Method and system for dynamic early warning of communication equipment standing wave faults based on time series deep learning

Abstract

The application discloses a communication equipment standing wave fault dynamic early warning method and system based on time series deep learning, and relates to the technical field of communication.The method comprises the following steps: acquiring forward power, reflected power, standing wave ratio time series data and environmental monitoring data; calculating a reflected energy sequence, constructing a standing wave mismatch evolution sequence, and forming a basic sequence through time series alignment; performing decomposition on the basic sequence, constructing an evolution feature vector; calculating an environmental gradient and fluctuation intensity, fusing the feature vector to form a coupling vector; combining structure information coupling coding, forming an evolution sequence and performing prediction, and constructing a fault risk index.The technical problem that the prior art cannot simultaneously consider the standing wave time series characteristics, energy accumulation effect and operation environment coupling effect, resulting in inaccurate capture of early evolution characteristics of the standing wave fault and early warning lag is solved, and the technical effect of realizing dynamic and accurate early warning of the standing wave fault based on time series deep learning and multi-source data fusion and improving the reliability and operation efficiency of the communication system is achieved.

Description

Technical Field

[0001] This invention relates to the field of communication technology, specifically to a dynamic early warning method and system for standing wave faults in communication equipment based on time-series deep learning. Background Technology

[0002] In communication equipment operating scenarios such as mobile communication base stations, broadcast communication systems, and dedicated communication networks, the radio frequency (RF) transmission link typically consists of multiple components, including power amplifiers, feeders, jumpers, and antennas. During long-term operation, factors such as aging connection interfaces, moisture in the feeder, poor contact, or loose structure can easily lead to impedance mismatch in the RF link, resulting in high voltage standing wave ratios (VSWR) and reflected power. In severe cases, this can cause power amplifier overload, decreased signal transmission efficiency, or even equipment damage. Therefore, continuous monitoring and early warning of VSWR faults in communication equipment has become a key technical issue in the operation and maintenance of communication equipment.

[0003] In the RF operation and maintenance scenarios of communication base stations, standing wave ratio (SWR) data exhibits significant temporal sequence, multi-scale fluctuations, and dynamic characteristics coupled with environmental conditions. For example, changes in temperature, humidity, and wind speed can cause slight drifts in the electrical characteristics of the RF connection interface, resulting in complex temporal evolution characteristics of SWR and reflected power. Traditional threshold monitoring methods or anomaly detection methods based on single indicators often struggle to accurately characterize the evolution process of SWR mismatch. These methods typically rely on fixed thresholds or single-moment detection mechanisms. While they can trigger alarms when the SWR is already significantly abnormal, their ability to identify early, slow degradation stages is weak, easily leading to delayed warnings or false alarms. Summary of the Invention

[0004] This application provides a dynamic early warning method and system for standing wave (SWR) faults in communication equipment based on temporal deep learning. The key aspect lies in addressing the technical obstacles posed by the multi-scale temporal fluctuations, strong environmental coupling effects, and concealed evolution trends in monitoring data such as SWR and reflected power in communication base station radio frequency transmission link operation monitoring scenarios. These obstacles make it difficult to accurately model the evolution law of SWR mismatch and to identify early faults in a timely manner. The method constructs a temporal deep prediction algorithm based on multi-scale temporal decomposition and multi-channel feature correlation modeling. This is combined with a data processing flow that includes joint temporal alignment of reflected energy sequences and SWR mismatch evolution sequences, fusion of environmental coupling features, and encoding of radio frequency transmission path structures. This process enables multi-time-step prediction of the SWR state of communication equipment and dynamic assessment of the fault risk index, thereby improving the accuracy, foresight, and reliability of SWR fault early warning.

[0005] The first aspect of this application provides a dynamic early warning method for standing wave faults in communication equipment using time-series deep learning, the method comprising:

[0006] The process involves acquiring time-series data on forward power, reflected power, and voltage standing wave ratio (VSWR) in the radio frequency transmission link of a communication device, while simultaneously collecting monitoring data of the device's operating environment. Based on the forward power and transmitted power, a reflected energy sequence is calculated, and a standing wave mismatch evolution sequence is constructed based on the VSWR time-series data. The reflected energy sequence and the standing wave mismatch evolution sequence are then jointly time-aligned to form a basic standing wave state sequence. Multi-scale time-series decomposition is performed on the basic standing wave state sequence to construct a multi-dimensional standing wave evolution feature vector. The environmental change gradient and environmental fluctuation intensity are calculated based on the monitoring data of the device's operating environment, and the calculation results are fused with the multi-dimensional standing wave evolution feature vector to form a coupled multi-dimensional standing wave evolution feature vector under environmental coupling. The transmission path structure description information of the communication device's radio frequency transmission is coupled and encoded with the coupled multi-dimensional standing wave evolution feature vector to form a standing wave evolution feature sequence. This sequence is used to perform standing wave state prediction at multiple time steps, and the standing wave mismatch growth rate and energy accumulation index are calculated to construct a fault risk index.

[0007] A second aspect of this application provides a dynamic early warning system for standing wave faults in communication equipment based on time-series deep learning, the system comprising:

[0008] The system comprises the following modules: a data acquisition module, a timing module, and a timing module. The data acquisition module acquires forward power, reflected power, and voltage standing wave ratio (VSWR) time-series data from the radio frequency transmission link of the communication equipment, and simultaneously collects monitoring data of the equipment's operating environment. A timing alignment module calculates the reflected energy sequence based on the forward power and transmitted power, constructs a standing wave mismatch evolution sequence based on the VSWR time-series data, and performs joint timing alignment on the reflected energy sequence and the standing wave mismatch evolution sequence to form a basic standing wave state sequence. A sequence decomposition module performs multi-scale timing decomposition on the basic standing wave state sequence to construct a multi-dimensional standing wave evolution feature vector. A feature fusion module calculates the environmental change gradient and environmental fluctuation intensity based on the equipment's operating environment monitoring data, fuses the calculation results with the multi-dimensional standing wave evolution feature vector, and forms a coupled multi-dimensional standing wave evolution feature vector under environmental coupling. A state prediction module couples the transmission path structure description information of the communication equipment's radio frequency transmission with the coupled multi-dimensional standing wave evolution feature vector to form a standing wave evolution feature sequence. It then uses this sequence to perform standing wave state prediction at multiple time steps, calculates the standing wave mismatch growth rate and energy accumulation index, and constructs a fault risk index.

[0009] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0010] First, time-series monitoring data such as forward power, reflected power, and voltage standing wave ratio (VSWR) of the communication equipment's radio frequency transmission link are collected, along with monitoring information of the equipment's operating environment. Then, a reflected energy change sequence is calculated based on the power data, and a standing wave mismatch evolution sequence is constructed by combining it with the VSWR data. The two sequences are time-aligned to form a basic standing wave state sequence. Next, the sequence is decomposed into multi-scale time series, extracting multi-dimensional features characterizing the standing wave variation. Environmental change features calculated from environmental data are then fused with the standing wave evolution features to form environmental coupling features. Finally, feature encoding is performed using the communication equipment's radio frequency transmission path structure information to generate a standing wave evolution feature sequence. This sequence is used to predict the standing wave state at multiple future time steps, while simultaneously calculating the standing wave mismatch growth rate and reflected energy accumulation. Ultimately, a risk index is constructed to assess equipment failure risk, thereby achieving early warning and risk quantification for standing wave faults. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 A schematic flowchart of a time-series deep learning-based dynamic early warning method for standing wave faults in communication equipment provided in an embodiment of this application.

[0013] Figure 2 A schematic diagram of the structure of a time-series deep learning-based communication equipment standing wave fault dynamic early warning system provided in an embodiment of this application.

[0014] Figure labeling: Data acquisition module 11, temporal alignment module 12, sequence decomposition module 13, feature fusion module 14, state prediction module 15. Detailed Implementation

[0015] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0016] Example 1, as Figure 1 As shown, this application provides a dynamic early warning method for standing wave faults in communication equipment based on time-series deep learning, wherein the method includes:

[0017] Acquire timing data of forward power, reflected power, and voltage standing wave ratio in the radio frequency transmission link of communication equipment, and simultaneously collect monitoring data of the equipment's operating environment.

