High-speed rail platform area intrusion detection and early warning alarm system

Abstract

The invention relates to the technical field of high-speed rail platform safety monitoring, and discloses a high-speed rail platform area intrusion detection and early warning alarm system, which comprises an acoustic and vibration sensor network deployed in a platform area and used for collecting environment signals in real time, the edge calculation unit extracts signal features and compares the signal features with a dynamically maintained environmental rhythm feature baseline to generate a detuning event mark, and the detuning mode analysis device identifies an abnormal mode through space-time correlation and triggers an alarm. According to the method, early perception of subtle anomalies is realized by constructing a multi-modal environment rhythm baseline, active perturbation injection and differential response analysis technologies are combined, the recognition capability of a silent target is remarkably improved, meanwhile, the monitoring continuity under an extreme working condition is ensured by using an elastic baseline adaptive mechanism, and the detection accuracy is improved. And a closed-loop security and protection system from passive sensing to active discrimination is formed.

Description

Technical Field

[0001] The present invention relates to a high-speed railway platform area intrusion detection and early warning alarm system, belonging to the technical field of high-speed railway platform safety monitoring. Background Art

[0002] At present, the field of high-speed railway platform security monitoring generally adopts a composite perception solution with video surveillance as the core and supplemented by infrared radiation or lidar. This type of technology triggers alarms by capturing human contours, movement trajectories or physical cross-border behaviors. Its core logic relies on the direct identification of explicit intrusion events. Under normal working conditions, this type of system can effectively deal with typical intrusion behaviors, but its recognition stability is significantly reduced in scenarios with surging passenger density, complex lighting conditions or active target avoidance. It is worth noting that the existing technology has long been limited to simple threshold alarms for the use of acoustic and vibration information contained in the platform environment itself, and has failed to deeply explore the potential rhythmic characteristics and abnormal evolution laws in environmental signals, resulting in significant blind spots in the system's perception of new security threats such as covert infiltration, group abnormal gatherings and static risk lurking.

[0003] Taking the passenger flow diversion scenario during peak hours as an example, although the existing system can detect individual crossings through video analysis, it is difficult to identify subtle imbalances or abnormal retention trends in the overall flow pattern of the crowd. When the intruder uses the crowd as cover to implement gradual approach or place objects, the traditional solution lacks the ability to continuously model environmental rhythm parameters such as soundscape energy distribution and vibration propagation timing, and it is very easy to misjudge such low-feature exposure behavior as background noise. In addition, in the face of sudden strong interference (such as abnormal noise caused by equipment failure and structural vibration caused by instantaneous strong wind), the existing static threshold model is often forced to enter a conservative operation mode due to the surge in false trigger rate, which in turn provides a time window for the real threat to lurk. Specifically, there is The following problems: 1. The passive response mechanism driven by explicit events makes the system insensitive to signal changes in the embryonic stage of anomalies, and it is impossible to build a complete perception chain of pre-warning, in-process tracking, and post-event review; 2. The environmental baseline model with fixed thresholds or offline training is difficult to adapt to the rhythmic feature drift caused by the dynamic switching of platform operation status (such as train arrival and departure, passenger flow tidal changes), resulting in the false alarm and missed alarm rate increasing exponentially with the increase in scene complexity; 3. There is a lack of spatiotemporal correlation modeling capabilities between multi-source signal processing modules. The independent analysis mode of acoustic and vibration data breaks the overall propagation characteristics of environmental disturbance events, causing weak abnormal signals to be systematically filtered out in the cross-sensor collaborative verification link.

[0004] To address these issues, the industry has explored optimizing video target recognition accuracy through reinforcement learning algorithms or enhancing environmental perception through the introduction of multi-sensor data fusion technology. However, these improvements remain stuck within the traditional paradigm of focusing on target features, failing to fundamentally address the core challenges of dynamically modeling environmental rhythmic features and analyzing abnormal propagation patterns. Therefore, the technical challenge addressed by this invention is how to build a platform safety warning system that is adaptive to the environment, can capture hidden detuning signals, and can withstand interference from complex operating conditions. Summary of the Invention

[0005] The present invention provides a high-speed railway platform area intrusion detection and early warning alarm system, the main purpose of which is to solve the problems of delayed hidden threat detection, high false alarm rate in complex working conditions and failure of multi-source signal coordination caused by reliance on explicit event recognition and static environmental modeling in the existing technology.

