Communication network hidden fault detection disaster recovery method and device

Abstract

The invention provides a communication network hidden fault detection disaster recovery method and device, and the method comprises the steps: receiving a calling session initiation request through a physical signaling forwarding layer composed of a session boundary controller, and transmitting the calling session initiation request to an operator physical mobile switching center to complete the actual call connection and obtain the actual response delay; and inputting service, network and environment side data corresponding to the request into a dialing delay prediction model pre-trained based on a Transform architecture in a digital twinborn layer of the virtual mobile switching center to obtain theoretical response delay. And then dynamic deviation and preset statistical characteristics of the two are calculated, normal congestion or hidden faults are identified according to rules, if the hidden faults are judged, disaster recovery operation is triggered to generate routing control instructions, the routing control instructions are sent to the session boundary controller, and subsequent requests are forwarded to the disaster recovery nodes by the session boundary controller, so that hidden fault detection and disaster recovery are realized. According to the scheme, normal congestion and hidden faults can be accurately distinguished, the faults are recognized in advance, routing switching is triggered, and the fault detection accuracy and disaster recovery timeliness are greatly improved.

Description

Technical Field

[0001] This invention relates to the field of communication network management technology, and in particular to a method and apparatus for detecting and mitigating latent faults in communication networks. Background Technology

[0002] In modern communication service scenarios such as cloud communication privacy numbers, the cloud platform, acting as middleware connecting the calling and called parties, relies heavily on the stable operation of the underlying basic operator's physical core network equipment, such as mobile switching centers, for high availability. With the periodic surges and sudden fluctuations in service concurrency, physical core network equipment often experiences performance jitter or even failures. Traditional technologies primarily rely on passive monitoring and circuit breaker mechanisms based on status codes, or static rules based on fixed performance thresholds for fault detection and switching. However, these methods have inherent drawbacks. For example, error code-based solutions have significant lag, only triggering a response after a fault has occurred and caused numerous service failures, failing to achieve preventative disaster recovery. Fixed threshold-based solutions struggle to adapt to the inherent tidal effects of communication networks; normal latency spikes during peak periods can easily lead to misjudgments, causing unnecessary routing switches and system instability. Furthermore, while network function virtualization (NFV) technology can replace some physical device functions in software form, it primarily serves as a hardware resource substitute and does not establish an independent reference system that can dynamically predict the ideal performance of devices, thus failing to accurately identify and assess early performance degradation of physical equipment.

[0003] Therefore, existing technologies lack an effective means to identify hidden network faults in advance, accurately, and adaptively, and to perform intelligent disaster recovery under complex load environments such as peak business periods. Summary of the Invention

[0004] In view of this, embodiments of the present invention provide a method and apparatus for detecting and mitigating hidden faults in communication networks, so as to eliminate or improve one or more defects existing in the prior art, and solve the problem that the prior art cannot accurately distinguish between normal congestion and hidden equipment faults under dynamic load and quickly mitigate disasters.

[0005] One aspect of the present invention provides a method for detecting and mitigating latent faults in communication networks. The method is executed based on a cloud communication platform and includes the following steps: Upon receiving a session initiation request initiated by the calling party, the session initiation request is sent to the operator's physical mobile switching center based on the physical signaling forwarding layer composed of session boundary controllers, the actual call setup is performed, and the actual response delay is obtained. The service-side data containing the session initiation request, the network-side data marking the network status, and the environmental data marking the external environment are input into the dialing delay prediction model pre-loaded in the digital twin layer of the virtual mobile switching center and the output is the theoretical dialing response delay; the dialing delay prediction model is obtained by pre-training based on the Transformer model architecture. Calculate the dynamic deviation between the actual response delay and the theoretical response delay, and calculate one or more preset statistical features of the dynamic deviation. Based on the preset statistical features, identify normal congestion state or latent fault state. In the state of the latent fault, a disaster recovery operation is triggered to generate a routing control instruction, which is then sent to the session boundary controller to forward the subsequent session initiation request to the disaster recovery node.

[0006] In some embodiments, the service-side data includes call type, call direction, calling terminal type, timestamp, current region, and operator information; The network-side data includes the current number of concurrent calls, as well as the CPU utilization, memory utilization, and signaling link load of the physical mobile switching center; The environmental data includes holiday markers, weather data, and large-scale event markers.

[0007] In some embodiments, the pre-training step of the dialing delay prediction model includes: Collect historical signaling logs of the communication network, which include multiple call samples from the target region during multiple sampling periods throughout the year. Each call sample contains the service-side data, the network-side data, and the environmental data. The call delay of each call sample is labeled as a tag to construct a training sample set. The training sample set is used to train an initial neural network model based on the Transformer model architecture. The initial neural network model includes a continuous input mapping layer, a multi-head self-attention mechanism layer, a residual connection layer, a normalization layer, and a feedforward neural network layer. The feedforward neural network adopts a multilayer perceptron. The initial neural network model takes the service-side data, the network-side data, and the environmental data as inputs and outputs a predicted value for the call delay. Based on the deviation between the predicted value and the label, a loss is constructed to update the parameters of the initial neural network model, thereby obtaining the dialing delay prediction model.

