Rail transit leaky cable multi-parameter real-time monitoring system and signal processing method

Abstract

The application discloses a kind of track traffic leak cable multi-parameter real-time monitoring system and signal processing method.System is constituted by multi-parameter composite monitoring terminal, edge processing node and cloud management platform.Monitoring terminal integrates microwave network parameter, vibration and temperature sensing module, synchronously acquires standing wave ratio, insertion loss, vibration and temperature signal, and is reported in time division multiplexing mode by differential signal bus.Edge processing node identifies interference conditions such as arc of pantograph-catenary according to vibration signal power spectrum density characteristics, and selects double-threshold asymmetric threshold function to wavelet decomposition coefficient adaptive noise reduction accordingly.Cloud management platform inputs long short-term memory network prediction model with noise reduction characteristics, outputs equivalent health index prediction sequence and graded early warning.The application realizes strong electromagnetic interference suppression, leak cable health multidimensional perception and fault early prediction.

Description

Technical Field

[0001] This invention belongs to the field of intelligent operation and maintenance technology of rail transit communication systems, and relates to a real-time monitoring system for multiple parameters of rail transit leaky cables and a signal processing method. Background Technology

[0002] In enclosed spaces such as railway tunnels, leaky cables, as coaxial cables that combine signal transmission and antenna transceiver functions, are widely used to ensure the coverage of train wireless communication signals. The health condition of leaky cables directly affects the reliability of train communication.

[0003] Existing online monitoring solutions for leaky cables mainly rely on measurement systems based on a single standing wave ratio (SWR) or signal attenuation. These systems extract the reflected power signal from the leaky cable transmission line using a directional coupler, convert it into an SWR or return loss indicator, and compare it with a preset fixed threshold. When the measured value exceeds the threshold, the system issues a fault alarm. Some systems can also estimate the location of the fault based on the time delay information of the reflected signal.

[0004] However, such existing technical solutions have the following drawbacks. First, the system only focuses on the deterioration of electrical performance indicators such as VSWR or signal attenuation, while mechanical damage such as sheath cracking and loose clamping parts of leaky cables often does not cause significant changes in electrical performance in the early stages of a fault. This causes the system to completely fail to detect the precursor stages of these mechanical damages, resulting in missed fault detection.

[0005] Second, there are strong transient electromagnetic pulse interferences such as pantograph-catenary arcs in rail transit tunnels. The common-mode suppression capability of the existing system's acquisition circuit is insufficient, and the back-end lacks targeted filtering processing algorithms. As a result, the random fluctuation noise in the VSWR measurement can reach a high level, completely drowning out the weak VSWR changes caused by early faults such as small cable cracks, resulting in a high false alarm rate and missed alarm rate.

[0006] Third, existing technologies generally adopt alarm mechanisms based on fixed thresholds, which can only issue a notification after the fault has developed to a certain extent and the electrical performance indicators exceed the preset threshold. This makes it impossible to predict and warn of slow performance degradation trends, resulting in a passive state of operation and maintenance and repair, which is difficult to meet the intelligent operation and maintenance needs of high-reliability vehicles. Summary of the Invention

[0007] To address the problems existing in the background technology, this invention proposes a real-time multi-parameter monitoring system and signal processing method for leaky cables in rail transit.

[0008] To achieve the above objectives, the technical solution adopted by the present invention is as follows: a real-time monitoring system for multiple parameters of leaky cables in rail transit, comprising a multi-parameter composite monitoring terminal, an edge processing node, and a cloud management platform;

[0009] The multi-parameter composite monitoring terminal synchronously collects the standing wave ratio monitoring signal, insertion loss monitoring signal, vibration monitoring signal and temperature monitoring signal of the leaky cable deployed in the rail transit scenario, and reports all kinds of monitoring signals to the edge processing node through the same differential signal bus in a time-division multiplexing manner;

[0010] The edge processing node acquires various monitoring signals reported by the multi-parameter composite monitoring terminal, identifies the current electromagnetic interference condition type based on the vibration monitoring signal, selects the corresponding threshold function based on the identified current electromagnetic interference condition type, and performs threshold processing on the wavelet decomposition coefficients of the standing wave ratio monitoring signal using the selected threshold function to obtain the noise-reduced standing wave ratio signal. The node then packages and uploads the feature data of the noise-reduced standing wave ratio signal, the insertion loss monitoring signal, and the vibration monitoring signal to the cloud management platform.

[0011] The cloud management platform acquires the feature data uploaded by the edge processing node, inputs the feature data into the trained long short-term memory network prediction model, obtains a prediction sequence of the equivalent health index of the leaky cable within a preset time period, and performs a graded early warning operation based on the prediction sequence.

[0012] Furthermore, the multi-parameter composite monitoring terminal includes a microwave network parameter module, a vibration sensing module, a temperature sensing module, and a differential acquisition circuit module, which are integrated into the same sealed cavity.

[0013] The microwave network parameter module includes a directional coupler and a logarithmic detector. The reverse coupling port of the directional coupler is electrically connected to the radio frequency input terminal of the logarithmic detector. The logarithmic detector converts the acquired reflected power signal into a DC voltage signal and outputs the DC voltage signal to the differential acquisition circuit module.

[0014] The vibration sensing module includes an inertial measurement unit, which collects vibration monitoring signals at the installation location of the leaky cable.

[0015] The temperature sensing module includes a temperature sensor, which collects the temperature monitoring signal at the installation location of the leaky cable.

[0016] The differential acquisition circuit module is an active instrumentation amplifier circuit. After conditioning the DC voltage signal, the differential acquisition circuit module shares the same differential signal bus with the vibration monitoring signal output by the vibration sensing module and the temperature monitoring signal output by the temperature sensing module, and reports to the edge processing node in a time-division manner according to the preset time slot length.

[0017] Furthermore, the edge processing node includes a condition classifier and an adaptive noise reduction module;

[0018] The operating condition classifier acquires the vibration monitoring signal, extracts the power spectral density features of the vibration monitoring signal within a specific time window, identifies the current electromagnetic interference operating condition type based on the power spectral density features, and outputs the identified current electromagnetic interference operating condition type to the adaptive noise reduction module.

[0019] The current electromagnetic interference conditions include intermittent strong pulse interference conditions and background noise conditions.

[0020] The intermittent strong pulse interference condition is defined as the condition in which the power spectral density exhibits discrete pulse peaks with amplitudes exceeding a first energy threshold within a preset frequency band, and the pulse interval is greater than a preset interval threshold.

[0021] The background noise condition is defined as the condition in which the amplitude of the power spectral density in the preset frequency band is continuously lower than the second energy threshold and the amplitude fluctuation is less than the preset fluctuation threshold.

[0022] The adaptive noise reduction module receives the current electromagnetic interference condition type. When the current electromagnetic interference condition type is the intermittent strong pulse interference condition, a double threshold asymmetric threshold function is selected to process the wavelet decomposition coefficients.

[0023] When the current electromagnetic interference condition is the background noise condition, a soft threshold function or a hard threshold function is selected to process the wavelet decomposition coefficients.

[0024] Furthermore, when the adaptive noise reduction module processes the current wavelet decomposition coefficients using the dual-threshold asymmetric threshold function, it sets an upper-level threshold and a lower-level threshold, and performs the following operations:

[0025] When the absolute value of the current wavelet decomposition coefficient is less than or equal to the lower-level threshold, the value of the current wavelet decomposition coefficient is set to zero;

[0026] When the absolute value of the current wavelet decomposition coefficient is greater than the upper threshold, the absolute value of the current wavelet decomposition coefficient is subtracted from the upper threshold to obtain the intermediate amplitude. Based on the sign of the current wavelet decomposition coefficient and the intermediate amplitude, the corrected wavelet decomposition coefficient is determined.

[0027] When the absolute value of the current wavelet decomposition coefficient is greater than the lower threshold and less than or equal to the upper threshold, the square of the lower threshold is calculated and divided by the absolute value of the current wavelet decomposition coefficient to obtain the reduction amount. The reduction amount is then subtracted from the absolute value of the current wavelet decomposition coefficient to obtain the intermediate amplitude. Based on the sign of the current wavelet decomposition coefficient and the intermediate amplitude, the corrected wavelet decomposition coefficient is determined.

[0028] Under the intermittent strong pulse interference condition, the upper threshold is dynamically set according to the product of the root mean square power of the current signal window and the first adjustment coefficient, and the lower threshold is dynamically set according to the product of the root mean square power of the current signal window and the second adjustment coefficient.

[0029] Furthermore, the cloud management platform includes an equivalent health index prediction module and a dynamic weight generation module;

[0030] The equivalent health index prediction module is equipped with the trained long short-term memory network prediction model. The trained long short-term memory network prediction model is an encoder-decoder structure. The encoder acquires the time series of the feature data within the past first preset time period and outputs an encoded state vector. The decoder decodes and outputs the prediction sequence of the equivalent health index within the future second preset time period based on the encoded state vector.

[0031] The dynamic weight generation module uses a logistic regression model trained based on historical fault case data to dynamically generate the fusion weight value corresponding to each monitoring parameter in the equivalent health index based on the current degree of degradation of each monitoring parameter.

[0032] Furthermore, the edge processing node also includes a data compression and transmission module;

[0033] The data compression and transmission module acquires the noise-reduced VSWR signal, the insertion loss monitoring signal, the vibration monitoring signal, and the temperature monitoring signal. When the change in the value of the noise-reduced VSWR signal exceeds a preset VSWR change threshold, the change in the value of the insertion loss monitoring signal exceeds a preset insertion loss change threshold, the vibration feature value extracted from the vibration monitoring signal exceeds a preset vibration threshold, the change in the value of the temperature monitoring signal exceeds a preset temperature change threshold, or a preset reporting time interval is reached, the feature data is extracted.