[0006] To achieve the above objectives, the present invention provides a high-speed railway platform area intrusion detection and early warning alarm system, the system comprising: Multiple acoustic sensors and vibration sensors are deployed in the high-speed rail platform area to continuously collect the ambient soundscape and vibration signals in the high-speed rail platform area; A baseline learning and maintenance device is used to extract statistical features from historically collected environmental soundscape and vibration signals to generate and dynamically maintain an environmental rhythm feature baseline representing the normal operation of the high-speed railway platform area under different predetermined operating modes. The environmental rhythm feature baseline includes expected statistical feature values ​​corresponding to each predetermined operating mode and tolerance thresholds set based on the expected statistical feature values; Multiple edge computing units, each communicating with at least one acoustic sensor or vibration sensor, are configured to extract statistical features of a currently collected signal in real time, compare the features with expected statistical feature values ​​corresponding to the current operating mode of the high-speed railway platform area in the environmental rhythm feature baseline, and generate a detuning event marker when the comparison result indicates that the deviation between the statistical features of the currently collected signal and the expected statistical feature values ​​exceeds a tolerance threshold; A detuning pattern analysis and early warning device is used to receive detuning event markers from multiple edge computing units, perform spatiotemporal correlation analysis on the detuning event markers to identify detuning patterns that indicate potential intrusion or anomaly, and trigger an alarm signal when a detuning pattern is identified.

[0007] Preferably, the statistical features included in the environmental rhythm feature baseline include: the average energy envelope of one or more frequency bands pre-configured for the baseline learning and maintenance device, the fluctuating rhythmic characteristics of the signal's short-time energy, and the basic time domain correlation pattern and basic frequency domain correlation pattern between multiple sensor signals; and, the baseline learning and maintenance device is also used to adaptively adjust the currently effective environmental rhythm feature baseline based on real-time platform operation information obtained from an external interface or the platform operation plan stored in the system, the real-time platform operation information or the platform operation plan including train arrival and departure time data and passenger flow level data.

[0008] Preferably, the detune event marker generated by the edge computing unit is a data structure with a predetermined format, which includes a detune type code for identifying the detune type, a precise timestamp of the detune occurrence, and a detune intensity level value representing the detune amplitude.

[0009] Preferably, the detuning pattern identified by the detuning pattern analysis and early warning device includes one of the following pattern types recorded in the pre-stored pattern library or any combination thereof: a continuous diffusion pattern of detuning events in spatial positions, a synchronous outbreak pattern of detuning events at multiple different positions in time, and a continuous occurrence pattern of the same type of detuning events corresponding to the detuning type code in the detuning event mark within a single monitoring area.

[0010] Preferably, the detuning mode analysis and early warning device is also used to, when the triggering conditions preset by the detuning mode analysis and early warning device for starting the active identification process are met: control at least one existing acoustic device capable of outputting wide-area acoustic signals or an existing vibration device capable of generating structural vibrations deployed in the high-speed railway platform area, synchronously inject a synchronous perturbation signal having a time-frequency spectrum characteristic predetermined for the detuning mode analysis and early warning device and whose output intensity is controlled to be at a level equivalent to the current environmental background noise into the target monitoring area and the reference area specified by the operator; utilize the acoustic sensors and vibration sensors deployed in the target monitoring area and the reference area to analyze and distinguish the target monitoring area and the reference area; The response signals during the injection of the synchronous perturbation signal are separately collected; a differential detuned response spectrum of the target monitoring area is generated by comparing the difference between the response signal of the target monitoring area and at least one of the following reference signals: a background signal before the injection of the synchronous perturbation signal, a response signal of the reference area, or a healthy state response model established by historical data learning or on-site calibration of the target monitoring area under normal conditions; and based on whether the differential detuned response spectrum presents identifiable characteristic changes related to the presence of the hidden silent target, it is determined whether there is a hidden silent target in the target monitoring area, and an alarm signal is generated or the level of the existing alarm signal is adjusted accordingly.

