis result formula:

[0139]

[0140] U b : Normalized behavior pattern, the previous step is calculated as 0.25;

[0141] U b The average value of U b The data set is {0.2, 0.25, 0.3}, then

[0142] Calculate V b :

[0143]

[0144] This result V b = 0 means no deviation, that is, in this example, U b is exactly equal to its average value, no anomaly, for different U b The deviation provides an intuitive indication of the degree of abnormality.

[0145] If V b Greater than the set threshold θ b , then it is considered that there is an abnormality, let θ b is 0.1, because V b =0, it does not exceed the threshold and there is no abnormality;

[0146] θ b : Judgment threshold, set based on historical data or expert advice, here 0.1, representing the maximum allowed deviation;

[0147] Since V b Not exceeding θ b , which indicates that the current data point behavior is normal and no anomaly is detected. The result directly reflects the state of the data point relative to the set threshold, providing a direct means of anomaly monitoring.

[0148] See also Figure 4 , the specific steps for obtaining the abnormal status verification results are:

[0149] Receive data from preliminary anomaly detection results and convert them into a format suitable for cluster analysis using the formula:

[0150]

[0151] Generate a weighted distance matrix where xc,i represents the data points obtained from anomaly detection, μ c represents the average value of the data points, n represents the total number of data points, D c represents the weighted distance matrix;

[0152] According to the weighted distance matrix, a deep clustering algorithm is applied to classify data points, using the formula:

[0153]

[0154] Calculate the classification labels of the data points and obtain the group classification results, where G c is the group classification result, w c,i is the weight of the data point, D c represents the weighted distance matrix;

[0155] The group classification results are analyzed to test the consistency of data points in each classification group, and multivariate statistical methods are used to analyze the variability within the group, determine the abnormal state of each group, and generate abnormal state verification results.

[0156] Weighted distance matrix formula:

[0157]

[0158] x c,i : A single data point, obtained from the anomaly detection results. For example, suppose the data point set is {120, 130, 140, 150, 160};

[0159] μ c : The center of the data set, calculated as the average value of the data point set, that is,

[0160] n: the total number of data points, here it is 5;

[0161] Refer to the specific data point x c,1 =120 and x c,2 =130, calculate D c :

[0162]

[0163] This result D c =200 indicates that in a given data set, the average square distance between each data point and the center is 200, which indicates the distribution and concentration of data points. The smaller the D c Values ​​indicate tighter clustering.

[0164] Group classification result formula:

[0165]

[0166] w c,i : The weight of the data point is assigned according to the importance of the data point. For example, the weight of all data points is 1;

[0167] D c : The distance calculated in the previous step, use the distance 200 calculated in the previous step;

[0168] Assume there are 5 data points, all with weights of 1, and the D of the five data points is c Both are 200, calculate G c :

[0169]

[0170] G c =5

[0171] This result G c =5 indicates that the classification score is 5, which means that all data points contribute evenly and significantly to the clustering. -200 Close to 0.

[0172] See also Figure 5 , the specific steps for obtaining the priority adjustment result are:

[0173] According to the abnormal status verification results, a preliminary priority is assigned to each data point according to the abnormal level, using the formula:

[0174]

[0175] The priority of generating data points, where k d is the adjustment factor, S d Indicates the abnormal state of the data point, P d is the priority of the data point;

[0176] Use the priority of the data point to perform the initial reordering of the data queue, and process the priority data through the function, using the formula:

[0177]

[0178] Generate a queue with adjusted priority in descending order, where R d Indicates the queues with descending priority after adjustment, P d is the priority of the data point, and sort means sorting the calculation results;

[0179] The queues arranged in descending order of priority after adjustment are applied, data is optimized through a parallel processing method, processing resources are allocated to each queue entry, and a priority adjustment result is obtained.

[0180] The priority formula for data points is:

[0181]

[0182] S d : The abnormal status score of the data point is obtained by analyzing the historical data monitoring results. The range is usually between 0 and 1. It is used to indicate the abnormality of each data point. The abnormal status score of a data point is set to S d =0.7;

[0183] k d : Adjustment coefficient, used to control the sensitivity of the formula to abnormal state scoring, determined according to the distribution characteristics of historical data, set k d =1.2, this value is used to balance the response intensity to different abnormal states and changes with the fluctuation of abnormal state data.