[0008] In some embodiments, the dynamic deviation between the actual response delay and the theoretical response delay is calculated, and one or more preset statistical characteristics of the dynamic deviation are calculated. Based on the preset statistical characteristics and predefined rules, normal congestion states or latent fault states are identified, including: The dynamic deviation is compared with a set threshold. If the deviation is less than the set threshold, it is determined to be a normal congestion state. If the deviation is greater than or equal to the set threshold, it is determined to be a latent fault state. A call log is established to record the time points of the normal congestion state and the latent fault state.

[0009] In some embodiments, the dynamic deviation between the actual response delay and the theoretical response delay is calculated, and one or more preset statistical characteristics of the dynamic deviation are calculated. Based on the preset statistical characteristics and predefined rules, normal congestion states or latent fault states are identified, including: The moving average, median, moving standard deviation, excess rate and / or number of consecutive excesses within a set time window are used as the preset statistical features for the dynamic deviation statistics. The preset statistical features are assigned values ​​using a preset standard value range, and a weighted sum is obtained based on preset weights to obtain a state score. The normal congestion state or the latent fault state is identified based on the state score.

[0010] In some embodiments, triggering a disaster recovery operation to generate a routing control instruction under the latent fault state, and sending the routing control instruction to the session boundary controller to forward subsequent session initiation requests to the disaster recovery node, includes: Generate the routing control command containing the faulty physical mobile switching center identifier, the target disaster recovery node identifier, the handover policy parameters, the command ID, and the timestamp; the handover policy parameters are used to mark a full handover or a handover according to a first set ratio; The routing control command is forwarded to the session boundary controller via an internal message queue based on a dedicated channel, and the session boundary controller updates the routing table and executes the routing switching policy. The routing switching strategy includes: performing a gradual traffic switching, switching to the first set ratio step by step according to the second set ratio and duration.

[0011] In some embodiments, the method further includes: classifying the latent fault state into three categories—minor fault, moderate fault, and severe fault—based on the preset statistical characteristics; logging and alerting the type of minor fault, without performing route switching; performing progressive traffic switching for the type of moderate fault; and performing full switching for the type of severe fault.

[0012] In some embodiments, the method further includes: when the routing switching policy is a full switch, reserving 1% to 5% of session traffic for forwarding via the original physical mobile switching center as a routing node, and continuously monitoring the actual response delay; when the cooling time is reached after the routing switching policy is executed, determining whether the actual response delay of the original physical mobile switching center meets the normal congestion state; if it does, performing a gradual traffic switchback back to the original physical mobile switching center; otherwise, continuing to maintain the status quo.

[0013] On the other hand, the present invention also provides a communication network hidden fault detection and disaster recovery device, including a processor, a memory, and a computer program or instructions stored in the memory, characterized in that the processor is used to execute the computer program or instructions, and when the computer program or instructions are executed, the device implements the steps of the above method.

[0014] On the other hand, the present invention also provides a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method.

[0015] The communication network latent fault detection and disaster recovery method and apparatus of this invention receive the caller's session initiation request through a physical signaling forwarding layer composed of a session boundary controller and send it to the operator's physical mobile switching center to complete the actual call connection, while obtaining the actual response delay. Then, the service-side, network-side, and environment-side data corresponding to the request are input into a dialing delay prediction model pre-trained based on the Transformer model architecture in the digital twin layer of the virtual mobile switching center to obtain the theoretical dialing response delay. Subsequently, the dynamic deviation between the actual and theoretical response delays and their preset statistical characteristics are calculated, and the network is identified as being in a normal congestion or latent fault state according to set rules. If a latent fault state is determined, a disaster recovery operation is triggered to generate routing control instructions and send them to the session boundary controller, which then forwards subsequent session initiation requests to the disaster recovery node, thereby realizing the detection and disaster recovery of latent faults in the communication network. This invention can accurately distinguish between normal congestion and latent faults, identify faults in advance, and trigger disaster recovery routing switching, significantly improving the accuracy of communication network fault detection and the timeliness of disaster recovery.

[0016] Furthermore, by limiting the specific forms of business-side data, network-side data, and environmental data, multi-dimensional factors affecting communication network latency are incorporated into the model input, making the prediction of theoretical response latency more consistent with actual network operation scenarios, significantly improving prediction accuracy, and laying a data foundation for accurately distinguishing between normal congestion and latent faults in the future.

[0017] Furthermore, threshold comparison enables rapid fault identification, reduces the computational complexity of decision logic, and improves the efficiency of hidden fault detection. The establishment of call logs can retain time information of fault and normal states, providing data support for subsequent tracing of network fault causes and analysis of network operation patterns, and also facilitates the optimization and adjustment of thresholds.

[0018] Furthermore, judging the network state based on multiple statistical characteristics of dynamic deviation breaks through the limitations of single threshold judgment. Analyzing dynamic deviation from multiple dimensions can more comprehensively and objectively reflect the actual operating state of the network and effectively avoid misjudgment caused by single deviation fluctuations.