[0011] Preferably, the rhythmic characteristics of the short-time energy fluctuations of the signal specifically include the periodic fluctuation characteristics and non-periodic burst characteristics of the signal energy envelope in the time series; the basic time domain correlation pattern and the basic frequency domain correlation pattern between multiple sensor signals specifically include the specific time sequence pattern of signal strength enhancement or weakening between different sensor acquisition signals expected based on the known deployment positions of each sensor, and the predetermined delay relationship pattern between these signal energy envelopes.

[0012] Preferably, the detuning mode analysis and early warning device is further configured to: when the time window preset for the detuning mode analysis and early warning device is Within, their respective strengths Exceeds a saturation intensity judgment threshold preset by a detuning mode analysis and warning device The number of detuning event markers The total number of edge computing units involved in monitoring proportion Meet the conditions When A global detuning saturation event trigger ratio threshold is pre-set for a detuning pattern analysis and early warning device to determine the occurrence of a global detuning saturation event; when a global detuning saturation event occurs, the macro-statistical characteristics of the initial detuning event flow that triggers the global detuning saturation event are extracted to generate saturation portrait data, and the dynamic change quantization value of the environmental rhythm feature baseline under the influence of the global detuning saturation event is evaluated to characterize its baseline elastic deformation modulus; and based on the evaluation results of the saturation portrait data and the baseline elastic deformation modulus, the interpretation logic parameters or alarm priority judgment rules for detuning events or detuning patterns that occur in the context of a global detuning saturation event or within a predetermined recovery period thereafter are dynamically adjusted to enhance the ability to identify real threat signals.

[0013] Preferably, the logic of the edge computing unit for determining whether the degree of deviation between the statistical characteristics of the currently collected signal and the expected statistical characteristic values ​​exceeds the tolerance threshold includes: determining whether the value of the average energy envelope currently collected, corresponding to one or more frequency band ranges pre-configured for the baseline learning and maintenance device in the environmental rhythm feature baseline, exceeds the normal fluctuation range defined by the tolerance threshold; determining whether the fluctuation pattern of the current signal energy envelope and the difference measure between it and the expected rhythm pattern corresponding to the current operating mode recorded in the environmental rhythm feature baseline exceeds the limit determined by the tolerance threshold; and determining whether the spatiotemporal correlation pattern between multiple sensor signals and the expected spatiotemporal correlation pattern corresponding to the current operating mode recorded in the environmental rhythm feature baseline exceeds the limit determined by the tolerance threshold.

[0014] Preferably, when the detuning pattern analysis and early warning device determines that the identified detuning pattern meets the pre-configured high-risk confirmation standard based on a comprehensive analysis of the identified detuning pattern, the system also links the existing video surveillance subsystem of the station through a predetermined communication interface to conduct a key target video review of the corresponding geographical area where the detuning pattern occurs. The video image or video clip data obtained by the video surveillance subsystem is used as supporting data for auxiliary judgment of the potential intrusion or the nature of the abnormal event.

[0015] Preferably, when dynamically maintaining the environmental rhythm feature baseline, the baseline learning and maintenance device also utilizes statistical data of environmental soundscapes and vibration patterns that have been identified and marked as originating from routine operations or known non-threatening interference sources in the system's historical operation data, as well as feedback learning data obtained after correlation analysis of historical alarm events and their subsequent verification results, to continuously iteratively update the expected statistical characteristic values ​​and tolerance thresholds in the environmental rhythm feature baseline, so as to improve the dynamic adaptability of the baseline to the normal state of the actual platform environment and the detection sensitivity to unknown real abnormal signals.

[0016] Compared with the background technology problems, the beneficial effects of the present invention are: 1. This system constructs a multi-dimensional environmental rhythm feature baseline, maps the statistical characteristics of high-speed rail platform soundscapes and vibration signals into dynamically changing normal fingerprints, and combines this with the real-time comparison mechanism of edge computing units to achieve early detection of subtle environmental anomalies. This collaborative design based on dynamic baseline maintenance and localized detuning judgment effectively solves the problem of traditional systems' delayed perception of sudden, low-feature exposure intrusion events, enabling the system to identify potential detuning precursors from the harmony of the environmental background, significantly improving the timeliness of early warnings.

