Network capacity and performance bottleneck prediction and analysis system, method, apparatus, equipment and storage medium

By using multi-level feature extraction and multi-task deep neural network models, the problem of low prediction accuracy in traditional network operation and maintenance modes has been solved, enabling accurate prediction of resource capacity and early warning of performance bottlenecks, thereby improving the level of intelligence in network operation and maintenance.

CN123123189APending Publication Date: 2026-11-17SHUOHUANG RAILWAY DEV +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611346295.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-01
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Traditional network operation and maintenance models cannot simultaneously capture the multi-scale patterns of network operation, the complex causal dependencies between devices, and the dynamic topology evolution characteristics, resulting in poor generalization ability and low prediction accuracy of prediction models, making it difficult to adapt to complex and ever-changing industrial network scenarios.

Method used

A multi-level feature extraction module is used to process multi-source heterogeneous data, generate multi-level fused features, and make predictions through a multi-task deep neural network model. It integrates IP address lifecycle management to achieve accurate prediction of resource capacity, early warning of performance bottlenecks, and assessment of network-wide impact.

Benefits of technology

It improves the generalization ability and accuracy of the prediction model, enables accurate prediction of resource capacity, early warning of performance bottlenecks, reduces the risk of network changes, and enhances the intelligence level of network operation and maintenance and the ability to ensure business operations.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123123189A_ABST
    Figure CN123123189A_ABST
Patent Text Reader

Abstract

This application relates to a network capacity and performance bottleneck prediction and analysis system, method, apparatus, computer equipment, readable storage medium, and program product, belonging to the field of computer network technology. This application can improve the generalization ability and prediction accuracy of the prediction model. The system includes: a data acquisition module for collecting multi-source heterogeneous data; a multi-source heterogeneous data preprocessing module for cleaning, normalizing, temporally aligning, and fusing the multi-source heterogeneous data to generate a standardized multimodal dataset; a multi-level feature extraction module for performing multi-level joint feature extraction on the standardized multimodal dataset to generate multi-level fused features; and a capacity and bottleneck prediction module for inputting real-time network data into a multi-task deep neural network model, processing it through a capacity prediction branch, a performance bottleneck prediction branch, and a network-wide impact assessment branch, and outputting equipment resource capacity consumption prediction results, performance bottleneck prediction results, and a network-wide service impact assessment report.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of computer network technology, and in particular to a network capacity and performance bottleneck prediction and analysis system, method, apparatus, computer equipment, computer-readable storage medium, and computer program product. Background Technology

[0002] With the deep integration of technologies such as cloud computing, big data, and artificial intelligence into industries such as energy, railways, and manufacturing, the scale and complexity of industrial data center networks are growing exponentially, and the core services carried by the networks have extremely high requirements for reliability and continuity.

[0003] In traditional network operation and maintenance models, only basic statistical features and simple time-series correlation features at a single time scale can be extracted. They cannot simultaneously capture the multi-scale patterns of network operation, the complex causal dependencies between devices, and the dynamic topology evolution characteristics. This results in poor generalization ability and low prediction accuracy of the prediction model, making it difficult to adapt to complex and ever-changing industrial network scenarios. Summary of the Invention

[0004] Based on this, it is necessary to provide a network capacity and performance bottleneck prediction and analysis system, method, apparatus, computer equipment, computer-readable storage medium, and computer program product to address the above-mentioned technical problems.

[0005] In a first aspect, this application provides a network capacity and performance bottleneck prediction and analysis system, the system comprising: a data acquisition module, a multi-source heterogeneous data preprocessing module, a multi-level feature extraction module, a multi-task prediction model training module, a capacity and bottleneck prediction module, an IP (Internet Protocol) address full lifecycle management integration module, and a visual intelligent alarm module;

[0006] The data acquisition module is used to collect multi-source heterogeneous data, including historical network operation data, dynamic network topology data, IP address lifecycle management data, and service traffic characteristic data.

[0007] The multi-source heterogeneous data preprocessing module is used to clean, normalize, time-series align, and fuse heterogeneous data to generate a standardized multimodal dataset.

[0008] The multi-level feature extraction module is used to perform multi-level joint feature extraction on the standardized multimodal dataset to generate multi-level fused features;

[0009] The multi-task prediction model training module is used to train a multi-task deep neural network model based on the multi-level fusion features; the multi-task deep neural network model includes a capacity prediction branch, a performance bottleneck prediction branch, and a network-wide impact assessment branch.

[0010] The capacity and bottleneck prediction module is used to input real-time network data into the multi-task deep neural network model, process it through the capacity prediction branch, the performance bottleneck prediction branch and the network-wide impact assessment branch respectively, and output the device resource capacity consumption prediction result, the performance bottleneck prediction result and the network-wide service impact assessment report.

[0011] The IP address lifecycle management integration module is used to uniformly manage the IP address resources of the entire network and synchronize the allocation, use and change data of IP addresses to the capacity and bottleneck prediction module in real time.

[0012] The visualization intelligent alarm module is used to visualize various prediction results and output graded alarms and optimization suggestions when network risks are predicted.

[0013] In one embodiment, the multi-task deep neural network model adopts an architecture with a fusion gating mechanism; the multi-task deep neural network model further includes a shared feature layer, which contains two fully connected layers for extracting general features across tasks; the capacity prediction branch is a regression task branch, used to output the resource consumption of each network device in the future for a preset period of time; the performance bottleneck prediction branch is a classification task branch, used to output the bottleneck level of each device and link, the bottleneck level including no risk, low risk, medium risk, and high risk; the network-wide impact assessment branch is a regression task branch, used to output the degree of impact of service changes on the performance of network-wide services.

