A communication fault early warning and resource scheduling method and device

Abstract

This invention discloses a communication fault early warning and resource scheduling method and apparatus, relating to the field of network security protection technology. The method includes: acquiring physical layer energy consumption parameters and network layer traffic data of the corresponding device; calculating the correlation degree between the physical layer energy consumption parameters and the network layer traffic data based on a unified time window, and determining strongly correlated samples based on the correlation degree; performing feature fusion on the strongly correlated samples to obtain a device state feature tensor; performing anomaly detection based on the device state feature tensor and a fault risk score obtained from time-series prediction, and obtaining fault detection results; adjusting the channel bandwidth ratio according to the fault detection results, and executing corresponding channel operation state control. This application achieves a synergistic improvement in the accuracy of fault early warning and the timeliness of resource scheduling by accurately identifying the causes of traffic anomalies and proactively perceiving fault trends, thereby dynamically adjusting channel resources.

Description

Technical Field

[0001] This application relates to the field of network security protection technology, specifically to a method and apparatus for communication fault early warning and resource scheduling. Background Technology

[0002] In power communication networks, 5G terminal equipment is widely deployed in complex electromagnetic environments such as substations and transmission towers, and needs to simultaneously support mixed services such as relay protection, differential control, and high-definition video inspection. Conventional operation and maintenance monitoring methods mainly employ network layer traffic threshold alarms, SNMP interface status polling, and system log keyword matching. However, these operation and maintenance monitoring methods operate independently and lack the effective ability to cross-domain correlate physical layer hardware electrical parameters (such as motherboard current and module power consumption) with network layer transmission behavior. This single-dimensional monitoring mode has the following defects: First, when traffic surges abnormally, the system cannot distinguish whether it is caused by a sudden video service outbreak, a DDoS attack, or 5G module / power supply hardware degradation, which easily leads to a large number of false alarms or missed alarms. At the same time, when a fault occurs, multiple systems such as CPU, traffic, and interfaces trigger alarms independently. Due to the lack of causal reasoning and suppression logic, alarm storms occur frequently, and root cause location is extremely difficult. Second, fixed threshold or static baseline methods cannot adapt to operational drift such as power grid load fluctuations and equipment aging, resulting in alarm threshold mismatch. Finally, the fault risk perception results are disconnected from the communication resource scheduling. When a signaling storm or a surge in services occurs, the channel resource guarantee mode cannot be dynamically adjusted in real time according to the real-time priority of power services, making it difficult to guarantee the deterministic latency of critical scheduling services. Summary of the Invention

[0003] This application addresses the problems of inaccurate identification of traffic anomaly causes and poor timeliness of communication resource scheduling in existing communication fault monitoring methods. It proposes a communication fault early warning and resource scheduling method and device, which accurately identifies traffic anomaly causes and proactively senses fault trends through cross-domain correlation analysis and time-series prediction of physical layer energy consumption parameters and network layer traffic data. Based on this, channel resources are dynamically adjusted to achieve a synergistic improvement in the accuracy of fault early warning and the timeliness of resource scheduling.

[0004] To achieve the above objectives, the technical solutions adopted in the embodiments of this application are as follows: In a first aspect, embodiments of this application provide a communication fault early warning and resource scheduling method, the method comprising: The system acquires the physical layer power consumption parameters and network layer traffic data of the corresponding device for the channel; calculates the correlation degree between the physical layer power consumption parameters and the network layer traffic data based on a unified time window, and determines strongly correlated samples based on the correlation degree; performs feature fusion on the strongly correlated samples to obtain a device state feature tensor; performs anomaly detection based on the device state feature tensor and the fault risk score obtained by time series prediction, and obtains fault detection results; adjusts the channel bandwidth ratio according to the fault detection results, and executes the corresponding channel operation state control.

[0005] This solution integrates physical layer energy consumption parameters and network layer traffic data into a unified time-series correlation analysis framework, establishing a mechanism for calculating the correlation degree between cross-domain parameters and screening strongly correlated samples. This allows the system to penetrate the logical appearance of simple traffic monitoring and capture the physical imprint of abnormal behavior from the perspective of hardware energy consumption, thus avoiding the problem of failing to accurately identify the causes of traffic anomalies due to single-dimensional monitoring. Based on the feature tensor fusion of strongly correlated samples, a multi-dimensional joint representation of electrical parameters, traffic characteristics, and equipment status is constructed, providing a complete input foundation for subsequent anomaly detection. The risk scoring mechanism combined with time-series prediction enables the system to perceive the evolution trend of anomalies before they fully develop, overcoming the lag in response to gradual faults caused by pure threshold judgment. Dynamically adjusting the bandwidth ratio and executing corresponding channel operation state control based on the detection results achieves an end-to-end closed loop from anomaly perception to resource adaptation, effectively improving the targeting and timeliness of communication resource scheduling.

[0006] Optionally, the step of calculating the correlation degree between the physical layer energy consumption parameters and the network layer traffic data based on a unified time window, and determining strongly correlated samples based on the correlation degree, includes: calculating the covariance between the traffic data sequence corresponding to the network layer traffic data and the energy consumption parameter sequence corresponding to the physical layer energy consumption parameters; obtaining the correlation degree based on the covariance combined with the standard deviation of the traffic data sequence and the standard deviation of the energy consumption parameter sequence; and marking the corresponding network layer traffic data and physical layer energy consumption parameters as strongly correlated samples when the absolute value of the correlation degree is greater than a preset correlation threshold.