[0019] Furthermore, by implementing graded management of latent faults, appropriate handling methods are adopted for faults of different severity levels. This avoids the waste of network resources and service fluctuations caused by full switching for minor faults, while ensuring that severe faults can be handled quickly, minimizing the impact of faults on the communication network, and improving the utilization efficiency of network resources and the flexibility of disaster recovery strategies.

[0020] Additional advantages, objects, and features of the invention will be set forth in part in the description which follows, and will also become apparent in part to those skilled in the art upon studying the description, or may be learned by practice of the invention. The objects and other advantages of the invention can be realized and obtained by means of the structures specifically pointed out in the description and drawings.

[0021] Those skilled in the art will understand that the objectives and advantages achievable with the present invention are not limited to those specifically described above, and that the above and other objectives achievable with the present invention will become clearer from the following detailed description. Attached Figure Description

[0022] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, are not intended to limit the scope of the invention. In the drawings: Figure 1 This is a schematic diagram of the structure of a communication network latent fault detection and disaster recovery method according to an embodiment of the present invention.

[0023] Figure 2 This is a logical schematic diagram of a communication network latent fault detection and disaster recovery method according to an embodiment of the present invention.

[0024] Figure 3 This is a parallel deduction and routing control timing diagram of the communication network latent fault detection and disaster recovery method according to an embodiment of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the embodiments and accompanying drawings. Here, the illustrative embodiments and descriptions of this invention are used to explain the invention, but are not intended to limit the invention.

[0026] It should also be noted that, in order to avoid obscuring the invention with unnecessary details, only the structures and / or processing steps closely related to the solution according to the invention are shown in the accompanying drawings, while other details that are not closely related to the invention are omitted.

[0027] It should be emphasized that the term "including / comprises" as used herein refers to the presence of a feature, element, step, or component, but does not exclude the presence or addition of one or more other features, elements, steps, or components.

[0028] Existing technologies for communication network fault detection and disaster recovery mainly rely on two approaches: one is passively monitoring error codes returned by physical devices and switching over only after a fault occurs, which has significant latency; the other is judging performance based on fixed thresholds, such as a delay greater than 2 seconds after dialing. However, this approach cannot adapt to network traffic fluctuations and often generates false alarms during peak periods due to normal latency increases, causing unnecessary routing oscillations. Furthermore, although solutions exist that use network function virtualization (NFV) technology to software-enable devices, these merely serve as replacements for physical devices and fail to establish an independent reference system that can dynamically predict ideal performance. Therefore, they cannot effectively identify and warn of early, latent performance degradation in devices. Overall, existing technologies lack an effective means to accurately and proactively distinguish between normal service congestion and latent device faults under dynamically changing real network loads, and to achieve rapid, adaptive disaster recovery.

[0029] It should be noted beforehand that the MSC mentioned in this application refers to the Mobile Switching Center, a key physical network element in the communication core network responsible for call control and signaling switching. A virtual MSC refers to a digital twin prediction model based on deep learning. It does not undertake actual call setup functions but serves as a logical reference system to output the theoretically expected performance indicators of the physical MSC under the current load. Latent faults refer to a micro-congestion state where the communication network has not been completely interrupted, such as not returning 503 / 408 error codes, but performance exhibits unexpected degradation such as latency drift and voice quality decline. SBC refers to the Session Border Controller, located at the edge of the cloud communication platform, responsible for signaling reception, forwarding, and routing control.

[0030] One aspect of the present invention provides a method for detecting and mitigating latent faults in communication networks, such as... Figure 1 and Figure 2 As shown, the method is executed based on a cloud communication platform, and includes the following steps S101~S104: Step S101: Receive the session initiation request initiated by the calling party, send the session initiation request to the operator's physical mobile switching center based on the physical signaling forwarding layer composed of session boundary controllers, perform actual call setup, and obtain the actual response delay.

[0031] Step S102: Input the service-side data containing the session initiation request, the network-side data marking the network status, and the environmental data marking the external environment into the dialing delay prediction model preloaded in the digital twin layer of the virtual mobile switching center and output the theoretical dialing response delay; the dialing delay prediction model is obtained by pre-training based on the Transformer model architecture.

[0032] Step S103: Calculate the dynamic deviation between the actual response delay and the theoretical response delay, and calculate one or more preset statistical features of the dynamic deviation. Based on the preset statistical features, identify normal congestion state or latent fault state.

[0033] Step S104: Trigger disaster recovery operation to generate routing control instructions under the condition of latent failure, so as to send routing control instructions to the session boundary controller to forward subsequent session initiation requests to the disaster recovery node.

[0034] In step S101, when the Session Border Controller (SBC) of the cloud communication platform receives a SIP Session Initiation Request (SIP INVITE) initiated by the calling party, it, as the core of the physical signaling forwarding layer, routes the request to the Physical Mobile Switching Center (MSC) of the corresponding operator for actual call setup. During this process, the system records key signaling events using high-precision timestamps and accurately calculates the actual response delay. This ensures the system's synchronous monitoring of the actual performance of the current communication network, providing a benchmark data source for subsequent comparisons.

