An AI-based fault routing rapid location system

Abstract

This invention discloses an AI-based fault routing rapid location system, relating to the field of wireless communication network technology. It includes: a network-wide flow topology data synchronization acquisition module for obtaining synchronized raw data from network elements in the wireless communication network; a disturbance feature temporal extraction module for generating feature time series based on the synchronized raw data; a flow topology coupling effect quantification module for combining the feature time series with network topology index data to generate a real-time fault fingerprint sequence; and a fault root cause reverse reasoning module for acquiring a preset fault fingerprint database, calculating fingerprint dissimilarity, and determining the prototype fingerprint with the smallest dissimilarity to generate a fault location report. This invention can globally and accurately locate fault root causes, sensitively detect soft faults, integrate multi-dimensional information to improve location accuracy and robustness, and adaptively optimize parameters and learn to expand.

Description

Technical Field

[0001] This invention relates to the field of wireless communication network technology, and more specifically to an AI-based fault routing rapid location system. Background Technology

[0002] With the deployment of next-generation wireless communication networks, the scale, complexity, and dynamism of networks have increased dramatically. Large-scale networks have a huge number of base stations and core network elements with intricate connections. Failures of a single network element often cause local or global performance disturbances in the entire network through behaviors such as service flow rerouting and user handover. Moreover, these disturbances exhibit complex propagation characteristics in time and space, posing a greater challenge to fault location.

[0003] Existing wireless network fault location methods rely heavily on network element alarm correlation analysis, making them highly dependent on the timeliness and accuracy of alarms. They are weak at locating "soft" faults without obvious alarms, such as external interference, improper parameter configuration, and hidden capacity bottlenecks. Furthermore, static correlation rules are ill-suited to the dynamic changes in network topology and service patterns, easily leading to false positives and false negatives. Anomaly detection methods based on single-base station KPIs offer an isolated perspective, only identifying the fault location but not its cause. For example, a decline in service quality at a base station might be caused by handover congestion due to a fault at an adjacent base station; traditional methods struggle to deduce the root cause from localized data. Current technologies lack mechanisms for fault tracing based on the spatiotemporal patterns formed by fault disturbances propagating through service flows and network topology from a global perspective, making it difficult to solve the technical challenges of large-scale network operation and maintenance. Summary of the Invention

[0004] The purpose of this invention is to provide an AI-based fault routing rapid location system, which solves the problems existing in the background technology.

[0005] To address the aforementioned technical problems, this invention provides an AI-based fault routing rapid location system, comprising: a network-wide flow topology data synchronous acquisition module, used to synchronously acquire service flow indicator data and network topology indicator data from network elements in the wireless communication network to obtain synchronized raw data;

[0006] The disturbance feature time series extraction module is used to calculate the time change rate and short-term volatility of business flow indicator data based on the synchronized raw data, so as to generate a feature time series that characterizes network state disturbances.

[0007] The flow topology coupling effect quantification module is used to combine the feature time series with the network topology index data to quantify the mutual influence strength between node pairs in the network, so as to generate a real-time fault fingerprint sequence characterizing the mutual influence strength.

[0008] The fault root cause reverse reasoning module is used to obtain a preset fault fingerprint database, which contains prototype fingerprints corresponding to various known fault types; calculate the fingerprint difference between the real-time fault fingerprint sequence and each prototype fingerprint in the fault fingerprint database; determine the prototype fingerprint with the smallest fingerprint difference, and use the fault type and root cause node corresponding to the prototype fingerprint with the smallest fingerprint difference as the location result to generate a fault location report.

[0009] Preferably, the perturbation feature temporal extraction module is specifically used for:

[0010] Perform first-order difference calculation on the throughput time series in the business flow indicator data to obtain the throughput change rate series;

[0011] The fluctuation series of active users in the business flow metric data is obtained by calculating the standard deviation of the active user count using a sliding window.

[0012] The throughput change rate sequence and the active user number fluctuation sequence are combined to generate the characteristic time series.