[0184] Actual example and calculation process:

[0185] Enter specific values ​​for calculation:

[0186]

[0187] First calculate the exponential part:

[0188] -1.2×0.7=-0.84

[0189] Then calculate the natural logarithm part of the exponential:

[0190] e -0.84 ≈0.431

[0191] Substitute the formula to continue calculating:

[0192]

[0193] Calculation result P d ≈0.699 indicates that the priority of the data point is 0.699. The closer the value is to 1, the higher the priority. The point has a relatively high processing priority.

[0194] The adjusted queue formula in descending order of priority is:

[0195]

[0196] P d : The priority of the data point, calculated from the previous step, let P d =0.699;

[0197] First, bring in P d The fractional part inside a value calculation expression:

[0198]

[0199] Compute the squared part:

[0200] 0.699 2 =0.4886

[0201] Substituting the squared result into the formula:

[0202]

[0203] The calculated result 0.672 is used as the sorting basis for the data points. The smaller the value, the higher the priority. Therefore, the queue will be sorted in descending order based on this value.

[0204] Sorting means sorting the data points according to the calculated priority values, specifically from high priority to low priority, to ensure that important data points are arranged according to urgency, so that high priority data points are processed first.

[0205] Parallel processing is performed on the queues arranged in descending order of priority after adjustment, and processing resources are allocated to each entry in the queue. The data points of different priorities are controlled by the scheduler to enter the processing stage to obtain the priority adjustment result. The priority adjustment result establishes the actual order of each data point in the processing queue and clarifies the priority of data processing.

[0206] See also Figure 6 , the specific steps for obtaining the scheduling optimization results are:

[0207] Based on the priority adjustment result, the data request is initially scheduled, the priority result of each data request is received, and the initial scheduling priority of each data request is calculated as a function of the data priority, weight and load status, through the formula:

[0208]

[0209] Generate preliminary scheduling priority, where P h Indicates the priority adjustment result, w h is the weight of each data request, m h is the current load state, γ h is the load regulation parameter, D h Indicates the preliminary scheduling priority;

[0210] Optimize the initial scheduling priority and sort the data requests according to the scheduling score using the formula:

[0211]

[0212] Get the scheduling score, where l h For real-time response delay, b h is the set delay benchmark, η his the scheduling preference parameter, c h is the additional priority factor, S h is the scheduling score, D h Indicates the preliminary scheduling priority;

[0213] Analyze the scheduling priority of multiple data requests based on the scheduling score, set the threshold to judge S h Whether the optimization conditions are met, re-evaluation and resource allocation are performed for data requests that do not meet the conditions, and the scheduling optimization result is obtained after the resource allocation is completed.

[0214] Preliminary scheduling priority formula:

[0215]

[0216] P h : Priority adjustment result, calculated from the previous step, represents the priority of each data request, with a value range between 0 and 1, determined by the urgency of the monitoring data and historical priority data, and the actual value is set to P h =0.75;

[0217] w h : The weight of the data request is determined based on the data request type (such as high priority task, low priority task, etc.) and historical statistical data. The weight value fluctuates between 1 and 5. The high priority task is set to 5 and the low priority task is set to 1. Here, w is set h =3;

[0218] m h : Current load status, measured in real time by monitoring tools, such as CPU usage, memory usage, etc., set a reasonable range of 0 to 100 according to the average load level of the data center, and set m k =60;

[0219] γ h : Load adjustment parameter, used to balance system response speed and resource utilization, determined by analyzing the system's historical load data, ranging from 1 to 10, set according to system sensitivity, set γ h =4;

[0220] Substitute actual numerical calculation formula:

[0221]

[0222] The absolute value and square root part of the inner part:

[0223]

[0224] Substitute the formula to continue calculating:

[0225]

[0226] Calculation result D h =0.28125 indicates that the preliminary scheduling priority is 0.28125. A higher value indicates a higher priority. This result reflects the preliminary priority of the data request under the current system load and request type.

