Communication channel regular checking method based on intelligent time delay map

Abstract

The invention is suitable for the technical field of communication network operation and maintenance, and provides a communication channel regular checking method based on an intelligent time delay map, and the method comprises the steps: collecting the performance data of a plurality of communication channels in a communication network; inputting the collected performance data into a pre-trained time delay intelligent evaluation model, and outputting an evaluation result, including a time delay predicted value, a time delay abnormal score and a channel health degree, of each communication channel; generating an intelligent time delay map of the communication network based on the evaluation result; visual elements of the intelligent time delay map are dynamically associated with the evaluation result, and the visual elements comprise colors, line thicknesses, data labels and early warning icons which dynamically change according to the evaluation result; in response to a regular checking instruction for the target communication channel on the intelligent time delay map, automatically generating a regular checking report in combination with an evaluation result corresponding to the target communication channel; according to the method, through fusion of the intelligent analysis model and the dynamic visual map, the operation and maintenance decision efficiency and the network reliability are effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of communication network operation and maintenance, and in particular to a communication channel fixed inspection method based on intelligent time delay map. BACKGROUND

[0002] At present, the conventional fixed inspection process is mainly based on the active operation and experience analysis of the operation and maintenance personnel. The operation and maintenance personnel first manually query or test instructions through the network management system to obtain the performance data of the specified channel, such as time delay, bit error rate, etc. The isolated data needs to be manually compared and analyzed to determine the on-off state, performance degradation, etc. of the channel, and finally the test report is manually sorted and written to complete a fixed inspection task.

[0003] However, this fixed inspection method relies more on manual operation and analysis, and the fixed inspection efficiency is low, which is difficult to meet the real-time monitoring demand of a large number of channels in a large-scale network. Secondly, the network management system provides mainly discrete numerical indicators, which lack a visual global view of the related topology, making it difficult for the operation and maintenance personnel to quickly locate the high time delay section or performance bottleneck. In addition, the current analysis is based on historical and current data, which lacks the ability to conduct in-depth mining and intelligent analysis of the data, and cannot effectively predict the future trend of channel performance, actively and accurately identify potential abnormal risks and root causes, and meet the technical needs of real-time, accurate and intelligent cognition and decision support for channel state in modern complex communication networks.

[0004] In view of this, a communication channel fixed inspection method based on intelligent time delay map is proposed. SUMMARY

[0005] The present application provides a communication channel fixed inspection method based on intelligent time delay map, which is used to solve the problem that it cannot meet the technical needs of real-time, accurate and intelligent cognition and decision support for channel state in modern complex communication networks.

[0006] The present application provides a communication channel fixed inspection method based on intelligent time delay map, which comprises: Collecting performance data of a plurality of communication channels in a communication network, the performance data including historical and real-time time delay data, bit error rate data and alarm data; Inputting the collected performance data into a pre-trained time delay intelligent evaluation model to output an evaluation result of each communication channel; the evaluation result includes a time delay prediction value, a time delay abnormal score and a channel health degree; Generating an intelligent time delay map of the communication network based on the evaluation result; wherein the visualization elements of the intelligent time delay map are dynamically associated with the evaluation result, and the visualization elements include dynamically changed colors, line thicknesses, data labels and warning icons according to the evaluation result; In response to an inspection instruction for a target communication channel on the intelligent latency map, an inspection report is automatically generated in combination with the evaluation result corresponding to the target communication channel.

[0007] Further, the performance data of the plurality of communication channels in the communication network is collected, including: The performance data is collected in a programmed manner by integrating an interface of a network management system or communicating with network element devices in a protocol; The performance data constitutes a multi-source data set, including real-time latency data collected from the network management system or the network element devices, historical latency data stored in a database, error rate data, and alarm data.

[0008] Further, the training method of the intelligent latency evaluation model includes: Obtaining training data, the training data including time series performance data of a plurality of communication channels in a historical time period and corresponding network topology data; Building an initial evaluation model, the initial evaluation model including a time series prediction sub-module, an anomaly detection sub-module, and a health degree evaluation sub-module connected in sequence; Iteratively training the initial evaluation model through a total loss function until a convergence condition is met, obtaining the intelligent latency evaluation model; The total loss function includes at least a first loss function and a second loss function, the first loss function being used to constrain the predicted latency value output by the time series prediction sub-module to comply with physical laws in the communication field, and the second loss function being used to constrain the anomaly score output by the anomaly detection sub-module to reflect the dependence transmission in the network topology.