[0013] Preferably, the flow topology coupling effect quantification module is specifically used for:

[0014] Calculate the normalized flow perturbation correlation of the node pairs;

[0015] Calculate the dynamic topology influence factor of the node pair;

[0016] The normalized flow disturbance correlation and the dynamic topology influence factor are weighted and summed to obtain the flow topology coupling index;

[0017] A coupling effect matrix is ​​constructed based on the flow topology coupling index to generate the real-time fault fingerprint sequence.

[0018] Preferably, the calculation of the normalized flow disturbance correlation is specifically as follows:

[0019] Within a preset time delay range, the feature time series of the node pair are slid, and for each delay time point, the Pearson correlation coefficient between the two series is calculated;

[0020] The maximum absolute value of the correlation coefficient among all delayed time points is selected as the normalized flow disturbance correlation.

[0021] Preferably, the calculation of the dynamic topology influence factor is specifically as follows:

[0022] The number of successful handovers between the node pairs within the current time window is obtained and divided by their historical average number of handovers to obtain a normalized handover behavior factor.

[0023] Obtain the reference signal received power from the source node reported by the user equipment of the target node, and subtract the preset signal quality reference threshold to obtain the signal quality difference factor;

[0024] The normalized handover behavior factor is multiplied by a preset handover impact weight, and the signal quality difference factor is multiplied by a preset signal quality impact weight.

[0025] The two products are then nonlinearly fused using a hyperbolic tangent function to generate the dynamic topological influence factor.

[0026] Preferably, the switching impact weight and the signal quality impact weight are hyperparameters obtained by training and optimizing a machine learning model based on historical fault case data.

[0027] Preferably, the calculation of the fingerprint difference is specifically as follows:

[0028] Within a preset fingerprint sequence length, the Frobenius norm between the real-time coupling effect matrix and the coupling effect matrix of the prototype fingerprint is calculated at each time point.

[0029] The fingerprint difference is obtained by taking the arithmetic mean of the Frobenius norm at all time points.

[0030] Preferably, the fault root cause reverse reasoning module is further used for:

[0031] The calculated minimum fingerprint difference is compared with a preset confidence threshold.

[0032] When the minimum fingerprint difference is lower than the confidence threshold, the localization result is output;

[0033] When the minimum fingerprint difference is not lower than the confidence threshold, a report indicating an unknown fault type is output.

[0034] Preferably, the confidence threshold is determined by plotting the subject operating characteristic curve on the validation dataset and based on preset positioning accuracy and recall targets.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] 1. By establishing a network-wide flow topology data synchronous acquisition module, the system is no longer limited to a single network element, but instead collects business flow index data and network topology index data from a global perspective. Based on this, the flow topology coupling effect quantification module calculates the flow topology coupling index and constructs a coupling effect matrix, generating a real-time fault fingerprint sequence that can depict the dynamic influence relationship between nodes across the entire network. This achieves global and root cause localization, breaking through the limitations of a local perspective.

[0037] 2. It has high sensitivity to "soft" faults and does not rely on hardware alarms. The disturbance feature time-series extraction module shifts the focus from the absolute value of performance indicators to their subtle dynamic changes by calculating the time change rate and short-term volatility. Based on the detection and location of subtle dynamic disturbances in service and network performance, it improves the ability to locate soft faults without alarms.

[0038] 3. By integrating the quantitative coupling effect of heterogeneous information from business flow and network topology, the constructed fault fingerprint sequence has higher dimensions and richer information content than a single information source. Its identification ability far exceeds that of a single type of indicator, thereby significantly improving the accuracy and robustness of the location.

[0039] 4. The system's key parameters are adaptively optimized through machine learning, and the fault fingerprint database is continuously learned and expanded, enabling it to adapt to network evolution and new fault modes. This allows the entire positioning system to adapt to network evolution and the emergence of new fault modes, possessing the ability to continuously evolve and ensuring its long-term effectiveness. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 This is a logical block diagram of the AI-based fault routing rapid location system of the present invention;

[0042] Figure 2 The flowchart shows the calculation process for the dynamic topological influence factor.