[0014] Secondly, this application also provides a network capacity and performance bottleneck prediction and analysis method, applied to the network capacity and performance bottleneck prediction and analysis system as described in the above embodiments, the method comprising:

[0015] Collect multi-source heterogeneous data, including historical network operation data, dynamic network topology data, IP address lifecycle management data, and service traffic characteristic data; clean, normalize, time-series align, and fuse the multi-source heterogeneous data to generate a standardized multimodal dataset;

[0016] Multi-level joint feature extraction is performed on the standardized multimodal dataset to generate multi-level fused features; a multi-task deep neural network model is trained based on the multi-level fused features; the multi-task deep neural network model includes a capacity prediction branch, a performance bottleneck prediction branch, and a network-wide impact assessment branch;

[0017] Real-time network data is input into the multi-task deep neural network model, and processed by the capacity prediction branch, the performance bottleneck prediction branch and the network-wide impact assessment branch respectively, outputting device resource capacity consumption prediction results, performance bottleneck prediction results and network-wide service impact assessment report;

[0018] It visualizes various prediction results and outputs tiered alerts and optimization suggestions when network risks are predicted.

[0019] In one embodiment, the multi-level joint feature extraction of the standardized multimodal dataset includes:

[0020] Multi-scale dynamic statistics and anomaly feature extraction are performed on the standardized multimodal dataset to output first-level multi-scale statistical features; based on the first-level multi-scale statistical features, multi-device temporal causal correlation and attention feature extraction are performed to output second-level device causal correlation features; based on the second-level device causal correlation features, dynamic spatiotemporal graph attention and business path weighted feature extraction are performed to output third-level dynamic spatiotemporal topology and business path weighted features.

[0021] In one embodiment, the step of performing multi-scale dynamic statistics and anomaly feature extraction on the standardized multimodal dataset to output first-level multi-scale statistical features includes:

[0022] The standardized multimodal dataset is divided into three time-scale subsets: fine-grained, medium-grained, and coarse-grained. For each time-scale subset, the basic statistical features of each network metric are calculated, and an improved isolated forest algorithm with time weights is used to identify abnormal features. The basic statistical features, the basic statistical features, and the abnormal features of the three time scales are concatenated to obtain the first-level multi-scale statistical features.

[0023] In one embodiment, the step of extracting multi-device temporal causal associations and attention features based on the first-level multi-scale statistical features, and outputting second-level device causal association features, includes:

[0024] The first-level multi-scale statistical features are grouped by device dimension to generate a temporal feature sequence for each device. The temporal feature sequence is encoded by a multi-scale temporal Transformer network to extract multi-scale temporal dependency features within each device. Granger causality test is used to analyze the temporal causal relationship between devices, and a device causal relationship matrix is ​​constructed. The device causal relationship matrix is ​​used as an attention mask, and the temporal feature sequence is fused through a causal attention mechanism to extract causal association features between devices. The multi-scale temporal dependency features and the causal association features are concatenated to obtain the second-level device causal association features.

[0025] In one embodiment, the step of extracting dynamic spatiotemporal graph attention and service path weighted features based on the second-level device causal correlation features, and outputting third-level dynamic spatiotemporal topology and service path weighted features, includes:

[0026] Based on the causal relationship features of the second-level devices, the spatial attention head of the spatiotemporal graph attention network is used to aggregate neighbor features in combination with the dynamic graph and causal relationship matrix. Multi-time-step spatial topological features are extracted and fused with the spatial topological features through the temporal attention head to extract spatiotemporal evolution features. The weights of each active service path are calculated based on service importance and traffic, and the path node features are aggregated to generate service path weighted features. The spatiotemporal evolution features and the service path weighted features are concatenated to obtain the third-level dynamic spatiotemporal topology and service path weighted features.

[0027] Thirdly, this application also provides a network capacity and performance bottleneck prediction and analysis device, applied to the network capacity and performance bottleneck prediction and analysis system as described in the above embodiments, the device comprising:

[0028] The data acquisition unit is used to collect multi-source heterogeneous data, including historical network operation data, dynamic network topology data, IP address lifecycle management data, and service traffic characteristic data; and to clean, normalize, time-series align, and fuse the multi-source heterogeneous data to generate a standardized multimodal dataset.

[0029] The feature extraction unit is used to perform multi-level joint feature extraction on the standardized multimodal dataset to generate multi-level fused features; and to train a multi-task deep neural network model based on the multi-level fused features; the multi-task deep neural network model includes a capacity prediction branch, a performance bottleneck prediction branch, and a network-wide impact assessment branch.

[0030] The result prediction unit is used to input real-time network data into the multi-task deep neural network model, process it through the capacity prediction branch, the performance bottleneck prediction branch and the network-wide impact assessment branch respectively, and output the device resource capacity consumption prediction result, the performance bottleneck prediction result and the network-wide service impact assessment report.

[0031] The results display unit is used to visualize various prediction results and output graded alarms and optimization suggestions when network risks are predicted.

[0032] Fourthly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0033] The system collects multi-source heterogeneous data, including historical network operation data, dynamic network topology data, IP address lifecycle management data, and service traffic characteristic data. It then cleans, normalizes, aligns, and fuses this heterogeneous data to generate a standardized multimodal dataset. Multi-level joint feature extraction is performed on the standardized multimodal dataset to generate multi-level fused features. A multi-task deep neural network model is trained based on these multi-level fused features. This model includes a capacity prediction branch, a performance bottleneck prediction branch, and a network-wide impact assessment branch. Real-time network data is input into the model and processed by these branches, outputting device resource capacity consumption prediction results, performance bottleneck prediction results, and a network-wide service impact assessment report. The prediction results are visualized, and tiered alarms and optimization suggestions are output when network risks are predicted.