[0007] Optionally, when calculating the correlation between the physical layer energy consumption parameters and the network layer traffic data, the following steps are also performed: A Granger causality test is performed on the traffic data sequence and the energy consumption parameter sequence; the abnormal cause of the data frame is determined based on the temporal leading relationship between the abnormal fluctuations of the traffic data and the abnormal fluctuations of the energy consumption parameters; when the abnormal fluctuations of the traffic data lead the abnormal fluctuations of the energy consumption parameters in time, the abnormal cause of the current data frame is determined to be a behavioral layer anomaly caused by changes in service load or network attacks; when the abnormal fluctuations of the energy consumption parameters lead the abnormal fluctuations of the traffic data in time, the abnormal cause of the current data frame is determined to be a physical layer fault caused by changes in hardware electrical characteristics; for behavioral layer anomalies, the correlation is multiplied by a first directional correction coefficient to obtain the correlation of the target; for physical layer faults, the correlation is multiplied by a second directional correction coefficient to obtain the correlation of the target; the first directional correction coefficient is greater than the second directional correction coefficient.

[0008] Optionally, the step of performing feature fusion on the strongly correlated samples to obtain the equipment state feature tensor includes: calculating the three-phase current balance based on the physical layer energy consumption parameters, and using a wavelet transform algorithm to extract the energy distribution ratio of the fundamental frequency component and harmonic components, calculating the harmonic content rate and total harmonic distortion rate, and integrating them to form an electrical parameter feature vector group; concatenating the electrical parameter feature vector group, the network layer flow data, and the equipment state data, and obtaining the equipment state feature tensor after normalization processing; wherein, the three-phase current balance is calculated based on the maximum, minimum, and average values ​​of the three-phase currents, and the total harmonic distortion rate is calculated based on the effective values ​​of the harmonic voltages and the effective value of the fundamental voltage.

[0009] Optionally, the anomaly detection based on the fault risk score obtained from the device state feature tensor and time-series prediction, and the acquisition of fault detection results, includes: acquiring historical operating data of the device under various operating conditions, wherein the historical operating data includes at least power consumption characteristic parameters, current harmonic characteristic parameters, and flow characteristic parameters; using a sliding time window statistical method to extract the dynamic distribution characteristics of the historical operating data under different operating conditions, and using the dynamic distribution characteristics as the adaptive threshold for the corresponding operating condition; when the device state feature tensor exceeds the adaptive threshold corresponding to the current operating condition, or when the fault risk score shows a monotonically increasing trend within a continuous preset time period and the growth rate exceeds a preset rate threshold, it is determined that there is a potential fault, and the resource scheduling process is triggered.

[0010] Optionally, the step of obtaining a fault risk score includes: based on a time-series prediction model, performing fault trend prediction on collected historical time-series data and real-time time-series data to obtain predicted channel operating status values ​​within a future preset time window, wherein the predicted channel operating status values ​​include voltage prediction values, current prediction values, and time delay prediction values; calculating the absolute value of a first difference between the voltage prediction value and the actual voltage measurement value, the absolute value of a second difference between the current prediction value and the actual current measurement value, and the absolute value of a third difference between the time delay prediction value and the actual time delay measurement value; and weighted summing the first absolute value of the difference, the second absolute value of the difference, and the third absolute value of the difference to obtain a fault risk score.

[0011] Optionally, adjusting the channel bandwidth ratio based on the fault detection result includes: when the fault risk score exceeds a preset switching threshold and continues to exceed a preset confirmation time window, switching to hard isolation mode, allocating independent physical time slots or frequency bands for critical services through a dedicated physical resource module; dynamically monitoring the traffic status of the primary channel and the backup channel and the fault risk score, and when the fault risk score of the primary channel exceeds the preset switching threshold, performing a bandwidth switch from the primary channel to the backup channel.

[0012] Optionally, the execution of the corresponding channel operation state control includes: When the fault risk score indicates a high-level fault hazard condition, basic fault handling control instructions are generated, including faulty component disconnection instructions and abnormal link disconnection instructions. When the fault risk score indicates a no-fault-hazard or low-level fault-hazard condition, operation optimization control instructions are generated based on the correlation between equipment operating status and energy consumption characteristics. These operation optimization control instructions include adjusting the processor operating frequency, adjusting the power output, and adjusting the link bandwidth allocation to regulate the operating status of each key component to the optimal range of the corresponding operating condition.

[0013] Optionally, when acquiring the physical layer energy consumption parameters of the device corresponding to the channel, the method further includes: collecting the air interface protocol parameters of the communication module, and constructing a virtual energy consumption feature sequence based on the air interface protocol parameters; the air interface protocol parameters include at least the modulation and coding scheme level, the hybrid automatic repeat request retransmission frequency, and the TCP retransmission rate; within the same time window, calculating the sliding window matching degree between the measured physical layer energy consumption parameter sequence and the virtual energy consumption feature sequence; when the sliding window matching degree is greater than or equal to a preset matching degree threshold, determining that the measured physical layer energy consumption parameters are reliable, and calculating the system deviation value between the measured physical layer energy consumption parameter sequence and the virtual module energy consumption feature sequence; if the absolute value of the system deviation value is greater than the deviation tolerance threshold, determining that the sensor has temperature drift or zero drift error, and calculating the drift compensation coefficient based on the system deviation value to correct the measured physical layer energy consumption parameters.