[0035] In step S102, three types of data are extracted and constructed simultaneously: service-side data, network-side data, and environmental data. Service-side data includes call type, call direction, calling terminal type, timestamp, current region, and operator information. Network-side data includes the current number of concurrent calls, as well as the CPU utilization, memory utilization, and signaling link load of the physical mobile switching center. Environmental data includes holiday identifiers, weather data, and large-scale event identifiers.

[0036] These multi-dimensional features constitute a panoramic view of the current network state and are fed in real time into a digital twin model, i.e., a virtual mobile switching center, pre-trained based on the Transformer architecture. This model, through its multi-head self-attention mechanism, deeply mines the complex nonlinear spatiotemporal relationships between load, time, resource allocation, and latency in historical data, thereby outputting the theoretical response latency that a physical mobile switching center should theoretically possess under the current specific context. The purpose of this step is to generate a dynamic, adaptive performance benchmark, replacing the rigid, fixed thresholds found in traditional approaches.

[0037] In some embodiments, the pre-training steps of the dialing delay prediction model include S1021 and S1022: Step S1021: Collect historical signaling logs of the communication network, which include multiple call samples from multiple sampling periods throughout the year for the target region. Each call sample contains service-side data, network-side data, and environmental data. The call delay of each call sample is labeled as a tag to construct a training sample set.

[0038] Step S1022: Train the initial neural network model based on the Transformer model architecture using the training sample set. The initial neural network model includes a continuous input mapping layer, a multi-head self-attention mechanism layer, a residual connection layer, a normalization layer, and a feedforward neural network layer. The feedforward neural network adopts a multilayer perceptron. The initial neural network model takes business-side data, network-side data, and environmental data as inputs and outputs a predicted value for call delay. Based on the deviation between the predicted value and the label, a loss is constructed to update the parameters of the initial neural network model, resulting in a dialing delay prediction model.

[0039] Steps S1021 and S1022 learn and extract the complex, non-linear mapping patterns between network load, resource configuration, external environment, and call latency from massive, multi-dimensional historical real-world data. Step S1021 first collects historical logs covering long-term periods and target regions from the cloud platform's signaling processing system. Through data cleaning, alignment, and labeling, each successful call instance is constructed as a structured training sample. This sample not only contains service-side metadata such as the calling and called parties, time, and operator, but also associates it with real-time network performance indicators and environmental context information at that moment. The precise time difference from signaling initiation to critical response is used as the label, thus forming a high-quality supervised learning dataset that comprehensively reflects the network's performance under various conditions. Step S1022 uses this dataset to drive the training of a deep learning model based on the Transformer architecture. The key to its implementation lies in the model's structural design: the input mapping layer first encodes heterogeneous features into a unified high-dimensional vector; subsequently, the encoder core, composed of multiple stacked multi-head self-attention mechanisms and feedforward neural network layers, can analyze the entire input feature sequence in parallel and efficiently, automatically identifying and establishing long-range dependencies across time and dimensions, and ensuring stable convergence of the training process through residual connections and layer normalization; finally, the regression output layer maps this deep spatiotemporal pattern learned by the model into a single delay prediction value. The training process iteratively updates the model parameters by continuously comparing the deviation between the model's prediction value and the true label, using the backpropagation algorithm, until it can robustly deduce a theoretical delay approaching the ideal performance of the physical network from complex features. The dialing delay prediction model ultimately obtained by this invention is not a simple fitting tool, but a digital twin brain with deep spatiotemporal understanding capabilities. This brain can dynamically and adaptively generate an accurate, context-aware performance benchmark in the face of real-time inflow of multidimensional state information during the online deduction phase. This provides an intelligent reference system for subsequent deviation calculation and fault determination, enabling a shift from fixed threshold false alarms to dynamic and accurate early warning.

[0040] Step S103 calculates the dynamic deviation, using the following formula: This deviation value isolates latency fluctuations caused by normal factors such as business activity fluctuations, directly reflecting the degree of abnormal deviation in the performance of the physical equipment itself. To further improve the robustness of the judgment, preset statistical characteristics, such as moving average, standard deviation, trend slope, or number of consecutive exceedances, are calculated for the continuously generated Δ value sequence, and a comprehensive judgment is made based on the set rules. Through this step, the system can accurately distinguish between normal network congestion and hidden equipment faults, achieving early and accurate fault identification.

[0041] In some embodiments, the dynamic deviation between the actual response delay and the theoretical response delay is calculated, and one or more preset statistical features of the dynamic deviation are calculated. Based on the preset statistical features, normal congestion state or latent fault state is identified, including: comparing the dynamic deviation with a preset threshold; if the deviation is less than the preset threshold, it is determined to be a normal congestion state; if the deviation is greater than or equal to the preset threshold, it is determined to be a latent fault state; and establishing a call log to record the time points of the normal congestion state and the latent fault state.