[0034] Fifthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0035] The system collects multi-source heterogeneous data, including historical network operation data, dynamic network topology data, IP address lifecycle management data, and service traffic characteristic data. It then cleans, normalizes, aligns, and fuses this heterogeneous data to generate a standardized multimodal dataset. Multi-level joint feature extraction is performed on the standardized multimodal dataset to generate multi-level fused features. A multi-task deep neural network model is trained based on these multi-level fused features. This model includes a capacity prediction branch, a performance bottleneck prediction branch, and a network-wide impact assessment branch. Real-time network data is input into the model and processed by these branches, outputting device resource capacity consumption prediction results, performance bottleneck prediction results, and a network-wide service impact assessment report. The prediction results are visualized, and tiered alarms and optimization suggestions are output when network risks are predicted.

[0036] Sixthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0037] The system collects multi-source heterogeneous data, including historical network operation data, dynamic network topology data, IP address lifecycle management data, and service traffic characteristic data. It then cleans, normalizes, aligns, and fuses this heterogeneous data to generate a standardized multimodal dataset. Multi-level joint feature extraction is performed on the standardized multimodal dataset to generate multi-level fused features. A multi-task deep neural network model is trained based on these multi-level fused features. This model includes a capacity prediction branch, a performance bottleneck prediction branch, and a network-wide impact assessment branch. Real-time network data is input into the model and processed by these branches, outputting device resource capacity consumption prediction results, performance bottleneck prediction results, and a network-wide service impact assessment report. The prediction results are visualized, and tiered alarms and optimization suggestions are output when network risks are predicted.

[0038] The aforementioned network capacity and performance bottleneck prediction and analysis system, method, device, computer equipment, computer-readable storage medium, and computer program product first collect multi-source heterogeneous data, including historical network operation data, dynamic network topology data, IP address lifecycle management data, and service traffic characteristic data, through a data acquisition module. Then, a multi-source heterogeneous data preprocessing module cleans, normalizes, aligns, and fuses the heterogeneous data to generate a standardized multimodal dataset. Next, a multi-level feature extraction module performs multi-level joint feature extraction on the standardized multimodal dataset to generate multi-level fused features. Finally, a capacity and bottleneck prediction module inputs real-time network data into a trained multi-task deep neural network model, which processes the data through capacity prediction, performance bottleneck prediction, and overall network impact assessment branches, outputting device resource capacity consumption prediction results, performance bottleneck prediction results, and an overall network service impact assessment report. This application achieves accurate prediction of resource capacity, early warning of performance bottlenecks, quantitative assessment of change risks, and chain reaction analysis of network-wide impact through a multi-level joint feature extraction system and a multi-task deep neural network model. At the same time, it integrates IP address lifecycle management to form a closed-loop operation and maintenance system of "prediction-alarm-handling-optimization", thereby improving the generalization ability and prediction accuracy of the prediction model, and enhancing the intelligence level and business assurance capability of network operation and maintenance, which can adapt to complex and ever-changing industrial network scenarios. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a block diagram of a network capacity and performance bottleneck prediction and analysis system in one embodiment.

[0041] Figure 2 This is a flowchart illustrating a network capacity and performance bottleneck prediction and analysis method in one embodiment;

[0042] Figure 3 This is a flowchart illustrating the multi-level joint feature extraction steps in one embodiment;

[0043] Figure 4 This is a structural block diagram of a network capacity and performance bottleneck prediction and analysis device in one embodiment;

[0044] Figure 5 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0046] In one embodiment, such as Figure 1 As shown, a network capacity and performance bottleneck prediction and analysis system 100 is provided. The system includes: a data acquisition module 101, a multi-source heterogeneous data preprocessing module 102, a multi-level feature extraction module 103, a multi-task prediction model training module 104, a capacity and bottleneck prediction module 105, an IP address full lifecycle management integration module 106, and a visual intelligent alarm module 107.

[0047] Data acquisition module 101 is used to collect multi-source heterogeneous data, including historical network operation data, dynamic network topology data, IP address lifecycle management data, and service traffic characteristic data.

[0048] The multi-source heterogeneous data preprocessing module 102 is used to clean, normalize, time-series align, and fuse heterogeneous data from multiple sources to generate a standardized multimodal dataset.

[0049] The multi-level feature extraction module 103 is used to perform multi-level joint feature extraction on the standardized multimodal dataset to generate multi-level fused features;

[0050] The multi-task prediction model training module 104 is used to train a multi-task deep neural network model based on multi-level fusion features. The multi-task deep neural network model includes a capacity prediction branch, a performance bottleneck prediction branch, and a network-wide impact assessment branch.

[0051] The capacity and bottleneck prediction module 105 is used to input real-time network data into a multi-task deep neural network model, and process it through the capacity prediction branch, the performance bottleneck prediction branch and the network-wide impact assessment branch respectively, and output the equipment resource capacity consumption prediction results, the performance bottleneck prediction results and the network-wide service impact assessment report.

[0052] The IP address lifecycle management integration module 106 is used to uniformly manage the IP address resources of the entire network and synchronize the allocation, use and change data of IP addresses to the capacity and bottleneck prediction module in real time.

[0053] The Visualized Intelligent Alarm Module 107 is used to visualize various prediction results and output graded alarms and optimization suggestions when network risks are predicted.

[0054] Historical network operation data may include CPU utilization, memory utilization, number of VSIs (Virtual Switch Instances), number of VRFs (Virtual Routing Forwarding), number of routing table entries, number of AC (Access Control) rules, port traffic, packet loss rate, latency, etc.; dynamic network topology data may include device connection relationships, link bandwidth, link status, interface status, etc.; IP address lifecycle data may include address allocation, usage records, change history, violation records, etc.; service traffic characteristic data may include service type, service path, traffic size, traffic direction, service importance level, etc.