[0009] Further, the first loss function is specifically used for: Calculating a data fitting loss between the predicted latency value output by the time series prediction sub-module and the real latency value; Calculating a constraint loss based on the physical laws, the constraint loss including a latency non-negativity loss, a latency change smoothness loss, and a traffic periodicity loss; According to a weighted sum of the data fitting loss and the constraint loss, a first loss function value is calculated.

[0010] Further, the second loss function is specifically used for: Inputting the time series performance data and the network topology data into the anomaly detection sub-module; The graph neural network in the anomaly detection sub-module performs message passing and aggregation according to the network topology data to generate a channel representation vector containing a topology dependence; According to the channel representation vector, a topology dependence reflecting loss is calculated;Figure One consistency loss; inputting the channel characterization vector into an isolation forest algorithm for anomaly judgment, and calculating an anomaly detection loss; Figure One a weighted sum of the consistency loss and the anomaly detection loss, to calculate a second loss function value.

[0011] Further, the total loss function further includes a third loss function, and the third loss function is specifically used for: inputting the time delay prediction value output by the time series prediction submodule, the time delay anomaly score and suspected root cause information output by the anomaly detection submodule, and the basic performance data into the health degree evaluation submodule; the attention network in the health degree evaluation submodule dynamically learns the weight of each input feature under the current network state, and outputs a channel health degree score; calculating the difference between the channel health degree score and the benchmark health degree labeled based on expert rules as the third loss function value.

[0012] Further, the intelligent time delay map of the communication network based on the evaluation result includes: establishing a mapping rule from the evaluation result to a visual element; based on the mapping rule, the following operations are performed: according to the channel health degree or the time delay prediction value, assigning a corresponding color to each communication channel link in a topology graph; adjusting the display thickness of the link according to a predefined channel traffic importance level; when the time delay anomaly score exceeds a preset threshold, superimposing a warning icon on the corresponding link for display.

[0013] Further, the mapping rule further includes: when the evaluation result includes suspected root cause information, displaying a root cause identifier on a network element device node corresponding to the suspected root cause information in the intelligent time delay map; the display style of the root cause identifier is different from the warning icon.

[0014] Further, the automatic generation of the inspection report includes: filling the time delay prediction value, the time delay anomaly score, the channel health degree, and the suspected root cause information in the evaluation result corresponding to the target communication channel into the corresponding fields of a preset inspection report template; based on the evaluation result, generating a natural language text including performance trend description, anomaly diagnosis conclusion and maintenance suggestion; ​Integrate the filled template and the natural language text to generate a complete inspection report containing channel state summary, trend analysis, diagnosis conclusion and maintenance suggestions, and output in a specified format.

[0015] Further, after collecting the performance data of the plurality of communication channels in the communication network, and before inputting the performance data into the intelligent delay evaluation model, the method further comprises: The collected performance data is preprocessed, and the preprocessing comprises: time stamp alignment and unit unification of delay data from different data sources, and classification and deduplication of alarm data according to a preset rule base.

[0016] From the above technical solutions, the present application has the following advantages: The present application automatically collects channel performance data, inputs it into the intelligent delay evaluation model that integrates time series prediction, anomaly detection and health degree evaluation, and obtains comprehensive evaluation results including delay prediction value, anomaly score and health degree. According to the results, an intelligent delay map is dynamically generated, which superimposes visual elements such as color, thickness and warning icons. Finally, the user can trigger an inspection instruction on the map, and the system automatically generates a detailed inspection report based on the evaluation results. The present application uses an artificial intelligence model to deeply analyze the data, realizes accurate prediction of future delay and intelligent positioning of abnormal root causes. Secondly, the intelligent delay map dynamically rendered based on the evaluation results converts scattered numerical indicators into an intuitive global topology view, which can clearly understand high delay sections, performance bottlenecks and fault root causes, and improves the intuitiveness and efficiency of problem positioning. In addition, the report is automatically generated by one-key map interaction, replacing the traditional manual query, analysis and writing process, effectively improving the automation degree and execution efficiency of the inspection work. It provides efficient intelligent decision support for the operation and maintenance of large-scale complex networks. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 An embodiment flowchart of the communication channel inspection method based on intelligent delay map in the present application; Figure 2 A flowchart of the training phase and application phase of the intelligent delay evaluation model in the present application; Figure 3 A flowchart of the automatic generation and output of the inspection report in the present application. DETAILED DESCRIPTION