[0014] Secondly, embodiments of this application provide a communication fault early warning and resource scheduling device, comprising: The data acquisition module is used to acquire physical layer power consumption parameters and network layer traffic data of the corresponding devices in the channel; the feature fusion module calculates the correlation between the physical layer power consumption parameters and the network layer traffic data based on a unified time window, and determines strongly correlated samples based on the correlation; it performs feature fusion on the strongly correlated samples to obtain a device state feature tensor; the risk assessment module performs anomaly detection based on the device state feature tensor and obtains a fault risk score by combining time series prediction; it adjusts the channel bandwidth ratio based on the fault risk score and executes the corresponding channel operation state control.

[0015] The beneficial effects of this application are: 1. By conducting cross-domain correlation analysis between physical layer energy consumption parameters and network layer traffic data, an anomaly cause timing direction discrimination mechanism based on Granger causality test was established. This mechanism can accurately distinguish whether traffic anomalies are caused by changes in service load, network attacks, or hardware electrical characteristic degradation. Compared with traditional single-dimensional threshold monitoring methods, this overcomes the problem of inaccurate anomaly cause identification, enabling subsequent fault handling to select targeted control strategies based on the anomaly propagation direction. This avoids the risk of misoperation caused by using the same handling method for behavioral layer anomalies and physical layer faults. 2. By combining adaptive dynamic threshold with time-series trend prediction, the dual-channel anomaly detection architecture avoids the problem of gradual mismatch of fixed threshold throughout the entire life cycle. The time-series prediction channel, through weighted monitoring of the deviation between predicted and measured values ​​of voltage, current, and time delay, can identify the deterioration trend of gradual faults before the values ​​reach the threshold boundary. This makes up for the inherent defect of threshold detection in responding to slow deterioration faults, and realizes the dual capabilities of rapid response to sudden faults and early warning of gradual faults. 3. A hierarchical and progressive channel resource scheduling mechanism is constructed based on fault risk scoring and abnormal cause classification results. At the same time, based on the different levels of fault risks, hard measures such as disconnecting faulty components and cutting off abnormal links are executed, while flexible optimization control commands such as processor frequency adjustment and power supply regulation are implemented. Under the premise of ensuring the safety bottom line of key power communication services, the energy efficiency optimization of equipment in non-critical scenarios is also taken into account, forming a complete closed loop from accurate early warning to hierarchical control and then to energy efficiency management. Attached Figure Description

[0016] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings. The drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings.

[0017] Figure 1 This is a flowchart of the communication fault early warning and resource scheduling method according to an embodiment of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description of this application is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely one preferred embodiment of this application and are only used to explain this application. They do not limit the scope of protection of this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] Example 1: As Figure 1 As shown, a communication fault early warning and resource scheduling method includes steps S1-S5, wherein: S1. Obtain the physical layer power consumption parameters and network layer traffic data of the device corresponding to the channel.

[0020] Specifically, the channel refers to the 5G hardware multiplexing channel in the power system. The physical layer energy consumption parameters include at least three-phase voltage, three-phase current, power factor, active power, reactive power, harmonic distortion rate, as well as the operating current of the core processor and the radio frequency operating current of the 5G communication module. The current data is read directly from the power management chip or shunt through a hardware-level independent sampling circuit. The network layer traffic data includes packet throughput, latency, packet loss rate, jitter, and service traffic type distribution.

[0021] As one implementation method, when obtaining the physical layer power consumption parameters of the device corresponding to the channel, the method further includes: Collect the air interface protocol parameters of the communication module, and construct a virtual energy consumption feature sequence based on the air interface protocol parameters; the air interface protocol parameters include at least the modulation and coding scheme level, the hybrid automatic repeat request retransmission frequency, and the TCP retransmission rate. Within the same time window, calculate the sliding window fit between the measured physical layer energy consumption parameter sequence and the virtual energy consumption feature sequence; When the sliding window matching degree is greater than or equal to the preset matching degree threshold, the measured physical layer energy consumption parameters are determined to be reliable, and the system deviation value between the measured physical layer energy consumption parameter sequence and the virtual module energy consumption characteristic sequence is calculated. If the absolute value of the system deviation is greater than the deviation tolerance threshold, it is determined that the sensor has temperature drift or zero drift error. The drift compensation coefficient is calculated based on the system deviation value to correct the measured physical layer energy consumption parameters.

[0022] It should be noted that when the modulation and coding strategy level increases, it indicates that the 5G communication module adopts a higher-order modulation method. Higher-order modulation has higher requirements for RF transmit power and linearity, which physically necessitates a larger RF power amplifier bias current. Therefore, a positive correlation is established between the change in modulation and coding strategy level and the expected change in RF current. When the hybrid automatic repeat request (ARRP) frequency increases, it indicates a decrease in air interface transmission quality. The module's baseband processor needs to repeatedly perform data merging and decoding operations, increasing the computational load and leading to an increase in baseband current. Therefore, a positive correlation is established between the retransmission frequency and the baseband processing current. When the TCP retransmission rate increases, it indicates congestion or packet loss in the end-to-end transmission path. The upper-layer protocol stack needs to consume more CPU resources to handle retransmission timers and congestion window adjustments. Therefore, a positive correlation is established between the TCP retransmission rate and the processor core current. The current changes obtained from these three mappings are superimposed within a unified time window to generate a virtual energy consumption characteristic sequence.