[0034] The extracted feature data is serialized into binary format using a protocol buffer and asynchronously reported to the cloud management platform via a message queue telemetry transmission protocol at a preset quality of service level.

[0035] Furthermore, the edge processing node is equipped with a timing module, which receives timing signals from the Global Navigation Satellite System and outputs pulse signals and time information;

[0036] When the various monitoring signals reported by the multi-parameter composite monitoring terminal enter the edge processing node, they are stamped with an absolute timestamp based on the pulse signal and the time information, thereby achieving time synchronization of data between different monitoring terminals.

[0037] Furthermore, the microwave network parameter module in the multi-parameter composite monitoring terminal performs a system-level calibration process before being put into online monitoring;

[0038] The system-level calibration process includes: connecting the open-circuit calibrator, short-circuit calibrator and matched load calibrator sequentially to the input port of the directional coupler in the microwave network parameter module; and the differential acquisition circuit module in the multi-parameter composite monitoring terminal records the corresponding digital code values ​​under the three calibration states as calibration reference points.

[0039] The edge processing node acquires the calibration reference point and establishes a calibration mapping relationship between digital code value and reflected power, including determining the corresponding curve or piecewise linear interpolation table between reflected power and digital code value.

[0040] In actual monitoring, the measured digital code value of the device under test is converted into a reflected power value according to the calibration mapping relationship, and then the value of the standing wave ratio monitoring signal is calculated based on the reflected power value.

[0041] Furthermore, the differential acquisition circuit module is a third-order active instrumentation amplifier circuit, comprising a buffer stage, a programmable gain amplification stage, and a low-pass filter stage cascaded in sequence.

[0042] The buffer stage is composed of a dual-channel precision operational amplifier, one channel of which is connected to the DC voltage signal output by the logarithmic detector, and the other channel is grounded.

[0043] The programmable gain amplifier stage automatically switches between preset gain levels according to the amplitude of the input signal, so that the amplitude of the output signal is kept within a preset ratio range of the full scale of the analog-to-digital converter.

[0044] The low-pass filter stage is a high-order elliptic low-pass filter, and the cutoff frequency of the high-order elliptic low-pass filter is automatically set according to the sampling rate of the system to filter out high-frequency noise components in the DC voltage signal.

[0045] This technical solution also provides a method for real-time multi-parameter monitoring signal processing of leaky cables in rail transit, applied to a monitoring system consisting of a multi-parameter composite monitoring terminal, edge processing nodes, and a cloud management platform, including:

[0046] The multi-parameter composite monitoring terminal synchronously collects the standing wave ratio monitoring signal, insertion loss monitoring signal, vibration monitoring signal and temperature monitoring signal of the leaky cable deployed in the rail transit scenario, and reports all kinds of monitoring signals to the edge processing node through the same differential signal bus in a time-division multiplexing manner;

[0047] The edge processing node acquires various monitoring signals reported by the multi-parameter composite monitoring terminal and extracts the power spectral density characteristics of the vibration monitoring signal within a specific time window;

[0048] The edge processing node identifies the current electromagnetic interference condition type based on the power spectral density characteristics. The current electromagnetic interference condition type includes intermittent strong pulse interference condition and background noise condition.

[0049] The intermittent strong pulse interference condition is a condition in which the power spectral density exhibits discrete pulse peaks with amplitudes exceeding a first energy threshold within a preset frequency band, and the pulse interval is greater than a preset interval threshold.

[0050] The background noise condition is the condition in which the amplitude of the power spectral density within the preset frequency band is continuously lower than the second energy threshold, and the amplitude fluctuation is less than the preset fluctuation threshold.

[0051] When the intermittent strong pulse interference condition is identified, the edge processing node selects a double-threshold asymmetric threshold function to perform threshold processing on the wavelet decomposition coefficients of the VSWR monitoring signal.

[0052] When the background noise condition is identified, a soft thresholding function or a hard thresholding function is used to perform thresholding on the wavelet decomposition coefficients to obtain the VSWR signal after noise reduction.

[0053] The edge processing node packages and uploads the characteristic data of the noise-reduced VSWR signal, the insertion loss monitoring signal, and the vibration monitoring signal to the cloud management platform;

[0054] The cloud management platform acquires the feature data uploaded by the edge processing node, inputs the feature data into the trained long short-term memory network prediction model, obtains a prediction sequence of the equivalent health index of the leaky cable within a preset time period, and performs a graded early warning operation based on the prediction sequence.

[0055] Compared with the prior art, the present invention has the following beneficial effects: by constructing a multi-parameter composite monitoring terminal integrating microwave network parameter module, vibration sensing module and temperature sensing module, multi-dimensional collaborative monitoring of the electrical performance, mechanical behavior and environmental parameters of leaky cables is realized.

[0056] By using the operating condition classifier in the edge processing node, the intermittent strong pulse interference conditions are identified by utilizing the power spectral density characteristics of the vibration monitoring signal. Furthermore, an adaptive noise reduction process is performed on the wavelet decomposition coefficients of the VSWR monitoring signal using a dual-threshold asymmetric threshold function. This achieves effective preservation of minor damage signals of leaky cables under strong transient electromagnetic pulse interference such as pantograph-catenary arc.

[0057] By using a long short-term memory network prediction model with an encoder-decoder structure in the cloud management platform, combined with dynamically generated fusion weight values ​​from a logistic regression model, a prediction sequence of the equivalent health index of leaky cables is output, enabling analysis of the degradation trend of leaky cables and graded early warning of potential faults. Attached Figure Description

[0058] Figure 1 This is an architecture diagram of a real-time multi-parameter monitoring system for leaky cables in rail transit according to the present invention;

[0059] Figure 2 This is a schematic diagram of the internal hardware module connection and data flow of the multi-parameter composite monitoring terminal of the present invention;

[0060] Figure 3 This is a flowchart of the adaptive improved wavelet threshold noise reduction algorithm for the working conditions of this invention.

[0061] Figure 4 This is a flowchart of the operation of the full-port calibration method of the present invention. Detailed Implementation

[0062] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0063] like Figures 1-4 As shown, the technical solution adopted by the present invention is as follows: a real-time monitoring system for multiple parameters of leaky cables in rail transit, comprising a multi-parameter composite monitoring terminal, an edge processing node, and a cloud management platform.

[0064] The multi-parameter composite monitoring terminal is a sensing layer device deployed in on-site environments such as rail transit tunnels. Its function is to simultaneously acquire raw monitoring signals of four different physical quantities characterizing the health status of leaky cables. Specifically, these include: VSWR and insertion loss monitoring signals reflecting radio frequency transmission performance; vibration monitoring signals reflecting abnormal mechanical behavior; and temperature monitoring signals reflecting changes in the physical environment. For data communication, the terminal internally employs a shielded low-voltage differential signal bus as the unified physical transmission medium and uses a time-division multiplexing mechanism as the channel multiplexing strategy for the data link layer.

[0065] Time-division multiplexing (TDM) is a technique that divides communication time into a series of periodic, fixed-length time slots, allocating a dedicated, non-overlapping time slot for each signal for transmission. Data collected by each sensor module is allocated according to the predetermined time slots, and the differential signal bus is used in a time-division manner for reporting. This completely avoids bus contention, data collisions, and crosstalk that can occur when multiple sensors transmit in parallel at the physical and data link layers. Ultimately, it transmits four types of heterogeneous signals with a high signal-to-noise ratio to the edge processing nodes for subsequent processing.

[0066] To achieve multi-dimensional collaborative sensing of electrical performance, mechanical and environmental parameters, the multi-parameter composite monitoring terminal integrates four major modules in its structure: microwave network parameter module, vibration sensing module, temperature sensing module and differential acquisition circuit module.

[0067] The system comprises several modules: a microwave network parameter module for generating and acquiring signals reflecting the cable's radio frequency characteristics; a vibration sensing module for detecting cable vibration and impact; a temperature sensing module for sensing ambient temperature; and a differential acquisition circuit module for high-precision conditioning of weak analog signals. All modules are physically integrated within a single sealed cavity. This integrated design shortens the analog signal path, reduces cable routing and connector count, effectively lowers the coupling path of external electromagnetic interference, and provides a robust electromagnetic shielding environment for subsequent high-precision differential signal acquisition. This structure forms the structural basis for solving the problem of high signal-to-noise ratio collaborative operation of multi-physics-based sensors under strong electromagnetic interference.

[0068] The microwave network parameter module consists of dual directional couplers (e.g., Mini-Circuits ZFDC-20-5+) and two logarithmic detectors. The dual directional couplers have a forward coupling port and a reverse coupling port, used to sample the incident power and reflected power on the transmission line, respectively. The reverse coupling port is used to extract the weak reflected power signal generated by impedance mismatch from the main transmission line of the leaky cable; the strength of this signal is directly related to the VSWR of the leaky cable. The reverse coupling port is electrically connected to the RF input of the first logarithmic detector, which converts the reflected power signal into a first DC voltage signal. The forward coupling port is used to couple the incident power signal generated by the monitoring terminal signal source or train communication signal from the main transmission line of the leaky cable. The forward coupling port is electrically connected to the RF input of the second logarithmic detector, which converts the incident power signal into a second DC voltage signal. Both DC voltage signals are output to the next-stage differential acquisition circuit module for further processing. The differential acquisition circuit module converts the received DC voltage signal into a differential signal, then conditions and transmits it. The differential signal transmission effectively suppresses common-mode noise, which is the first-level signal conditioning for this system to achieve high-precision measurement in a high-interference environment.