[0018] The terms "first", "second", "third", "fourth" and the like in the description of this application and in the claims hereof, if any, are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the use of the terms so termed herein is solely for descriptive purposes and not for pronouncing the limitations of the application described except as set forth in the claims. Furthermore, the terms "comprise", "comprising", "corresponding" and "corresponds" and any variations thereof are intended to cover a non-exclusive inclusion, for example, a process, method, system, product or apparatus that comprises a list of steps or units can not necessarily be limited to those steps or units which are expressly listed or to method combinations consisting only of the steps or units but can include additional steps or units, which are neither expressly set forth nor essential to such processes, methods, systems, products or apparatus.

[0019] Embodiment one The method implemented in this embodiment can be implemented in a system, and can be implemented in a server or in a terminal, and the specific implementation is not limited. The method in this application will be introduced from the perspective of system implementation. Please refer to Figure 1 The method provided by the embodiment of the application comprises the following steps: S1. Collecting performance data of a plurality of communication channels in a communication network, the performance data comprising historical and real-time delay data, error rate data and alarm data; In this embodiment, the performance data is collected in a programmed manner by integrating the interface of a network management system or communicating with network element devices; wherein the performance data constitutes a multi-source data set, including real-time delay data collected from the network management system or the network element devices, historical delay data stored in the database, error rate data and alarm data. The delay data includes the latest round-trip delay or one-way delay measurement value collected in real time from the network management system or the network element devices, and the historical delay records of all collection periods in the past continuous time stored in the performance database. The error rate data is the index data reflecting the transmission error code situation of the channel. The alarm data is the event notification information about the performance degradation of the channel, device failure or protocol exception generated by the network management system or the device.

[0020] In this embodiment, the collected performance data is preprocessed, and the preprocessing includes: time stamp alignment and unit unification of the delay data from different data sources, and classification and deduplication of the alarm data according to the preset rule base.

[0021] S2. Inputting the collected performance data into a pre-trained intelligent delay evaluation model to output an evaluation result of each communication channel; the evaluation result comprising a delay prediction value, a delay anomaly score and a channel health degree; Please refer to Figure 2The pre-trained delay intelligent evaluation model is used for comprehensive analysis of the original scattered performance data to obtain a structured evaluation result. In this embodiment, the training method of the delay intelligent evaluation model includes the following sub-steps: 201. Obtain training data, which includes time series performance data and corresponding network topology data of multiple communication channels in a historical time period; 202. Construct an initial evaluation model, which includes a time series prediction submodule, an anomaly detection submodule, and a health degree evaluation submodule connected in sequence; 203. Iteratively train the initial evaluation model through a total loss function until a convergence condition is met to obtain the delay intelligent evaluation model; wherein the total loss function at least includes a first loss function and a second loss function, the first loss function is used to constrain the predicted delay value output by the time series prediction submodule to conform to the physical law in the communication field, and the second loss function is used to constrain the anomaly score output by the anomaly detection submodule to reflect the dependence transmission in the network topology.

[0022] Specifically, the constructed initial evaluation model is an end-to-end deep learning architecture, which includes three logically connected and information-sharing submodules; the time series prediction submodule is constructed based on a time series neural network, which is used to learn the time-dependent pattern in the performance data and output the predicted value of the channel delay in the future period. The anomaly detection submodule is used to identify abnormal delay behavior deviating from the normal mode. This module first models the input network topology data using a graph neural network. The graph neural network enables the representation vectors of each node (network element) and edge (channel) to aggregate the neighbor information, thereby generating a channel representation vector containing the topological dependence. This vector not only contains the historical features of the channel itself, but also encodes its context in the network structure. Secondly, these representation vectors are sent to unsupervised anomaly detection algorithms such as Isolation Forest to determine whether they belong to an anomaly and output a quantitative delay anomaly score. The health degree evaluation submodule is a multilayer perceptron containing an attention mechanism, which is used to comprehensively evaluate the overall running state of the channel. The module takes the outputs of the two aforementioned submodules, as well as the basic performance data and suspected root cause information provided by the anomaly detection submodule, as inputs to calculate a comprehensive channel health score.