[0023] In some embodiments, the similarity calculation method employs a similarity metric based on error energy normalization. Specifically, within the same time window, the difference between the measured value and the virtual value is calculated point by point. The sum of the squares of all differences yields the error energy, while the sum of the squares of the measured values ​​themselves yields the signal energy. The similarity is equal to 1 minus the ratio of the error energy to the signal energy. The similarity threshold can be set based on the device's factory calibration data, such as the historical average similarity.

[0024] Specifically, the system deviation value is calculated using a least-squares linear fitting method. A linear regression line is fitted using the virtual energy consumption characteristic sequence as the independent variable and the measured physical layer energy consumption parameter sequence as the dependent variable. The intercept term of this line is extracted as the system deviation value. This intercept reflects the fixed offset of the sensor throughout the entire measurement range. The slope obtained from the linear fitting is used as the first compensation coefficient, and the intercept is used as the second compensation coefficient. The corrected measured value is equal to the original measured value minus the second compensation coefficient, divided by the first compensation coefficient.

[0025] In this embodiment, the virtual energy consumption characteristics obtained by inverting the air interface protocol parameters are essentially a reference benchmark built based on the communication physical layer mechanism. Through mutual verification of heterogeneous data sources, sensor drift or even faults can be identified. The slope and intercept are obtained by least squares fitting as compensation coefficients and linear correction is performed to ensure that the accuracy of the measured parameters is restored to near the factory state, providing a reliable guarantee for the input data source of the entire fault warning system.

[0026] S2. Based on a unified time window, calculate the correlation between the physical layer energy consumption parameters and the network layer traffic data, and determine strongly correlated samples based on the correlation.

[0027] In one implementation, step S2 includes: Calculate the covariance between the traffic data sequence corresponding to the network layer traffic data and the energy consumption parameter sequence corresponding to the physical layer energy consumption parameters; The correlation degree is obtained based on the covariance combined with the standard deviation of the flow data sequence and the standard deviation of the energy consumption parameter sequence. When the absolute value of the correlation degree is greater than the preset correlation threshold, the corresponding network layer traffic data and physical layer energy consumption parameters are marked as strongly correlated samples.

[0028] Specifically, the correlation degree is obtained by multiplying the standard deviation of the flow data sequence and the standard deviation of the energy consumption parameter sequence, and then dividing the covariance by this product. The preset correlation threshold is set based on the statistical distribution of historical normal samples.

[0029] As one implementation method, when calculating the correlation between the physical layer energy consumption parameters and the network layer traffic data, the following is also performed: Perform Granger causality test on the traffic data sequence and the energy consumption parameter sequence, and determine the abnormal cause of the data frame based on the leading relationship between the abnormal fluctuations of the traffic data and the abnormal fluctuations of the energy consumption parameters in the time series. When abnormal fluctuations in traffic data precede abnormal fluctuations in energy consumption parameters in time sequence, the abnormal cause of the current data frame is determined to be a behavioral layer abnormality caused by changes in business load or network attacks. When the abnormal fluctuation of energy consumption parameters precedes the abnormal fluctuation of traffic data in time, it is determined that the abnormal cause of the current data frame belongs to a physical layer fault caused by changes in hardware electrical characteristics. For behavioral layer anomalies, the correlation degree is multiplied by the first directional correction coefficient to obtain the correlation degree of the target; For physical layer faults, the correlation degree is multiplied by the second direction correction coefficient to obtain the correlation degree of the target; The first direction correction factor is greater than the second direction correction factor.

[0030] Specifically, the steps for performing Granger causality tests on the flow data series and energy consumption parameter series include: First, performing a stationarity test on the two time series. If the series is non-stationary, it is first-order differencing is performed until the stationarity requirement is met. Then, a vector autoregressive model containing lagged terms of both series is constructed, with the model order determined according to the Akaike information content criterion. Within this model framework, constrained and unconstrained regression equations are established respectively. The unconstrained equation includes lagged terms of both energy consumption parameters and flow data, while the constrained equation only includes lagged terms of the energy consumption parameter itself. An F-statistic is constructed from the sum of squared residuals of the two equations to test whether the coefficients of the lagged terms of the flow data are jointly significant and not zero. If significant, it indicates that the flow data is statistically a Granger cause of the energy consumption parameter, meaning that changes in the flow data lead changes in the energy consumption parameter in time. Conversely, if not, it indicates that changes in the energy consumption parameter lead changes in the flow data.

[0031] Specifically, when traffic data precedes energy consumption parameters, it means that the anomaly first manifests at the network layer and then is reflected in hardware energy consumption. This situation corresponds to behavioral layer causes such as sudden increases in service load or network attacks. When energy consumption parameters precede traffic data, it means that abnormal fluctuations in hardware electrical characteristics first occur, which then lead to a deterioration in network layer transmission quality. This situation corresponds to physical layer causes such as power supply jitter and module aging.

[0032] Specifically, the first direction correction coefficient can be set to a value greater than 1, and the second direction correction coefficient can be set to a value greater than 0 and less than 1.