[0017] 2. By integrating the time-domain and frequency-domain correlation characteristics of acoustic and vibration signals, the system avoids the limitations of single-sensor signal analysis. When an edge node detects a local detuning event, the central platform accurately distinguishes between benign interference and real threats by analyzing the coupling relationship between the event's spatial diffusion path, temporal synchronization, and intensity distribution. This cross-modal spatiotemporal correlation analysis mechanism maintains high robustness in scenarios such as dense crowds and complex sound fields, avoiding misjudgments caused by target occlusion in traditional vision systems.

[0018] 3. The system innovatively reuses station broadcast equipment to inject controllable perturbation signals. By comparing the response differences between the target area and the reference area, it stimulates changes in the acoustic vibration propagation characteristics caused by silent targets. This mechanism utilizes the nonlinear effects of hidden objects on perturbation signals, such as absorption and reflection, to form a characteristic differential spectrum under disturbances close to the background noise level, avoiding the blind spot of traditional passive monitoring where static suspicious objects continue to merge into the environmental baseline. In the face of instantaneous signal overload caused by extreme events, the system constructs an environmental elastic deformation model by analyzing the macro-statistical characteristics of the detuned event flow and the dynamic deformation of the baseline. While ensuring the continuity of routine monitoring, this mechanism dynamically adjusts the sensitivity threshold of the alarm logic, allowing the system to lock on abnormal signals related to threats even in a strong interference background, effectively suppressing the surge in false alarms caused by baseline drift in traditional solutions, and enhancing the reliability of decision-making under complex working conditions.

[0019] 4. Based on historical alarm verification results and known non-threatening interference patterns, the system continuously iterates and optimizes the statistical characteristics and tolerance range of the rhythm baseline. This process of embedding operation and maintenance feedback into baseline learning enables the system to autonomously distinguish between occasional environmental fluctuations and persistent abnormal patterns, gradually improving the generalized recognition capability of new threats, forming a closed-loop evolutionary path of monitoring-learning-optimization, and significantly reducing manual maintenance costs and false alarm rates. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a comparison diagram of the differential detuning response spectrum of the present invention; Figure 2 This is a functional flow diagram of the high-speed railway platform area intrusion detection and early warning alarm system of the present invention; Figure 3 This is a flowchart of the high-speed railway platform area multimodal rhythm analysis and differential response decision-making process of the present invention.

[0021] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0022] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0023] The present invention provides a high-speed railway platform area intrusion detection and early warning alarm system, the system comprising: Multiple acoustic sensors and vibration sensors are deployed in the high-speed rail platform area to continuously collect the ambient soundscape and vibration signals in the high-speed rail platform area; A baseline learning and maintenance device is used to extract statistical features from historically collected environmental soundscape and vibration signals to generate and dynamically maintain an environmental rhythm feature baseline representing the normal operation of the high-speed railway platform area under different predetermined operating modes. The environmental rhythm feature baseline includes expected statistical feature values ​​corresponding to each predetermined operating mode and tolerance thresholds set based on the expected statistical feature values; Multiple edge computing units, each communicating with at least one acoustic sensor or vibration sensor, are used to extract the statistical features of the current collected signal in real time, compare it with the expected statistical feature values ​​corresponding to the current operating mode of the high-speed rail platform area in the environmental rhythm feature baseline, and generate a detuning event marker when the comparison result indicates that the deviation between the statistical features of the current collected signal and the expected statistical feature values ​​exceeds the tolerance threshold; at the same time, in order to further enhance the consistency of the system parameter model and engineering reproducibility, the statistical feature parameters used in the environmental rhythm feature baseline specifically include: the average energy envelope of each frequency band extracted from the short-time Fourier transform (STFT) analysis results based on a 512-point Hamming window (unit: dB), the periodic fluctuation frequency (unit: Hz) and the non-periodic burst frequency (unit: times / minute) of the energy envelope, the average delay (unit: ms) and spectral correlation coefficient of the signal response between multiple nodes, and the upper and lower bounds of the feature distribution under different modes. The synchronous perturbation signal adopts dual-frequency amplitude modulation, with the main modulation frequencies at 40Hz and 90Hz respectively. The overall frequency band of the signal is controlled between 30Hz and 200Hz, and the power level is kept within the range of ±3dB of the background noise mean to ensure that there is no perceptible interference to passengers while having the ability to stimulate structural response. In the construction process of the differential detuned response spectrum, the system sequentially completes the acquisition of the target and reference area response signals, the extraction of characteristic indicators, the comparison of the baseline response spectrum differences and the calculation of similarity. All steps are executed synchronously with the sampling timestamp as the main sequence. For the variables involved in determining the global detuned saturation event, (global trigger ratio threshold, unit: %), (Number of high-intensity events, unit: piece), (Total number of monitoring units currently, unit: units), uniformly configured in the analysis module parameter set during system initialization and dynamically updated during real-time determination. The energy envelope is calculated based on a 2kHz sampling rate, a 10-second sliding window width, and a 1-second step length to ensure sufficient time resolution and frequency recognition capabilities.