[0042] This embodiment effectively eliminates the interference of normal fluctuations such as service surges on network performance judgment by utilizing a Transformer-based digital twin prediction model that runs in parallel with the actual physical network to generate dynamic performance benchmarks in real time. This shifts the focus of fault detection from absolute latency levels to whether performance deviates from its theoretical expected level, achieving high-precision, low-false-report early identification of latent faults in communication networks. Furthermore, by analyzing the statistical characteristics of deviation values ​​and automatically triggering graded and progressive intelligent routing switching, it upgrades the traditional reactive circuit breaker based on error codes to proactive disaster recovery based on performance deviation prediction. This achieves a closed-loop self-healing process at the minute or even second level, from fault perception and decision-making to execution. Ultimately, this significantly improves the service availability, operational intelligence, and overall service reliability of the communication system under complex dynamic loads.

[0043] In some embodiments, the dynamic deviation between the actual response delay and the theoretical response delay is calculated, and one or more preset statistical features of the dynamic deviation are calculated. Based on the preset statistical features, normal congestion state or latent fault state is identified according to the set rules. This includes: using the moving average, median, moving standard deviation, excess ratio and / or number of consecutive excesses within the time window range of the dynamic deviation as preset statistical features; assigning values ​​to the preset statistical features using a preset standard value range interval, performing weighted summation based on preset weights to obtain a state score, and identifying normal congestion state or latent fault state based on the state score.

[0044] This embodiment significantly improves the robustness and accuracy of fault status identification by introducing a weighted scoring mechanism based on multi-dimensional statistical features. By comprehensively calculating multi-dimensional features such as the moving average, median, moving standard deviation, exceedance rate, and number of consecutive exceedances of dynamic deviation within a set time window, and using preset value ranges and weights for standardized weighted fusion, the system can construct a comprehensive, quantitative, and interpretable network health status score. This method effectively overcomes the shortcomings of single-threshold judgment, which is easily affected by instantaneous jitter or misjudgment of periodic congestion. It makes the distinction between normal congestion and latent faults no longer dependent on a single isolated data point, but rather on intelligent decision-making based on the overall pattern and statistical regularity of deviation behavior over a period of time. This significantly reduces false alarms and false negatives, achieving more stable and reliable automated fault diagnosis.

[0045] In step S104, once a latent fault state is determined, the disaster recovery process is automatically triggered immediately. It generates structured routing control instructions, which are sent to the session boundary controller via a reliable channel. The session boundary controller then dynamically updates its routing table according to the strategy in the instructions, such as progressive traffic switching or tiered disaster recovery, smoothly and quickly redirecting subsequent call requests to pre-prepared off-site disaster recovery nodes. This achieves a minute-level or even second-level automated closed loop from fault detection to service recovery, fundamentally solving the problem of long service interruption times and slow recovery caused by existing technologies relying on manual intervention or passive circuit breakers.

[0046] In some embodiments, triggering disaster recovery operations under a latent fault state generates routing control instructions to send routing control instructions to the session boundary controller to forward subsequent session initiation requests to the disaster recovery node. This includes generating routing control instructions containing the faulty physical mobile switching center identifier, the target disaster recovery node identifier, handover policy parameters, instruction ID, and timestamp; the handover policy parameters are used to mark a full handover or a handover according to a first predetermined ratio. The routing control instructions are forwarded to the session boundary controller via an internal message queue on a dedicated channel, and the session boundary controller updates the routing table and executes the routing handover policy; wherein, the routing handover policy includes: performing a gradual traffic handover, switching step-by-step to the first predetermined ratio according to a second predetermined ratio step size and duration.

[0047] This implementation achieves high reliability, precise control, and smooth service transition during the disaster recovery process by designing structured disaster recovery commands and controllable execution strategies. First, by generating routing control commands containing complete numerical parameters, it ensures precise targeting and full-process traceability of disaster recovery actions. Second, the command issuance mechanism based on dedicated channels and internal message queues guarantees high reliability and low latency in control signal transmission, avoiding handover failures caused by command loss or delays. By adopting a routing handover strategy centered on gradual traffic switching, the system can migrate traffic step-by-step according to preset steps and durations, thereby avoiding the impact on the disaster recovery node caused by a sudden complete traffic switch and potential secondary service disruptions. This achieves a smooth, stable, and controllable service takeover from the faulty node to the disaster recovery node, ensuring high availability while maximizing the continuity of user experience.

[0048] In some embodiments, the method further includes: classifying latent fault states into three categories—minor faults, moderate faults, and severe faults—based on preset statistical characteristics; logging and alerting for minor fault types without performing route switching; performing gradual traffic switching for moderate fault types; and performing full switching for severe fault types.