[0043] Figure 3 This is a flowchart for calculating fingerprint differences. Detailed Implementation

[0044] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0045] Example 1:

[0046] Please see Figures 1 to 3This invention provides an AI-based fault routing rapid location system, including: a full network flow topology data synchronous acquisition module, used to synchronously acquire service flow indicator data and network topology indicator data from network elements in the wireless communication network to obtain synchronized raw data;

[0047] The disturbance feature time-series extraction module is used to calculate the time change rate and short-term volatility of the service flow indicator data based on the synchronized original data, so as to generate a feature time series characterizing network state disturbances.

[0048] The flow topology coupling effect quantification module is used to combine the feature time series with the network topology index data to quantify the mutual influence strength between node pairs in the network, so as to generate a real-time fault fingerprint sequence characterizing the mutual influence strength.

[0049] The fault root cause reverse reasoning module is used to obtain a preset fault fingerprint database, which contains prototype fingerprints corresponding to various known fault types; calculate the fingerprint difference between the real-time fault fingerprint sequence and each prototype fingerprint in the fault fingerprint database; determine the prototype fingerprint with the smallest fingerprint difference, and use the fault type and root cause node corresponding to the prototype fingerprint with the smallest fingerprint difference as the location result to generate a fault location report.

[0050] This invention provides a systematic fault location framework. First, a network-wide flow topology data synchronous acquisition module captures network dynamics from a global perspective, avoiding the limitations of traditional methods that rely solely on isolated alarms or single network element performance indicators. Next, a disturbance feature time-series extraction module transforms raw performance data into features that better reflect network state changes, laying the foundation for in-depth analysis. Its core lies in a flow topology coupling effect quantification module, which innovatively constructs fault fingerprints, revealing the complex propagation patterns of faults across both service flows and network topology. Finally, a fault root cause reverse reasoning module compares the fault with pre-defined typical fault patterns, achieving precise tracing from phenomenon to root cause. The value of this framework lies in its ability to penetrate the chain reaction phenomena caused by faults, directly pinpointing the root cause and achieving root cause localization of network faults.

[0051] Example 2:

[0052] The perturbation feature temporal extraction module is specifically used for:

[0053] The throughput time series in the business flow index data is subjected to first-order difference calculation to obtain the throughput change rate series.

[0054] The fluctuation series of active users in the business flow metric data is obtained by calculating the standard deviation of the active user count using a sliding window.

[0055] The throughput change rate sequence and the active user number fluctuation sequence are combined to generate the feature time series;

[0056] The flow topology coupling effect quantification module is specifically used for:

[0057] Calculate the normalized flow perturbation correlation of the node pairs;

[0058] Calculate the dynamic topology influence factor of the node pair;

[0059] The normalized flow disturbance correlation and the dynamic topology influence factor are weighted and summed to obtain the flow topology coupling index;

[0060] A coupling effect matrix is ​​constructed based on the flow topology coupling index to generate the real-time fault fingerprint sequence;

[0061] In this embodiment, the perturbation feature temporal extraction module operates by transforming the original performance metrics into more sensitive feature metrics. It performs first-order difference calculation on throughput, thereby transforming the absolute value measure into a rate of change measure. This transformation enables the system to more sensitively capture instantaneous changes in performance. At the same time, by calculating the sliding window standard deviation of the number of active users, the short-term volatility of user behavior is effectively quantified. The ingenuity of this processing method lies in the fact that it filters out the stationary part of the performance metrics and amplifies the dynamic information that can characterize network perturbations.

[0062] The flow topology coupling effect quantification module integrates these disturbance information with network topology changes to construct a comprehensive impact metric. It calculates the flow topology coupling index and uses it to construct a coupling effect matrix, thereby generating a real-time fault fingerprint sequence describing the overall network relationships. The significance of this is that it no longer views the performance of each node in isolation, but rather quantifies the intensity of mutual influence between nodes when a fault occurs from the perspective of the entire network. This provides high-dimensional and highly identifiable input for subsequent fault tracing, significantly improving the accuracy of location.