[0069] The vibration sensing module consists of an inertial measurement unit (ITU) (TDK InvenSense ICM-20689). This ITU is a six-axis sensor based on microelectromechanical systems (MEMS) technology, integrating a three-axis accelerometer and a three-axis gyroscope. Its function is to be fixed to the installation location of the leaky cable (such as the cable clamp or the cable itself), sensing and acquiring in real time the changes in linear acceleration and angular velocity at that location caused by factors such as train operation, pantograph-catenary contact, loosening of the cable clamp, or sheath stress release. These physical quantity changes collectively constitute the vibration monitoring signal. Acquiring this signal enables the system to detect early signs of mechanical damage not directly coupled to electrical performance (such as loose clamps or sheath cracking).

[0070] The temperature sensing module includes an on-chip thermometer, serving as a temperature sensor. This sensor is integrated within the monitoring terminal and has a thermal conduction path to the external environment. Its function is to collect the ambient temperature at the location where the leaky cable is installed in real time and generate a temperature monitoring signal. Temperature is a crucial environmental factor affecting the aging rate of cable materials and the drift of radio frequency characteristics. The introduction of this signal provides an environmental dimension input to the cloud-based equivalent health index model, helping the model eliminate the interference of temperature fluctuations on electrical parameter measurements, thereby more accurately decoupling and identifying the true performance degradation caused by physical damage.

[0071] Remote Monitoring Node: To achieve accurate measurement of insertion loss, this system also includes at least one remote monitoring node deployed at the far end of the leaky cable (the end furthest from the multi-parameter composite monitoring terminal). The remote monitoring node includes a remote directional coupler and a remote logarithmic detector; the remote directional coupler is connected in series along the transmission path of the leaky cable to sample the transmitted power signal reaching the far end after passing through the entire leaky cable; the remote logarithmic detector converts the transmitted power signal into a DC voltage signal, and after analog-to-digital conversion, obtains the digital code value of the transmitted power. The remote monitoring node also includes a wireless communication module (e.g., LoRa or 4G module) or a wired communication interface (e.g., RS-485) to transmit the digital code value of the transmitted power back to the edge processing node. The edge processing node calculates the value of the insertion loss monitoring signal based on the incident power digital code value collected at the near end and the transmitted power digital code value transmitted back from the far end. ,in For the linear value of the incident power, This represents the linear value of the transmitted power.

[0072] The differential acquisition circuit module is designed as an active instrumentation amplifier, a precision amplifier circuit specifically designed to amplify the voltage difference between two input terminals and suppress common-mode signals, featuring high input impedance and high common-mode rejection ratio. It processes the reflected power DC voltage signal from the first logarithmic detector and the incident power DC voltage signal from the second logarithmic detector. After conditioning, the reflected power digital signal, the incident power digital signal, the vibration monitoring signal output from the vibration sensing module, and the temperature monitoring signal output from the temperature sensing module are all fed into the same differential signal bus within the terminal. Then, under the scheduling of the microcontroller, the bus is used in a time-division multiplexing manner according to the preset time slot length, and the signals are sequentially reported to the edge processing node.

[0073] The differential acquisition circuit is a third-order active instrumentation amplifier. This circuit consists of three cascaded functional stages in the signal path: a buffer stage, a programmable gain amplifier stage, and a low-pass filter stage. This three-stage cascaded architecture corresponds to the three basic tasks of signal conditioning: impedance transformation, amplitude calibration, and anti-aliasing / noise filtering. Each stage contains active components to provide gain or impedance matching. This design is the hardware core that ensures the weak μV-level DC voltage signal output from the logarithmic detector is reliably and accurately conditioned before entering the analog-to-digital converter.

[0074] The buffer stage consists of a dual-channel precision operational amplifier chip. This chip integrates two independent, parameter-matched operational amplifier channels, each configured as a voltage follower (buffer). The non-inverting input of one channel is connected to the positive terminal of the DC voltage signal from the logarithmic detector, while the non-inverting input of the other channel is grounded (or connected to a reference voltage). The voltage follower has extremely high input impedance and extremely low output impedance. Its function is to provide a near-ideal light load for the preceding logarithmic detector, preventing voltage division and distortion of the weak signal source due to the limited input impedance of the subsequent circuitry. It also provides the low-impedance output capability required to drive the subsequent programmable gain amplifier stage, a crucial preliminary step for achieving high-precision weak signal acquisition.

[0075] The gain of the programmable gain amplifier stage is not fixed, but can dynamically and automatically switch between multiple preset gain levels. Its control logic is as follows: the system judges the magnitude of the input signal amplitude in real time. If the DC voltage signal amplitude output by the logarithmic detector is large due to strong cable reflection, the system will automatically switch to a lower gain level to prevent saturation of the subsequent analog-to-digital converter (ADC). If the signal is extremely weak, the system will automatically switch to a higher gain level to amplify signal details. The goal of this automatic switching is to ensure that the final output signal amplitude of the programmable gain amplifier stage is always stably maintained within a preset percentage range of the ADC's full-scale input range (e.g., 80% to 90% of full scale). This function greatly expands the system's dynamic range, ensuring that even minute changes in reflected signals are not lost due to quantization noise, regardless of the fault state of the leaky cable, thus guaranteeing measurement accuracy.

[0076] The low-pass filter stage is chosen to be a high-order elliptic low-pass filter. Elliptic filters have the narrowest transition band for their order and the steepest roll-off from the passband to the stopband, effectively separating the useful signal from near-frequency interference. The higher order further enhances its attenuation depth in the stopband. Its cutoff frequency is not fixed but automatically set according to the system sampling rate to conform to the Nyquist sampling theorem, serving as an anti-aliasing filter. Its functional objective is to filter out specific interference, namely the high-frequency carrier interference components used in the track circuits. In the rail transit environment, to ensure train positioning and control, track circuits transmit carrier signals of specific frequencies on the rails (such as audio or frequency shift keying signals from uninsulated track circuits). These signals may enter the monitoring system through spatial coupling or conducted interference, becoming strong sources of interference at fixed frequencies. This filter, by placing this interference frequency band in its stopband, highly suppresses the high-frequency carrier interference components in the track circuits while ensuring a flat response within the required low-frequency reflected signal passband, ultimately outputting a clean signal for sampling by the analog-to-digital converter.

[0077] Before being put into online monitoring, the microwave network parameter module in the multi-parameter composite monitoring terminal performs a system-level calibration process. This calibration is performed before the module is officially put into online health monitoring of leaky cables in rail transit. This design allows the system to measure and calculate the inherent fixed system errors of the measurement link (including directional couplers, logarithmic detectors, differential acquisition circuit modules, and connecting cables) using standard components in its initial state before the leaky cable under test is connected. This lays the foundation for subtracting these errors from any subsequent field measurements to obtain the true reflection characteristics of the device under test.

[0078] For the insertion loss measurement link, system-level calibration is also performed. In calibration mode, a matched load is connected to the remote port of the leaky cable (i.e., the output of the remote monitoring node). The edge processing node records the correspondence between the incident power digital code value and the transmitted power digital code value at this time, eliminating the fixed gain error and offset error of the directional coupler and logarithmic detector in the remote monitoring node, and ensuring the accuracy of the insertion loss calculation.

[0079] The system-level calibration process uses three standard components with known and stable reflection coefficients: an open-circuit calibrator, a short-circuit calibrator, and a matched-load calibrator. During operation, the leaky cable is disconnected from the input port of the directional coupler, and the three calibrators are connected sequentially and individually to the input port of the directional coupler. When the open-circuit calibrator is connected, the signal is totally reflected, with a reflection coefficient of +1; when the short-circuit calibrator is connected, the signal is also totally reflected, with a reflection coefficient of -1; when the matched-load calibrator is connected, the incident power is completely absorbed, and the reflection coefficient is 0. Under each stable calibration state formed by the connection of a calibrator, the differential acquisition circuit module located in the terminal performs analog-to-digital conversion and data acquisition on the received DC voltage signal after logarithmic detector conversion, recording the corresponding digital code values. This yields the digital code values ​​under the open-circuit calibration state, the short-circuit calibration state, and the matched-load calibration state, respectively. These digital code values ​​under these three states serve as three calibration reference points, providing a known and ideal reflection standard and a precise mapping relationship between the actual system response for subsequent calibration mapping.

[0080] The calibration reference point data is reported via the differential signal bus between the terminal and the edge processing node, and then acquired by the edge processing node. The edge processing node acquires the calibration reference points and establishes a calibration mapping relationship between the digital code value and the reflected power. Specifically, since the output digital code value of the logarithmic detector is linearly related to the input RF power (dBm), three known reflected power values ​​can be obtained through three standard states: open circuit, short circuit, and matched load (the reflected power is approximately 0 under matched load, and is maximum and equal under open circuit and short circuit). Using these three calibration reference points, the linear mapping relationship between the digital code value and the reflected power can be determined (e.g., using a two-point calibration curve or a piecewise linear interpolation table), thereby eliminating the fixed gain error and offset error of the logarithmic detector and the differential acquisition circuit.

[0081] After the system completes calibration and connects to the device under test (DUT), it enters the actual monitoring state. At this time, the raw response of the DUT acquired by the multi-parameter composite monitoring terminal is represented as a measured digital code value. The edge processing node converts this measured digital code value into a reflected power value according to the calibration mapping relationship. Since the reflected power is proportional to the square of the reflection coefficient (reflection coefficient = sqrt(reflected power / incident power)), and the standing wave ratio (VSWR) = (1 + reflection coefficient) / (1 - reflection coefficient), the VSWR monitoring signal value can be further calculated. This calibration method is based on scalar power measurement, does not require phase information, and conforms to the physical characteristics of a logarithmic detector.

[0082] The edge processing node receives various monitoring signals reported by the multi-parameter composite monitoring terminal via a differential signal bus in a time-division multiplexing manner, including VSWR monitoring signals, insertion loss monitoring signals, vibration monitoring signals, and temperature monitoring signals.

[0083] The function of the edge processing node is to perform adaptive noise reduction on the VSWR monitoring signal. Its processing chain is as follows: First, the node extracts the vibration monitoring signal from the received signal and automatically identifies the current electromagnetic interference condition type based on the signal's time-frequency characteristics. After identifying the current electromagnetic interference condition type, the node dynamically selects the threshold function corresponding to that condition type according to a preset mapping relationship.