[0049] This embodiment significantly improves the precision and resource utilization efficiency of communication network operation and maintenance by introducing a severity-based hierarchical disaster recovery response mechanism. Specifically, multiple predefined value ranges based on preset statistical characteristics are matched to three categories of faults: minor, moderate, and severe. After obtaining the preset statistical characteristics, the corresponding fault category can be directly determined. The system can intelligently distinguish the severity level of faults based on statistical analysis of dynamic deviations: for minor faults with only slight deviations, only recording and alarms are generated without triggering a switchover, avoiding operational interference and resource waste caused by overreaction; for moderate faults with clear and continuous abnormal signs, a gradual traffic switchover is adopted, ensuring a smooth service transition while verifying the disaster recovery node's capabilities, achieving controllable risk recovery; and for severe faults with severely degraded performance, a full switchover is immediately executed to maximize service continuity. This hierarchical strategy enables the system to perform disaster recovery in the optimal way that matches the actual impact of the fault, effectively reducing unnecessary routing oscillations and operational costs while ensuring high availability.

[0050] In some embodiments, the method further includes: when the routing switching policy is a full switch, reserving 1% to 5% of session traffic to be forwarded via the original physical mobile switching center as a routing node, and continuously monitoring the actual response delay; when the cooling time is reached after the routing switching policy is executed, determining whether the actual response delay of the original physical mobile switching center meets the normal congestion state; if it does, performing a gradual traffic switchback to the original physical mobile switching center; otherwise, continuing to maintain the status quo.

[0051] This embodiment achieves a qualitative leap in disaster recovery by retaining a small amount of probe traffic during a full switchover and introducing an intelligent rollback mechanism, transforming the process from a one-way switch to a closed-loop self-healing system. When a full switchover is performed due to a severe fault, 1% to 5% of session traffic is retained to continuously pass through the original faulty node, which is equivalent to implanting a real-time performance probe. This allows the system to continuously obtain the true recovery status data of the original node without affecting the stability of the main business, thus solving the problems of blind switching and delayed rollback in traditional disaster recovery. Combined with cooling time and automatic judgment based on real-time performance data, after confirming that the performance of the original node has indeed stabilized and recovered to a normal congestion state, it can proactively and safely perform gradual traffic rollback. This ensures that the fault is completely eliminated and automatically optimizes the service routing back to the optimal path, which not only significantly reduces the cost of manual operation and maintenance intervention but also realizes the dynamic optimal utilization of network resources and intelligent guarantee of service continuity.

[0052] On the other hand, the present invention also provides a communication network hidden fault detection and disaster recovery device, including a processor, a memory, and a computer program or instructions stored in the memory, characterized in that the processor is used to execute the computer program or instructions, and when the computer program or instructions are executed, the device implements the steps of the above method.

[0053] On the other hand, the present invention also provides a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method.

[0054] The present invention will now be described with reference to a specific embodiment: This embodiment provides a method for detecting and mitigating latent faults in communication networks based on real-time digital twin parallel simulation. Based on a Turing test-style real-time comparison, when each real call is initiated, it is not only sent to the physical network but also its characteristics are input into the virtual twin model. The virtual twin model, acting as a judgment subject unaffected by physical interference, calculates the theoretical delay of each real call in real time, i.e., the theoretical connection time. The method compares the deviation between the actual response delay and the theoretical response delay (…). This method isolates the impact of environmental factors such as peak seasons and holidays, allowing for precise identification of equipment malfunctions. Specifically, the system composition of this embodiment is described in [link to system composition documentation]. Figure 1The system comprises a physical signaling forwarding layer, a parallel digital twin layer, and a differential decision layer. The physical signaling forwarding layer includes an SBC cluster and a traffic mirroring module, responsible for actual interaction with the operator's physical MSC. The parallel digital twin layer includes a spatiotemporal sequence prediction engine and a virtual signaling generator; the spatiotemporal sequence prediction engine deploys a pre-trained Transformer model to generate a dynamic baseline based on the current time slot, concurrency per second (CPS), and historical patterns. The virtual signaling generator simulates the MSC's signaling interaction logic. The differential decision layer includes a deviation calculation unit and a routing controller, responsible for calculating... Then, the routing instructions are issued. The specific steps are as follows: Step S1: In the offline training phase, construct a predictive virtual MSC model for physical mobile switching centers, i.e., a digital twin prediction model.

[0055] Based on 6-12 months of signaling logs from the cloud platform, a feature vector X = {timestamp, region, operator, current concurrent call count, historical average post-dialing latency} is extracted. A model M is trained using the Transformer algorithm to fit the load-response curve of the physical MSC. The model's goal is to output the theoretical response latency under a specific load. .

[0056] Step S2: As Figure 3 As shown, during the online simulation phase, the call request is processed in a dual-track parallel manner.

[0057] When the Session Border Controller (SBC) receives a SIP INVITE request initiated by the calling party, the Traffic Mirroring module copies the request.

[0058] Path A (real path): The original request is sent to the operator's physical mobile switching center for actual call setup.

[0059] Path B (virtual path): The mirroring characteristic parameters of the original request are input into the virtual MSC model of the virtual mobile switching center. This process is strictly parallel to Path A in real time.

[0060] Step S3: As Figure 3 As shown, implicit fault identification based on deviation values ​​is performed, capturing the actual time taken for the physical MSC to return critical signaling (such as 180 Ringing). And the theoretical time consumption derived from the virtual MSC model. Calculate the dynamic deviation value: .