[0055] Specifically, refer to Figure 1In response to the received prediction and analysis command, the network capacity and performance bottleneck prediction and analysis system 100 collects data including historical network operation data, dynamic network topology data, and IP addresses via the data acquisition module 101. Multi-source heterogeneous data including address lifecycle management data and business traffic characteristic data; the total collected data volume is approximately 15.6TB, including 8.2TB of raw log data and 7.4TB of time-series data; the multi-source heterogeneous data preprocessing module 102 cleans, normalizes, aligns the time series data, and fuses the heterogeneous data to generate a standardized multimodal dataset; the multi-level feature extraction module 103 performs multi-level joint feature extraction on the standardized multimodal dataset, generating first-level features, second-level features, and third-level features in sequence; the multi-task prediction model training module 104 trains a multi-task deep neural network model that simultaneously supports capacity prediction, bottleneck identification, and network-wide impact assessment based on the multi-level fused features; the capacity and bottleneck prediction module 105 inputs the real-time collected network data into the trained multi-task deep neural network model, processes it through capacity prediction branch, performance bottleneck prediction branch, and network-wide impact assessment branch respectively, and outputs device resource capacity consumption prediction results, performance bottleneck prediction results, and network-wide business impact assessment report; the IP address lifecycle management integration module 106 manages all network IPs in a unified manner. Address resources are allocated, and the allocation, usage, and change data of IP addresses are synchronized to the capacity and bottleneck prediction module in real time. The visualization and intelligent alarm module 107 visualizes and displays various prediction results, and outputs graded alarms and provides optimization and handling suggestions when network risks are predicted.

[0056] Furthermore, the capacity and bottleneck prediction module includes: a resource capacity prediction unit, a performance bottleneck prediction unit, and a network-wide service impact assessment unit;

[0057] Resource capacity prediction unit: used to predict the consumption of resources such as VSI, VRF, routing table entries, AC rules, CPU, and memory of each device in the next 7 days, and generate a capacity trend report;

[0058] Performance bottleneck prediction unit: used to predict the performance indicators of each link and device in the next 24 hours, and identify the location, severity and expected time of potential performance bottlenecks;

[0059] Network-wide Business Impact Assessment Unit: Used to assess the scope and extent of the impact of new services or network changes on all services across the network, based on IP address mapping and business path analysis.

[0060] Furthermore, the IP address lifecycle management integration module provides functions such as IP address planning, automatic discovery, usage record tracking, violation alarms, historical backtracking, and IPv6 transition support, and synchronizes IP address allocation data, usage data, and change data to the data acquisition module in real time.

[0061] Furthermore, the visual intelligent alarm module provides topology visualization, trend visualization, comparison visualization, and alarm visualization functions, supports a four-level alarm mechanism, and issues alarms via SMS, email, system pop-ups, and automatic dispatching through the work order system.

[0062] The beneficial effects of the above embodiments are as follows:

[0063] (1) A three-level feature extraction system of “multi-scale statistical features - device causal relationship features - dynamic spatiotemporal topology features” is constructed to capture the multi-scale laws of network operation, complex causal dependencies between devices and topological evolution characteristics. This solves the defects of traditional technology in single scale and shallow relationship, and improves the generalization ability and prediction accuracy of prediction model.

[0064] (2) It can accurately predict the consumption trend of all dimensions of device resources such as VSI, VRF, routing table entries, CPU, and memory in the next 7 days, generate a quantitative capacity report, improve resource utilization, and avoid waste caused by over-configuration of resources and business interruption caused by insufficient resources.

[0065] (3) Identify the location, severity and expected occurrence time of potential performance bottlenecks 24 hours in advance, reduce the average fault recovery time from several hours to less than 15 minutes, and realize the transformation of operation and maintenance mode from passive response to proactive prevention.

[0066] (4) Reduce the time spent on network change risk assessment, improve the accuracy of assessment, significantly reduce network accidents caused by human configuration errors, and reduce the occurrence rate of change-related failures.

[0067] (5) Based on IP address mapping and business path analysis, the impact of new services or network changes on all services across the network can be comprehensively assessed, effectively preventing cascading failures and ensuring the continuity of critical services.

[0068] (6) Integrate IP address lifecycle management functions to achieve real-time synchronization of address resource data and predictive analysis data; provide multi-dimensional visualization and four-level intelligent alarms, support multi-channel automatic notification and work order dispatch, and form a closed-loop operation and maintenance system of "prediction-alarm-handling-optimization".

[0069] In one embodiment, the multi-task deep neural network model adopts an architecture with a fusion gating mechanism. The multi-task deep neural network model also includes a shared feature layer, which contains two fully connected layers for extracting general features across tasks. The capacity prediction branch is a regression task branch, used to output the resource consumption of each network device in the future for a preset period of time. The performance bottleneck prediction branch is a classification task branch, used to output the bottleneck level of each device and link, including no risk, low risk, medium risk, and high risk. The network-wide impact assessment branch is a regression task branch, used to output the degree of impact of service changes on the performance of network-wide services.

[0070] Specifically, the multi-task prediction model training module 104 adopts a multi-task deep neural network model with a fusion gating mechanism, inputting three levels of fused features into the model. The model includes a shared feature layer and three task-specific branches:

[0071] Shared feature layer: Contains two fully connected layers for extracting common features across tasks;

[0072] Capacity prediction branch: regression task, outputs the resource consumption of each device for the next 7 days;

[0073] Performance bottleneck prediction branch: Classify tasks and output the bottleneck level of each device and link; among which, the bottleneck level includes: no risk, low risk, medium risk and high risk;

[0074] The network-wide impact assessment branch is a regression task that outputs the degree of impact of business changes on the overall network business performance.