[0084] Subsequently, the node performs wavelet transform on the VSWR monitoring signal, decomposing it to multiple scales to obtain a set of wavelet decomposition coefficients. A selected threshold function is then used to threshold each of these coefficients, selectively suppressing noise components and preserving or shrinking signal components. The thresholded coefficients are then reconstructed using inverse wavelet transform to obtain the denoised VSWR signal.

[0085] Finally, the node serializes the noise-reduced VSWR signal, the insertion loss monitoring signal, and the feature data extracted from the vibration monitoring signal into binary according to the protocol buffer format, encapsulates them into data packets, and asynchronously uploads them to the cloud management platform through the message queue telemetry transmission protocol.

[0086] The edge processing node includes a condition classifier and an adaptive noise reduction module. The condition classifier is specifically responsible for real-time automatic identification and classification of environmental conditions using vibration monitoring signals; the adaptive noise reduction module is specifically responsible for dynamically configuring and executing the corresponding wavelet threshold noise reduction algorithm based on the classification results. The two modules are logically connected in series, with the output of the condition classifier serving as the input to the adaptive noise reduction module.

[0087] The software of the operating condition classifier implements a pre-trained lightweight fully connected neural network model. Its workflow is as follows: First, it extracts data from a specific time window of 0.5 seconds from the vibration monitoring signal; second, it estimates the power spectral density of the vibration signal within this window and extracts its frequency domain energy distribution as the power spectral density feature; third, it inputs the power spectral density feature vector into the pre-trained lightweight fully connected neural network, which calculates and outputs the current electromagnetic interference (EMI) operating condition type at the current moment. The identified EMI operating condition type is then output to the subsequent adaptive noise reduction module. Specifically, the pantograph-catenary arc event simultaneously generates strong electromagnetic pulses and mechanical vibrations, which are synchronous in the time domain. By collecting historical vibration data containing pantograph-catenary arc events and corresponding EMI labels for offline training, a mapping classification model between vibration power spectral density features and EMI operating condition types can be established.

[0088] The current electromagnetic interference (EMI) conditions consist of two mutually exclusive categories. The first is the intermittent strong pulse interference condition, whose physical source represents the strong transient EMI interference environment in rail transit scenarios, typically represented by pantograph-catenary arcs. The second is the background noise condition, which represents the normal electromagnetic environment without strong interference such as pantograph-catenary arcs.

[0089] Intermittent strong pulse interference conditions have two characteristics that must be met simultaneously in the time and frequency domain: first, within a preset frequency band, the power spectral density exhibits discrete pulse peaks, the energy amplitude of which exceeds a preset first energy threshold of the system; second, the time interval between these discrete pulse peaks is greater than a preset interval threshold, thus manifesting as discontinuous, intermittent energy bursts in the time domain. This dual criterion characterizes the inherent pattern of electromagnetic pulse interference, such as pantograph-catenary arcing, that distinguishes it from continuous wave interference from both energy intensity and time-domain distribution dimensions.

[0090] The background noise condition in the time-frequency domain is characterized by the following: within a preset frequency band, the amplitude of the power spectral density is continuously lower than a preset second energy threshold of the system, and the amplitude fluctuation over time is less than a preset fluctuation threshold. This definition describes an electromagnetic environment with a low energy level and statistical stability, indicating that there are no intermittent strong pulse interference sources at this time.

[0091] The adaptive noise reduction module receives the classification results from the operating condition classifier. When the current electromagnetic interference condition is determined to be intermittent strong pulse interference, the module activates a dual-threshold asymmetric threshold function specifically designed for this condition. This function operates on the wavelet decomposition coefficients obtained after wavelet transform of the VSWR monitoring signal.

[0092] The processing characteristics of this dual-threshold asymmetric threshold function are determined by the upper threshold and the lower threshold. Dual threshold refers to independently set upper and lower thresholds, and asymmetric means that the upper and lower thresholds have different values.

[0093] When the current electromagnetic interference condition is determined to be background noise, it indicates that there is no strong pulse interference and the electromagnetic environment is relatively stable. In this case, the module uses a classic threshold function to process the wavelet decomposition coefficients. A soft threshold function or a hard threshold function can be selected according to a preset strategy. The soft threshold function sets coefficients with absolute values ​​below the threshold to zero and shrinks coefficients with absolute values ​​above the threshold by the threshold value to obtain a smooth reconstructed signal. The hard threshold function sets coefficients with absolute values ​​below the threshold to zero and retains the original values ​​of coefficients with absolute values ​​above the threshold to preserve the signal's abrupt changes to the greatest extent possible. Under stable noise conditions, both classic functions provide good noise reduction effects.

[0094] The dual-threshold asymmetric threshold function is the core noise reduction operator selected by the adaptive noise reduction module under intermittent strong impulse interference. Its input is the wavelet decomposition coefficients to be processed, and its output is the corrected wavelet decomposition coefficients after threshold processing. The function has two threshold parameters: an upper threshold and a lower threshold. Based on the relationship between the absolute value of the current wavelet decomposition coefficients and these two thresholds, it performs different processing operations in three cases.

[0095] The first scenario is the shearing region: when the absolute value of the current wavelet decomposition coefficient is less than or equal to the lower threshold, it is determined that the coefficient is mainly composed of a noise floor. The processing operation is to directly set the value of the current wavelet decomposition coefficient to zero, thereby completely eliminating the low-amplitude noise component.

[0096] The second scenario is the soft-threshold contraction region: When the absolute value of the current wavelet decomposition coefficient is greater than the upper threshold, it is determined that the coefficient may simultaneously contain abrupt changes in the useful signal and strong impulse interference. The processing operation involves subtracting the upper threshold from the absolute value of the current wavelet decomposition coefficient to obtain an intermediate amplitude. Based on the sign of the current wavelet decomposition coefficient and this intermediate amplitude, the corrected wavelet decomposition coefficient is determined. Using the soft-threshold contraction method can avoid the erroneous retention of large amplitude coefficients from strong impulse interference, while still preserving significant changes in the useful signal.

[0097] The third scenario is the transition zone: when the absolute value of the current wavelet decomposition coefficient is greater than the lower threshold and less than or equal to the upper threshold, it is determined that the coefficient consists of potentially useful signal components superimposed with interference, requiring smoothing and contraction processing. The processing steps are as follows: calculate the square of the lower threshold divided by the absolute value of the current wavelet decomposition coefficient to obtain the reduction amount; subtract the reduction amount from the absolute value of the current wavelet decomposition coefficient to obtain the intermediate amplitude; determine the corrected wavelet decomposition coefficient based on the original sign of the current wavelet decomposition coefficient and the intermediate amplitude, i.e., the sign of the corrected wavelet decomposition coefficient is the same as the original coefficient, and the amplitude is equal to the intermediate amplitude. This contraction amount decreases as the absolute value of the coefficient increases, making the contraction curve smooth and continuous, and the intermediate amplitude is always non-negative. While suppressing the noise response in the transition zone, it preserves the effective signal components reflecting minor damage to the leaky cable within this interval to the maximum extent.

[0098] The upper and lower thresholds are not globally fixed, but rather adaptively adjusted according to the statistical characteristics of the signal. Specifically, the system calculates the root mean square power of the data within the current signal window in real time and uses this value as a benchmark.

[0099] The upper-level threshold is obtained by multiplying the root mean square (RMS) power of the current signal window by a first adjustment factor of 3.5; the lower-level threshold is obtained by multiplying the RMS power of the current signal window by a second adjustment factor of 1.5. The RMS power is the arithmetic square root of the squared mean of the signal within the window, representing the overall energy level of the signal during that period. When intermittent strong pulse interference occurs, the signal energy increases sharply, and the RMS power rises accordingly. The upper and lower thresholds are also raised proportionally, thus avoiding incorrectly classifying strong interference pulses into the soft threshold contraction or transition region. When the pulse subsides, the RMS power falls back, and the threshold decreases accordingly, restoring sensitivity to changes in weak signals. This dynamic mechanism ensures that the threshold always matches the real-time signal strength, guaranteeing that the function maintains strong suppression of impulse noise and effective preservation of slightly damaged signals under different interference intensities.

[0100] In addition to the existing condition classifier and adaptive noise reduction module, the edge processing node is further equipped with a data compression and transmission module. The function of the data compression and transmission module is to efficiently encapsulate the feature data processed by the edge processing node and report it remotely. It serves as the communication outlet connecting the edge side and the cloud management platform.

[0101] The data compression and transmission module receives the denoised VSWR signal output by the adaptive denoising module and the feature data extracted from the vibration monitoring signal. The data compression and transmission module acquires the denoised VSWR signal, the insertion loss monitoring signal, the vibration monitoring signal, and the temperature monitoring signal. When the change in the value of the denoised VSWR signal exceeds a preset VSWR change threshold, the change in the value of the insertion loss monitoring signal exceeds a preset insertion loss change threshold, the vibration feature value extracted from the vibration monitoring signal exceeds a preset vibration threshold, the change in the value of the temperature monitoring signal exceeds a preset temperature change threshold, or a preset reporting time interval is reached, the feature data is extracted.

[0102] Standing Wave Ratio (SWR) Change Trigger: The current VSWR signal value after noise reduction is compared with the value reported last time. If the absolute value of the difference between the two exceeds the preset VSWR change threshold, it is determined that the electrical performance of the leaky cable has changed significantly, and data extraction is immediately triggered.

[0103] Insertion loss change trigger: The current insertion loss monitoring signal value is compared with the value at the last reported time. If the change exceeds the preset insertion loss change threshold, it is determined that the transmission attenuation is abnormal and data extraction is triggered immediately.