[0023] The first loss function is used to constrain the time series prediction submodule to ensure that its prediction result not only fits the historical data, but also conforms to the physical law in the communication field. The second loss function is used to constrain the anomaly detection submodule to ensure that its output anomaly score can effectively reflect the real fault dependence and transmission relationship in the network topology. The model training process minimizes the total loss through the back propagation algorithm until the model performance converges on the validation set, and finally obtains a trained delay intelligent evaluation model that can be used in a production environment.

[0024] In this embodiment, the first loss function is specifically used to optimize the time series prediction sub-module, and the calculation includes the following processes: 1. Calculate the data fitting loss between the predicted delay value output by the time series prediction sub-module and the real delay value; 2. Calculate the constraint loss based on the physical law, including the delay non-negativity loss, the delay change smoothness loss and the traffic periodicity loss; 3. Calculate the first loss function value according to the weighted sum of the data fitting loss and the constraint loss.

[0025] Specifically, using the mean square error standard loss function, the difference between the predicted delay value output by the time series prediction sub-module and the corresponding real delay value in the training data is calculated, which ensures that the model learns the basic pattern of the historical data. In order to integrate prior knowledge in the field of communication into the model, a constraint loss based on physical law is constructed, which includes delay non-negativity loss, delay change smoothness loss and traffic periodicity loss; wherein, the delay non-negativity loss punishes the predicted negative delay value, because the physical delay cannot be negative; the delay change smoothness loss constrains the change rate of adjacent time predicted delay within a reasonable range, avoiding the occurrence of sharp jump which does not conform to the characteristics of signal transmission; the traffic periodicity loss: by comparing the difference between the predicted delay curve and the known network traffic day / week cycle template, the model is encouraged to learn and fit the periodicity of network traffic. The above data fitting loss and constraint loss are weighted and summed according to the preset weight coefficient to obtain the final first loss function value. This step combines data-driven and knowledge-driven, which significantly improves the accuracy and reasonableness of the prediction.

[0026] In this embodiment, the second loss function is specifically used to optimize the anomaly detection sub-module, and the calculation includes the following processes: 1. Input the time series performance data and network topology data into the anomaly detection sub-module; 2. The graph neural network in the anomaly detection sub-module performs message passing and aggregation according to the network topology data to generate a channel representation vector containing topological dependency; 3. Calculate the Figure One consistency loss reflecting the topological dependency according to the channel representation vector; 4. Input the channel representation vector into the Isolation Forest algorithm for anomaly judgment to calculate the anomaly detection loss; 5. Calculate the second loss function value according to the weighted sum of the Figure One consistency loss and the anomaly detection loss.

[0027] Specifically, time-series performance data is used as features for nodes and edges, and network topology data defines the connection relationships between nodes and edges; both are input into the anomaly detection submodule. The graph neural network in this submodule performs multi-layer message passing and feature aggregation based on the topology structure, generating a channel representation vector containing topological dependencies for each channel. To enhance the model's utilization of topological relationships, a... Figure One A consistency loss is used to encourage topologically adjacent or related channels to have relatively similar representation vectors in the feature space; conversely, the representations of unrelated channels should be differentiated. The channel representation vectors generated in step 2 are input into the Isolation Forest algorithm. Multiple trees are constructed to isolate samples, and the path length required for each sample (channel) to be isolated is calculated. Shorter paths are more likely to indicate anomalies. Based on this principle, an anomaly detection loss is defined such that the latency anomaly score output by the model is consistent with the anomaly severity determined by the Isolation Forest algorithm. Figure One The consistency loss and the anomaly detection loss are weighted and summed to obtain the final second loss function value. This forces the model to consider both the anomaly of the data itself and the transmission logic of the network topology when detecting anomalies.

[0028] In this embodiment, the total loss function also includes a third loss function, used to optimize the health assessment submodule, which is specifically used for: 1. Input the latency prediction value output by the timing prediction submodule, the latency anomaly score and suspected root cause information output by the anomaly detection submodule, and the basic performance data into the health assessment submodule; 2. In the health assessment submodule, the attention network dynamically learns the weights of each input feature in the current network state and outputs a channel health score; 3. Calculate the difference between the channel health score and the baseline health score based on expert rule annotation, and use it as the third loss function value.