[0018] In this embodiment, a corresponding data acquisition unit is first deployed on the radio frequency (RF) transmission link of the communication device. This RF transmission link includes RF transmission components such as power amplifiers, duplexers, feeders, jumpers, combiners, and antennas. A data connection is established with the device's built-in power detection unit, VSWR detection unit, or an external RF monitoring device. Timing data of forward power, reflected power, and voltage standing wave ratio (VSWR) are continuously acquired according to a preset sampling period. Forward power characterizes the amount of RF energy output by the device towards the antenna; reflected power characterizes the amount of energy returning in the reverse direction due to impedance mismatch; and VSWR characterizes the degree of mismatch in the RF transmission link. To ensure the effectiveness of subsequent timing analysis, timestamps are uniformly added to all acquired RF parameters, and they are synchronously recorded according to the same clock reference to form a continuous and corresponding RF state timing dataset, including forward power, reflected power, and VSWR timing data. Meanwhile, environmental monitoring sensors are deployed in the operating environment of the equipment to collect real-time monitoring data on the equipment operating environment related to standing wave changes, such as temperature, humidity, and wind speed. These environmental data are also timestamped and stored synchronously according to a consistent time base that is consistent with or can be mapped to the radio frequency parameters, forming equipment operating environment monitoring data, which provides basic data support for subsequent reflection energy calculation, standing wave mismatch evolution modeling, and fault dynamic early warning.

[0019] For example, in one operational monitoring session, key parameters were continuously collected at a sampling period of 1 second. At timestamp 2025-05-12 10:15:01, the RF transmission link detected a forward power of 42.6W, a reflected power of 1.18W, and a voltage standing wave ratio (VSWR) of 1.32. Simultaneously, the ambient temperature near the equipment rack was recorded as 31.4℃, relative humidity as 58%RH, and wind speed as 2.1m / s. At timestamp 2025-05-12 10:15:02, the forward power was 42.8W, the reflected power as 1.21W, the VSWR as 1.34, the ambient temperature as 31.5℃, the relative humidity as 58.3%RH, and the wind speed as 2.0m / s.

[0020] Furthermore, the equipment operating environment monitoring data includes temperature data, humidity data, and wind speed data.

[0021] Preferably, the equipment operating environment monitoring data is used to reflect changes in the external environmental conditions of the communication equipment during actual operation. Specifically, this includes temperature data, humidity data, and wind speed data. Temperature data characterizes changes in the ambient air temperature or the temperature inside the cabinet. Temperature changes may alter the electrical properties of feeder, connector, or RF component materials, thus affecting the standing wave ratio (SWR). Humidity data reflects the water vapor content in the environment. High humidity can easily lead to moisture or oxidation at RF connection interfaces, causing impedance mismatch. Wind speed data reflects the airflow intensity in the area where the equipment is located. Higher wind speeds may cause slight swaying of the antenna structure or changes in the stress on connecting components, thus affecting the stability of RF transmission. During actual data acquisition, each environmental parameter is collected by environmental monitoring sensors according to a preset sampling period, and a corresponding timestamp is added to each set of data. This allows temperature, humidity, and wind speed data to be recorded synchronously with RF power and SWR data on the same timeline, forming a complete equipment operating environment monitoring data sequence. This provides fundamental data support for subsequent analysis of the impact of environmental changes on the SWR evolution process.

[0022] The reflected energy sequence is calculated based on the forward power and the transmitted power, and the standing wave mismatch evolution sequence is constructed based on the voltage standing wave ratio time series data. The reflected energy sequence and the standing wave mismatch evolution sequence are then jointly time-aligned to form the basic sequence of standing wave states.

[0023] In one embodiment, after obtaining continuous monitoring data of forward power, reflected power, and voltage standing wave ratio (VSWR), energy calculation processing is first performed on the RF power parameters. Specifically, based on the forward power and transmit power data corresponding to each sampling time step, the reflected power generated by impedance mismatch within that time step is calculated. This reflected power is then converted into reflected energy values ​​for the corresponding time period, forming a reflected energy sequence on a continuous time axis. This reflected energy sequence reflects the energy return changes caused by impedance mismatch during RF signal transmission and is one of the important indicators characterizing the changes in the standing wave state. Simultaneously, by calculating the amplitude, gradient, and trend of the VSWR time-series data within adjacent time steps or sliding time windows, a standing wave mismatch evolution sequence reflecting the degree of standing wave mismatch over time can be obtained. This standing wave mismatch evolution sequence describes the dynamic process of the standing wave state gradually evolving from a stable state to a mismatched state. Since the reflected energy sequence and the standing wave mismatch evolution sequence usually originate from different monitoring units, their sampling periods, data processing delays, and acquisition mechanisms may differ, easily causing a certain time offset on the time axis, thus affecting the correspondence between the two types of sequences. To ensure the accuracy of subsequent standing wave state analysis, joint time-series alignment is performed on the two types of sequences to establish a stable correspondence between changes in reflected energy and standing wave mismatch under a unified time reference. After time-series alignment, the aligned reflected energy sequence and the standing wave mismatch evolution sequence are jointly encoded at a unified time step to form a basic standing wave state sequence that can simultaneously characterize the features of reflected energy change and standing wave mismatch evolution. This process effectively reduces the impact of time asynchrony between different data sources on feature analysis, allowing for a more accurate expression of the dynamic correlation between reflected energy change and standing wave mismatch evolution. This provides a reliable data foundation for subsequent multi-scale time-series decomposition, standing wave evolution feature extraction, and standing wave state prediction.

[0024] Furthermore, the basic sequence of standing wave states includes:

[0025] The reflected power at each sampling time is calculated based on the forward power and transmitted power, and the reflected energy value at the corresponding time step is calculated based on the reflected power to form a reflected energy sequence. A sliding window analysis is performed on the voltage standing wave ratio (VSWR) time series data to calculate the amplitude and gradient of the VSWR change within adjacent time windows, constructing a VSWR mismatch evolution sequence characterizing the VSWR mismatch change process. The sequence time offset is calculated based on the correlation between the VSWR mismatch evolution sequence and the reflected energy sequence over a continuous time period, and dynamic time alignment processing is performed on the reflected energy sequence and the VSWR mismatch evolution sequence based on the sequence time offset. The dynamically time-aligned reflected energy sequence and the VSWR mismatch evolution sequence are jointly encoded to form a basic VSWR state sequence.

[0026] Preferably, the collected RF power data is first processed by calculating the reflected energy. For each sampling time step, the corresponding reflected power is calculated based on the forward power and transmit power data at that moment. Specifically, the reflected power value can be obtained by comparing the power difference between the transmit power and the forward power. This is because under ideal matching conditions, the forward power and transmit power are basically the same. When impedance mismatch exists, some energy will return in the reverse direction, thus forming reflected power. Subsequently, according to the sampling time interval, the reflected power is converted into the reflected energy value within the corresponding time step. That is, the reflected energy within that time period is obtained by multiplying the reflected power by the sampling time interval, thereby obtaining a reflected energy sequence on a continuous time axis. This reflected energy sequence can reflect the changes in energy return caused by impedance mismatch in the RF link. Next, a sliding window analysis is performed on the voltage standing wave ratio (VSWR) time series data. In this process, a fixed-length time window is first set, for example, a sliding window consisting of k consecutive sampling points. Within the window, the amplitude and gradient of the VSWR change are calculated. The amplitude can be calculated by the difference between the maximum and minimum values ​​within the window, representing the fluctuation of the standing wave state during that time period. The gradient can be calculated by the ratio of the VSWR change between adjacent sampling points to the time interval, representing the rate of change of standing wave mismatch. As the sliding window moves gradually along the time axis, a set of characteristic value sequences reflecting the trend of standing wave mismatch changes is obtained, thus constructing a standing wave mismatch evolution sequence. This sequence describes the dynamic process of the degree of standing wave mismatch changing over time. Since the reflected energy sequence and the standing wave mismatch evolution sequence may originate from different sampling mechanisms, their time axes may be offset. Therefore, time alignment processing is performed on the two types of sequences. Specifically, the correlation between the reflected energy sequence and the standing wave mismatch evolution sequence is calculated within a continuous time interval. For example, the correlation coefficient between the two sequences under different time offsets is determined through sliding correlation analysis or cross-correlation calculation methods. The time offset corresponding to the highest correlation is selected as the optimal alignment offset for the sequences. Then, based on the calculated time offset, time shifting, interpolation, or resampling is performed on one type of sequence to establish a correspondence between the two types of sequences under a unified time reference, thereby completing dynamic time series alignment. After time series alignment, the aligned reflected energy sequence and the standing wave mismatch evolution sequence are jointly encoded according to a unified time step. Through vector concatenation or feature combination, the reflected energy value and standing wave mismatch feature corresponding to the same time step are combined into a joint feature vector. With the continuous arrangement of time steps, a basic standing wave state sequence can be formed. This basic standing wave state sequence can simultaneously characterize the reflected energy change characteristics and the standing wave mismatch evolution characteristics, providing a unified and complete data foundation for subsequent multi-scale time series decomposition, standing wave evolution feature extraction, and fault early warning model construction.