[0063] The core of this module is the calculation of the flow topology coupling index, which is defined as follows:

[0064] ;

[0065] The flow topology coupling index is a dimensionless scalar value that represents the coupling strength between the perturbation states of node i and node j at time point t. The larger the value, the closer the association between the two nodes, and the more likely the perturbation is to propagate between them.

[0066] Let represent the flow weight and topology weight, two dimensionless positive constants, and These are adjustable hyperparameters of the system, used to balance the importance of traffic flow factors and network topology factors in the calculation of the coupling index; they are derived from initial settings based on expert experience and can be optimized through subsequent machine learning model training.

[0067] Represents the correlation of normalized flow disturbances, a dimensionless scalar value that quantifies the synchronization or latency correlation of service flow disturbances between node i and node j.

[0068] Represents the dynamic topology influence factor, a dimensionless scalar value that quantifies the tightness or drastic change of the topological connection between nodes i and j at time point t.

[0069] Subscripts i and j represent two different nodes in the network, such as the base station in the following text;

[0070] The subscript t represents the current time point in the calculation;

[0071] The technical motivation behind this formula is that neither simple service flow correlation nor topology correlation is sufficient to fully describe the propagation of a fault. A fault first occurs at the topology level, such as signal degradation, or at the service level, such as link interruption, and then propagates through the network's inherent mechanisms, intertwining at both levels. For example, a fault at node i may cause a decrease in the quality of user service flow under its coverage, i.e., flow disturbance. These users will try to switch to the neighboring node j, i.e., a topology event, which in turn causes service flow congestion at node j, i.e., a new flow disturbance. This formula creatively combines the propagation of flow disturbance with the influence of topology structure in a weighted manner, aiming to capture this complex causal transmission effect, thereby constructing a more identifiable fault fingerprint.

[0072] This module periodically calculates the values ​​of all monitored node pairs (i,j) in the network. Thus, a coupling effect matrix is ​​generated at each time snapshot t. The element in the i-th row and j-th column of the matrix is When a network failure occurs, the coupling index between affected node pairs changes. A unique, high-dimensional dynamic pattern is formed in the time series; this dynamic matrix sequence is the "flow-topology" perturbation propagation fingerprint defined in this invention. "Flow-topology" is an abbreviation for service flow and network topology, and their dynamic coupling relationship is the core basis for reverse reasoning of fault root causes in this invention; the output of this module is the fault fingerprint sequence. This will serve as the direct input to the final decision-making module;

[0073] Hyperparameters in the system, including And in the subsequent formulas All of these can be optimized using supervised learning methods. Specifically, using a labeled dataset containing known fault cases and their root causes, the final fault location accuracy is used as the optimization objective. Algorithms such as gradient descent are employed to adjust these parameters in reverse order to find the optimal combination. Specifically, since the accuracy is non-differentiable, a differentiable surrogate loss function, such as the cross-entropy loss function, is typically used in model training. The system treats the fault location problem as a multi-classification task (classifying the current fault fingerprint into a certain prototype fingerprint in the database). The gradient descent algorithm is driven by minimizing the cross-entropy between the fault type probability distribution predicted by the model and the actual fault label. End-to-end optimization of hyperparameters and other configuration parameters, such as... The parameters, etc., are set based on prior knowledge of network engineering to ensure the rationality of their physical meaning.

[0074] Example 3:

[0075] The calculation of the normalized flow disturbance correlation is as follows:

[0076] Within a preset time delay range, the feature time series of the node pair are slid, and for each delay time point, the Pearson correlation coefficient between the two series is calculated;

[0077] The maximum absolute value of the correlation coefficient among all delayed time points is selected as the normalized flow disturbance correlation.

[0078] The calculation of the dynamic topology influence factor is as follows:

[0079] The number of successful handovers between the node pairs within the current time window is obtained and divided by their historical average number of handovers to obtain a normalized handover behavior factor.