[0029] Specifically, the time delay prediction value output by the time prediction sub-module, the time delay anomaly score and suspected root cause information output by the anomaly detection sub-module, and the original basic performance data are collectively input to the health assessment sub-module. The attention network in the sub-module dynamically analyzes the current input feature set and automatically learns the weight of each input feature under the current network state, rather than using fixed weights. For example, when predicting that the future time delay will deteriorate sharply, the attention network will increase the weight of the time delay prediction value feature; when detecting a high anomaly score and a clear root cause, the weight of the anomaly score and root cause information will be increased. Based on the learned dynamic weights, the model comprehensively calculates and outputs the final channel health score. In order to train this sub-module, a reference standard is needed; by annotating the historical channel state in the training data according to expert experience rules, a baseline health score is generated. The third loss function is to calculate the difference between the channel health score output by the model and this baseline health score, and use the difference to guide the parameter update of the attention network and the evaluation model.

[0030] S3. Generating an intelligent time delay map of the communication network based on the evaluation results; wherein the visualization elements of the intelligent time delay map are dynamically associated with the evaluation results, and the visualization elements include colors, line thicknesses, data labels, and warning icons that dynamically change according to the evaluation results; Referring to Figure 3 This step converts the evaluation results output by the above steps into a visual time delay map. The principle is to dynamically map abstract numerical values and diagnostic conclusions to visual elements such as colors, thicknesses, icons, etc. that are directly visible on the topology map, and the map is automatically refreshed as the background evaluation results are periodically updated or changed in real time, thereby providing network operation and maintenance personnel with real-time and intuitive global situational awareness of network time delay status, performance bottlenecks, and potential risks.

[0031] In this embodiment, mapping rules from evaluation results to visualization elements are established; based on the mapping rules, the following operations are performed: 1. Assign a corresponding color to each communication channel link in the topology map according to the channel health score or the time delay prediction value; 2. Adjust the display thickness of the link according to the pre-defined channel business importance level; 3. When the time delay anomaly score exceeds the preset threshold, superimpose and display a warning icon on the corresponding link.

[0032] Specifically, the state-based color mapping assigns a color to each communication channel link in the topology graph. The specific mapping rule is: taking the channel health score or latency prediction value as input, mapping through a continuous color gradient; for example, from green representing health, to yellow representing warning, to red representing serious problems; the higher the health score, the closer the link color to green; the larger the predicted latency value, the closer the link color to red. Importance-based thickness mapping: adjusting the display thickness of the link to intuitively reflect the importance of the traffic it carries or the historical traffic level. According to the pre-defined channel traffic importance level in the network management system, different levels of links are set to different rendering thickness. For example, the channel carrying core traffic is displayed with the thickest line, so as to guide the operation and maintenance personnel to quickly focus on the key resources in the complex network topology. Abnormal threshold-based icon warning: when the latency anomaly score exceeds the preset threshold, i.e. there is a high probability of abnormal behavior of the channel, the system will automatically display a warning icon on the corresponding position of the channel link. This provides direct visual alarm, surpassing the warning strength of simple color change.

[0033] In the embodiment, the mapping rule further includes visual positioning of fault root cause, specifically as follows: 1. When the evaluation result contains suspected root cause information, a root cause identifier is displayed on the network element device node corresponding to the suspected root cause information in the intelligent latency map. 2. The display style of the root cause identifier is different from the warning icon.

[0034] Root cause identifier mapping: when the evaluation result contains suspected root cause information, i.e. the list output by the anomaly detection submodule pointing to a specific network element device or link, the system will render a special root cause identifier, such as a red bullseye icon or a fault wrench icon, on the network element device node corresponding to the suspected root cause information in the intelligent latency map. Style differentiation design: the display style of the root cause identifier is designed to be significantly different from the warning icon indicating where the abnormal phenomenon occurs, while the root cause identifier indicates where the abnormality may originate from.

[0035] Through the execution of the above mapping rules, an intelligent latency map integrating performance status, traffic weight, anomaly alarm and root cause positioning is generated. The map serves as an interactive visualization platform, enabling complex network performance evaluation results to be intuitively and quickly understood and operated by operation and maintenance personnel.