[0227] The formula for scheduling score is:

[0228]

[0229] D h : Preliminary scheduling priority, calculated from the previous step, set to D h =0.28125;

[0230] l h : Actual response delay, obtained through system logs or real-time monitoring tools, ranging from 0 to 1000 milliseconds, set the current value to l h =200;

[0231] b h : The set delay benchmark is determined based on historical performance data and represents the ideal delay value, ranging from 50 to 500 milliseconds, set to b h =150;

[0232] η h : Scheduling preference parameter, reflecting the scheduling preference of a specific data request priority, set by business rules, ranging from 0.1 to 2, and is set as η here h =1.5;

[0233] c h : Additional priority factor, used to increase the priority of specific types of data requests, usually an integer value between 1 and 10, set to c h =4;

[0234] Substitute actual numerical calculation formula:

[0235]

[0236] Calculate the inner absolute value and square root part:

[0237]

[0238] Continue calculating the denominator and fraction:

[0239]

[0240] Calculate the priority factor part:

[0241]

[0242] Substitute the formula to get the scheduling score:

[0243] S h =0.03485+3=3.03485

[0244] Calculation result S h =3.03485 represents the optimized comprehensive scheduling score, which reflects the final priority of the data request under the current conditions. The larger the value, the higher the priority.

[0245] According to the scheduling score S obtained in the previous step h =3.03485, analyze the priority of each data request, and judge S by the set threshold h Whether the optimization conditions are met, re-evaluation and resource allocation are performed for data requests that do not meet the conditions, and the scheduling optimization result is obtained after the resource allocation is completed;

[0246] The final scheduling optimization result determines the order of data requests under the current system state, and reflects the priority order and resource allocation of each data request in the overall scheduling process.

[0247] See also Figure 7 , the specific steps for obtaining the fault point analysis results are:

[0248] According to the scheduling optimization results, monitor the real-time data of multiple performance indicators, and calculate the deviation of each performance indicator after capturing the data, using the formula:

[0249]

[0250] Generate the deviation value of the monitoring indicator, where x g To monitor data values ​​in real time, μ g is the mean of previous data, σ g is the standard deviation of previous data, n g is the sample size, Δ g Indicates the deviation value of the monitoring indicator;

[0251] Call the deviation value of the monitoring indicator, combine the load level and status data, and calculate the failure probability of each monitoring point through the formula:

[0252]

[0253] Generate potential failure probability, where λ g is the fault sensitivity coefficient, ω g is the state influence factor, L g is the load level, F g represents the potential failure probability, Δg Indicates the deviation value of the monitoring indicator;

[0254] Analyze the potential failure probability, compare and classify the fault types through the data of the machine learning model, determine the location of the fault point, set the judgment threshold, determine whether each monitoring point is a fault point, and generate the fault point analysis results.

[0255] Deviation value formula of monitoring indicator:

[0256]

[0257] x g :Real-time monitoring of data values, through real-time monitoring of the system performance data collected by the equipment, set the current performance data point method g =85 units;

[0258] μ g : Historical mean, calculated based on long-term monitoring data. The mean of historical monitoring data is set to μ g =75 units;

[0259] σ g : Historical standard deviation, reflecting the fluctuation of performance data. The standard deviation of historical data is set to σ g =5 units;

[0260] n a : Sample size, that is, the number of historical samples used to calculate the deviation value. The number of samples of historical monitoring data is set to n a =100;

[0261] Substitute actual values ​​for calculation:

[0262]

[0263] Compute the absolute value and the multiplication part:

[0264] 85-75|=10

[0265] 10·5=50

[0266] Substitute the calculated results into the formula:

[0267]

[0268] Calculation result Δ a ≈0.707 indicates the degree of deviation of the monitoring indicator relative to the historical mean. This result is used to determine the existence of abnormal conditions. The higher the value, the greater the deviation.

[0269] Potential failure probability formula:

[0270]

[0271] Δ g : The monitoring indicator deviation value calculated from the previous step is set to Δ a =0.707;

[0272] λ g : Fault sensitivity coefficient, obtained by analyzing historical fault data, indicating the sensitivity of this indicator to faults, with a reasonable range of 1 to 10, which is set as λ here g =2;

[0273] ω g : State impact factor, which reflects the influence of load on the probability of failure. Its value range is 0.1 to 1. It is set according to the relationship between load and failure in historical data and is set to ω g =0.5;

[0274] L a : The current load level is measured in real time by the monitoring equipment in percentage. The load level is set to L n =80; Substitute the actual numerical value into the calculation formula:

[0275]

[0276] Compute the inner multiplication and absolute value part:

[0277] 0.5 80 = 40

[0278] |40|=40

[0279] Let's continue with the square root part:

[0280]

[0281] Substitute the formula to continue calculating:

[0282]

[0283] Calculation result F a =0.221 means that the potential failure probability of the monitoring point is 22.1%. The higher the value, the greater the possibility of failure of the monitoring point.