[0080] Obtain the reference signal received power from the source node reported by the user equipment of the target node, and subtract the preset signal quality reference threshold to obtain the signal quality difference factor;

[0081] The normalized handover behavior factor is multiplied by a preset handover impact weight, and the signal quality difference factor is multiplied by a preset signal quality impact weight.

[0082] The two products are then nonlinearly fused using a hyperbolic tangent function to generate the dynamic topological influence factor.

[0083] This embodiment further refines the calculation method for the two core components of the flow topology coupling index;

[0084] Normalized flow disturbance correlation The calculation aims to capture the causal relationship of business flow disturbances over time; its calculation formula is:

[0085] ;

[0086] This indicates that node i is within the time window. Time series of business flow disturbance characteristics within the region;

[0087] This indicates that node j has elapsed a time delay. The time window of the sliding feature time series is Both sequences originate from the output of the perturbation feature temporal extraction module;

[0088] This indicates the size of the time window, a system configuration parameter that defines the length of the time series considered when calculating correlation. Its origin is based on network service characteristics and fault response time, for example, 5 minutes.

[0089] Indicates a time delay, one in The variables searched within the range are used to find the time difference corresponding to the strongest correlation between two perturbation sequences;

[0090] This represents the maximum search latency, a system configuration parameter that indicates the maximum possible time for a disturbance to propagate in the network. It is estimated based on the network's physical size and protocol processing latency, for example, 2 seconds.

[0091] represents the Pearson correlation coefficient function, a standard statistical function known in the field, used to measure the degree of linear correlation between two variables, with output values ​​between [-1, 1].

[0092] The significance of this formula lies in finding the strongest correlation between the business flow disturbances of two nodes within a reasonable delay range, capturing the propagation relationship whether it is a positive correlation, such as congestion propagation, or a negative correlation, such as the ebb and flow of business.

[0093] Dynamic topological influence factor The calculation quantifies the strength and changes in network topology; its calculation formula is:

[0094] ;

[0095] Dynamic topological influence factor, a dimensionless scalar; Hyperbolic tangent function; These are the switching impact weight and the signal quality impact weight, respectively, which are dimensionless adjustable hyperparameters. : Normalized handover behavior factor, obtained by comparing the current number of handovers with the historical average; Standardized signal quality factor, dimensionless; Two different nodes in a network; : The current time point in the calculation;

[0096] Through the above-mentioned refined calculations, the system not only considers the correlation at the business level, but also incorporates topological factors at the physical and protocol layers, such as handover behavior and signal coverage, thereby constructing a more comprehensive and accurate model of inter-node coupling relationships.

[0097] Example 4:

[0098] The switching impact weight and the signal quality impact weight are hyperparameters obtained by training and optimizing a machine learning model based on historical fault case data.

[0099] The calculation of the fingerprint difference is as follows:

[0100] Within a preset fingerprint sequence length, the Frobenius norm between the real-time coupling effect matrix and the coupling effect matrix of the prototype fingerprint is calculated at each time point.

[0101] The fingerprint difference is obtained by taking the arithmetic mean of the Frobenius norm at all time points.

[0102] This embodiment reveals the system's intelligence and adaptive capabilities; the switching impact weight and signal quality impact weight are not statically configured, but are obtained through machine learning models trained and optimized on a large amount of historical fault case data; this means that the system can learn from real network operation and maintenance experience and automatically adjust parameters to achieve the best fault location effect;

[0103] Meanwhile, the method for calculating fingerprint dissimilarity is clearly defined; the system quantifies the distance between the real-time generated coupling effect matrix sequence and the prototype fingerprints in the fault database by calculating the Frobenius norm; this method is mathematically rigorous and can effectively measure the similarity between two high-dimensional matrix sequences; the specific calculation is as follows:

[0104] ;

[0105] Indicates the degree of fingerprint difference;

[0106] This represents the coupling effect matrix generated in real time;

[0107] This represents the prototype fingerprint matrix corresponding to the k-th fault type in the fault fingerprint database.