[0061] During the judgment process, if It's very tall, but It is also very high, then If the congestion level is below the preset standard, it is considered normal peak congestion and no intervention is taken. For example... It takes 500ms. If the timeframe is 480ms, the deviation is 20ms, which is less than the preset standard of 100ms. Therefore, it is considered normal peak congestion and no intervention is initiated. It's very tall, but Very low, then The magnitude is significant, classifying it as a latent fault, meaning an abnormal degradation in the performance of the physical equipment; for example... It takes 500ms. If the value is 50ms, then the deviation is 450ms, which is greater than the preset standard of 100ms, and is judged as a latent fault.

[0062] Step S4: As Figure 3 As shown, dynamic routing takeover and self-healing occur when there are N consecutive calls. When the value exceeds the preset confidence interval, the differentiated decision layer triggers a disaster recovery instruction. The SBC dynamically adjusts the routing weights according to the instruction, smoothly migrating subsequent calls to remote disaster recovery nodes, such as switching from the Beijing node to the Hebei node, achieving self-repair within minutes.

[0063] This embodiment introduces a virtual MSC as a dynamic reference frame. When the overall network is busy, the prediction value of the virtual model also increases, thereby offsetting the increase in physical network latency. The value remains stable. This effectively avoids false alarms caused by the natural increase in latency during peak periods in existing technologies, improving diagnostic accuracy to 98%. By monitoring behavioral deviations, it can identify and switch over in advance when physical devices exhibit latent faults but have not yet refused service, advancing the fault detection window by 3-5 minutes. This embodiment uses a bypass mirror of the actual service flow for calculation, eliminating the need to send additional active probe packets to the operator network and thus avoiding increased network load.

[0064] Corresponding to the above method, the present invention also provides an apparatus / system including a computer device, the computer device including a processor and a memory, the memory storing computer instructions, the processor executing the computer instructions stored in the memory, and when the computer instructions are executed by the processor, the apparatus / system performs the steps of the method as described above.

[0065] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the aforementioned edge computing server deployment method. The computer-readable storage medium can be a tangible storage medium, such as random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, floppy disks, hard disks, removable storage disks, CD-ROMs, or any other form of storage medium known in the art.

[0066] In summary, the communication network latent fault detection and disaster recovery method and device of the present invention rely on a dual-track parallel architecture of physical signaling forwarding layer and virtual mobile switching center digital twin layer. While processing actual call connection and obtaining actual physical side response delay, it inputs multi-dimensional data of service, network and environment corresponding to the call request into a dialing delay prediction model pre-trained based on Transformer architecture to obtain theoretical response delay. By calculating the dynamic deviation between the two and preset statistical characteristics, the network status is determined. If a latent fault is identified, disaster recovery operation is triggered. The session boundary controller performs routing switching of the disaster recovery node, forming a complete technical closed loop including parallel inference, deviation analysis, fault identification and dynamic disaster recovery.

[0067] This invention, on the one hand, accurately isolates the influence of environmental factors such as network tidal effects through parallel inference using digital twins and multi-dimensional data-driven prediction models, effectively distinguishing between normal congestion and latent faults, and significantly improving the accuracy of fault detection. On the other hand, it can identify latent faults in advance without waiting for error codes to appear in the physical network, and combined with rapid dynamic routing switching, it significantly improves the timeliness of disaster recovery response, achieving preventive disaster recovery. In addition, it reduces computational complexity and improves fault detection efficiency through a simplified judgment method using threshold comparison, while relying on call logs to retain status information, providing data support for network operation and maintenance analysis and threshold optimization. It judges network status based on multi-dimensional statistical characteristics of dynamic deviation, breaking through the limitations of a single threshold and avoiding misjudgments caused by single deviation fluctuations, making status identification more comprehensive and objective. Through a hierarchical management strategy for latent faults, it matches differentiated processing methods, avoiding the waste of network resources and business shocks caused by full switching for minor faults, while quickly handling severe faults, minimizing the impact of faults, and improving network resource utilization and the flexibility of disaster recovery strategies.

[0068] Those skilled in the art will understand that the exemplary components, systems, and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software, or a combination of both. Whether implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this invention. When implemented in hardware, it can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this invention are programs or code segments used to perform the desired tasks. The programs or code segments can be stored in a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried in a carrier wave.

[0069] It should be clarified that the present invention is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of the present invention.

[0070] In this invention, features described and / or exemplified for one embodiment may be used in the same or similar manner in one or more other embodiments, and / or combined with or replace features of other embodiments. The above descriptions are merely preferred embodiments of the present invention and are not intended to limit the invention. For those skilled in the art, various modifications and variations of the embodiments of the present invention are possible. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for detecting and mitigating latent faults in communication networks, characterized in that, The method is executed based on a cloud communication platform and includes the following steps: Upon receiving a session initiation request initiated by the calling party, the session initiation request is sent to the operator's physical mobile switching center based on the physical signaling forwarding layer composed of session boundary controllers, the actual call setup is performed, and the actual response delay is obtained. The service-side data containing the session initiation request, the network-side data marking the network status, and the environmental data marking the external environment are input into the dialing delay prediction model pre-loaded in the digital twin layer of the virtual mobile switching center and the output is the theoretical dialing response delay; the dialing delay prediction model is obtained by pre-training based on the Transformer model architecture. Calculate the dynamic deviation between the actual response delay and the theoretical response delay, and calculate one or more preset statistical features of the dynamic deviation. Based on the preset statistical features, identify normal congestion state or latent fault state. In the state of the latent fault, a disaster recovery operation is triggered to generate a routing control instruction, which is then sent to the session boundary controller to forward the subsequent session initiation request to the disaster recovery node.