[0024] A detuning pattern analysis and early warning device is used to receive detuning event markers from multiple edge computing units, perform spatiotemporal correlation analysis on the detuning event markers to identify detuning patterns that indicate potential intrusion or anomaly, and trigger an alarm signal when a detuning pattern is identified.

[0025] Preferably, the statistical features included in the environmental rhythm feature baseline include: the average energy envelope of one or more frequency bands pre-configured for the baseline learning and maintenance device, the fluctuating rhythmic characteristics of the signal's short-time energy, and the basic time domain correlation pattern and basic frequency domain correlation pattern between multiple sensor signals; and, the baseline learning and maintenance device is also used to adaptively adjust the currently effective environmental rhythm feature baseline based on real-time platform operation information obtained from an external interface or the platform operation plan stored in the system, the real-time platform operation information or the platform operation plan including train arrival and departure time data and passenger flow level data.

[0026] Preferably, the detune event marker generated by the edge computing unit is a data structure with a predetermined format, which includes a detune type code for identifying the detune type, a precise timestamp of the detune occurrence, and a detune intensity level value representing the detune amplitude.

[0027] Preferably, the detuning pattern identified by the detuning pattern analysis and early warning device includes one of the following pattern types recorded in the pre-stored pattern library or any combination thereof: a continuous diffusion pattern of detuning events in spatial positions, a synchronous outbreak pattern of detuning events at multiple different positions in time, and a continuous occurrence pattern of the same type of detuning events corresponding to the detuning type code in the detuning event mark within a single monitoring area.

[0028] Preferably, the detuning mode analysis and early warning device is also used to, when the triggering conditions preset by the detuning mode analysis and early warning device for starting the active identification process are met: control at least one existing acoustic device capable of outputting wide-area acoustic signals or an existing vibration device capable of generating structural vibrations deployed in the high-speed railway platform area, synchronously inject a synchronous perturbation signal having a time-frequency spectrum characteristic predetermined for the detuning mode analysis and early warning device and whose output intensity is controlled to be at a level equivalent to the current environmental background noise into the target monitoring area and the reference area specified by the operator; utilize the acoustic sensors and vibration sensors deployed in the target monitoring area and the reference area to analyze and distinguish the target monitoring area and the reference area; The response signals during the injection of the synchronous perturbation signal are separately collected; a differential detuned response spectrum of the target monitoring area is generated by comparing the difference between the response signal of the target monitoring area and at least one of the following reference signals: a background signal before the injection of the synchronous perturbation signal, a response signal of the reference area, or a healthy state response model established by historical data learning or on-site calibration of the target monitoring area under normal conditions; and based on whether the differential detuned response spectrum presents identifiable characteristic changes related to the presence of the hidden silent target, it is determined whether there is a hidden silent target in the target monitoring area, and an alarm signal is generated or the level of the existing alarm signal is adjusted accordingly.

[0029] Preferably, the rhythmic characteristics of the short-time energy fluctuations of the signal specifically include the periodic fluctuation characteristics and non-periodic burst characteristics of the signal energy envelope in the time series; the basic time domain correlation pattern and the basic frequency domain correlation pattern between multiple sensor signals specifically include the specific time sequence pattern of signal strength enhancement or weakening between different sensor acquisition signals expected based on the known deployment positions of each sensor, and the predetermined delay relationship pattern between these signal energy envelopes.