[0108] This indicates the length of the fingerprint sequence used for comparison and is a configuration parameter.

[0109] This represents the Frobenius norm, a standard matrix norm well-known in the art, used to calculate the difference between two matrices;

[0110] In this way, the system transforms the complex fault location problem into a clear pattern matching problem, namely, finding the known fault mode that is most similar to the current network state.

[0111] Example 5:

[0112] The fault root cause reverse reasoning module is also used for:

[0113] The calculated minimum fingerprint difference is compared with a preset confidence threshold.

[0114] When the minimum fingerprint difference is lower than the confidence threshold, the localization result is output;

[0115] When the minimum fingerprint difference is not lower than the confidence threshold, a report indicating an unknown fault type is output.

[0116] The confidence threshold is determined by plotting the subject operating characteristic curve on the validation dataset and based on preset positioning accuracy and recall targets.

[0117] This embodiment illustrates the system's decision-making logic, ensuring the reliability of the output results. After finding the prototype fingerprint with the highest matching degree, the fault root cause reverse reasoning module does not immediately output the result, but introduces a confidence threshold for secondary verification. Only when the minimum fingerprint difference is lower than this threshold, i.e., the similarity is high enough, does the system confirm the location result and output it. Conversely, if all known fault modes differ greatly from the current network state, the system will output a "fault type unknown" report. This avoids making misjudgments when there is insufficient evidence.

[0118] The scientific basis of this mechanism lies in the method for determining the confidence threshold; it is not set based on experience, but is scientifically determined by plotting the receiver operating characteristic (ROC) curve on the validation dataset. Operation and maintenance personnel can select the most suitable balance point on the ROC curve as the threshold according to different requirements for positioning accuracy and recall. This data-driven decision-making mechanism enables the final output of the system to achieve a quantifiable and adjustable balance between reliability and coverage, greatly enhancing the practicality and robustness of the system.

[0119] Suppose that in a 5G SA network composed of multiple gNodeBs (base stations), the transmission optical cable of gNodeB-1 (gNB-1) experiences jitter, resulting in intermittent packet loss and increased latency for the user plane services it carries.

[0120] Data Acquisition and Feature Extraction: The network-wide flow topology data synchronous acquisition module collected data such as a decrease in gNB-1 throughput, an increase in the number of active users in neighboring gNB-2 / gNB-3, and an increase in the number of handovers; the disturbance feature time series extraction module transformed these raw data into time series features such as the rate of change. ;

[0121] Coupling effect quantization: The flow topology coupling effect quantization module receives the above feature sequence and the original topology data;

[0122] calculate That is, between gNB-1 and gNB-2: the module calculates The high value indicates a strong latency correlation between the decrease in throughput of gNB-1 and the increase in the number of users of gNB-2. The value is also high due to the number of times gNB-1 switches to gNB-2. Surge; final The result was a significantly high value;

[0123] Similarly, calculate It is also a significantly high value;

[0124] Since there is no direct strong causal relationship between gNB-2 and gNB-3, the calculated... Relatively low;

[0125] Finally, the generated coupling effect matrix In the fingerprint, the elements related to gNB-1, namely the first row and the first column, show a sudden high value, forming a "star" divergent fingerprint pattern centered on gNB-1;

[0126] Reverse reasoning localization: The fault root cause localization module receives this "star-shaped" fault fingerprint sequence. ;

[0127] The module compared it with the fault fingerprint database and found that it matched the prototype fingerprint in the database marked as "single base station transmission failure". The difference D is the smallest and is less than the confidence threshold. ;

[0128] The system finally outputs the diagnostic conclusion: "Fault root cause location: gNB-1; Fault type: suspected transmission link fault (high confidence)"; Based on this, maintenance personnel can directly check the transmission link of gNB-1, which greatly shortens the fault handling time.