[0036] S4. In response to the inspection instruction on the target communication channel in the intelligent latency map, an inspection report is automatically generated in combination with the evaluation result corresponding to the target communication channel.

[0037] The step works on the principle of converting the above-mentioned intelligent analysis results into standardized documents that can be directly used to guide operation and maintenance actions through human-computer interaction. When the maintenance personnel issues an inspection instruction on the intelligent time delay map for a certain target communication channel, the system will immediately respond to the instruction, automatically retrieve and integrate all relevant evaluation results generated for the channel in step S2, and through the integrated report generation module, without any manual data sorting and text writing, directly output a detailed and structured inspection report.

[0038] In this embodiment, the operation of automatically generating an inspection report is realized through the following three sub-steps: 1. Fill the time delay prediction value, time delay anomaly score, channel health degree and suspected root cause information in the evaluation results of the target communication channel into the corresponding fields of the preset inspection report template; 2. Based on the evaluation results, generate natural language text containing performance trend description, abnormal diagnosis conclusion and maintenance suggestion; 3. Integrate the filled template and natural language text to generate a complete inspection report containing channel state summary, trend analysis, diagnosis conclusion and maintenance suggestion, and output in a specified format.

[0039] The specific inspection report template includes a document framework of multiple standardized fields. The system automatically extracts the corresponding time delay prediction value, time delay anomaly score, channel health degree and suspected root cause information and other key data from the evaluation results of the channel, and fills them into the corresponding fields of the template. In order to generate report content beyond simple data listing and with analysis depth, the system drives a natural language generation component based on the same evaluation results. The component converts numerical indicators into descriptive text that is easy to understand according to predefined logical rules and knowledge base. For example, generate performance trend description according to time delay prediction value, generate abnormal diagnosis conclusion according to anomaly score and root cause information, and generate specific maintenance suggestion combined with network operation and maintenance knowledge base.

[0040] Finally, the system intelligently integrates the report template filled with structured data generated in the first step and the natural language text containing analytical description generated in the second step to form a complete inspection report. The report usually contains channel state summary, performance trend analysis, AI diagnosis conclusion and specific maintenance suggestion, etc. After the report is generated, the system automatically renders and packages it in a specified format, and can directly display and provide a download link through the interface, or automatically distribute it to the designated maintenance personnel through email and work order system, thereby completing the whole process from instruction triggering to report delivery.

[0041] The application converts the communication channel inspection from a passive response mode depending on artificial experience to an active early warning and accurate diagnosis mode driven by data intelligence through the fusion of intelligent analysis models and dynamic visual maps, and greatly improves the efficiency of operation and maintenance decision and network reliability.

[0042] It can be understood that those skilled in the art can combine various embodiments in the above embodiments to obtain various embodiments of the technical solutions.

[0043] The above only describes the preferred embodiments of the application and is not intended to limit the application. Any modification, equivalent replacement and improvement made within the spirit and principle of the application shall be included in the protection scope of the application.

Claims

1. A communication channel periodic inspection method based on an intelligent time-delay map, characterized in that, include: The system collects performance data from multiple communication channels in the communication network, including historical and real-time latency data, bit error rate data, and alarm data. The collected performance data is input into a pre-trained intelligent latency evaluation model, which outputs the evaluation results for each communication channel. The evaluation results include the latency prediction value, latency anomaly score, and channel health. A smart latency map of the communication network is generated based on the evaluation results; wherein, the visualization elements of the smart latency map are dynamically associated with the evaluation results, and the visualization elements include colors, line thicknesses, data labels, and warning icons that change dynamically according to the evaluation results; In response to the scheduled inspection command for the target communication channel on the intelligent latency map, and in conjunction with the evaluation results corresponding to the target communication channel, a scheduled inspection report is automatically generated.

2. The communication channel periodic inspection method based on intelligent time-delay map according to claim 1, characterized in that, The performance data of multiple communication channels in the collected communication network includes: The performance data is collected periodically in a programmed manner by integrating the network management system interface or communicating with network element devices via protocol. The performance data constitutes a multi-source data set, including real-time latency data collected from network management systems or network element devices, historical latency data stored in the database, bit error rate data, and alarm data.