[0033] In this embodiment, the original correlation degree is numerically corrected by the direction correction coefficient. The correlation degree of behavioral layer anomalies is amplified, making them easier to be marked as strong correlations. The correlation degree of physical layer faults is compressed, so that their contribution in subsequent multi-source fusion is adjusted accordingly. Essentially, prior knowledge of the fault propagation direction is embedded in the feature extraction stage, which improves the physical semantic expression capability of the feature tensor.

[0034] S3. Perform feature fusion on the strongly correlated samples to obtain the device state feature tensor.

[0035] In one implementation, step S3 includes: The three-phase current balance is calculated based on the physical layer energy consumption parameters, and the energy distribution ratio of the fundamental frequency component and harmonic components is extracted using the wavelet transform algorithm. The harmonic content rate and total harmonic distortion rate are calculated and integrated to form an electrical parameter feature vector group. The electrical parameter feature vector group, the network layer traffic data, and the device status data are concatenated and normalized to obtain the device status feature tensor. The three-phase current balance is calculated based on the maximum, minimum and average values ​​of the three-phase currents, and the total harmonic distortion rate is calculated based on the effective values ​​of each harmonic voltage and the effective value of the fundamental voltage.

[0036] Specifically, the three-phase current balance is calculated by subtracting the minimum value from the maximum value of the three-phase current and then dividing by the average value of the three-phase current. Simultaneously, a wavelet transform algorithm is used to decompose the current and voltage waveforms into multiple scales, extracting the energy distribution ratio of the fundamental frequency component (50Hz power frequency component) and each harmonic component. Based on this, the nth harmonic content rate is calculated according to the ratio of the effective value of each harmonic voltage to the effective value of the fundamental voltage, and the total harmonic distortion rate is calculated by taking the square root of the sum of the squares of the harmonic content rates. The harmonic content rate and the total harmonic distortion rate together constitute the spectral feature vector of the current and voltage. The three-phase current balance is integrated with the spectral feature vector to form a set of electrical parameter feature vectors.

[0037] In this embodiment, the three-phase current balance can directly reflect whether there is an abnormality at the power supply end of the equipment. The introduction of harmonic spectrum characteristics enables the system to distinguish between the nonlinear characteristics of normal load and the harmonic distortion caused by fault. The electrical parameter feature vector group, network layer traffic data, and equipment status data are fused by tensor to achieve joint judgment of fault risk from three perspectives: electrical health, network transmission quality, and equipment operating environment, thus avoiding inaccurate early warning caused by a single dimension.

[0038] S4. Based on the fault risk score obtained from the device state feature tensor and time series prediction, perform anomaly detection and obtain fault detection results.

[0039] In one implementation, step S4 includes: Acquire historical operating data of the equipment under various operating conditions. The historical operating data includes at least power consumption characteristic parameters, current harmonic characteristic parameters, and flow characteristic parameters. A sliding time window statistical method is used to extract the dynamic distribution characteristics of historical operating data under different working conditions, and the dynamic distribution characteristics are used as the adaptive threshold for the corresponding working conditions. When the device state feature tensor exceeds the adaptive threshold corresponding to the current operating condition, or when the fault risk score shows a monotonically increasing trend within a continuous preset time period and the growth rate exceeds the preset rate threshold, it is determined that there is a potential fault and the resource scheduling process is triggered.

[0040] Specifically, the operating conditions are distinguished based on the temperature and load rate of core components. Typical operating conditions include three types: peak load, normal operation, and low load. Historical operating data includes at least power consumption characteristic parameters, current harmonic characteristic parameters, and flow characteristic parameters. For each operating condition, a sliding time window statistical method is used to extract dynamic distribution characteristics. Specifically, a fixed-length historical data window is truncated backward from the current time, and the mean and standard deviation of each characteristic parameter within the window are calculated. The combination of the mean and standard deviation is used as the adaptive threshold for that operating condition at the current time.

[0041] It should be noted that state detection based on adaptive thresholds and trend detection based on time-series prediction run in parallel. If either of the two channels is triggered, the subsequent fault handling process will begin.

[0042] Specifically, the steps to obtain a fault risk score include: Based on the time-series prediction model, fault trend prediction is performed on the collected historical time-series data and real-time time-series data to obtain the channel operation status prediction value within a future preset time window. The channel operation status prediction value includes voltage prediction value, current prediction value and time delay prediction value. Calculate the absolute value of the first difference between the predicted voltage value and the actual measured voltage value, the absolute value of the second difference between the predicted current value and the actual measured current value, and the absolute value of the third difference between the predicted time delay value and the actual measured time delay value, respectively. The fault risk score is obtained by weighted summing of the absolute values ​​of the first, second, and third differences.

[0043] Specifically, the time-series prediction model employs a Long Short-Term Memory (LSTM) network structure. This network consists of an input layer, a single-layer LSTM hidden layer, and an output layer. The number of neurons in the hidden layer is determined based on the length of the input sequence and the dimensions of the device parameters. The model's input data consists of historical time-series data extending back a preset time window from the current time point. This includes historical voltage values, historical current values, historical latency values, and corresponding historical values ​​of the device state feature tensors. It should be understood that the time-series prediction model uses a standard algorithm structure; as long as it can predict voltage, current, and latency, it is acceptable, and no further specific limitations are imposed here.