[0027] Multi-scale temporal decomposition is performed on the basic sequence of the standing wave state to construct a multi-dimensional standing wave evolution feature vector.

[0028] In one embodiment, after obtaining the basic sequence of standing wave states, multi-scale time series decomposition is required to more comprehensively characterize the changing characteristics of the standing wave states at different time scales. Since the standing wave state in the actual operation of communication equipment radio frequency transmission links often simultaneously contains multiple time-scale features such as short-term disturbances, periodic fluctuations, and long-term degradation trends, analyzing the standing wave state from only a single time scale can easily lead to insufficient characterization of the standing wave evolution law. Therefore, the basic sequence of standing wave states is first organized according to a unified time step to form continuous time series data, and then multiple analysis windows at different time scales are set, such as short-period, medium-period, and long-period scales. By decomposing the basic sequence of standing wave states at different time scales, the original sequence can be split into multiple components reflecting different changing laws, such as short-period disturbance components, medium-period fluctuation components, and long-term trend components. Then, at the short time scale, the rapid fluctuations of the standing wave state between adjacent sampling times are analyzed to extract disturbance features reflecting transient mismatches or instantaneous reflection fluctuations at the radio frequency connection interface. At a medium timescale, the periodic fluctuations of the basic sequence of standing wave states are analyzed to extract medium-period fluctuation features reflecting the periodic changes in standing waves. At a long timescale, the overall trend of standing wave states over time is analyzed to extract trend features reflecting the gradual degradation of RF links or the slow changes in impedance matching. After obtaining multiple feature parameters at different timescales, these features are concatenated in a preset order to form a multi-dimensional standing wave evolution feature vector. This multi-scale feature construction method allows for a more complete expression of the changes in standing wave states at multiple time levels, thereby improving the characterization capability of standing wave state features and providing a more stable and accurate feature foundation for subsequent environmental coupling feature fusion, standing wave state prediction, and fault risk assessment.

[0029] Furthermore, performing multi-scale temporal decomposition on the basic standing wave state sequence includes:

[0030] The standing wave state base sequence is decomposed into a multi-scale sliding window at different time scales to obtain short-period disturbance components, medium-period fluctuation components, and long-period trend components. The reflection energy change gradient of the short-period disturbance component within adjacent time windows is calculated, and the reflection energy change gradient is used to characterize the transient mismatch degree of the radio frequency connection interface. The periodic fluctuation amplitude and periodic stability of the medium-period fluctuation component are calculated. The slope and drift amplitude of the long-period trend component within continuous time segments are calculated to construct degradation trend parameters. A multidimensional standing wave evolution feature vector is generated based on the combination of the reflection energy change gradient, periodic fluctuation amplitude, periodic stability, and degradation trend parameters.

[0031] Preferably, the established standing wave state base sequence is first arranged along a unified time axis, where each time step corresponds to state data composed of reflection energy characteristics and standing wave mismatch characteristics. To extract the variation patterns of the standing wave state across different time ranges, multiple sliding windows of varying lengths are used to perform multi-scale decomposition processing on the standing wave state base sequence. In this process, a shorter first time window is set to extract short-period disturbance information, a medium-length second time window is set to extract medium-period fluctuation information, and a longer third time window is set to extract long-period trend information. Each sliding window slides continuously along the time axis at a preset step size. Within the time interval covered by each window, the sequence variation characteristics at the corresponding scale are calculated. For example, the short-period window length can be set to 10–30 sampling points, the medium-period window length can be set to 50–100 sampling points, and the long-period window length can be set to 200–500 sampling points. Each sliding window moves gradually along the time axis by 1–5 sampling points. In this way, the original standing wave state basic sequence can be decomposed into short-period disturbance components, medium-period fluctuation components, and long-period trend components. Among them, the short-period disturbance components mainly reflect the rapid fluctuation characteristics in a short period of time; the medium-period fluctuation components mainly reflect the repeated fluctuation patterns within a certain period; and the long-period trend components mainly reflect the slow changes and degradation processes over a longer operating period.

[0032] After obtaining the short-period perturbation components, their changes between adjacent time windows are analyzed. Specifically, the mean and peak values ​​of reflected energy within two adjacent short-period windows are compared, and the rate of change of reflected energy over time is calculated based on the window interval. For example, suppose the statistical values ​​of reflected energy corresponding to two adjacent short-period windows are respectively... and The corresponding time interval is Then you can This serves as the gradient of reflected energy variation within this segment. A larger gradient value indicates a more drastic change in reflected energy over a short period, typically suggesting a more pronounced transient mismatch or momentary contact anomaly at the RF connection interface. Therefore, it can be used to characterize the degree of transient mismatch at the RF connection interface. For example, with a window length of 10 seconds, the reflected energy sequence within the time segment 10:20:01 to 10:20:10 is {1.18J, 1.21J, 1.23J, 1.30J, 1.34J, 1.29J, 1.26J, 1.24J, 1.27J, 1.31J}. Calculating the average reflected energy within this window yields an average value of approximately 1.26J. Within the next adjacent window from 10:20:11 to 10:20:20, the reflected energy sequence is {1.32J, 1.36J, 1.39J, 1.41J, 1.38J, 1.35J, 1.33J, 1.30J, 1.28J, 1.27J}, with an average value of approximately 1.34J. The gradient of reflected energy variation, calculated from the average value of adjacent windows, is approximately 0.008J / s. This gradient reflects a certain degree of transient mismatch at the RF interface during this time period.

[0033] After obtaining the mid-cycle fluctuation components, their periodic variation patterns are analyzed. Specifically, within each mid-cycle analysis window, the amplitude of the standing wave state's fluctuation around the local mean is statistically analyzed. For example, the difference between peak and trough values, standard deviation, or mean square fluctuation can be used to represent the periodic fluctuation amplitude, reflecting the strength of the standing wave state's fluctuation within that time period. Simultaneously, to measure the stability of the periodic changes, the time interval between peaks or troughs is statistically analyzed across multiple consecutive mid-cycle windows. For example, let the times of two consecutive peaks be... and Then the period length can be calculated. By comparing the degree of variation among multiple period lengths and calculating the standard deviation of the period lengths, a period stability index is obtained. This period stability index characterizes whether the periodic changes of standing waves are regular and consistent. When the periodic fluctuations maintain similar period lengths and amplitude characteristics across multiple time intervals, it indicates high period stability; conversely, it indicates that the periodic changes are more disordered. Through the above processing, the evolution characteristics of periodic standing waves caused by environmental changes, equipment load fluctuations, or structural responses can be effectively extracted.

[0034] After obtaining the long-period trend component, its overall change trend over a relatively long continuous time period is analyzed. This involves trend fitting processing of the long-period trend component, for example, using linear regression and least squares to calculate the slope of the sequence over the continuous time period. This slope characterizes whether the standing wave state continuously increases, continuously decreases, or remains relatively stable over time. A continuously increasing slope indicates that the RF link mismatch is gradually worsening. Furthermore, the drift amplitude can be calculated, which is the difference between the initial and final values ​​of the trend component within a preset long-period period. This drift amplitude reflects the degree to which the standing wave state deviates from the normal operating range over a long period. By combining the slope and the drift amplitude, a degradation trend parameter describing the long-term degradation state of the equipment can be formed.