[0284] Analyze the potential failure probability F obtained in the previous step a =0.221, compare the result with the set threshold, use the historical data of the machine learning model to determine the actual location and type of the fault point, and generate the fault point analysis result;

[0285] The failure point analysis results determine the failure point and its nature, helping the system identify potential performance problems or failure areas and perform corresponding repairs or maintenance.

[0286] See also Figure 8 , the specific steps for obtaining the multi-network fault response results are:

[0287] According to the fault point analysis results, determine the network paths affected by multi-network parallel acceleration, calculate the performance loss of each path in turn, and evaluate the impact on the multi-network environment using the formula:

[0288]

[0289] Get the performance loss rate of each path, where p v is the real-time data transmission rate, q v is the expected transmission rate, σ v is the path weight parameter, κ v is the redundancy coefficient of the path, L v represents the performance loss rate of each path;

[0290] According to the performance loss rate of each path, evaluate the impact of each path on the fault recovery operation and calculate the impact factor using the formula:

[0291]

[0292] Generate the failure recovery impact factor, where δ v is the path reliability coefficient, γ v is the recovery sensitivity parameter, α v is the fault impact factor, τ v is the recovery time, I v represents the failure recovery impact factor, L v represents the performance loss rate of each path;

[0293] Use the fault recovery impact factor, set the priority threshold, sort the recovery priority of each path in turn, select the best priority path for recovery operation, arrange resource allocation and scheduling sequence, and establish multi-network fault response results.

[0294] The performance loss rate formula for each path is:

[0295]

[0296] p v : Actual data transmission rate, the data transmission rate monitored in real time using network monitoring tools, in megabits per second, the actual value is set to p v =75;

[0297] q v : Expected transmission rate, the ideal data transmission rate set based on historical network performance data, in megabits per second, set to q v =100;

[0298] σ v : Path weight parameter, which is set based on the importance analysis of network topology and historical transmission data. Its value range is between 0 and 1. It is used to reflect the importance of the path in the network and is set to σ v =0.8;

[0299] κ v : Path redundancy coefficient, calculated by the redundancy assessment tool, represents the redundancy of each path in a multi-network environment, set to κ v =1.2;

[0300] Substitute actual values ​​for calculation:

[0301]

[0302] Calculate the absolute value part:

[0303] |75-100|=25

[0304] Calculate the square root part:

[0305]

[0306] Substitute the formula to continue calculating:

[0307]

[0308] Calculation result L v ≈20.41 indicates that the performance loss rate of the current network path is 20.41%. This result is used to analyze the impact of each path in multi-network parallel acceleration. The larger the value, the more serious the path performance loss.

[0309] Fault recovery impact factor formula:

[0310]

[0311] L v : The path performance loss rate calculated from the previous step is set to L v =20.41;

[0312] δ v : Path reliability coefficient, which is set by analyzing the stability of the path through historical data, with a value range of 0.5 to 1.5, and is set to δ v =1.2;

[0313] γ v : Recovery sensitivity parameter, used to adjust the flexibility of the recovery operation, ranging from 0 to 5, set according to the network recovery history data, set to γ v =3;

[0314] α v : Fault impact factor, determined according to the impact degree in the fault analysis results, ranging from 1 to 10, set to α v =5;

[0315] τ v : Recovery time, measured by the actual monitored fault recovery time, in seconds, set to τ v =30;

[0316] Substitute actual numerical calculation formula:

[0317]

[0318] Compute the inner multiplication and absolute value part:

[0319] 5.30=150

[0320] |150|=150

[0321] Let's continue with the square root part:

[0322]

[0323] Substitute the formula to continue calculating:

[0324]

[0325] Calculation results I v ≈1.98 represents the impact factor of each path on fault recovery. A larger value indicates that the path has a more significant impact on the fault recovery operation.

[0326] Use the impact factor I obtained in the previous step n =1.98, set the priority threshold of path recovery, determine the recovery order according to the impact factor, arrange the operation order of resource allocation and fault recovery, and establish the multi-network fault response result;

[0327] The multi-network fault response results reflect the recovery priority and resource allocation strategy of each path in the multi-network parallel acceleration environment, and provide an optimized fault response sequence based on the calculation results in the previous steps.