[0075] An adaptive weighted loss function is used during training to dynamically adjust the loss weights for the three tasks.

[0076] In one embodiment, such as Figure 2 As shown, a method for predicting and analyzing network capacity and performance bottlenecks is provided. Taking the application of this method to the network capacity and performance bottleneck prediction and analysis system 100 as described in the above embodiment as an example, the method includes the following steps:

[0077] Step S201: Collect multi-source heterogeneous data, including historical network operation data, dynamic network topology data, IP address lifecycle management data, and service traffic characteristic data; clean, normalize, time-series align, and fuse the multi-source heterogeneous data to generate a standardized multimodal dataset.

[0078] Step S202: Perform multi-level joint feature extraction on the standardized multimodal dataset to generate multi-level fused features; train a multi-task deep neural network model based on the multi-level fused features.

[0079] Step S203: Input real-time network data into the multi-task deep neural network model, process it through the capacity prediction branch, performance bottleneck prediction branch and network-wide impact assessment branch respectively, and output the device resource capacity consumption prediction results, performance bottleneck prediction results and network-wide service impact assessment report.

[0080] Step S204: Visualize various prediction results and output graded alarms and optimization suggestions when network risks are predicted.

[0081] Specifically, the network capacity and performance bottleneck prediction and analysis system 100 first collects multi-source heterogeneous data, including historical network operation data, dynamic network topology data, IP address lifecycle management data, and service traffic characteristic data. It then cleans, normalizes, aligns the time series of the heterogeneous data, and fuses the heterogeneous data to generate a standardized multimodal dataset. Next, it performs multi-level joint feature extraction on the standardized multimodal dataset to generate multi-level fused features, and trains a multi-task deep neural network model based on these features. Real-time network data is then input into the multi-task deep neural network model, which processes the data through capacity prediction, performance bottleneck prediction, and network-wide impact assessment branches, respectively. The system outputs device resource capacity consumption prediction results, performance bottleneck prediction results, and a network-wide service impact assessment report. Simultaneously, it visualizes various prediction results and outputs tiered alarms and optimization suggestions when network risks are predicted.

[0082] The aforementioned network capacity and performance bottleneck prediction and analysis method achieves accurate resource capacity prediction, early warning of performance bottlenecks, quantitative assessment of change risks, and chain reaction analysis of network-wide impacts through a multi-level joint feature extraction system and a multi-task deep neural network model. At the same time, it integrates IP address lifecycle management to form a closed-loop operation and maintenance system of "prediction-alarm-handling-optimization", thereby improving the generalization ability and prediction accuracy of the prediction model, enhancing the intelligence level of network operation and maintenance and business assurance capabilities, and adapting to complex and ever-changing industrial network scenarios.

[0083] In one embodiment, such as Figure 3 As shown, step S202 above, which involves multi-level joint feature extraction of the standardized multimodal dataset, may include the following steps:

[0084] Step S301: Perform multi-scale dynamic statistics and anomaly feature extraction on the standardized multimodal dataset, and output the first-level multi-scale statistical features.

[0085] Step S302: Based on the first-level multi-scale statistical features, perform multi-device temporal causal association and attention feature extraction, and output the second-level device causal association features.

[0086] Step S303: Based on the second-level device causal relationship features, extract dynamic spatiotemporal graph attention and business path weighted features, and output the third-level dynamic spatiotemporal topology and business path weighted features.

[0087] Specifically, the network capacity and performance bottleneck prediction and analysis system 100 first performs multi-scale dynamic statistics and anomaly feature extraction on the standardized multimodal dataset, and outputs the first-level multi-scale statistical features; then, based on the first-level multi-scale statistical features, it performs multi-device temporal causal correlation and attention feature extraction, and outputs the second-level device causal correlation features; finally, based on the second-level device causal correlation features, it performs dynamic spatiotemporal graph attention and service path weighted feature extraction, and outputs the third-level dynamic spatiotemporal topology and service path weighted features.

[0088] In one embodiment, step S301 above, which involves performing multi-scale dynamic statistics and anomaly feature extraction on the standardized multimodal dataset to output first-level multi-scale statistical features, may include the following steps:

[0089] The standardized multimodal dataset is divided into three time-scale subsets: fine-grained, medium-grained, and coarse-grained. For each time-scale subset, the basic statistical features of each network metric are calculated, and an improved isolated forest algorithm with time weights is used to identify anomalous features. The basic statistical features, basic statistical features, and anomalous features of the three time scales are concatenated to obtain the first-level multi-scale statistical features.

[0090] The specific steps in this embodiment are as follows:

[0091] 1. Multi-scale partitioning: The standardized multimodal dataset is divided into three time-scale subsets:

[0092] Fine-grained: 1-minute time windows, totaling 43,200 windows;

[0093] Medium granularity: 15-minute time windows, totaling 2880 windows;

[0094] Coarse-grained: 1-hour time window, 720 windows in total.

[0095] 2. Calculation of basic statistical characteristics: For each subset of data at each time scale, calculate the following statistical characteristics for each indicator:

[0096] Central tendency: mean, median (50th percentile);

[0097] Dispersion: maximum value, minimum value, standard deviation, variance, interquartile range (75th quartile - 25th quartile);

[0098] Distribution characteristics: 25th percentile, 75th percentile, 95th percentile, skewness, kurtosis.

[0099] 3. Calculation of dynamic change characteristics: Calculate the following change characteristics of adjacent time windows:

[0100] Difference: Current window value - Previous window value;

[0101] Rate of change: (Current window value - Previous window value) / Previous window value;

[0102] Cumulative change: the sum of the differences over the past 10 windows;

[0103] Trend slope: The linear trend slope of the past 20 windows was fitted using the least squares method.