[0035] After extracting features at each scale, the reflection energy change gradient corresponding to the short-period disturbance component, the periodic fluctuation amplitude and periodic stability corresponding to the medium-period fluctuation component, and the degradation trend parameters corresponding to the long-period trend component are uniformly combined and spliced ​​into feature vectors corresponding to the same time step or the same analysis segment according to a preset order. As the sliding window continuously advances along the time axis, multiple continuous multidimensional feature vectors can be generated, thus forming a multidimensional standing wave evolution feature vector sequence. This multidimensional standing wave evolution feature vector can simultaneously reflect the evolution information of the standing wave state at three levels: short-term disturbance, medium-term fluctuation, and long-term degradation, providing a more complete feature basis for subsequent environmental coupling analysis, state prediction, and fault risk assessment.

[0036] The environmental change gradient and environmental fluctuation intensity are calculated based on the equipment operating environment monitoring data. The calculation results are then fused with the multidimensional standing wave evolution feature vector to form a coupled multidimensional standing wave evolution feature vector under environmental coupling.

[0037] In one embodiment, after obtaining the multidimensional standing wave evolution feature vector, it is also necessary to consider the impact of changes in the equipment operating environment on the standing wave state. Therefore, the collected equipment operating environment monitoring data is processed by calculating environmental change characteristics. Specifically, temperature data, humidity data, and wind speed data are first organized according to a time axis consistent with the basic sequence of the standing wave state to form continuous environmental monitoring time series data, including temperature series, humidity series, and wind speed series. Subsequently, environmental change gradients are calculated for various types of environmental monitoring data. The environmental change gradient is used to characterize the rate of change of environmental parameters in adjacent time periods. It can be obtained by calculating the ratio between the change in environmental parameter values ​​and the time interval between two adjacent time steps. For example, in the temperature series, the ratio of the temperature difference between adjacent time points to the time interval can be calculated as the temperature change gradient. By statistically analyzing these environmental change gradients, the changing trend of environmental conditions in a short period of time can be reflected, thereby identifying the possible impact of environmental factors on the radio frequency link state. Afterward, within a preset time window, the temperature, humidity, and wind speed data are statistically processed to calculate indicators such as the standard deviation, range, or average fluctuation amplitude of the data within the window to reflect the stability of environmental parameters within that time period. When environmental fluctuations are significant, it indicates drastic changes in the surrounding environment, potentially having a more pronounced impact on the stability of the RF connection interface and the standing wave (SWR) state. After obtaining the environmental change gradient and environmental fluctuation intensity, these environmental features are fused with the previously extracted multidimensional SWR evolution feature vector. Specifically, the environmental change gradient, environmental fluctuation intensity, and multidimensional SWR evolution feature vector are combined at a unified time step. Through vector concatenation or feature expansion, environmental features are added as a new dimension to the original SWR feature vector, forming an extended feature vector that incorporates environmental influences. This approach allows for the joint expression of environmental change information and SWR state change information, forming a coupled multidimensional SWR evolution feature vector under environmental coupling. This coupled multidimensional SWR evolution feature vector not only reflects the SWR evolution characteristics of the RF link itself but also reflects the impact of environmental changes on the SWR state, providing more comprehensive feature input for subsequent SWR state prediction and fault risk assessment.

[0038] Based on the transmission path structure description information of the radio frequency transmission of communication equipment and the coupled multidimensional standing wave evolution feature vector, a standing wave evolution feature sequence is formed. The standing wave evolution feature sequence is used to perform standing wave state prediction at multiple time steps, and the standing wave mismatch growth rate and energy accumulation index are calculated to construct a fault risk index.

[0039] In one embodiment, after obtaining the coupled multidimensional standing wave evolution feature vector under environmental coupling, the standing wave features are structured and encoded by combining the structural information of the communication equipment's radio frequency transmission link. This structural information characterizes the transmission relationship of the radio frequency signal between various components within the equipment, such as the connection order and signal propagation path between power amplifiers, feeders, jumpers, and antennas. Since the radio frequency transmission link of communication equipment typically has well-defined node connection relationships and signal propagation paths, the structural positions and propagation path lengths between different nodes will affect the propagation mode of the reflected wave and the standing wave evolution process. If the analysis is based solely on the standing wave timing features while ignoring the transmission path structural information, the propagation characteristics of the standing wave state changes between different structural nodes may not be fully expressed. Therefore, during the coupling encoding process, the coupled multidimensional standing wave evolution feature vector corresponding to each time step is first combined with the radio frequency transmission path structural information on a unified time axis. For example, the path structural features are mapped to the standing wave evolution feature space through structural feature embedding or feature splicing, thereby generating a standing wave evolution feature representation containing structural information. By arranging time steps sequentially, a standing wave evolution characteristic sequence reflecting the change of standing wave state over time can be formed. This standing wave evolution characteristic sequence is then input into the time-series deep prediction channel for multi-time-step prediction processing, resulting in a standing wave prediction state sequence. This sequence reflects the changing trend of the standing wave state over a future period. Next, this prediction state sequence is analyzed. By comparing the changes in standing wave mismatch between adjacent prediction time steps, a standing wave mismatch growth rate index is calculated. Simultaneously, based on the changes in reflected energy during corresponding time periods in the prediction sequence, the reflected energy is cumulatively calculated to obtain an energy accumulation index. Finally, the standing wave mismatch growth rate and energy accumulation index are comprehensively processed. By normalizing each index and fusing them according to preset weights, a fault risk index is obtained, comprehensively characterizing the degree of equipment fault risk. When the fault risk index reaches or exceeds a preset threshold, it can be determined that the communication equipment's radio frequency transmission link has a high standing wave fault risk, thus providing an early warning basis for equipment operation and maintenance, and improving the accuracy and reliability of standing wave fault prediction and risk assessment for communication equipment.

[0040] Furthermore, standing wave state prediction at multiple time steps is performed using standing wave evolution characteristic sequences, including:

[0041] The standing wave evolution feature sequence is used to construct a continuous time-series input segment according to a preset time window. This continuous time-series input segment is then input into a time-series depth prediction channel to obtain a standing wave prediction state sequence for multiple time steps. The increment of the standing wave mismatch ratio change between adjacent prediction time steps is calculated based on the standing wave prediction state sequence, and the increment of the standing wave mismatch ratio change is accumulated to form a standing wave mismatch evolution trajectory. The standing wave mismatch growth rate is calculated based on the standing wave mismatch evolution trajectory, and an energy accumulation index is calculated by combining the reflected energy change data corresponding to the standing wave prediction state sequence. The standing wave mismatch growth rate and the energy accumulation index are normalized and fused to construct a fault risk index.

[0042] Preferably, the already formed standing wave evolution feature sequence is first divided into time windows according to a preset time window length L and sliding step size. The standing wave evolution feature sequence is divided into multiple consecutive time-series input segments according to continuous time order. For example, let the standing wave evolution feature sequence be... , of which each Given the standing wave evolution feature vector corresponding to the time step, input segments of length L can be constructed sequentially. , Multiple consecutive time segments are input. Each input segment maintains its original temporal order and reflects the continuous change characteristics of the standing wave state within that time interval. These consecutive time-series input segments are then fed into a time-series deep prediction channel for prediction processing. This channel can employ a deep learning model suitable for time series prediction, such as a recurrent neural network, a long short-term memory network, or an attention-based time-series prediction structure. By learning the temporal correlations between historical standing wave evolution characteristics, it predicts the standing wave state at multiple future time steps, thus obtaining a predicted standing wave state sequence.

[0043] After obtaining the predicted standing wave (SWR) state sequence, an SWR mismatch change analysis is performed on the prediction results. Specifically, based on the SWR mismatch ratios corresponding to adjacent time steps in the prediction sequence, the increment of the SWR mismatch ratio change between adjacent prediction time steps is calculated by subtracting the values. By gradually accumulating the increments of change between multiple consecutive prediction time steps, the change path of the SWR mismatch over time can be obtained, thus forming the SWR mismatch evolution trajectory. This trajectory describes the overall trend of the SWR mismatch degree within a future time period. After obtaining the SWR mismatch evolution trajectory, the degree of increase in SWR mismatch per unit time is obtained by trend fitting or rate of change calculation. For example, the average growth rate of the SWR mismatch can be obtained by calculating the difference between the trajectory's endpoint and its initial value and dividing it by the corresponding time span. Simultaneously, based on the reflected energy change data corresponding to each time step in the standing wave (SWR) prediction state sequence, the reflected energy is cumulatively calculated. That is, the reflected energy values ​​at each time step within the prediction time interval are progressively added together to obtain the total reflected energy within that time interval, thus forming an energy accumulation index. This energy accumulation index reflects the degree of reflected energy accumulation caused by SWR mismatch within a future time range. Finally, the SWR mismatch growth rate and energy accumulation index are normalized to eliminate the influence between different index dimensions, bringing them within a unified numerical range. Then, the normalized SWR mismatch growth rate and energy accumulation index are fused according to preset weights to generate a fault risk index. When the obtained fault risk index exceeds a preset threshold, it can be determined that the communication equipment's radio frequency transmission link has a high SWR fault risk, thereby triggering a corresponding early warning mechanism and providing a basis for equipment operation and maintenance.