[0328] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A multi-network parallel acceleration system, It is characterized in that The system comprises: The network traffic monitoring module uses the autoregressive moving average algorithm to analyze the data items based on the network traffic and delay data, generates a traffic trend prediction model, and predicts the traffic and delay in the future time period through the traffic trend prediction model to obtain the traffic and delay trend prediction results; The anomaly detection and response module uses the traffic and delay trend prediction results to analyze the behavior patterns of data points, compares the degree of deviation of data points, generates preliminary anomaly detection results, and refines the anomaly points through a deep clustering method based on the preliminary anomaly detection results to obtain an abnormal state verification result; The data queue management module evaluates the request priority according to the abnormal data points in the abnormal state verification result, performs queue reordering, optimizes the data flow through parallel processing, calculates the data flow priority, generates a priority adjustment result, and dynamically schedules the data request based on the priority adjustment result to obtain a scheduling optimization result; The fault response operation module monitors performance and potential faults based on the scheduling optimization results, analyzes fault points using a machine learning model, generates fault point analysis results, optimizes fault recovery operations in multi-network parallel acceleration based on the fault point analysis results, and establishes multi-network fault response results.

2. The multi-network parallel acceleration system according to claim 1, It is characterized in that The steps for obtaining the traffic and delay trend prediction results are specifically as follows: Based on network traffic and delay data, an autoregressive moving average algorithm is used to calculate the autoregressive coefficient and moving average coefficient using the formula: Get the preliminary trend analysis results, where α a represents the autoregressive coefficient, β a represents the moving average coefficient, X a Represents the previous traffic data, Z a Represents the previous delay data, Y a Represents preliminary trend analysis results; The preliminary trend analysis results are refined using the formula: Q a =log(γ a ·Y a +1) Introducing logarithmic transformation parameters to generate refined trend analysis results, where γ a represents the logarithmic transformation enhancement parameter, Q a Represents the refined trend analysis results, Y a Represents preliminary trend analysis results; The refined trend analysis results are combined with real-time network monitoring data, using the weight adjustment coefficient, using the formula: Get the traffic and delay trend prediction results, where θ a represents the trend adjustment coefficient, W a Represents real-time monitoring data, T a represents the traffic and delay trend prediction results, Q a Represents the refined trend analysis results.

3. The multi-network parallel acceleration system according to claim 2, It is characterized in that The steps for obtaining the preliminary anomaly detection results are specifically as follows: The data point behavior pattern analysis is performed on the traffic and delay trend prediction results, and normalization is performed, through the formula: Generate a normalized behavior model where T a Represents the traffic and delay trend prediction results, ∈ b is a small constant added for stability calculation, U b Represents the normalized behavior pattern; The normalized behavior model is used to perform outlier analysis, using the formula: Get the deviation analysis results, where V b is the deviation analysis result, It's U b The average value, U b Represents the normalized behavior pattern; The deviation analysis result is compared with the threshold value, and the judgment threshold value is set as θ b , by comparing V b Is it greater than θ b , determine whether each data point is an anomaly and generate preliminary anomaly detection results.

4. The multi-network parallel acceleration system according to claim 3, It is characterized in that The steps for obtaining the abnormal state verification result are specifically as follows: Receive data from the preliminary anomaly detection results and convert them into a format that matches the cluster analysis, using the formula: Generate a weighted distance matrix where x c,i represents the data points obtained from anomaly detection, μ c represents the average value of the data points, n represents the total number of data points, D c represents the weighted distance matrix; According to the weighted distance matrix, a deep clustering algorithm is applied to classify data points, using the formula: Calculate the classification labels of the data points and obtain the group classification results, where G c is the group classification result, w c,i is the weight of the data point, D c represents the weighted distance matrix; The group classification results are analyzed to check the consistency of the data points in each classification group, and the variability within the group is analyzed using multivariate statistical methods to determine the abnormal state of each group and generate abnormal state verification results.