[0129] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. An AI-based fault routing rapid location system, characterized in that, include: The whole network flow topology data synchronization acquisition module is used to synchronously acquire service flow indicator data and network topology indicator data from network elements in the wireless communication network to obtain the synchronized raw data. The disturbance feature time series extraction module is used to calculate the time change rate and short-term volatility of business flow indicator data based on the synchronized raw data, so as to generate a feature time series that characterizes network state disturbances. The flow topology coupling effect quantification module is used to combine the feature time series with the network topology index data to quantify the mutual influence strength between node pairs in the network, so as to generate a real-time fault fingerprint sequence characterizing the mutual influence strength. The fault root cause reverse reasoning module is used to obtain a preset fault fingerprint database, which contains prototype fingerprints corresponding to various known fault types. It also calculates the fingerprint difference between the real-time fault fingerprint sequence and each prototype fingerprint in the fault fingerprint database; And determine the prototype fingerprint with the smallest fingerprint difference, and use the fault type and root cause node corresponding to the prototype fingerprint with the smallest fingerprint difference as the location result to generate a fault location report. The perturbation feature temporal extraction module is specifically used for: Perform first-order difference calculation on the throughput time series in the business flow indicator data to obtain the throughput change rate series; The fluctuation series of active users in the business flow metric data is obtained by calculating the standard deviation of the active user count using a sliding window. The throughput change rate sequence and the active user number fluctuation sequence are combined to generate the characteristic time series.

2. The AI-based fault routing rapid location system according to claim 1, characterized in that, The flow topology coupling effect quantification module is specifically used for: Calculate the normalized flow perturbation correlation of the node pairs; Calculate the dynamic topology influence factor of the node pair; The normalized flow disturbance correlation and the dynamic topology influence factor are weighted and summed to obtain the flow topology coupling index; A coupling effect matrix is ​​constructed based on the flow topology coupling index to generate the real-time fault fingerprint sequence.

3. The AI-based fault routing rapid location system according to claim 2, characterized in that, The calculation of the normalized flow disturbance correlation is as follows: Within a preset time delay range, the feature time series of the node pair are slid, and for each delay time point, the Pearson correlation coefficient between the two series is calculated; The maximum absolute value of the correlation coefficient among all delayed time points is selected as the normalized flow disturbance correlation.

4. The AI-based fault routing rapid location system according to claim 3, characterized in that, The calculation of the dynamic topology influence factor is as follows: The number of successful handovers between the node pairs within the current time window is obtained and divided by their historical average number of handovers to obtain a normalized handover behavior factor. Obtain the reference signal received power from the source node reported by the user equipment of the target node, and subtract the preset signal quality reference threshold to obtain the signal quality difference factor; The normalized handover behavior factor is multiplied by a preset handover impact weight, and the signal quality difference factor is multiplied by a preset signal quality impact weight. The two products are then nonlinearly fused using a hyperbolic tangent function to generate the dynamic topological influence factor.

5. The AI-based fault routing rapid location system according to claim 4, characterized in that, The switching impact weight and the signal quality impact weight are hyperparameters obtained by training and optimizing a machine learning model based on historical fault case data.

6. The AI-based fault routing rapid location system according to claim 1, characterized in that, The calculation of the fingerprint difference is as follows: Within a preset fingerprint sequence length, the Frobenius norm between the real-time coupling effect matrix and the coupling effect matrix of the prototype fingerprint is calculated at each time point. The fingerprint difference is obtained by taking the arithmetic mean of the Frobenius norm at all time points.

7. The AI-based fault routing rapid location system according to claim 1, characterized in that, The fault root cause reverse reasoning module is also used for: The calculated minimum fingerprint difference is compared with a preset confidence threshold. When the minimum fingerprint difference is lower than the confidence threshold, the localization result is output; When the minimum fingerprint difference is not lower than the confidence threshold, a report indicating an unknown fault type is output.

8. The AI-based fault routing rapid location system according to claim 7, characterized in that, The confidence threshold is determined by plotting the subject operating characteristic curve on the validation dataset and based on preset positioning accuracy and recall targets.

Images