[0104] Vibration feature triggering: Extract features such as root mean square value, kurtosis or preset frequency band energy from vibration monitoring signals in real time. If any feature value exceeds the corresponding preset vibration threshold, it is determined that there are signs of mechanical damage (such as loose clamping parts or cracked sheath), and data extraction is triggered immediately.

[0105] Temperature change trigger: The current temperature monitoring signal value is compared with the value reported last time. If the change exceeds the preset temperature change threshold, it is determined that the ambient temperature has changed suddenly or the cable is overheated, and data extraction is triggered immediately.

[0106] Time-triggered: If the cumulative time since the last report reaches the preset reporting interval, even if none of the above signal changes exceed the corresponding threshold, a data extraction will still be triggered to ensure that the cloud management platform can obtain the system status regularly.

[0107] When any of the above conditions are met, the module performs a data extraction operation to extract the characteristic values ​​of the noise-reduced VSWR signal, insertion loss monitoring signal, vibration monitoring signal, and temperature monitoring signal from the current signal as the characteristic data to be uploaded.

[0108] The extracted feature data is serialized into binary format using a protocol buffer and asynchronously reported to the cloud management platform via a message queue telemetry transmission protocol at a preset quality of service level.

[0109] The first processing step is data serialization: The module uses a protocol buffer format as its data serialization protocol. Protocol buffering is a language-neutral, platform-independent structured data serialization mechanism. Its principle is to encode structured data into a compact binary byte stream based on a predefined message structure description file. Using this format for binary serialization of feature data results in a smaller data load compared to transmitting the original text stream. The second processing step is data delivery: The module uses a message queue telemetry transport protocol as its communication protocol. Message queue telemetry transport protocol is a lightweight IoT message transmission protocol based on a publish / subscribe model, suitable for low-bandwidth and unstable network environments. The module publishes serialized message data asynchronously, without blocking the main signal processing thread.

[0110] A preset Quality of Service (QoS) level is specified when publishing a message. The Message Queuing Telemetry Transport Protocol (MQTP) defines three QoS levels: Level 0 indicates that the message will be delivered at most once, with no acknowledgment mechanism; Level 1 indicates that the message will be delivered at least once, but duplicates are possible; and Level 2 indicates that the message will be delivered exactly once, with a four-way handshake ensuring no duplication or loss. The preset QoS level can be selected from these three levels based on the system's requirements for data transmission reliability. Data is asynchronously reported to the cloud management platform via the network using this protocol. This serialization and transmission mechanism reduces the average data transmission bandwidth requirement from approximately 15kbps in the original streaming mode to below 6kbps.

[0111] The edge processing node is equipped with a timing module, which receives timing signals from the Global Navigation Satellite System and outputs pulse signals and time information.

[0112] The specific hardware for the timing module is the u-blox NEO-M9N Global Navigation Satellite System timing module. This module's function is to receive timing signals broadcast by the Global Navigation Satellite System, which contain high-precision satellite atomic clock time base information. After receiving and locking onto the timing signal, the timing module outputs two time reference signals: one is a pulse signal with extremely high edge accuracy; the other is time information containing absolute date and time in Coordinated Universal Time (UTC) format. This pulse signal provides a nanosecond-level time edge reference, and the time information provides the absolute year, month, day, hour, minute, and second values ​​corresponding to that pulse edge.

[0113] When various monitoring signals reported by the multi-parameter composite monitoring terminal reach the edge processing node, the signal receiving logic inside the node uses the pulse signal output by the timing module as the time reference edge, and combines it with the absolute time value given by the time information corresponding to the pulse to generate and attach an absolute timestamp to each frame of monitoring data received. For multiple multi-parameter composite monitoring terminals deployed at different physical locations along the entire line, the data reported by each terminal is timestamped with an absolute timestamp based on Coordinated Universal Time (UTC) obtained from the same global navigation satellite system constellation when it enters its respective edge processing node. Thus, the data collected by all terminals is synchronized to the same time coordinate system, and the time deviation between them is controlled within microseconds, thereby achieving time synchronization of data between different monitoring terminals. This synchronization mechanism provides a time reference for the cloud management platform to perform multi-source data alignment, fusion, and accurate time series prediction of the Long Short-Term Memory (LSTM) network model.

[0114] The cloud management platform is the top layer of the entire system, responsible for long-term data storage, multi-dimensional model computation, and trend prediction. This platform acquires feature data from edge processing nodes. This feature data has undergone adaptive noise reduction processing and is encapsulated in a protocol buffer format, then uploaded via a message queue telemetry transmission protocol. The cloud management platform uses the received feature data as input, feeding it into a pre-trained Long Short-Term Memory (LSTM) network prediction model. The model's output is a predicted sequence of the Equivalent Health Index (EHI) for leaky cables within a preset future time period. The EHI is a comprehensive evaluation index proposed in this invention, with values ​​between the closed interval [0,1], where 1 represents complete health and 0 represents complete failure. Its function is to integrate multi-dimensional information on electrical performance, mechanical behavior, and environmental parameters to quantitatively characterize the immediate degradation state of leaky cables. After obtaining the predicted sequence of the EHI, the cloud management platform executes tiered early warning operations based on the numerical trend of the sequence and preset multi-level early warning thresholds, thereby notifying maintenance personnel to take intervention measures before the actual occurrence of a fault.

[0115] The cloud management platform includes a dynamic weight generation module and an equivalent health index prediction module.

[0116] Dynamic Weight Generation Module: This module utilizes a logistic regression model trained based on historical failure case data to dynamically generate fusion weight values ​​for each monitoring parameter (VSWR, insertion loss, RMS vibration value, temperature, etc.) according to their degradation level. The normalized value of each monitoring parameter is weighted and summed with this fusion weight value to obtain the equivalent health index at the current moment. The equivalent health index is set within the closed interval [0,1], where 1 represents complete health and 0 represents complete failure. This equivalent health index at the current moment is used for real-time status display and serves as a label value for subsequent training of the Long Short-Term Memory (LSTM) network prediction model.

[0117] Equivalent Health Index Prediction Module: This module deploys a trained Long Short-Term Memory (LSTM) network prediction model, employing an encoder-decoder sequence-to-sequence architecture. The model's input is an encoder, which acquires a sequence of multi-parameter feature vectors (including raw monitoring parameters such as VSWR, insertion loss, RMS vibration value, and temperature, but not the equivalent health index) over a first preset time period (e.g., the past 24 hours) at 1-minute intervals. The encoder recursively processes this input sequence, outputting a fixed-length encoded state vector that captures the long-term dependencies related to the leakage cable's degradation trend within the input sequence. The decoder receives this encoded state vector and, based on it, decodes step-by-step to generate a predicted sequence of the leakage cable's equivalent health index for a second preset time period (e.g., the next 60 minutes).

[0118] During the training phase, the Long Short-Term Memory (LSTM) network takes a historical multi-parameter feature vector sequence as input and a historical equivalent health index sequence calculated by the dynamic weight generation module as training labels, employing a dynamic time warping loss function for supervised learning. During the prediction phase, the model directly outputs a predicted sequence of future equivalent health indices based on the currently input multi-parameter feature sequence, without relying on the dynamic weight module.

[0119] The output of the Long Short-Term Memory (LSTM) network prediction model is a decoder. Its function is to receive the encoded state vector from the encoder and, based on this vector, decode it time-step by time to generate a predicted sequence of the equivalent health index of leaky cables for a second preset duration in the future. The first preset duration is the past 24 hours, and the second preset duration is the next 60 minutes. The hyperparameters used during model training are as follows: Adam optimizer, initial learning rate 0.0005, batch size 64, and an early stopping strategy where training stops if the loss does not decrease after 20 rounds of validation. A dynamic time warping loss function is used to make the model focus more on the accuracy of predicting sequence degradation trends during training.

[0120] The dynamic weight generation module uses a logistic regression model trained based on historical fault case data to dynamically generate the fusion weight value corresponding to each monitoring parameter in the equivalent health index based on the current degree of degradation of each monitoring parameter.

[0121] The equivalent health index is a weighted fusion of the degradation characteristic function values ​​of multiple monitoring parameters. The fusion weight vector is not a fixed value but is dynamically generated in real-time by a dynamic weight generation module. This module internally deploys a logistic regression model trained on historical failure case data. The training data source consists of at least 80 complete degradation cycle data points from healthy to failed, including typical failure modes such as connector corrosion, clamp loosening, and sheath aging. The function of this logistic regression model is, during the online monitoring phase, to receive real-time measurements of each monitoring parameter, evaluate and output the contribution of each parameter to the current overall degradation state, which is reflected in the fusion weight value corresponding to each monitoring parameter. Its dynamic adjustment mechanism is as follows: when physical damage associated with a certain type of parameter enters its initial stage, even if other parameters have not yet shown significant changes, the logistic regression model will automatically increase the fusion weight value of that parameter, thereby reflecting the anomaly in advance in the calculation of the equivalent health index, causing the equivalent health index value to be lowered in advance and triggering an early warning. This mechanism enables the system to detect early signs of failure by utilizing abnormal mechanical behavior or before electrical performance deteriorates, overcoming the limitations of traditional monitoring that relies solely on electrical parameter thresholds.

[0122] This technical solution also provides a method for real-time multi-parameter monitoring signal processing of leaky cables in rail transit, applied to a monitoring system consisting of a multi-parameter composite monitoring terminal, edge processing nodes, and a cloud management platform, including:

[0123] The multi-parameter composite monitoring terminal synchronously collects the standing wave ratio monitoring signal, insertion loss monitoring signal, vibration monitoring signal and temperature monitoring signal of the leaky cable deployed in the rail transit scenario, and reports all kinds of monitoring signals to the edge processing node through the same differential signal bus in a time-division multiplexing manner;

[0124] The edge processing node acquires various monitoring signals reported by the multi-parameter composite monitoring terminal and extracts the power spectral density characteristics of the vibration monitoring signal within a specific time window;

[0125] The edge processing node identifies the current electromagnetic interference condition type based on the power spectral density characteristics. The current electromagnetic interference condition type includes intermittent strong pulse interference condition and background noise condition.