[0030] Preferably, the detuning mode analysis and early warning device is further configured to: when the time window preset for the detuning mode analysis and early warning device is Within, their respective strengths Exceeds a saturation intensity judgment threshold preset by a detuning mode analysis and warning device The number of detuning event markers The total number of edge computing units involved in monitoring proportion Meet the conditions When A global detuning saturation event trigger ratio threshold is pre-set for a detuning pattern analysis and early warning device to determine the occurrence of a global detuning saturation event; when a global detuning saturation event occurs, the macro-statistical characteristics of the initial detuning event flow that triggers the global detuning saturation event are extracted to generate saturation portrait data, and the dynamic change quantization value of the environmental rhythm feature baseline under the influence of the global detuning saturation event is evaluated to characterize its baseline elastic deformation modulus; and based on the evaluation results of the saturation portrait data and the baseline elastic deformation modulus, the interpretation logic parameters or alarm priority judgment rules for detuning events or detuning patterns that occur in the context of a global detuning saturation event or within a predetermined recovery period thereafter are dynamically adjusted to enhance the ability to identify real threat signals.

[0031] Preferably, the logic of the edge computing unit for determining whether the degree of deviation between the statistical characteristics of the currently collected signal and the expected statistical characteristic values ​​exceeds the tolerance threshold includes: determining whether the value of the average energy envelope currently collected, corresponding to one or more frequency band ranges pre-configured for the baseline learning and maintenance device in the environmental rhythm feature baseline, exceeds the normal fluctuation range defined by the tolerance threshold; determining whether the fluctuation pattern of the current signal energy envelope and the difference measure between it and the expected rhythm pattern corresponding to the current operating mode recorded in the environmental rhythm feature baseline exceeds the limit determined by the tolerance threshold; and determining whether the spatiotemporal correlation pattern between multiple sensor signals and the expected spatiotemporal correlation pattern corresponding to the current operating mode recorded in the environmental rhythm feature baseline exceeds the limit determined by the tolerance threshold.

[0032] Preferably, when the detuning pattern analysis and early warning device determines that the identified detuning pattern meets the pre-configured high-risk confirmation standard based on a comprehensive analysis of the identified detuning pattern, the system also links the existing video surveillance subsystem of the station through a predetermined communication interface to conduct a key target video review of the corresponding geographical area where the detuning pattern occurs. The video image or video clip data obtained by the video surveillance subsystem is used as supporting data for auxiliary judgment of the potential intrusion or the nature of the abnormal event.

[0033] Preferably, when dynamically maintaining the environmental rhythm feature baseline, the baseline learning and maintenance device also utilizes statistical data of environmental soundscapes and vibration patterns that have been identified and marked as originating from routine operations or known non-threatening interference sources in the system's historical operation data, as well as feedback learning data obtained after correlation analysis of historical alarm events and their subsequent verification results, to continuously iteratively update the expected statistical characteristic values ​​and tolerance thresholds in the environmental rhythm feature baseline, so as to improve the dynamic adaptability of the baseline to the normal state of the actual platform environment and the detection sensitivity to unknown real abnormal signals, and the global decision logic adopted by the system when handling the intensive triggering of large-scale detuning events involves the following quantitative indicator relationships: ,in, Indicates that within the set time window, the detuning intensity exceeds the saturation threshold The proportion of event markers to the total number of monitoring units, The global detuning trigger threshold ratio configured for the system is used to determine whether to enter the elastic baseline adjustment process. The units of each variable and the judgment rules are configured in the system software platform. In terms of signal processing, the system extracts signal features in the frequency band of 30Hz to 200Hz. The default sampling rate is 2kHz. Fast Fourier transform (FFT) is performed based on a 512-point short-time window (Hamming window). Statistics such as the energy envelope are extracted within a sliding window width of 10 seconds to ensure sufficient resolution for frequency change trends and sudden disturbances. In determining deviations exceeding 25%, the system uses the normalized difference measurement formula: ,in is the current signal characteristic value, is the baseline expectation value. In the detuning event type coding, the system uses an internal coding structure to distinguish event types such as abnormal frequency band enhancement, cross-node delay drift, and sudden vibration enhancement. The specific coding system is embedded in the analysis module to support scalable event category management. In the differential detuning response spectrum, the system defaults to performing comparative analysis in the frequency domain of the response signals before and after perturbation injection, extracting the differential spectrum structure, and combining the cross-node delay and resonance peak change trends for pattern recognition. The relevant calculations do not rely on a single analytical model, but are adaptively modeled based on the statistical structural variation trends in the response data.