[0044] Furthermore, the weights corresponding to the absolute values ​​of the three differences are dynamically set according to the priority of the business scenario. For example, when the business carried by the device is sensitive to latency, the weight corresponding to the absolute value of the latency difference is higher; when the fault is mainly manifested as abnormal electrical parameters, the weights corresponding to the absolute values ​​of the voltage and current differences are higher.

[0045] In this embodiment, due to the periodic fluctuations in grid load, seasonal changes in ambient temperature, and the slow aging of equipment over time, a fixed threshold will inevitably become invalid at some point. However, the mean and standard deviation of a dynamic distribution are updated in real time with the operating status, encompassing both the statistical regularity of short-term load fluctuations and the slow drift of long-term equipment operating characteristics. This ensures that the threshold always matches the current true health status of the equipment, avoiding the contradiction of fixed thresholds being prone to false alarms during low load periods and false alarms during high load periods. The time-series prediction trend detection channel compensates for the limitation of threshold detection, which can only determine whether the current state exceeds the limit. For gradual faults such as slowly rising current and gradually decreasing module efficiency, the values ​​may not exceed the threshold boundary in the early stages, but the trend already points to future risks. The trend detection channel achieves early identification of such sub-health states by monitoring the slope of the change. The two channels work in parallel, retaining the rapid response capability of threshold detection to sudden faults while enhancing the early warning capability for gradual degradation.

[0046] S5. Adjust the channel bandwidth ratio based on the fault detection results and execute the corresponding channel operation status control.

[0047] As one implementation method, step S5, adjusting the channel bandwidth ratio based on the fault detection result, includes: When the fault risk score exceeds the preset switching threshold and continues to exceed the preset confirmation time window, the system switches to hard isolation mode and allocates independent physical time slots or frequency bands for critical services through a dedicated physical resource module. The system dynamically monitors the traffic status of the primary channel and the backup channel, as well as the fault risk score. When the fault risk score of the primary channel exceeds the preset switching threshold, the bandwidth of the primary channel is switched to the backup channel.

[0048] Specifically, under normal operating conditions, the system adopts a logically isolated soft slicing mode. Different service types are divided into traffic types at the software level through software-defined network flow table rules. Various services share the same physical bandwidth resources but are logically isolated from each other through priority queues and traffic shaping mechanisms. The preset switching threshold is obtained based on historical fault sample statistics. For example, based on confirmed fault event data from the equipment's historical operation, a fault risk score is calculated within a preset time window before each fault event occurs. The critical value among these scores that can effectively distinguish between fault events and normal events is taken as the switching threshold. Hard isolation mode refers to allocating independent physical time slots or frequency bands for critical services through dedicated physical resource modules in the 5G network. These resources are not shared with other services, achieving exclusive access to communication resources at the physical layer. No other service traffic can intrude into this resource range, thus ensuring that critical services are not affected by network congestion or interference.

[0049] Specifically, when the failure risk score of the primary channel exceeds a preset switching threshold, the bandwidth of the primary channel is switched to the backup channel. During the switching process, the primary and backup channels adopt a dual-link parallel transmission mode. That is, the primary channel continues to operate until the newly established backup channel fully carries the service traffic. The primary channel resources are released only after the backup channel has completed synchronization and all service flows have been migrated. This avoids communication interruption at the moment of switching and ensures both reliability and service continuity.

[0050] In this embodiment, the soft slicing mode fully utilizes physical bandwidth resources during daily operation. Various services share the same physical channel through software-level flow table rules, avoiding the bandwidth waste caused by exclusive physical resource allocation in the hard isolation mode, and improving the overall utilization of network resources. When the system detects an increased risk of failure, it switches to hard isolation mode, ensuring that critical services receive exclusive physical layer resource guarantees. Network congestion or interference cannot affect their transmission quality, fundamentally guaranteeing the deterministic latency requirements of critical power services under abnormal conditions.

[0051] As one implementation method, step S5 involves performing corresponding channel operation state control, including: When the fault risk score determines that the operating condition is a high-level fault hazard, a basic fault handling control instruction is generated, which includes a faulty component isolation instruction and an abnormal link disconnection instruction. When the fault risk score determines that the operating condition is free of fault risks or has low-level fault risks, an operation optimization control instruction is generated based on the correlation between the equipment operating status and energy consumption characteristics. The operation optimization control instruction includes adjusting the processor operating frequency, adjusting the power output power, and adjusting the link bandwidth allocation, so as to regulate the operating status of each key component to the optimal range of the corresponding operating condition.