[0044] Furthermore, the continuous time-series input segments are input into the time-series depth prediction channel to obtain a standing wave prediction state sequence at multiple time steps, including:

[0045] The standing wave evolution features in the continuous time-series input segment are divided into multiple feature subsequences according to disturbance variation features, periodic fluctuation features, and trend degradation features. These feature subsequences are then input into a mapping feature processing subchannel, which includes a disturbance feature processing subchannel, a periodic feature processing subchannel, and a trend feature processing subchannel. The disturbance time-series features characterizing the short-period reflection fluctuation mode of the standing wave are extracted from the disturbance feature processing subchannel; the periodic evolution features characterizing the periodic variation law of the standing wave are extracted from the periodic feature processing subchannel; and the trend feature processing subchannel extracts… The trend evolution characteristics characterize the long-term degradation trend of standing waves; cross-sub-channel feature correlation modeling is performed on the disturbance time series characteristics, periodic evolution characteristics, and trend evolution characteristics; a standing wave evolution characteristic correlation matrix is ​​constructed by calculating the correlation weights between features; and correlation enhancement processing of each feature is performed based on the standing wave evolution characteristic correlation matrix; unified time series fusion encoding is performed on each feature after correlation enhancement processing to generate a standing wave evolution fusion characteristic sequence; the standing wave evolution fusion characteristic sequence is input into the prediction sub-channel to perform multi-time step standing wave state prediction to generate a standing wave prediction state sequence for multiple time steps.

[0046] Optionally, the continuous time-series input segment constructed according to a preset time window is first subjected to feature decomposition. This continuous time-series input segment includes a standing wave evolution feature vector for each time step, which contains multiple dimensions such as the gradient of reflected energy change, periodic fluctuation amplitude, periodic stability, degradation trend parameters, and environmental coupling characteristics. Based on the physical meaning and time scale attributes reflected by each feature, the features in each dimension of the standing wave evolution feature vector are classified. Features reflecting rapid changes within a short time range are classified as perturbation change features, features reflecting repetitive fluctuation behavior are classified as periodic fluctuation features, and features reflecting long-term slow drift or degradation trends are classified as trend degradation features, thus forming multiple feature subsequences. Subsequently, the above multiple feature subsequences are respectively input into the corresponding mapping feature processing sub-channels in the time-series depth prediction channel. These mapping feature processing sub-channels include perturbation feature processing sub-channels, periodic feature processing sub-channels, and trend feature processing sub-channels to achieve specialized processing for features at different time scales.

[0047] In the perturbation feature processing sub-channel, short-period variation modeling is performed on the perturbation feature sub-sequence. Specifically, within continuous time steps, local sequence convolution is used to extract the perturbation temporal features characterizing the short-period reflection wave pattern of the standing wave. Specifically, the perturbation variation feature sub-sequence is input into a one-dimensional local convolution processing unit in chronological order. A fixed-length convolution window is set within the continuous time step range, for example, a local time window composed of several adjacent sampling points. The convolution window slides gradually along the time axis, performing convolution operations on the reflection energy variation data within the window. By weighted combining the variation patterns between adjacent time steps, local response values ​​reflecting the rapid fluctuation characteristics of reflection energy within a short period are extracted. During the convolution processing, different convolution kernels can be used to capture different forms of short-period wave structures, such as abrupt fluctuations, continuous oscillating fluctuations, or gradually changing fluctuations. After convolution, a set of disturbance characteristic response sequences characterizing the local reflection energy change pattern can be obtained. Through nonlinear mapping or normalization, a disturbance time sequence feature vector is formed, which is used to describe the short-period standing wave reflection wave pattern caused by transient impedance changes or small disturbances at the connection interface in the radio frequency transmission link.

[0048] In the periodic feature processing sub-channel, periodic variation modeling is performed on the periodic feature sub-sequence. That is, the periodicity of the standing wave variation is detected within a certain time interval. For example, the period length is identified by calculating the sequence autocorrelation function or the statistical peak-to-trough interval, and the periodic fluctuation amplitude and periodic stability are further calculated. The periodic fluctuation amplitude can be represented by the difference between the peak and trough values; the periodic stability can be calculated by the degree of change between multiple adjacent period lengths, thereby extracting periodic evolution features that can characterize the periodic variation law of the standing wave. These features typically reflect the periodic standing wave changes caused by changes in ambient temperature, changes in equipment operating load, or the periodic response of the antenna structure.

[0049] In the trend feature processing sub-channel, long-term trend modeling is performed on the trend degradation feature sub-sequence. That is, trend fitting processing is performed on the standing wave feature within a longer time window. The slope of the sequence change is calculated through linear regression or local trend fitting, and the overall drift amplitude of the sequence within this time interval is also calculated. For example, if in the time interval... The slope of the internal trend change is If the value is calculated, it can be used as one of the indicators of the standing wave degradation trend. At the same time, the offset of the trend sequence relative to the historical benchmark value is calculated to reflect the degree of long-term mismatch accumulation, thereby extracting the trend evolution characteristics that can characterize the long-term degradation trend of the standing wave.

[0050] After extracting features from each sub-channel, cross-sub-channel feature correlation modeling is performed on the perturbation time-series features, periodic evolution features, and trend evolution features. Since actual standing wave evolution is typically formed by the combined effects of short-term perturbations, periodic fluctuations, and long-term degradation, it is necessary to analyze the synergistic changes among different features. At this point, the correlation weights between features are obtained by calculating the correlation coefficients or similarity indices between each feature sequence, and a standing wave evolution feature correlation matrix is ​​constructed based on these correlation weights. For example, suppose the three extracted features are... Then, the correlation between each pair of elements can be calculated and an association matrix can be constructed. Subsequently, correlation enhancement processing is performed on various features based on the correlation matrix. That is, features are weighted and combined according to their correlation weights to strengthen the expression of features with a high degree of correlation with the current evolutionary state, thereby improving the effectiveness of feature representation. After completing the correlation enhancement processing, the perturbation time series features, periodic evolution features, and trend evolution features are fused and encoded in a unified time order. That is, the three types of features corresponding to the same time step are combined into a fused feature vector by concatenating feature vectors, and then arranged continuously with time steps to form a standing wave evolution fused feature sequence. This standing wave evolution fused feature sequence can simultaneously express the comprehensive change information of the standing wave state at three levels: short-period perturbation, medium-period fluctuation, and long-term degradation.

[0051] After obtaining the standing wave evolution fusion feature sequence, this sequence is input into the prediction sub-channel to perform multi-time-step standing wave state prediction. This prediction sub-channel is constructed using a Long Short-Term Memory (LSTM) network structure. For this prediction sub-channel, the sample standing wave evolution fusion feature sequences from the sample library are used as the input sequence. The input sequence is then fed into the first-layer LSTM unit for temporal feature extraction. The first-layer LSTM unit can be set to 64 hidden units to extract short-term and medium-term temporal relationships from the standing wave evolution feature sequence. The output of this layer is the hidden state sequence. To enhance the model's ability to express complex temporal features, a Dropout layer can be added after the first-layer LSTM unit, with a dropout ratio of 0.2 to reduce the risk of overfitting. Next, the hidden state sequence output from the first layer is input into the second-layer LSTM unit to further extract higher-level time-dependent features. The second-layer LSTM unit can be set to 32 hidden units and only outputs the hidden state vector of the final time step, representing the comprehensive state of the standing wave evolution within the current time window. Then, the hidden state vector is input into a fully connected mapping layer for state prediction. Specifically, a fully connected layer can be constructed, with the output dimension set to the number of prediction time steps, for example, 5, indicating the prediction of the standing wave state values ​​for the next 5 time steps. The activation function of this layer can be a linear function to output a continuous numerical sequence of predicted standing wave states. During model training, the mean squared error loss function can be used as the evaluation index for prediction error. The optimization algorithm can use the Adam optimizer, with a learning rate of 0.001, a batch size of 32, and 50-100 training epochs. After training, the prediction sub-channel can learn the temporal mapping relationship between the standing wave evolution feature sequence and the future standing wave state, thereby achieving the prediction of the standing wave state for multiple time steps and providing a predictive data basis for subsequent calculations of the standing wave mismatch growth rate and fault risk index.