3. The communication channel periodic inspection method based on intelligent time-delay map according to claim 2, characterized in that, The training method for the latency intelligent evaluation model includes: Acquire training data, which includes time-series performance data of multiple communication channels over a historical period and corresponding network topology data; Construct an initial evaluation model, which includes a time-series prediction submodule, an anomaly detection submodule, and a health assessment submodule connected in sequence; The initial evaluation model is iteratively trained using the total loss function until the convergence condition is met, thus obtaining the time-delay intelligent evaluation model. The total loss function includes at least a first loss function and a second loss function. The first loss function is used to constrain the predicted delay value output by the time-series prediction submodule to conform to the physical laws of the communication field. The second loss function is used to constrain the anomaly score output by the anomaly detection submodule to reflect the propagation of dependencies in the network topology.

4. The communication channel periodic inspection method based on intelligent time-delay map according to claim 3, characterized in that, The first loss function is specifically used for: Calculate the data fitting loss between the predicted delay value output by the time series prediction submodule and the actual delay value; Calculate the constraint loss constructed based on the physical laws, the constraint loss including the time delay non-negativity loss, the time delay change smoothness loss, and the flow periodicity loss; The first loss function value is calculated based on the weighted sum of the data fitting loss and the constraint loss.

5. The communication channel periodic inspection method based on intelligent time-delay map according to claim 3, characterized in that, The second loss function is specifically used for: The timing performance data and the network topology data are input into the anomaly detection submodule; The graph neural network in the anomaly detection submodule performs message passing and aggregation based on the network topology data to generate channel representation vectors containing topological dependencies. The graph consistency loss, reflecting topological dependencies, is calculated based on the channel representation vector. The channel representation vector is input into the isolated forest algorithm for anomaly detection, and the anomaly detection loss is calculated. The second loss function value is calculated based on the weighted sum of the graph consistency loss and the anomaly detection loss.

6. The communication channel periodic inspection method based on intelligent time-delay map according to claim 3, characterized in that, The total loss function also includes a third loss function, which is specifically used for: The latency prediction value output by the timing prediction submodule, the latency anomaly score and suspected root cause information output by the anomaly detection submodule, and the basic performance data are input into the health assessment submodule. The attention network in the health assessment submodule dynamically learns the weights of each input feature in the current network state and outputs a channel health score. The difference between the channel health score and the baseline health score based on expert rule annotation is calculated and used as the value of the third loss function.

7. The communication channel periodic inspection method based on intelligent time-delay map according to claim 1, characterized in that, The process of generating an intelligent latency map of the communication network based on the evaluation results includes: Establish a mapping rule from the evaluation results to the visualization elements; based on the mapping rule, perform the following operations: Assign a corresponding color to each communication channel link in the topology diagram based on the channel health status or the latency prediction value; The display thickness of the link is adjusted according to the predefined channel service importance level; When the latency anomaly score exceeds a preset threshold, a warning icon is overlaid on the corresponding link.

8. The communication channel periodic inspection method based on intelligent delay map according to claim 7, characterized in that, The mapping rules also include: When the evaluation results contain suspected root cause information, the root cause identifier is displayed on the network element device node corresponding to the suspected root cause information in the intelligent latency map; The root cause identifier is displayed differently from the warning icon.

9. The communication channel periodic inspection method based on intelligent time-delay map according to claim 8, characterized in that, The automatic generation of scheduled inspection reports includes: The predicted latency value, latency anomaly score, channel health and suspected root cause information from the evaluation results corresponding to the target communication channel are filled into the corresponding fields of the preset scheduled inspection report template. Based on the evaluation results, natural language text containing performance trend descriptions, anomaly diagnosis conclusions, and maintenance suggestions is generated. The filled template and the natural language text are integrated to generate a complete scheduled inspection report containing a channel status summary, trend analysis, diagnostic conclusions and maintenance recommendations, and output in a specified format.

10. The communication channel periodic inspection method based on intelligent time-delay map according to claim 2, characterized in that, After collecting performance data from multiple communication channels in the communication network, and before inputting the performance data into the latency intelligent evaluation model, the method further includes: The collected performance data is preprocessed, including: aligning the timestamps and unifying the units of latency data from different data sources, and classifying and deduplicating alarm data according to a preset rule base.

Images