[0052] Specifically, when the fault risk score is greater than the first threshold, it is determined to be a high-level fault hazard condition. When the fault risk score is less than or equal to the first threshold but greater than the second threshold, the abnormal cause classification results output by the Granger causality test are further referenced. If the cause is classified as a physical layer fault, it is determined to be a high-level fault hazard; if the cause is classified as a behavioral layer anomaly, it is determined to be a low-level fault hazard. When the fault risk score is less than or equal to the second threshold, it is determined to be a fault-free condition. The first threshold is higher than the second threshold, and both can be determined based on the score distribution of historical fault samples. When a high-level fault hazard condition is determined, the system generates basic fault handling control instructions. For example, when the Granger causality test determines that the abnormal cause is a change in hardware electrical characteristics and the fault risk score continues to rise, the system determines it to be hardware degradation, generates a fault component isolation instruction to disconnect the abnormal 5G communication module from the working circuit, and simultaneously generates an abnormal link disconnection instruction to disconnect the network connection associated with the module, preventing the fault from spreading to other normally operating equipment. When an operating condition is determined to be free of potential faults or have a low-level potential fault, operational optimization is performed. The goal of the optimized control instructions is to adjust the operating status of key components such as the processor, 5G communication module, and power supply module to the optimal power consumption range corresponding to the current operating condition, minimizing energy consumption without affecting service carrying capacity. For example, the system generates instructions to adjust the processor's operating frequency, reducing the CPU clock speed to the minimum frequency required to maintain the current service throughput. At the same time, it generates instructions to adjust the power supply output power, lowering the supply voltage to a lower level within the chip's normal operating range to reduce heat dissipation.

[0053] In this embodiment, differentiated processing logic enables the system to adopt matching control strategies when facing anomalies of different types and severity. Under high-level operating conditions, hard measures such as isolating faulty components and severing abnormal links are directly executed to prioritize overall system safety and prevent local faults from escalating into global paralysis. Under low-level and fault-free operating conditions, operating parameters are optimized to explore energy-saving potential while ensuring business continuity, thus balancing security and operational economy. The logic of prioritizing high-level handling for physical layer faults fully considers the rapid deterioration of hardware anomalies, avoiding complete equipment failure due to untimely handling.

[0054] Example 2, based on the same inventive concept, also provides a communication fault early warning and resource scheduling device corresponding to the communication fault early warning and resource scheduling method, including: The data acquisition module is used to acquire the physical layer power consumption parameters and network layer traffic data of the device corresponding to the channel; The feature fusion module calculates the correlation between the physical layer energy consumption parameters and the network layer traffic data based on a unified time window, determines strongly correlated samples based on the correlation, and performs feature fusion on the strongly correlated samples to obtain a device state feature tensor. The risk assessment module performs anomaly detection based on the device state feature tensor and obtains a fault risk score by combining it with time series prediction; it adjusts the channel bandwidth ratio according to the fault risk score and executes the corresponding channel operation state control.

[0055] In this embodiment, by collecting physical layer energy consumption parameters and network layer traffic data of the corresponding devices of the 5G hardware multiplexing channel, the correlation between the two in the time dimension is calculated to generate fusion state characteristics. The cross-verification mechanism of physical energy consumption and network traffic is used to accurately assess the degree of fault risk, and the resource allocation and operation status of the 5G multiplexing channel are dynamically adjusted accordingly, thereby achieving accurate fault warning and communication reliability assurance.

[0056] The above-described embodiments are preferred embodiments of this application and are not intended to limit the specific scope of this application. The scope of this application includes but is not limited to the specific embodiments described above. All equivalent changes made in accordance with the shape, structure, and method of this application are within the protection scope of this application.

Claims

1. A method for communication failure pre-warning and resource scheduling, characterized in that: Includes the following steps: Obtain the physical layer power consumption parameters and network layer traffic data of the device corresponding to the channel; Based on a unified time window, the correlation between the physical layer energy consumption parameters and the network layer traffic data is calculated, and strongly correlated samples are determined based on the correlation. Feature fusion is performed on the strongly correlated samples to obtain the device state feature tensor; Anomaly detection is performed based on the fault risk score obtained from the device state feature tensor and time series prediction to obtain fault detection results; The bandwidth ratio of the channel is adjusted based on the fault detection results, and the corresponding channel operation status control is executed. 2.The method of claim 1, wherein: The process of calculating the correlation between the physical layer energy consumption parameters and the network layer traffic data based on a unified time window, and determining strongly correlated samples based on the correlation, includes: Calculate the covariance between the traffic data sequence corresponding to the network layer traffic data and the energy consumption parameter sequence corresponding to the physical layer energy consumption parameters; The correlation degree is obtained based on the covariance combined with the standard deviation of the flow data sequence and the standard deviation of the energy consumption parameter sequence. When the absolute value of the correlation degree is greater than the preset correlation threshold, the corresponding network layer traffic data and physical layer energy consumption parameters are marked as strongly correlated samples.

3. The method of claim 2, wherein: When calculating the correlation between the physical layer energy consumption parameters and the network layer traffic data, the following also applies: Perform Granger causality test on the traffic data sequence and the energy consumption parameter sequence, and determine the abnormal cause of the data frame based on the leading relationship between the abnormal fluctuations of the traffic data and the abnormal fluctuations of the energy consumption parameters in the time series. When abnormal fluctuations in traffic data precede abnormal fluctuations in energy consumption parameters in time sequence, the abnormal cause of the current data frame is determined to be a behavioral layer abnormality caused by changes in business load or network attacks. When the abnormal fluctuation of energy consumption parameters precedes the abnormal fluctuation of traffic data in time, it is determined that the abnormal cause of the current data frame belongs to a physical layer fault caused by changes in hardware electrical characteristics. For behavioral layer anomalies, the correlation degree is multiplied by the first directional correction coefficient to obtain the correlation degree of the target; For physical layer faults, the correlation degree is multiplied by the second direction correction coefficient to obtain the correlation degree of the target; The first direction correction factor is greater than the second direction correction factor.