[0104] 4. Anomaly Feature Extraction: An improved Isolation Forest algorithm is used to identify anomalies. Improvements include:

[0105] Introducing time weighting, recent outliers have higher weighting;

[0106] Filter false alarms by incorporating business rules;

[0107] Calculate the following characteristics of outliers: frequency of occurrence, duration, degree of deviation (multiple of standard deviation from the mean), and type of outlier (sudden increase, sudden decrease, fluctuation).

[0108] 5. Feature concatenation: Concatenate all features from the three time scales to obtain the first-level multi-scale statistical features.

[0109] In one embodiment, step S302 above, which involves extracting multi-device temporal causal correlation and attention features based on first-level multi-scale statistical features and outputting second-level device causal correlation features, may include the following steps:

[0110] The first-level multi-scale statistical features are grouped by device dimension to generate temporal feature sequences for each device. The temporal feature sequences are encoded using a multi-scale temporal Transformer network to extract multi-scale temporal dependency features within each device. Granger causality test is used to analyze the temporal causal relationships between devices, and a device causal relationship matrix is ​​constructed. The device causal relationship matrix is ​​used as an attention mask, and the temporal feature sequences are fused through a causal attention mechanism to extract causal association features between devices. The multi-scale temporal dependency features and causal association features are concatenated to obtain the second-level device causal association features.

[0111] The specific steps in this embodiment are as follows:

[0112] 1. Device time series grouping: The first-level multi-scale statistical features are grouped into 168 network devices (8 core devices + 32 aggregation devices + 128 access devices), and a time series feature sequence for each device is generated. The sequence length is 168 time steps (corresponding to 7 days × 24 hours).

[0113] 2. Construction of Equipment Causality Matrix: Using the Granger causality test algorithm with a lag order of 5, the temporal causal relationship between any two devices is analyzed, and a 168×168 equipment causality matrix is ​​constructed. ,in Indicates device The change is in the equipment The Granger cause of the change, otherwise 0;

[0114] 3. Extraction of temporal dependency features within the device: Construct a multi-scale temporal Transformer network with 3 encoder layers. Each encoder layer includes: a multi-head self-attention layer (8 attention heads), a feedforward neural network layer (2048 hidden layers), layer normalization, and residual connections; Encode the temporal feature sequence of each device to extract multi-scale temporal dependency features within the device.

[0115] 4. Extraction of causal relationships between devices: Design a causal attention mechanism based on the device causal relationship matrix. Calculate the attention weights using the attention weight mask:

[0116]

[0117] In the above formula, For Hadama accumulation, The dimension of the key vector;

[0118] We perform weighted fusion of the temporal dependency features of all devices to extract causal relationship features between devices;

[0119] 5. Feature splicing: The internal temporal dependency features of the device and the causal relationship features between the devices are spliced ​​together to obtain the second-level device causal relationship features.

[0120] In one embodiment, step S303 above, which involves extracting dynamic spatiotemporal graph attention and service path weighted features based on the second-level device causal relationship features, and outputting the third-level dynamic spatiotemporal topology and service path weighted features, may include the following steps:

[0121] Based on the causal relationship characteristics of the second-level devices, the spatial attention head of the spatiotemporal graph attention network is used to aggregate neighbor features in combination with the dynamic graph and causal relationship matrix. Multi-time step spatial topological features are extracted and fused with the spatial topological features through the temporal attention head to extract spatiotemporal evolution features. The weights of each active business path are calculated based on business importance and traffic, and the path node features are aggregated to generate business path weighted features. The spatiotemporal evolution features and business path weighted features are spliced ​​together to obtain the third-level dynamic spatiotemporal topology and business path weighted features.

[0122] The specific steps in this embodiment are as follows:

[0123] 1. Dynamic Graph Structure Construction: Based on real-time network topology data, a dynamic graph structure is constructed. ,in: Let t be the set of device nodes, consisting of 168 nodes. Let t be the set of links, consisting of 320 links. This is a link weight matrix, where the weights are calculated based on link bandwidth and real-time traffic in a 7:3 ratio. ,in This represents the actual bandwidth of the link. This is the maximum bandwidth of the link. For real-time traffic on the link, Maximum traffic on the link;

[0124] 2. Spatiotemporal graph attention network construction: Construct a network containing two spatiotemporal graph attention layers. Each layer includes: spatial attention heads (4), temporal attention heads (4), layer normalization, and residual connections;

[0125] 3. Spatial topological feature extraction: The spatial attention layer is based on a dynamic graph structure. Causal relationship matrix of devices Calculate the attention weights of each node and its neighboring nodes, aggregate the features of neighboring nodes, and extract spatial topological features;

[0126] 4. Spatiotemporal evolution feature extraction: The temporal attention layer performs weighted fusion of the spatial topological features of 12 consecutive time steps to extract the spatiotemporal evolution features of the network topology;

[0127] 5. Weighted feature extraction of business paths:

[0128] All active business paths were identified, totaling 32, including 8 core business paths, 12 important business paths, and 12 general business paths.

[0129] Calculate the weight of each path based on its business importance and traffic volume: ,in The importance level is determined by the business level (1 for core business, 0.7 for important business, and 0.3 for general business). For path traffic, Total network traffic;

[0130] Aggregate the features of all nodes on the path to generate a weighted feature for the business path.

[0131] 6. Feature splicing: The spatiotemporal evolution features and business path weighted features are spliced ​​together to obtain the third-level dynamic spatiotemporal topology and business path weighted features.