[0126] The intermittent strong pulse interference condition is a condition in which the power spectral density exhibits discrete pulse peaks with amplitudes exceeding a first energy threshold within a preset frequency band, and the pulse interval is greater than a preset interval threshold.

[0127] The background noise condition is the condition in which the amplitude of the power spectral density within the preset frequency band is continuously lower than the second energy threshold, and the amplitude fluctuation is less than the preset fluctuation threshold.

[0128] When the intermittent strong pulse interference condition is identified, the edge processing node selects a double-threshold asymmetric threshold function to perform threshold processing on the wavelet decomposition coefficients of the VSWR monitoring signal.

[0129] When the background noise condition is identified, a soft thresholding function or a hard thresholding function is used to perform thresholding on the wavelet decomposition coefficients to obtain the VSWR signal after noise reduction.

[0130] The edge processing node packages and uploads the characteristic data of the noise-reduced VSWR signal, the insertion loss monitoring signal, and the vibration monitoring signal to the cloud management platform;

[0131] The cloud management platform acquires the feature data uploaded by the edge processing node, inputs the feature data into the trained long short-term memory network prediction model, obtains a prediction sequence of the equivalent health index of the leaky cable within a preset time period, and performs a graded early warning operation based on the prediction sequence.

[0132] A method for real-time monitoring signal processing of leaky cables in rail transit is disclosed. This method operates on a system consisting of a multi-parameter composite monitoring terminal, an edge processing node, and a cloud management platform connected sequentially. The steps cover four stages in sequence: data acquisition, working condition identification and adaptive noise reduction, data uploading, and prediction and early warning.

[0133] During the data acquisition phase, the multi-parameter composite monitoring terminal simultaneously acquires four types of signals from leaky cables deployed in rail transit scenarios such as tunnels: standing wave ratio (VSWR), insertion loss, vibration, and temperature. All monitoring signals are transmitted from the terminal to the edge processing node via the same differential signal bus, using time-division multiplexing, and are reported in a time-division manner according to preset time slots to avoid bus contention and crosstalk.

[0134] During the condition identification and adaptive noise reduction stage, after receiving the aforementioned monitoring signals, the edge processing node extracts the power spectral density characteristics of the vibration monitoring signal within a specific time window and identifies the current electromagnetic interference condition type based on these characteristics. There are two types of conditions: intermittent strong pulse interference is defined as discrete pulse peaks with amplitudes exceeding a first energy threshold appearing in the power spectral density within a preset frequency band, and the pulse interval being greater than a preset interval threshold; background noise is defined as the power spectral density continuously having an amplitude below a second energy threshold within a preset frequency band, and the amplitude fluctuation being less than a preset fluctuation threshold.

[0135] When the intermittent strong pulse interference is identified, the edge processing node uses a dual-threshold asymmetric threshold function to perform threshold processing on the wavelet decomposition coefficients of the VSWR monitoring signal, so as to strongly suppress pulse noise and preserve the minor damage signal.

[0136] When the background noise condition is identified, a soft or hard thresholding function is used to threshold the wavelet decomposition coefficients. After thresholding, the signal is reconstructed by inverse wavelet transform to obtain the denoised VSWR signal.

[0137] During the data upload phase, the edge processing node packages the noise-reduced VSWR signal, the insertion loss monitoring signal, and the feature data extracted from the vibration monitoring signal together and uploads them to the cloud management platform.

[0138] During the prediction and early warning phase, the cloud management platform receives feature data and inputs it into the trained Long Short-Term Memory Network prediction model. The model outputs a prediction sequence of the equivalent health index of the leaky cable within a preset time period in the future, and performs graded early warning operations based on the numerical change trend of the prediction sequence.

[0139] In one specific embodiment, a real-time multi-parameter monitoring system and signal processing method for leaky cables in rail transit are applied to a scenario of monitoring the health status of leaky cables in a tunnel section of a subway line in a certain city. The tunnel section is 1.8 kilometers long and is equipped with leaky cables for train communication signal coverage, operating at a frequency of 900MHz. The system consists of a three-tier architecture: a multi-parameter composite monitoring terminal, edge processing nodes, and a cloud management platform.

[0140] Regarding the deployment of multi-parameter composite monitoring terminals, a monitoring access point is set up every 400 meters along the leaky cable, for a total of 5 multi-parameter composite monitoring terminals. The multi-parameter composite monitoring terminals utilize IP67-rated sealed cavities and are fixed to cable supports on the tunnel sidewall. Each multi-parameter composite monitoring terminal integrates a microwave network parameter module, a vibration sensing module, a temperature sensing module, and a differential acquisition circuit module. The directional coupler in the microwave network parameter module is a Mini-Circuits BDCN-20-13+ type, with its main port connected in series to the leaky cable transmission path, and its reverse coupling port connected to the RF input of the logarithmic detector ADI AD8318. The vibration sensing module uses a TDK InvenSense ICM-20689 six-axis inertial measurement unit, which is rigidly fixed to the base of the leaky cable clamp by bolts to directly sense vibrations and impacts at the clamp. The temperature sensor of the temperature sensing module is attached to the inner wall of the sealed cavity, establishing a heat conduction path with the cavity wall through thermally conductive silicone grease.

[0141] Before online monitoring, the microwave network parameter module in each multi-parameter composite monitoring terminal performs a system-level calibration process. After disconnecting the leaky cable, the open-circuit calibrator, short-circuit calibrator, and matched-load calibrator are sequentially connected to the input port of the directional coupler. The differential acquisition circuit module records the digital code values ​​corresponding to the open-circuit calibration state, the short-circuit calibration state, and the matched-load calibration state, respectively, as three calibration reference points. The edge processing node acquires the above three calibration reference points and establishes a calibration mapping relationship between the digital code value and the reflected power. Specifically, since the output digital code value of the logarithmic detector is linearly related to the input RF power (dBm), three known reflected power values ​​can be obtained through the three standard states of open circuit, short circuit, and matched load (the reflected power is approximately 0 under matched load, and is maximum and equal under open circuit and short circuit). Using these three calibration reference points, the linear mapping relationship between the digital code value and the reflected power can be determined (e.g., using a two-point calibration curve or a piecewise linear interpolation table), thereby eliminating the fixed gain error and offset error of the logarithmic detector and the differential acquisition circuit. The calibration mapping is stored in the non-volatile memory of the edge processing node.

[0142] After calibration, the leaky cable was reconnected, and the system entered online monitoring mode. Simultaneously, a remote monitoring node was installed at the far end of each leaky cable section (approximately 1.8 km from the near-end multi-parameter composite monitoring terminal). This remote monitoring node uses the same directional coupler (Mini-Circuits BDCN-20-13+) and logarithmic detector as the near-end terminal to form a transmitted power sampling link, and reports the transmitted power data to the edge processing node every 5 minutes via a LoRa wireless communication module. The multi-parameter composite monitoring terminal synchronously collects the VSWR monitoring signal, insertion loss monitoring signal, vibration monitoring signal, and temperature monitoring signal of the leaky cable. The insertion loss monitoring signal is calculated by the edge processing node based on the incident power and the transmitted power returned from the far end.

[0143] The directional coupler in the microwave network parameter module samples the reflected power signal from the leaky cable transmission line. A logarithmic detector converts this reflected power signal into a DC voltage signal, which is then output to the differential acquisition circuit module. The differential acquisition circuit module is a third-order active instrumentation amplifier circuit, consisting of a cascaded buffer stage, a programmable gain amplifier stage, and a low-pass filter stage. The buffer stage is composed of a dual-channel precision operational amplifier, connected to the positive and negative terminals of the DC voltage signal respectively, providing high input impedance. The programmable gain amplifier stage automatically switches between eight preset gain levels (1 to 128 times, increasing in powers of 2) based on the input signal amplitude, ensuring the output signal amplitude remains within 80% to 90% of the analog-to-digital converter's full-scale range. If the input signal amplitude is large, it switches to a lower gain level; if the input signal amplitude is small, it switches to a higher gain level. The low-pass filter stage is a fifth-order elliptic low-pass filter, whose cutoff frequency is automatically set according to the system sampling rate. In this embodiment, the analog-to-digital converter has a sampling rate of 10 kHz. According to the Nyquist sampling theorem, the cutoff frequency is set to 4 kHz to filter out high-frequency carrier interference components that may exist in the audio track circuit, while ensuring a flat response in the passband of low-frequency reflected signals.

[0144] The inertial measurement unit (IMU) acquires triaxial acceleration and triaxial angular velocity signals at the installation location of the leaky cable, synthesizing them into a vibration monitoring signal. A temperature sensor acquires the temperature monitoring signal at the installation location of the leaky cable. After acquisition and conditioning, the four types of monitoring signals are reported via the same differential signal bus using time-division multiplexing. The time-division multiplexing frame period is set to 100 milliseconds, divided into four equal time slots, each 25 milliseconds long, allocated sequentially to the VSWR monitoring signal, insertion loss monitoring signal, vibration monitoring signal, and temperature monitoring signal. Under the scheduling of the microcontroller, each module only occupies the bus to transmit data within its allocated time slot.