[0052] After receiving the standing wave evolution fusion feature sequence, the prediction sub-channel will gradually extract the temporal correlation features through two layers of LSTM to generate standing wave state values ​​for multiple future time steps, thereby generating a standing wave prediction state sequence. This standing wave prediction state sequence is used to characterize the development trend of the standing wave state in the future time segment and provides a predictive data basis for subsequent standing wave mismatch growth rate calculation, reflection energy accumulation analysis, and fault risk assessment, thereby improving the accuracy and stability of standing wave fault prediction for communication equipment.

[0053] Furthermore, the construction of the transmission path structure description information includes:

[0054] The power amplifiers, feeders, jumpers, and antennas in the radio frequency (RF) transmission link of the communication equipment are abstracted into nodes and mapped to RF transmission nodes. A set of node connection relationships is established based on the physical connection relationships between each RF transmission node, and an RF transmission topology is constructed based on the signal transmission direction between the RF transmission nodes. The signal propagation path and path length parameters of the RF transmission nodes are calculated based on the RF transmission topology, and the hierarchical relationship is determined. A transmission path structure matrix is ​​constructed based on the signal propagation path and path length parameters and the hierarchical relationship, and the transmission path structure matrix is ​​used as the transmission path structure description information.

[0055] Preferably, the key components of the communication equipment's RF transmission link, such as power amplifiers, feeders, jumpers, and antennas, are first structurally represented. Each device with signal transmission or connection functions is abstracted as an RF transmission node. For example, the power amplifier can be abstracted as a signal output node, the feeder and jumper as signal transmission nodes, and the antenna as a signal radiation node. This node abstraction transforms the originally complex RF hardware connections into a structural network composed of multiple nodes, thus providing a foundation for subsequent topology modeling. After completing the node abstraction, a set of node connection relationships is established based on the physical connections between each RF transmission node. Specifically, the actual connection relationships between each node are determined by reading the equipment structure configuration file, equipment design drawing, or field connection relationship record. For example, the power amplifier node is connected to the feeder node via a jumper, and the feeder node is then connected to the antenna node. Subsequently, based on the propagation direction of the radio frequency (RF) signal in the link, the connection relationships between nodes are marked as directional signal transmission relationships, thereby constructing an RF transmission topology. This RF transmission topology can be represented as a directed network structure, where nodes represent RF device units, and directed edges represent the paths of RF signals from one node to another. After obtaining the RF transmission topology, the signal propagation paths between each RF transmission node are calculated. That is, starting from the signal source node, the propagation paths of the RF signal from the source node to each target node are determined by traversing the directed connections in the topology level by level. Simultaneously, the path length parameters of each propagation path are calculated based on parameters such as the physical connection distance between nodes, feeder length, or jumper length. By traversing the entire topology and calculating the path lengths, the path position and propagation distance information of each node in the signal transmission link can be obtained. Furthermore, based on the relative positions of nodes in the signal propagation paths, the hierarchical relationships between nodes can be determined, such as the signal source layer, transmission layer, and terminal radiation layer, thus forming a hierarchical RF transmission path model.

[0056] After obtaining the signal propagation path, path length parameters, and node hierarchy, this information is processed into a structured representation. Specifically, a transmission path structure matrix is ​​constructed based on the connection relationships between nodes, the propagation path length, and the node hierarchy. For example, matrix elements can represent information such as whether there is a connection between nodes, the length of the connection path, or the hierarchy difference between nodes. In this way, the structural topology information of the radio frequency transmission link can be uniformly represented in matrix form, thus forming a transmission path structure matrix. This matrix can completely describe the structural relationships of the radio frequency transmission link of the communication equipment and serves as the transmission path structure description information for subsequent coupled analysis of standing wave evolution characteristics and structural information, as well as state prediction modeling. This more accurately reflects the propagation characteristics of standing wave changes between different radio frequency nodes, improving the accuracy of standing wave state analysis and fault prediction.

[0057] Furthermore, the system performs early warning level matching based on the fault risk index, configures early warning signal issuance, and uses the early warning signal issuance to perform early warning issuance management.

[0058] Preferably, after obtaining the fault risk index, the calculated fault risk index is matched against pre-defined risk level classification rules. For example, the risk index can be divided into multiple risk ranges of different levels, such as a normal state range, a low-risk warning range, a medium-risk warning range, and a high-risk warning range. When the fault risk index falls into a different risk range, the communication equipment is determined to be in the corresponding level of operational risk, thus completing the warning level matching. After completing the warning level matching, a corresponding early warning signal is generated based on the determined warning level. This early warning signal includes the warning level identifier, equipment identifier information, the current risk index value, and the corresponding warning time. The early warning signal can be transmitted to the equipment operation monitoring platform through an internal data interface, or it can be converted into visual alarm information, system log records, or alarm prompts for output. For example, under a low-risk warning level, a prompt warning signal can be generated to remind maintenance personnel to pay attention to the equipment's operating status; under a medium-risk warning level, an alarm signal can be generated and equipment inspection can be recommended; under a high-risk warning level, an emergency alarm signal can be triggered and equipment maintenance or fault diagnosis can be prompted. After generating an early warning signal, the warning information is managed uniformly using this signal. This includes recording, classifying, storing, and archiving historical warning data. Simultaneously, the warning information is pushed to the equipment operation and maintenance management system or remote monitoring platform, enabling maintenance personnel to obtain real-time information on the equipment's operational risk status. Furthermore, different warning handling strategies can be implemented based on the warning level. For example, in cases of higher risk, the frequency of warning alerts can be increased, key equipment status monitoring can be initiated, or subsequent maintenance procedures can be triggered. Through this early warning management method, timely alerts and tiered management of communication equipment standing wave fault risks can be achieved, thereby improving the efficiency of equipment operation monitoring and maintenance management.

[0059] Furthermore, an alarm response window is configured according to the alarm dispatch signal. If no alarm response data is detected in the alarm response window, the alarm dispatch signal is upgraded, and alarm dispatch management is performed.

[0060] Preferably, after generating an early warning signal and completing initial early warning management, to ensure timely response to the early warning information, a corresponding alarm response window is configured for each early warning message based on its corresponding warning level. This alarm response window is a preset time interval used to monitor the processing feedback of maintenance personnel or the management system to the early warning information. For example, a longer response time window can be set for low-level early warnings, while a shorter response time window can be set for high-level early warnings to facilitate quick confirmation of the handling status. After the alarm response window is opened, the system continuously monitors for the existence of corresponding alarm response data. This alarm response data includes confirmation operation records from maintenance personnel on the monitoring platform, remote inspection commands, equipment status verification data, or processing feedback information automatically generated by the system. When the corresponding alarm response data is detected within the preset alarm response window, the early warning is considered to have been processed or confirmed in a timely manner, and the response information can be recorded and the regular early warning management process can continue. If no corresponding alarm response data is detected before the alarm response window ends, the current early warning information is determined to have not been processed in a timely manner. In this scenario, to prevent potential fault risks from being overlooked, the existing early warning signals will be escalated. For example, the current warning level will be raised to a higher level, and a new escalated early warning signal will be generated and re-push to the monitoring platform or operation and maintenance management system. Furthermore, the scope of early warning notifications can be expanded, for example, by sending alarm prompts to more maintenance personnel or management terminals and increasing the frequency of alarm notifications. Finally, the early warning management process continues based on the escalated early warning signal, recording, monitoring, and tracking subsequent responses to the escalated early warning information. By setting up alarm response windows and early warning escalation mechanisms, it can be ensured that critical early warning information can be automatically escalated to a higher level if it is not processed in a timely manner, thereby improving the timeliness and reliability of communication equipment standing wave fault early warning management.