2. The method for detecting and mitigating latent faults in communication networks according to claim 1, characterized in that, The service-side data includes call type, call direction, calling terminal type, timestamp, current region, and operator information; The network-side data includes the current number of concurrent calls, as well as the CPU utilization, memory utilization, and signaling link load of the physical mobile switching center; The environmental data includes holiday markers, weather data, and large-scale event markers.

3. The method for detecting and mitigating latent faults in communication networks according to claim 2, characterized in that, The pre-training steps of the dialing delay prediction model include: Collect historical signaling logs of the communication network, which include multiple call samples from the target region during multiple sampling periods throughout the year. Each call sample contains the service-side data, the network-side data, and the environmental data. The call delay of each call sample is labeled as a tag to construct a training sample set. The training sample set is used to train an initial neural network model based on the Transformer model architecture. The initial neural network model includes a continuous input mapping layer, a multi-head self-attention mechanism layer, a residual connection layer, a normalization layer, and a feedforward neural network layer. The feedforward neural network adopts a multilayer perceptron. The initial neural network model takes the service-side data, the network-side data, and the environmental data as inputs and outputs a predicted value for the call delay. Based on the deviation between the predicted value and the label, a loss is constructed to update the parameters of the initial neural network model, thereby obtaining the dialing delay prediction model.

4. The method for detecting and mitigating latent faults in communication networks according to claim 1, characterized in that, Calculate the dynamic deviation between the actual response delay and the theoretical response delay, and calculate one or more preset statistical characteristics of the dynamic deviation. Based on the preset statistical characteristics and predefined rules, identify normal congestion states or latent fault states, including: The dynamic deviation is compared with a set threshold. If the deviation is less than the set threshold, it is determined to be a normal congestion state. If the deviation is greater than or equal to the set threshold, it is determined to be a latent fault state. A call log is established to record the time points of the normal congestion state and the latent fault state.

5. The method for detecting and mitigating latent faults in communication networks according to claim 1, characterized in that, Calculate the dynamic deviation between the actual response delay and the theoretical response delay, and calculate one or more preset statistical characteristics of the dynamic deviation. Based on the preset statistical characteristics and predefined rules, identify normal congestion states or latent fault states, including: The moving average, median, moving standard deviation, excess rate and / or number of consecutive excesses within a set time window are used as the preset statistical features for the dynamic deviation statistics. The preset statistical features are assigned values ​​using a preset standard value range, and a weighted sum is obtained based on preset weights to obtain a state score. The normal congestion state or the latent fault state is identified based on the state score.

6. The method for detecting and mitigating latent faults in communication networks according to claim 1, characterized in that, In the state of latent failure, a disaster recovery operation is triggered to generate a routing control command, which is then sent to the session boundary controller to forward subsequent session initiation requests to the disaster recovery node. This includes: Generate the routing control command containing the faulty physical mobile switching center identifier, the target disaster recovery node identifier, the handover policy parameters, the command ID, and the timestamp; the handover policy parameters are used to mark a full handover or a handover according to a first set ratio; The routing control command is forwarded to the session boundary controller via an internal message queue based on a dedicated channel, and the session boundary controller updates the routing table and executes the routing switching policy. The routing switching strategy includes: performing a gradual traffic switching, switching to the first set ratio step by step according to the second set ratio and duration.

7. The method for detecting and mitigating latent faults in communication networks according to claim 6, characterized in that, The method further includes: Based on the preset statistical characteristics, the latent fault states are divided into three categories: mild faults, moderate faults, and severe faults; mild faults are logged and alarms are generated, but no routing switch is performed; moderate faults are handled with a gradual traffic switch; and severe faults are handled with a full switch.

8. The method for detecting and mitigating latent faults in communication networks according to claim 7, characterized in that, The method further includes: when the routing switching strategy is a full switch, reserving 1% to 5% of session traffic to be forwarded via the original physical mobile switching center as a routing node, and continuously monitoring the actual response delay; when the cooling time is reached after the routing switching strategy is executed, determining whether the actual response delay of the original physical mobile switching center meets the normal congestion state; if it does, performing a gradual traffic switchback back to the original physical mobile switching center; otherwise, continuing to maintain the status quo.

9. A communication network latent fault detection and disaster recovery device, comprising a processor, a memory, and a computer program or instructions stored in the memory, characterized in that, The processor is configured to execute the computer program or instructions, and when the computer program or instructions are executed, the device implements the steps of the method as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method as described in any one of claims 1 to 8.

Images