[0132] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0133] Based on the same inventive concept, this application also provides a network capacity and performance bottleneck prediction and analysis device for implementing the network capacity and performance bottleneck prediction and analysis method described above. The solution provided by this device is similar to the implementation scheme described in the above method. Therefore, the specific limitations of one or more network capacity and performance bottleneck prediction and analysis device embodiments provided below can be found in the limitations of the network capacity and performance bottleneck prediction and analysis method described above, and will not be repeated here.

[0134] In one exemplary embodiment, such as Figure 4 As shown, a network capacity and performance bottleneck prediction and analysis device is provided, applied to the network capacity and performance bottleneck prediction and analysis system 100 as described in the above embodiment. The device may include:

[0135] The data acquisition unit 401 is used to collect multi-source heterogeneous data, including historical network operation data, dynamic network topology data, IP address lifecycle management data, and service traffic characteristic data; and to clean, normalize, time-series align, and fuse the multi-source heterogeneous data to generate a standardized multimodal dataset.

[0136] Feature extraction unit 402 is used to perform multi-level joint feature extraction on standardized multimodal datasets to generate multi-level fused features; a multi-task deep neural network model is trained based on the multi-level fused features; the multi-task deep neural network model includes a capacity prediction branch, a performance bottleneck prediction branch, and a network-wide impact assessment branch;

[0137] The result prediction unit 403 is used to input real-time network data into a multi-task deep neural network model, process it through a capacity prediction branch, a performance bottleneck prediction branch and a network-wide impact assessment branch, and output the device resource capacity consumption prediction result, the performance bottleneck prediction result and the network-wide service impact assessment report.

[0138] The results display unit 404 is used to visualize various prediction results and output graded alarms and optimization suggestions when network risks are predicted.

[0139] In one embodiment, the feature extraction unit 402 is further configured to perform multi-scale dynamic statistics and anomaly feature extraction on the standardized multimodal dataset, and output first-level multi-scale statistical features; perform multi-device temporal causal association and attention feature extraction based on the first-level multi-scale statistical features, and output second-level device causal association features; and perform dynamic spatiotemporal graph attention and business path weighted feature extraction based on the second-level device causal association features, and output third-level dynamic spatiotemporal topology and business path weighted features.

[0140] In one embodiment, the feature extraction unit 402 is further configured to divide the standardized multimodal dataset into three time-scale subsets: fine-grained, medium-grained, and coarse-grained; calculate the basic statistical features of each network index for each time-scale subset, and use an improved isolated forest algorithm with time weights to identify abnormal features; and concatenate the basic statistical features, basic statistical features, and abnormal features of the three time-scale subsets to obtain the first-level multi-scale statistical features.

[0141] In one embodiment, the feature extraction unit 402 is further configured to group the first-level multi-scale statistical features by device dimension to generate a temporal feature sequence for each device; encode the temporal feature sequence through a multi-scale temporal Transformer network to extract multi-scale temporal dependency features within the device; analyze the temporal causal relationship between devices using Granger causality test to construct a device causal relationship matrix; use the device causal relationship matrix as an attention mask to fuse the temporal feature sequence through a causal attention mechanism to extract causal association features between devices; and concatenate the multi-scale temporal dependency features and causal association features to obtain the second-level device causal association features.

[0142] In one embodiment, the feature extraction unit 402 is further configured to: extract multi-time-step spatial topology features based on the second-level device causal relationship features through the spatial attention head of the spatiotemporal graph attention network, combine the dynamic graph and causal relationship matrix to aggregate neighbor features, and fuse the spatial topology features through the temporal attention head to extract spatiotemporal evolution features; calculate the weight of each active business path based on business importance and traffic, aggregate path node features to generate business path weighted features; and concatenate the spatiotemporal evolution features and business path weighted features to obtain the third-level dynamic spatiotemporal topology and business path weighted features.

[0143] Each module in the aforementioned network capacity and performance bottleneck prediction and analysis device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0144] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 5 As shown, the computer device includes a processor, memory, input / output interfaces, a communication interface, a display unit, and an input device. The processor, memory, and input / output interfaces are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interfaces. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interfaces are used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When executed by the processor, the computer program implements a network capacity and performance bottleneck prediction and analysis method. The display unit is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0145] Those skilled in the art will understand that Figure 5 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0146] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0147] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0148] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0149] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0150] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0151] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0152] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A network capacity and performance bottleneck prediction and analysis system, characterized in that, The system includes: a data acquisition module, a multi-source heterogeneous data preprocessing module, a multi-level feature extraction module, a multi-task prediction model training module, a capacity and bottleneck prediction module, an IP address full lifecycle management integration module, and a visual intelligent alarm module. The data acquisition module is used to collect multi-source heterogeneous data, including historical network operation data, dynamic network topology data, IP address lifecycle management data, and service traffic characteristic data. The multi-source heterogeneous data preprocessing module is used to clean, normalize, time-series align, and fuse heterogeneous data to generate a standardized multimodal dataset. The multi-level feature extraction module is used to perform multi-level joint feature extraction on the standardized multimodal dataset to generate multi-level fused features; The multi-task prediction model training module is used to train a multi-task deep neural network model based on the multi-level fusion features; the multi-task deep neural network model includes a capacity prediction branch, a performance bottleneck prediction branch, and a network-wide impact assessment branch. The capacity and bottleneck prediction module is used to input real-time network data into the multi-task deep neural network model, process it through the capacity prediction branch, the performance bottleneck prediction branch and the network-wide impact assessment branch respectively, and output the device resource capacity consumption prediction result, the performance bottleneck prediction result and the network-wide service impact assessment report. The IP address lifecycle management integration module is used to uniformly manage the IP address resources of the entire network and synchronize the allocation, use and change data of IP addresses to the capacity and bottleneck prediction module in real time. The visualization intelligent alarm module is used to visualize various prediction results and output graded alarms and optimization suggestions when network risks are predicted.