4. The communication fault early warning and resource scheduling method according to claim 2, characterized in that: The step of fusing features from the strongly correlated samples to obtain the device state feature tensor includes: The three-phase current balance is calculated based on the physical layer energy consumption parameters, and the energy distribution ratio of the fundamental frequency component and harmonic components is extracted using the wavelet transform algorithm. The harmonic content rate and total harmonic distortion rate are calculated and integrated to form an electrical parameter feature vector group. The electrical parameter feature vector group, the network layer traffic data, and the device status data are concatenated and normalized to obtain the device status feature tensor. The three-phase current balance is calculated based on the maximum, minimum and average values ​​of the three-phase currents, and the total harmonic distortion rate is calculated based on the effective values ​​of each harmonic voltage and the effective value of the fundamental voltage.

5. The communication fault early warning and resource scheduling method according to claim 1, characterized in that: The fault risk score obtained based on the device state feature tensor and time series prediction is used for anomaly detection to obtain fault detection results, including: Acquire historical operating data of the equipment under various operating conditions. The historical operating data includes at least power consumption characteristic parameters, current harmonic characteristic parameters, and flow characteristic parameters. A sliding time window statistical method is used to extract the dynamic distribution characteristics of historical operating data under different working conditions, and the dynamic distribution characteristics are used as the adaptive threshold for the corresponding working conditions. When the device state feature tensor exceeds the adaptive threshold corresponding to the current operating condition, or when the fault risk score shows a monotonically increasing trend within a continuous preset time period and the growth rate exceeds the preset rate threshold, it is determined that there is a potential fault and the resource scheduling process is triggered.

6. The communication fault early warning and resource scheduling method according to claim 5, characterized in that: The steps to obtain a fault risk score include: Based on the time-series prediction model, fault trend prediction is performed on the collected historical time-series data and real-time time-series data to obtain the channel operation status prediction value within a future preset time window. The channel operation status prediction value includes voltage prediction value, current prediction value and time delay prediction value. Calculate the absolute value of the first difference between the predicted voltage value and the actual measured voltage value, the absolute value of the second difference between the predicted current value and the actual measured current value, and the absolute value of the third difference between the predicted time delay value and the actual measured time delay value, respectively. The fault risk score is obtained by weighted summing of the absolute values ​​of the first, second, and third differences.

7. The communication fault early warning and resource scheduling method according to claim 1, characterized in that: The step of adjusting the channel bandwidth ratio based on the fault detection result includes: When the fault risk score exceeds the preset switching threshold and continues to exceed the preset confirmation time window, the system switches to hard isolation mode and allocates independent physical time slots or frequency bands for critical services through a dedicated physical resource module. The system dynamically monitors the traffic status of the primary and backup channels and the fault risk score. When the fault risk score of the primary channel exceeds the preset switching threshold, the bandwidth of the primary channel is switched to the backup channel.

8. The communication fault early warning and resource scheduling method according to claim 7, characterized in that: The execution of the corresponding channel operation state control includes: When the fault risk score determines that the operating condition is a high-level fault hazard, a basic fault handling control instruction is generated, which includes a faulty component isolation instruction and an abnormal link disconnection instruction. When the fault risk score determines that the operating condition is free of fault risks or has low-level fault risks, an operation optimization control instruction is generated based on the correlation between the equipment operating status and energy consumption characteristics. The operation optimization control instruction includes adjusting the processor operating frequency, adjusting the power output power, and adjusting the link bandwidth allocation, so as to regulate the operating status of each key component to the optimal range of the corresponding operating condition.

9. The communication fault early warning and resource scheduling method according to claim 1, characterized in that: When obtaining the physical layer power consumption parameters of the device corresponding to the channel, the following is also included: Collect the air interface protocol parameters of the communication module, and construct a virtual energy consumption feature sequence based on the air interface protocol parameters; the air interface protocol parameters include at least the modulation and coding scheme level, the hybrid automatic repeat request retransmission frequency, and the TCP retransmission rate. Within the same time window, calculate the sliding window fit between the measured physical layer energy consumption parameter sequence and the virtual energy consumption feature sequence; When the sliding window matching degree is greater than or equal to the preset matching degree threshold, the measured physical layer energy consumption parameters are determined to be reliable, and the system deviation value between the measured physical layer energy consumption parameter sequence and the virtual module energy consumption characteristic sequence is calculated. If the absolute value of the system deviation is greater than the deviation tolerance threshold, it is determined that the sensor has temperature drift or zero drift error. The drift compensation coefficient is calculated based on the system deviation value to correct the measured physical layer energy consumption parameters.

10. A communication fault early warning and resource scheduling device, characterized in that: The communication fault early warning and resource scheduling method applicable to any one of claims 1-9 includes: The data acquisition module is used to acquire the physical layer power consumption parameters and network layer traffic data of the device corresponding to the channel; The feature fusion module calculates the correlation between the physical layer energy consumption parameters and the network layer traffic data based on a unified time window, determines strongly correlated samples based on the correlation, and performs feature fusion on the strongly correlated samples to obtain a device state feature tensor. The risk assessment module performs anomaly detection based on the device state feature tensor and obtains a fault risk score by combining it with time series prediction; it adjusts the channel bandwidth ratio according to the fault risk score and executes the corresponding channel operation state control.

Images