[0061] In summary, the embodiments of this application have at least the following technical effects:

[0062] First, time-series data of forward power, reflected power, and voltage standing wave ratio (VSWR) in the radio frequency transmission link of the communication equipment are acquired, and equipment operating environment monitoring data are collected simultaneously. Then, the reflected energy sequence is calculated based on the forward power and transmitted power, and a standing wave mismatch evolution sequence is constructed based on the VSWR time-series data. The reflected energy sequence and the standing wave mismatch evolution sequence are jointly time-aligned to form a basic standing wave state sequence. Next, multi-scale time-series decomposition is performed on the basic standing wave state sequence to construct a multi-dimensional standing wave evolution feature vector. Then, the environmental change gradient and environmental fluctuation intensity are calculated based on the equipment operating environment monitoring data, and the calculation results are fused with the multi-dimensional standing wave evolution feature vector to form a coupled multi-dimensional standing wave evolution feature vector under environmental coupling. Finally, the transmission path structure description information of the communication equipment's radio frequency transmission is coupled and encoded with the coupled multi-dimensional standing wave evolution feature vector to form a standing wave evolution feature sequence. The standing wave evolution feature sequence is used to perform standing wave state prediction at multiple time steps, and the standing wave mismatch growth rate and energy accumulation index are calculated to construct a fault risk index. This invention addresses the technical problem that existing methods struggle to simultaneously consider the temporal characteristics of standing waves, energy accumulation effects, and the coupled influence of the operating environment, leading to inaccurate capture of early evolution characteristics of standing wave faults and delayed early warnings. It achieves the technical effect of realizing dynamic and accurate early warning of standing wave faults based on temporal deep learning and multi-source data fusion, thereby improving the reliability and operation and maintenance efficiency of communication systems.

[0063] Example 2, based on the same inventive concept as the time-series deep learning-based dynamic early warning method for communication equipment standing wave faults in the aforementioned examples, such as... Figure 2 As shown, this application provides a dynamic early warning system for standing wave faults in communication equipment based on time-series deep learning, wherein the system includes:

[0064] Data acquisition module 11: Acquires forward power, reflected power, and voltage standing wave ratio (VSWR) time-series data from the radio frequency transmission link of the communication equipment, and simultaneously collects equipment operating environment monitoring data; Timing alignment module 12: Calculates the reflected energy sequence based on the forward power and transmitted power, constructs a standing wave mismatch evolution sequence based on the VSWR time-series data, and performs joint timing alignment on the reflected energy sequence and the standing wave mismatch evolution sequence to form a basic standing wave state sequence; Sequence decomposition module 13: Performs multi-scale timing decomposition on the basic standing wave state sequence to construct multi-dimensional standing wave evolution features. Vector; Feature fusion module 14: Calculates the environmental change gradient and environmental fluctuation intensity based on the equipment operating environment monitoring data, and fuses the calculation results with the multidimensional standing wave evolution feature vector to form a coupled multidimensional standing wave evolution feature vector under environmental coupling; State prediction module 15: Couples and encodes the transmission path structure description information of the communication equipment radio frequency transmission with the coupled multidimensional standing wave evolution feature vector to form a standing wave evolution feature sequence, uses the standing wave evolution feature sequence to perform standing wave state prediction at multiple time steps, and calculates the standing wave mismatch growth rate and energy accumulation index to construct a fault risk index.

[0065] Furthermore, the data acquisition module 11 is used to perform the following methods:

[0066] The equipment operating environment monitoring data includes temperature data, humidity data, and wind speed data.

[0067] Furthermore, the timing alignment module 12 is used to perform the following method:

[0068] The reflected power at each sampling time is calculated based on the forward power and transmitted power, and the reflected energy value at the corresponding time step is calculated based on the reflected power to form a reflected energy sequence. A sliding window analysis is performed on the voltage standing wave ratio (VSWR) time series data to calculate the amplitude and gradient of the VSWR change within adjacent time windows, constructing a VSWR mismatch evolution sequence characterizing the VSWR mismatch change process. The sequence time offset is calculated based on the correlation between the VSWR mismatch evolution sequence and the reflected energy sequence over a continuous time period, and dynamic time alignment processing is performed on the reflected energy sequence and the VSWR mismatch evolution sequence based on the sequence time offset. The dynamically time-aligned reflected energy sequence and the VSWR mismatch evolution sequence are jointly encoded to form a basic VSWR state sequence.

[0069] Furthermore, the sequence decomposition module 13 is used to perform the following method:

[0070] The standing wave state base sequence is decomposed into a multi-scale sliding window at different time scales to obtain short-period disturbance components, medium-period fluctuation components, and long-period trend components. The reflection energy change gradient of the short-period disturbance component within adjacent time windows is calculated, and the reflection energy change gradient is used to characterize the transient mismatch degree of the radio frequency connection interface. The periodic fluctuation amplitude and periodic stability of the medium-period fluctuation component are calculated. The slope and drift amplitude of the long-period trend component within continuous time segments are calculated to construct degradation trend parameters. A multidimensional standing wave evolution feature vector is generated based on the combination of the reflection energy change gradient, periodic fluctuation amplitude, periodic stability, and degradation trend parameters.

[0071] Furthermore, the state prediction module 15 is used to perform the following method:

[0072] The standing wave evolution feature sequence is used to construct a continuous time-series input segment according to a preset time window. This continuous time-series input segment is then input into a time-series depth prediction channel to obtain a standing wave prediction state sequence for multiple time steps. The increment of the standing wave mismatch ratio change between adjacent prediction time steps is calculated based on the standing wave prediction state sequence, and the increment of the standing wave mismatch ratio change is accumulated to form a standing wave mismatch evolution trajectory. The standing wave mismatch growth rate is calculated based on the standing wave mismatch evolution trajectory, and an energy accumulation index is calculated by combining the reflected energy change data corresponding to the standing wave prediction state sequence. The standing wave mismatch growth rate and the energy accumulation index are normalized and fused to construct a fault risk index.

[0073] Furthermore, the state prediction module 15 is used to perform the following method:

[0074] The standing wave evolution features in the continuous time-series input segment are divided into multiple feature subsequences according to disturbance variation features, periodic fluctuation features, and trend degradation features. These feature subsequences are then input into a mapping feature processing subchannel, which includes a disturbance feature processing subchannel, a periodic feature processing subchannel, and a trend feature processing subchannel. The disturbance time-series features characterizing the short-period reflection fluctuation mode of the standing wave are extracted from the disturbance feature processing subchannel; the periodic evolution features characterizing the periodic variation law of the standing wave are extracted from the periodic feature processing subchannel; and the trend feature processing subchannel extracts… The trend evolution characteristics characterize the long-term degradation trend of standing waves; cross-sub-channel feature correlation modeling is performed on the disturbance time series characteristics, periodic evolution characteristics, and trend evolution characteristics; a standing wave evolution characteristic correlation matrix is ​​constructed by calculating the correlation weights between features; and correlation enhancement processing of each feature is performed based on the standing wave evolution characteristic correlation matrix; unified time series fusion encoding is performed on each feature after correlation enhancement processing to generate a standing wave evolution fusion characteristic sequence; the standing wave evolution fusion characteristic sequence is input into the prediction sub-channel to perform multi-time step standing wave state prediction to generate a standing wave prediction state sequence for multiple time steps.

[0075] Furthermore, the state prediction module 15 is used to perform the following method:

[0076] The power amplifiers, feeders, jumpers, and antennas in the radio frequency (RF) transmission link of the communication equipment are abstracted into nodes and mapped to RF transmission nodes. A set of node connection relationships is established based on the physical connection relationships between each RF transmission node, and an RF transmission topology is constructed based on the signal transmission direction between the RF transmission nodes. The signal propagation path and path length parameters of the RF transmission nodes are calculated based on the RF transmission topology, and the hierarchical relationship is determined. A transmission path structure matrix is ​​constructed based on the signal propagation path and path length parameters and the hierarchical relationship, and the transmission path structure matrix is ​​used as the transmission path structure description information.

[0077] Furthermore, the state prediction module 15 is used to perform the following method:

[0078] Based on the fault risk index, the early warning level is matched, an early warning signal is configured, and the early warning signal is used to perform early warning management.

[0079] Furthermore, the state prediction module 15 is used to perform the following method:

[0080] Configure an alarm response window based on the alarm dispatch signal. If no alarm response data is detected in the alarm response window, then perform an alarm escalation of the alarm dispatch signal and perform alarm dispatch management.