[0145] The edge processing nodes utilize NVIDIA Jetson Xavier NX AI computing modules, installed within the communication equipment cabinets along the tunnel section. They connect to each multi-parameter composite monitoring terminal via a shielded low-voltage differential signal bus. Each edge processing node houses a u-blox NEO-M9N global navigation satellite system timing module. This module receives timing signals from GPS and BeiDou satellite navigation systems, locks onto them, and outputs a 1Hz pulse signal along with the corresponding Coordinated Universal Time (UTC) information. When monitoring signals arrive at the edge processing node, the node's receiving logic uses the rising edge of the pulse signal as a time reference. Combined with the year, month, day, hour, minute, and second values ​​from the corresponding time information, an absolute timestamp is added to each frame of data, with a timestamp accuracy of 100 nanoseconds. Thus, the data collected by the five multi-parameter composite monitoring terminals distributed at different locations along the tunnel are all synchronized to a unified UTC coordinate system.

[0146] The edge processing nodes' GPUs deploy a condition classifier and an adaptive noise reduction module. The condition classifier acquires vibration monitoring signals, extracts a specific time window of 0.5 seconds, estimates the power spectral density of the vibration signals within this window, and extracts the frequency domain energy distribution as the power spectral density feature. The power spectral density feature vector is input into a pre-trained lightweight fully connected neural network. This network contains one input layer, two hidden layers, and one output layer. The hidden layer has 64 and 32 neurons, respectively, and uses a linear rectified function as the activation function. The output layer uses a flexible maximum function to output the binary classification result. The network is deployed after offline training, with training data consisting of labeled background noise condition samples and intermittent strong impulse interference condition samples.

[0147] The current electromagnetic interference (EMI) condition type output by the condition classifier is divided into two categories. The first is intermittent strong pulse interference, identified by the following criteria: within a preset frequency band of 100 Hz to 2 kHz, discrete pulse peaks appear in the power spectral density, the energy amplitude of these peaks exceeds a first energy threshold, and the time interval between adjacent pulse peaks is greater than a preset interval threshold. The first energy threshold is 10 times (10 dB) of the average power spectral density of the background noise, and the preset interval threshold is 10 milliseconds. The second is background noise, identified by the following criteria: within a preset frequency band of 100 Hz to 2 kHz, the amplitude of the power spectral density is continuously lower than a second energy threshold, and the amplitude fluctuation is less than a preset fluctuation threshold. The second energy threshold is set to 1.5 times the long-term statistical average of the background noise, and the preset fluctuation threshold is set to 20% of the second energy threshold.

[0148] The adaptive noise reduction module receives the current electromagnetic interference condition type output by the condition classifier and selects the corresponding threshold function based on the identification result. When identified as an intermittent strong pulse interference condition, a dual-threshold asymmetric threshold function is selected; when identified as a background noise condition, a classic soft threshold function is selected.

[0149] Under intermittent strong pulse interference, the adaptive noise reduction module performs a 5-level discrete wavelet transform on the VSWR monitoring signal, using the Daubechies 4th order wavelet as the wavelet basis function. For the high-frequency wavelet decomposition coefficients of each level, a dual-threshold asymmetric thresholding function is applied. The upper threshold of the dual-threshold asymmetric thresholding function is dynamically set based on the product of the root mean square power of the current signal window and a first adjustment coefficient of 3.5; the lower threshold is dynamically set based on the product of the root mean square power of the current signal window and a second adjustment coefficient of 1.5. For the current wavelet decomposition coefficients to be processed, perform the following operations: If the absolute value of the current wavelet decomposition coefficient is less than or equal to the lower threshold, set the value of the current wavelet decomposition coefficient to zero; if the absolute value of the current wavelet decomposition coefficient is greater than the upper threshold, subtract the upper threshold from the absolute value of the current wavelet decomposition coefficient to obtain the intermediate amplitude, and determine the corrected wavelet decomposition coefficient based on the original sign of the current wavelet decomposition coefficient and the intermediate amplitude; if the absolute value of the current wavelet decomposition coefficient is greater than the lower threshold and less than or equal to the upper threshold, calculate the square of the lower threshold and divide it by the absolute value of the current wavelet decomposition coefficient to obtain the reduction amount, subtract the reduction amount from the absolute value of the current wavelet decomposition coefficient to obtain the intermediate amplitude, and determine the corrected wavelet decomposition coefficient based on the original sign of the current wavelet decomposition coefficient and the intermediate amplitude, that is, the sign of the corrected wavelet decomposition coefficient is consistent with the original coefficient, and the amplitude is the intermediate amplitude.

[0150] Under background noise conditions, the adaptive noise reduction module employs a globally unified threshold, calculated based on the estimated standard deviation of the signal and noise and the number of wavelet decomposition levels. A soft thresholding function is used: wavelet decomposition coefficients with absolute values ​​less than or equal to the threshold are set to zero; for coefficients with absolute values ​​greater than the threshold, the threshold is subtracted from the absolute value, and the original sign is retained. After thresholding, the coefficients at each level are reconstructed using inverse wavelet transform to obtain the denoised VSWR signal.

[0151] The edge processing node also includes a data compression and transmission module. This module acquires the denoised VSWR signal and monitors its changes. Data extraction is triggered when the change in the denoised VSWR signal exceeds a preset threshold of 0.05, or when the accumulated time since the last report reaches a preset reporting interval of 60 seconds. The extracted feature data includes the current denoised VSWR signal value, insertion loss monitoring signal value, vibration root mean square value, and temperature monitoring signal value. This feature data is serialized into binary data using a protocol buffer format, resulting in a data packet size of approximately 80 bytes. The data packet is asynchronously reported to the cloud management platform via a message queue telemetry transmission protocol at Quality of Service Level 1 (at least once, possibly repeatedly). Compared to the approximately 15kbps bandwidth required for directly transmitting raw streaming data, this mechanism reduces the average uplink bandwidth to approximately 1.6kbps.

[0152] The cloud management platform is deployed on a server cluster in the subway control center, receiving feature data reported by each edge processing node. The cloud management platform includes an equivalent health index prediction module and a dynamic weight generation module.

[0153] The dynamic weight generation module deploys a logistic regression model trained on historical fault case data. The data used to train this model comes from historical data of over 80 leaky cables throughout their complete degradation cycle from healthy to failure, covering three typical fault modes: connector corrosion, clamp loosening, and sheath aging. In actual operation, the dynamic weight generation module receives four monitoring parameters: VSWR, insertion loss, vibration RMS, and temperature. It then calls the logistic regression model to calculate and output the fused weight values ​​corresponding to each monitoring parameter. In the fused weight vector, the sum of the fused weight values ​​for VSWR, insertion loss, vibration, and temperature is always 1. In the early stages of cable clamp loosening, even if the VSWR monitoring signal has not yet changed significantly, the logistic regression model will automatically increase the vibration-related fused weight value from the usual 0.2 to above 0.5 based on the increase in the vibration RMS value, while simultaneously decreasing the other fused weight values ​​accordingly.

[0154] The equivalent health index prediction module is equipped with a trained long short-term memory network prediction model, which employs an encoder-decoder sequence-to-sequence network. The encoder receives a sequence of multi-parameter feature vectors from the past 24 hours, with 1-minute time intervals. Each feature vector contains VSWR, insertion loss, vibration root mean square (RMS), and temperature monitoring signal values. The encoder recursively processes the input sequence and outputs an encoded state vector. Based on the encoded state vector, the decoder decodes and generates an equivalent health index prediction sequence for the next 60 minutes, with 1-minute intervals between adjacent prediction points. The equivalent health index is calculated by multiplying the normalized degradation function values ​​of each monitoring parameter by the corresponding fusion weight values ​​output by the dynamic weight generation module, and then summing the results. The equivalent health index value is within the closed interval [0,1], where 1 represents complete health and 0 represents complete failure. The training hyperparameters of this Long Short-Term Memory Network prediction model are as follows: the optimizer is Adam, the initial learning rate is 0.0005, the batch size is 64, an early stopping strategy is adopted, and training is stopped when the loss does not decrease for 20 consecutive rounds; the loss function is the dynamic time warping loss function.

[0155] When any predicted point in the equivalent health index prediction sequence falls below the first-level warning threshold of 0.6, the cloud management platform generates a yellow warning, prompting maintenance personnel to arrange planned repairs. When any predicted point in the prediction sequence falls below the second-level warning threshold of 0.3, a red warning is generated, prompting immediate fault repair. In the actual operation of this embodiment, the system experienced an increase in the fusion weight value corresponding to vibration due to abnormal vibration monitoring signals. The equivalent health index prediction sequence dropped below 0.6 approximately 420 seconds before the actual fault occurred, issuing a yellow warning, thus completing the transformation from passive alarm to early predictive warning.

[0156] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A real-time multi-parameter monitoring system for leaky cables in rail transit, characterized in that, This includes a multi-parameter composite monitoring terminal, edge processing nodes, and a cloud management platform; The multi-parameter composite monitoring terminal synchronously collects the standing wave ratio monitoring signal, insertion loss monitoring signal, vibration monitoring signal and temperature monitoring signal of the leaky cable deployed in the rail transit scenario, and reports all kinds of monitoring signals to the edge processing node through the same differential signal bus in a time-division multiplexing manner; The edge processing node acquires various monitoring signals reported by the multi-parameter composite monitoring terminal, identifies the current electromagnetic interference condition type based on the vibration monitoring signal, selects the corresponding threshold function based on the identified current electromagnetic interference condition type, and performs threshold processing on the wavelet decomposition coefficients of the standing wave ratio monitoring signal using the selected threshold function to obtain the noise-reduced standing wave ratio signal. The node then packages and uploads the feature data of the noise-reduced standing wave ratio signal, the insertion loss monitoring signal, and the vibration monitoring signal to the cloud management platform. The cloud management platform acquires the feature data uploaded by the edge processing node, inputs the feature data into the trained long short-term memory network prediction model, obtains a prediction sequence of the equivalent health index of the leaky cable within a preset time period, and performs a graded early warning operation based on the prediction sequence.