2. The system according to claim 1, characterized in that, The multi-task deep neural network model adopts an architecture with a fusion gating mechanism; The multi-task deep neural network model also includes a shared feature layer, which comprises two fully connected layers for extracting general features across tasks. The capacity prediction branch is a regression task branch, used to output the resource consumption of each network device in the future preset time. The performance bottleneck prediction branch is a classification task branch, which is used to output the bottleneck level of each device and link. The bottleneck level includes no risk, low risk, medium risk and high risk. The network-wide impact assessment branch is a regression task branch, used to output the degree of impact of business changes on the performance of network-wide services.

3. A method for predicting and analyzing network capacity and performance bottlenecks, characterized in that, The method, applied to the network capacity and performance bottleneck prediction and analysis system as described in any one of claims 1 to 2, comprises: Collect multi-source heterogeneous data, including historical network operation data, dynamic network topology data, IP address lifecycle management data, and service traffic characteristic data; clean, normalize, time-series align, and fuse the multi-source heterogeneous data to generate a standardized multimodal dataset; Multi-level joint feature extraction is performed on the standardized multimodal dataset to generate multi-level fused features; a multi-task deep neural network model is trained based on the multi-level fused features; the multi-task deep neural network model includes a capacity prediction branch, a performance bottleneck prediction branch, and a network-wide impact assessment branch; Real-time network data is input into the multi-task deep neural network model, and processed by the capacity prediction branch, the performance bottleneck prediction branch and the network-wide impact assessment branch respectively, outputting device resource capacity consumption prediction results, performance bottleneck prediction results and network-wide service impact assessment report; It visualizes various prediction results and outputs tiered alerts and optimization suggestions when network risks are predicted.

4. The method according to claim 3, characterized in that, The multi-level joint feature extraction of the standardized multimodal dataset includes: Multi-scale dynamic statistics and anomaly feature extraction are performed on the standardized multimodal dataset to output the first-level multi-scale statistical features; Based on the first-level multi-scale statistical features, multi-device temporal causal association and attention feature extraction are performed, and the second-level device causal association features are output. Based on the second-level device causal relationship features, dynamic spatiotemporal graph attention and business path weighted features are extracted, and the third-level dynamic spatiotemporal topology and business path weighted features are output.

5. The method according to claim 4, characterized in that, The process of performing multi-scale dynamic statistics and anomaly feature extraction on the standardized multimodal dataset, and outputting first-level multi-scale statistical features, includes: The standardized multimodal dataset is divided into three time-scale subsets: fine-grained, medium-grained, and coarse-grained. For each time scale subset, the basic statistical characteristics of each network metric are calculated, and an improved isolated forest algorithm with time weights is used to identify abnormal features. By concatenating the basic statistical features, the basic statistical features, and the anomaly features of the three time scales, the first-level multi-scale statistical features are obtained.

6. The method according to claim 5, characterized in that, The step of extracting multi-device temporal causal correlation and attention features based on the first-level multi-scale statistical features, and outputting second-level device causal correlation features, includes: The first-level multi-scale statistical features are grouped according to the device dimension to generate a time-series feature sequence for each device. The temporal feature sequence is encoded using a multi-scale temporal Transformer network to extract multi-scale temporal dependency features within the device; Granger causality test is used to analyze the temporal causal relationship between devices, and a device causality matrix is ​​constructed. The device causality matrix is ​​used as an attention mask, and the temporal feature sequence is fused through a causal attention mechanism to extract the causal association features between devices. By concatenating the multi-scale temporal dependency features with the causal association features, the second-level device causal association features are obtained.

7. The method according to claim 6, characterized in that, The process of extracting dynamic spatiotemporal graph attention and service path weighted features based on the second-level device causal relationship features, and outputting third-level dynamic spatiotemporal topology and service path weighted features, includes: Based on the causal relationship features of the second-level devices, the spatial attention head of the spatiotemporal graph attention network is used to combine the dynamic graph and the causal relationship matrix to aggregate neighbor features, extract multi-time step spatial topological features, and fuse the spatial topological features through the time attention head to extract spatiotemporal evolution features. The weights of each active business path are calculated based on business importance and traffic, and the characteristics of path nodes are aggregated to generate weighted features of business paths. By splicing the spatiotemporal evolution features and the business path weighted features, the third-level dynamic spatiotemporal topology and business path weighted features are obtained.

8. A network capacity and performance bottleneck prediction and analysis device, characterized in that, The apparatus, applied to the network capacity and performance bottleneck prediction and analysis system as described in any one of claims 1 to 2, comprises: The data acquisition unit is used to collect multi-source heterogeneous data, including historical network operation data, dynamic network topology data, IP address lifecycle management data, and service traffic characteristic data; and to clean, normalize, time-series align, and fuse the multi-source heterogeneous data to generate a standardized multimodal dataset. The feature extraction unit is used to perform multi-level joint feature extraction on the standardized multimodal dataset to generate multi-level fused features; and to train a multi-task deep neural network model based on the multi-level fused features; the multi-task deep neural network model includes a capacity prediction branch, a performance bottleneck prediction branch, and a network-wide impact assessment branch. The result prediction unit is used to input real-time network data into the multi-task deep neural network model, process it through the capacity prediction branch, the performance bottleneck prediction branch and the network-wide impact assessment branch respectively, and output the device resource capacity consumption prediction result, the performance bottleneck prediction result and the network-wide service impact assessment report. The results display unit is used to visualize various prediction results and output graded alarms and optimization suggestions when network risks are predicted.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 3 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 3 to 7.