[0081] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A dynamic early warning method for standing wave faults in communication equipment based on temporal deep learning, characterized in that, The method includes: Acquire timing data of forward power, reflected power, and voltage standing wave ratio in the radio frequency transmission link of communication equipment, and simultaneously collect monitoring data of the equipment's operating environment; The reflected energy sequence is calculated based on the forward power and the transmitted power, and the standing wave mismatch evolution sequence is constructed based on the voltage standing wave ratio time series data. The reflected energy sequence and the standing wave mismatch evolution sequence are jointly time-aligned to form the basic sequence of standing wave state. Perform multi-scale temporal decomposition on the basic sequence of standing wave states to construct a multi-dimensional standing wave evolution feature vector; The environmental change gradient and environmental fluctuation intensity are calculated based on the equipment operating environment monitoring data. The calculation results are then fused with the multidimensional standing wave evolution feature vector to form a coupled multidimensional standing wave evolution feature vector under environmental coupling. Based on the transmission path structure description information of the radio frequency transmission of communication equipment and the coupled multidimensional standing wave evolution feature vector, a standing wave evolution feature sequence is formed. The standing wave evolution feature sequence is used to perform standing wave state prediction at multiple time steps, and the standing wave mismatch growth rate and energy accumulation index are calculated to construct a fault risk index. Standing wave state prediction at multiple time steps is performed using standing wave evolution characteristic sequences, including: The standing wave evolution feature sequence is constructed into a continuous time series input segment according to a preset time window. The continuous time series input segment is input into the time series depth prediction channel to obtain the standing wave prediction state sequence at multiple time steps. The standing wave mismatch ratio change increment between adjacent prediction time steps is calculated based on the standing wave prediction state sequence, and the standing wave mismatch ratio change increment is accumulated to form the standing wave mismatch evolution trajectory. The standing wave mismatch growth rate is calculated based on the standing wave mismatch evolution trajectory, and the energy accumulation index is calculated by combining the reflected energy change data corresponding to the standing wave predicted state sequence. The standing wave mismatch growth rate and energy accumulation index are normalized and fused to construct a fault risk index; The continuous time-series input segments are input into the time-series depth prediction channel to obtain a standing wave prediction state sequence at multiple time steps, including: The standing wave evolution characteristics in the continuous time-series input segment are divided into multiple feature subsequences according to the perturbation change characteristics, periodic fluctuation characteristics, and trend degradation characteristics. The multiple feature subsequences are then input into the mapping feature processing subchannel, which includes a perturbation feature processing subchannel, a periodic feature processing subchannel, and a trend feature processing subchannel. In the disturbance feature processing sub-channel, disturbance time series features characterizing the short-period reflection wave pattern of the standing wave are extracted; in the periodic feature processing sub-channel, periodic evolution features characterizing the periodic change law of the standing wave are extracted; and in the trend feature processing sub-channel, trend evolution features characterizing the long-term degradation trend of the standing wave are extracted. Cross-sub-channel feature correlation modeling is performed on the disturbance time series features, periodic evolution features, and trend evolution features. A standing wave evolution feature correlation matrix is ​​constructed by calculating the correlation weights between features, and correlation enhancement processing of each feature is performed based on the standing wave evolution feature correlation matrix. After the correlation enhancement process, each feature is subjected to unified temporal fusion coding to generate a standing wave evolution fusion feature sequence. The standing wave evolution fusion feature sequence is input into the prediction sub-channel to perform multi-time-step standing wave state prediction, generating a standing wave prediction state sequence for multiple time steps.

2. The method for dynamic early warning of standing wave faults in communication equipment using time-series deep learning as described in claim 1, characterized in that, The basic sequence of standing wave states is formed, including: The reflected power at each sampling time is calculated based on the forward power and the transmitted power, and the reflected energy value at the corresponding time step is calculated based on the reflected power to form a reflected energy sequence. A sliding window analysis is performed on the voltage standing wave ratio time series data to calculate the change amplitude and gradient of voltage standing wave ratio within adjacent time windows, and to construct a standing wave mismatch evolution sequence characterizing the standing wave mismatch change process. The sequence time offset is calculated based on the correlation of the changes in the standing wave mismatch evolution sequence and the reflected energy sequence over a continuous time period, and dynamic time alignment processing is performed on the reflected energy sequence and the standing wave mismatch evolution sequence based on the sequence time offset. The reflected energy sequence after dynamic temporal alignment and the standing wave mismatch evolution sequence are jointly encoded to form the basic sequence of standing wave state.

3. The method for dynamic early warning of standing wave faults in communication equipment using time-series deep learning as described in claim 1, characterized in that, Perform multi-scale temporal decomposition on the basic sequence of standing wave states, including: The standing wave state basic sequence is decomposed into a multi-scale sliding window decomposition according to different time scales to obtain short-period disturbance components, medium-period fluctuation components and long-period trend components. Calculate the gradient of the reflected energy change of the short-period disturbance component within adjacent time windows. The gradient of the reflected energy change is used to characterize the degree of transient mismatch at the radio frequency connection interface. Calculate the periodic fluctuation amplitude and periodic stability of the periodic fluctuation component; Calculate the slope and drift amplitude of the long-period trend component within a continuous time interval to construct the degradation trend parameter; A multidimensional standing wave evolution feature vector is generated based on the combination of parameters such as the reflection energy change gradient, periodic fluctuation amplitude, periodic stability, and degradation trend.

4. The method for dynamic early warning of standing wave faults in communication equipment using time-series deep learning as described in claim 1, characterized in that, The construction of the transmission path structure description information includes: The power amplifiers, feeders, jumpers, and antennas in the radio frequency transmission link of communication equipment are abstracted into nodes and mapped to radio frequency transmission nodes. Establish a set of node connection relationships based on the physical connection relationships between each radio frequency transmission node, and construct a radio frequency transmission topology based on the signal transmission direction between radio frequency transmission nodes; Calculate the signal propagation path and path length parameters of the radio frequency transmission nodes based on the radio frequency transmission topology, and determine the hierarchical relationship; A transmission path structure matrix is ​​constructed based on the signal propagation path, path length parameters, and hierarchical relationships, and the transmission path structure matrix is ​​used as the transmission path structure description information.

5. The method for dynamic early warning of standing wave faults in communication equipment using time-series deep learning as described in claim 1, characterized in that, The equipment operating environment monitoring data includes temperature data, humidity data, and wind speed data.

6. The method for dynamic early warning of standing wave faults in communication equipment using time-series deep learning as described in claim 1, characterized in that, Based on the fault risk index, the early warning level is matched, an early warning signal is configured, and the early warning signal is used to perform early warning management.

7. The method for dynamic early warning of standing wave faults in communication equipment using time-series deep learning as described in claim 6, characterized in that, Configure an alarm response window based on the alarm dispatch signal. If no alarm response data is detected in the alarm response window, then perform an alarm escalation of the alarm dispatch signal and perform alarm dispatch management.

8. A dynamic early warning system for standing wave faults in communication equipment based on time-series deep learning, characterized in that, The system is used to implement the dynamic early warning method for standing wave faults in communication equipment based on time-series deep learning as described in any one of claims 1-7, the system comprising: Data acquisition module: Acquires forward power, reflected power and voltage standing wave ratio timing data in the radio frequency transmission link of communication equipment, and simultaneously collects equipment operating environment monitoring data; Timing alignment module: Calculates the reflected energy sequence based on the forward power and the transmitted power, and constructs the standing wave mismatch evolution sequence based on the voltage standing wave ratio timing data. Performs joint timing alignment on the reflected energy sequence and the standing wave mismatch evolution sequence to form the basic sequence of standing wave state. Sequence decomposition module: Performs multi-scale temporal decomposition on the basic sequence of the standing wave state to construct a multi-dimensional standing wave evolution feature vector; Feature fusion module: Calculates the environmental change gradient and environmental fluctuation intensity based on the equipment operating environment monitoring data, and fuses the calculation results with the multidimensional standing wave evolution feature vector to form a coupled multidimensional standing wave evolution feature vector under environmental coupling; State prediction module: Based on the transmission path structure description information of the communication equipment radio frequency transmission and coupled multidimensional standing wave evolution feature vector, a standing wave evolution feature sequence is formed. The standing wave evolution feature sequence is used to perform standing wave state prediction at multiple time steps, and the standing wave mismatch growth rate and energy accumulation index are calculated to construct a fault risk index.

Images