2. The real-time multi-parameter monitoring system for leaky cables in rail transit according to claim 1, characterized in that, The multi-parameter composite monitoring terminal includes a microwave network parameter module, a vibration sensing module, a temperature sensing module, and a differential acquisition circuit module, all of which are integrated into the same sealed cavity. The microwave network parameter module includes a directional coupler and a logarithmic detector. The reverse coupling port of the directional coupler is electrically connected to the radio frequency input terminal of the logarithmic detector. The logarithmic detector converts the acquired reflected power signal into a DC voltage signal and outputs the DC voltage signal to the differential acquisition circuit module. The vibration sensing module includes an inertial measurement unit, which collects vibration monitoring signals at the installation location of the leaky cable. The temperature sensing module includes a temperature sensor, which collects the temperature monitoring signal at the installation location of the leaky cable. The differential acquisition circuit module is an active instrumentation amplifier circuit. After conditioning the DC voltage signal, the differential acquisition circuit module shares the same differential signal bus with the vibration monitoring signal output by the vibration sensing module and the temperature monitoring signal output by the temperature sensing module, and reports to the edge processing node in a time-division manner according to the preset time slot length.

3. The real-time multi-parameter monitoring system for leaky cables in rail transit according to claim 1, characterized in that, The edge processing node includes a condition classifier and an adaptive noise reduction module; The operating condition classifier acquires the vibration monitoring signal, extracts the power spectral density features of the vibration monitoring signal within a specific time window, identifies the current electromagnetic interference operating condition type based on the power spectral density features, and outputs the identified current electromagnetic interference operating condition type to the adaptive noise reduction module. The current electromagnetic interference conditions include intermittent strong pulse interference conditions and background noise conditions. The intermittent strong pulse interference condition is defined as the condition in which the power spectral density exhibits discrete pulse peaks with amplitudes exceeding a first energy threshold within a preset frequency band, and the pulse interval is greater than a preset interval threshold. The background noise condition is defined as the condition in which the amplitude of the power spectral density in the preset frequency band is continuously lower than the second energy threshold and the amplitude fluctuation is less than the preset fluctuation threshold. The adaptive noise reduction module receives the current electromagnetic interference condition type. When the current electromagnetic interference condition type is the intermittent strong pulse interference condition, a double threshold asymmetric threshold function is selected to process the wavelet decomposition coefficients. When the current electromagnetic interference condition is the background noise condition, a soft threshold function or a hard threshold function is selected to process the wavelet decomposition coefficients.

4. The real-time multi-parameter monitoring system for leaky cables in rail transit according to claim 3, characterized in that, When the adaptive noise reduction module processes the current wavelet decomposition coefficients using the dual-threshold asymmetric threshold function, it sets an upper-level threshold and a lower-level threshold, and performs the following operations: When the absolute value of the current wavelet decomposition coefficient is less than or equal to the lower-level threshold, the value of the current wavelet decomposition coefficient is set to zero; When the absolute value of the current wavelet decomposition coefficient is greater than the upper threshold, the absolute value of the current wavelet decomposition coefficient is subtracted from the upper threshold to obtain the intermediate amplitude. Based on the sign of the current wavelet decomposition coefficient and the intermediate amplitude, the corrected wavelet decomposition coefficient is determined. When the absolute value of the current wavelet decomposition coefficient is greater than the lower threshold and less than or equal to the upper threshold, the square of the lower threshold is calculated and divided by the absolute value of the current wavelet decomposition coefficient to obtain the reduction amount. The reduction amount is then subtracted from the absolute value of the current wavelet decomposition coefficient to obtain the intermediate amplitude. Based on the sign of the current wavelet decomposition coefficient and the intermediate amplitude, the corrected wavelet decomposition coefficient is determined. Under the intermittent strong pulse interference condition, the upper threshold is dynamically set according to the product of the root mean square power of the current signal window and the first adjustment coefficient, and the lower threshold is dynamically set according to the product of the root mean square power of the current signal window and the second adjustment coefficient.

5. The real-time multi-parameter monitoring system for leaky cables in rail transit according to claim 1, characterized in that, The cloud management platform includes an equivalent health index prediction module and a dynamic weight generation module; The dynamic weight generation module uses a logistic regression model trained based on historical fault case data to dynamically generate a fusion weight value for each monitoring parameter based on the current degree of degradation of each monitoring parameter. The module then weights and sums the values ​​of each monitoring parameter at the current moment according to the fusion weight value to obtain the equivalent health index at the current moment. The equivalent health index prediction module is equipped with the trained long short-term memory network prediction model. The trained long short-term memory network prediction model is an encoder-decoder structure. The encoder acquires the time series of the feature data within the past first preset time period and outputs an encoded state vector. The decoder decodes and outputs the prediction sequence of the equivalent health index within the future second preset time period based on the encoded state vector. In the training phase, the Long Short-Term Memory Network prediction model uses historical feature data sequences as input and historical equivalent health index sequences calculated by the dynamic weight generation module as training labels for supervised learning.

6. The real-time multi-parameter monitoring system for leaky cables in rail transit according to claim 1, characterized in that, The edge processing node also includes a data compression and transmission module; The data compression and transmission module acquires the noise-reduced VSWR signal, the insertion loss monitoring signal, the vibration monitoring signal, and the temperature monitoring signal; When the change in the value of the VSWR signal after noise reduction exceeds a preset VSWR change threshold, the change in the value of the insertion loss monitoring signal exceeds a preset insertion loss change threshold, the vibration feature value extracted from the vibration monitoring signal exceeds a preset vibration threshold, the change in the value of the temperature monitoring signal exceeds a preset temperature change threshold, or a preset reporting time interval is reached, the feature data is extracted. The extracted feature data is serialized into binary format using a protocol buffer and asynchronously reported to the cloud management platform via a message queue telemetry transmission protocol at a preset quality of service level.

7. The real-time multi-parameter monitoring system for leaky cables in rail transit according to claim 1, characterized in that, The edge processing node is equipped with a timing module, which receives timing signals from the global navigation satellite system and outputs pulse signals and time information. When the various monitoring signals reported by the multi-parameter composite monitoring terminal enter the edge processing node, they are stamped with an absolute timestamp based on the pulse signal and the time information, thereby achieving time synchronization of data between different monitoring terminals.

8. The real-time multi-parameter monitoring system for leaky cables in rail transit according to claim 2, characterized in that, Before being put into online monitoring, the microwave network parameter module in the multi-parameter composite monitoring terminal performs a system-level calibration process. The system-level calibration process includes: connecting the open-circuit calibrator, short-circuit calibrator and matched load calibrator sequentially to the input port of the directional coupler in the microwave network parameter module; and the differential acquisition circuit module in the multi-parameter composite monitoring terminal records the corresponding digital code values ​​under the three calibration states as calibration reference points. The edge processing node acquires the calibration reference point and establishes a calibration mapping relationship between digital code value and reflected power, including determining the corresponding curve or piecewise linear interpolation table between reflected power and digital code value. In actual monitoring, the measured digital code value of the device under test is converted into a reflected power value according to the calibration mapping relationship, and then the value of the standing wave ratio monitoring signal is calculated based on the reflected power value.

9. The real-time multi-parameter monitoring system for leaky cables in rail transit according to claim 2, characterized in that, The differential acquisition circuit module is a third-order active instrumentation amplifier circuit, including a buffer stage, a programmable gain amplification stage and a low-pass filter stage cascaded in sequence. The buffer stage is composed of a dual-channel precision operational amplifier, one channel of which is connected to the DC voltage signal output by the logarithmic detector, and the other channel is grounded. The programmable gain amplifier stage automatically switches between preset gain levels according to the amplitude of the input signal, so that the amplitude of the output signal is kept within a preset ratio range of the full scale of the analog-to-digital converter. The low-pass filter stage is a high-order elliptic low-pass filter, and the cutoff frequency of the high-order elliptic low-pass filter is automatically set according to the sampling rate of the system to filter out high-frequency noise components in the DC voltage signal.

10. A method for real-time multi-parameter monitoring signal processing of leaky cables in rail transit, applied to a monitoring system consisting of a multi-parameter composite monitoring terminal, an edge processing node, and a cloud management platform, characterized in that... include: The multi-parameter composite monitoring terminal synchronously collects the standing wave ratio monitoring signal, insertion loss monitoring signal, vibration monitoring signal and temperature monitoring signal of the leaky cable deployed in the rail transit scenario, and reports all kinds of monitoring signals to the edge processing node through the same differential signal bus in a time-division multiplexing manner; The edge processing node acquires various monitoring signals reported by the multi-parameter composite monitoring terminal and extracts the power spectral density characteristics of the vibration monitoring signal within a specific time window; The edge processing node identifies the current electromagnetic interference condition type based on the power spectral density characteristics. The current electromagnetic interference condition type includes intermittent strong pulse interference condition and background noise condition. The intermittent strong pulse interference condition is a condition in which the power spectral density exhibits discrete pulse peaks with amplitudes exceeding a first energy threshold within a preset frequency band, and the pulse interval is greater than a preset interval threshold. The background noise condition is the condition in which the amplitude of the power spectral density within the preset frequency band is continuously lower than the second energy threshold, and the amplitude fluctuation is less than the preset fluctuation threshold. When the intermittent strong pulse interference condition is identified, the edge processing node selects a double-threshold asymmetric threshold function to perform threshold processing on the wavelet decomposition coefficients of the VSWR monitoring signal. When the background noise condition is identified, a soft thresholding function or a hard thresholding function is used to perform thresholding on the wavelet decomposition coefficients to obtain the VSWR signal after noise reduction. The edge processing node packages and uploads the characteristic data of the noise-reduced VSWR signal, the insertion loss monitoring signal, and the vibration monitoring signal to the cloud management platform; The cloud management platform acquires the feature data uploaded by the edge processing node, inputs the feature data into the trained long short-term memory network prediction model, obtains a prediction sequence of the equivalent health index of the leaky cable within a preset time period, and performs a graded early warning operation based on the prediction sequence.

Images