5. The multi-network parallel acceleration system according to claim 4, It is characterized in that The steps for obtaining the priority adjustment result are specifically as follows: According to the abnormal status verification results, a preliminary priority is assigned to each data point according to the abnormal level, using the formula: The priority of generating data points, where k d is the adjustment factor, S d Indicates the abnormal state of the data point, P d is the priority of the data point; Using the priority of the data point, the data queue is initially reordered, and the priority data is processed by the function, using the formula: Generate a queue with adjusted priority in descending order, where R d Indicates the queues with adjusted priorities in descending order, P d is the priority of the data point, and sort means sorting the calculation results; The queues arranged in descending order of priority after adjustment are applied, data is optimized by a parallel processing method, and processing resources are allocated to each queue entry to obtain a priority adjustment result.

6. The multi-network parallel acceleration system according to claim 5, It is characterized in that The steps for obtaining the scheduling optimization result are specifically as follows: Based on the priority adjustment result, the data request is initially scheduled, the priority result of each data request is received, and the initial scheduling priority of each data request is calculated as a function of the data priority, weight and load status, through the formula: Generate preliminary scheduling priority, where P h Indicates the priority adjustment result, w h is the weight of each data request, m h is the current load state, γ h is the load regulation parameter, D h Indicates the preliminary scheduling priority; The preliminary scheduling priority is optimized, and the data requests are sorted according to the scheduling score using the formula: Get the scheduling score, where l h For real-time response delay, b h is the set delay benchmark, η h is the scheduling preference parameter, c h is the additional priority factor, S h is the scheduling score, D h Indicates the preliminary scheduling priority; According to the scheduling score, the scheduling priority of multiple data requests is analyzed, and a threshold is set to determine S h Whether the optimization conditions are met, re-evaluation and resource allocation are performed for data requests that do not meet the conditions, and the scheduling optimization result is obtained after the resource allocation is completed.

7. The multi-network parallel acceleration system according to claim 6, It is characterized in that The steps for obtaining the fault point analysis result are specifically as follows: According to the scheduling optimization results, the real-time data of multiple performance indicators are monitored, and the deviation of each performance indicator is calculated after capturing the data, using the formula: Generate the deviation value of the monitoring indicator, where x g To monitor data values ​​in real time, μ g is the mean of previous data, σ g is the standard deviation of previous data, n g is the sample size, Δ g Indicates the deviation value of the monitoring indicator; The deviation value of the monitoring indicator is called, combined with the load level and status data, to calculate the failure probability of each monitoring point, through the formula: Generate potential failure probability, where λ g is the fault sensitivity coefficient, ω g is the state influence factor, L g is the load level, F g represents the potential failure probability, Δ g Indicates the deviation value of the monitoring indicator; Analyze the potential failure probability, compare and classify the failure type through the data of the machine learning model, determine the location of the failure point, set the judgment threshold, determine whether each monitoring point is a failure point, and generate the failure point analysis result.

8. The multi-network parallel acceleration system according to claim 7, It is characterized in that The steps for obtaining the multi-network fault response result are specifically as follows: According to the fault point analysis results, determine the network paths affected in multi-network parallel acceleration, calculate the performance loss of each path in turn, and evaluate the impact on the multi-network environment using the formula: Get the performance loss rate of each path, where p v is the real-time data transmission rate, q v is the expected transmission rate, σ v is the path weight parameter, κ v is the redundancy coefficient of the path, L v represents the performance loss rate of each path; According to the performance loss rate of each path, the impact of each path on the fault recovery operation is evaluated and the impact factor is calculated using the formula: Generate the failure recovery impact factor, where δ v is the path reliability coefficient, γ v is the recovery sensitivity parameter, α v is the fault impact factor, τ v is the recovery time, I v represents the failure recovery impact factor, L v represents the performance loss rate of each path; The fault recovery impact factor is used to set a priority threshold, sort the recovery priority of each path in turn, select the best priority path for recovery operation, arrange resource allocation and scheduling sequence, and establish a multi-network fault response result.

Citation Information

Patent Citations

  • Parallel network traffic classification method based on ontology knowledge inference

    CN105516020A

  • Inference acceleration method and system based on CPU multi-core parallel edge depth model

    CN117875384A

  • Dynamic queue scheduling method and system

    CN118331708A

  • Operation and maintenance system of automatic industrial control system

    CN118348872A

  • Cryogenic hydrogen storage tank fault diagnosis system based on machine learning

    CN118551330A

Cited By

  • Cross-cloud resource migration method and system

    CN119440855A

  • A Cross-Cloud Resource Migration Method and System

    CN119440855B