[0034] Example 1: In an actual pilot project at a comprehensive hub-type high-speed railway station, the deployed system is suitable for a typical twelve-platform interconnected area, which covers the platform edge protection area, the passenger waiting area, and the lower space of the structure at the train entrance. During the deployment process, the influence of the coupling characteristics between the sound wave propagation path and the structural vibration on the signal perception effect is fully considered. Based on this consideration, no less than six high-fidelity broadband acoustic sensor nodes and four low-frequency and high-sensitivity vibration sensor nodes are selected in each platform unit, and are respectively arranged on the platform floor, the screen door track, the bottom of the waiting bench, and the base of the platform column, etc., which are conducive to capturing environmental responses. The above-mentioned sensor nodes are connected to the edge computing unit in this area through a ring network topology structure composed of Ethernet, and maintain periodic two-way handshake verification with the platform main control platform to ensure the integrity of the signal link and transmission synchronization.

[0035] At the initial stage of deployment, each edge computing unit will continuously collect the original acoustic and vibration signal data obtained by the sensors in this area during the complete operating cycle (with 72 hours as the initial cycle). Subsequently, energy estimation will be performed through the short-time window function to calculate the statistical characteristic indicators such as the mean, standard deviation, periodic fluctuation frequency and sudden change frequency of the energy envelope of each frequency band, and classified and sorted in combination with the platform operation status label (such as train entering the station, opening doors, passengers getting on and off the train, etc.), and finally form a set of rhythmic characteristic parameters under multiple typical operating modes. On this basis, the baseline learning and maintenance device will use the range analysis and confidence interval limit learning algorithm of the sliding time window to extract the normal fluctuation range of each rhythmic indicator under different modes, and determine the corresponding tolerance threshold range accordingly. The whole process is controlled by the system. Automatic execution and periodic iterative updates ensure that the generated environmental rhythm feature baseline can reflect the dynamic operating rhythm of the platform area in real time. After the system is put into actual operation, the signal data obtained by any edge computing unit in each sampling period (such as every half second) will be compared in real time with the rhythm feature baseline matching the current system state. For example, in the acoustic data processing process, if it is detected that the deviation between the energy envelope mean of a certain frequency band and its baseline set value exceeds the allowable threshold of 25%, and the deviation lasts for more than two sampling periods (that is, more than three seconds), the edge unit will generate a detuning event marker, which contains structured fields such as event type code (such as abnormal enhancement of frequency band), event intensity level, and the precise timestamp of the event, and will be immediately uploaded to the detuning pattern analysis and early warning device.

[0036] After receiving the detuning event markers uploaded by multiple edge computing units, the detuning pattern analysis device starts a parallel event clustering and pattern recognition process. First, it performs preliminary clustering based on the geographic location tag and time information attached to each marker, and combines the dynamic threshold clustering strategy to determine whether the current event has a trend of spatial continuous diffusion or temporal synchronous outbreak. If, within the specified time window, detuning events of the same type are found in three or more adjacent monitoring units, and their time interval does not exceed the set threshold (such as no more than three seconds), the system can preliminarily determine that there is a potential abnormal expansion pattern. On the premise that the above-mentioned detuning pattern is preliminarily identified, the system will actively start the perturbation injection mechanism. Specifically, the system controls the station broadcast subsystem in the area, and at the same time injects a perturbation acoustic signal with a specific time-frequency spectrum modulation structure into the high-risk area and its corresponding reference area. The frequency range of the signal is controlled between ten and three hundred hertz, and the signal strength is maintained at the same level. A low interference level equivalent to the background noise level is maintained to avoid any perceptible impact on passengers while effectively stimulating the nonlinear acoustic-vibration coupling effect between potential abnormal objects and the platform structure. During signal injection, the edge computing unit synchronously enters the perturbation response acquisition mode, recording the response curves before, during, and after injection in real time. It then performs differential processing on the response data of the target area and the reference area to generate a differential detuned response spectrum. The system then compares several key characteristic parameters of the differential detuned response spectrum, including changes in frequency band structure, resonance peak drift trends, and cross-sensor delay variations, with the historical silent target feature database established in the system for similarity. When the confidence match of the comparison result exceeds a preset threshold (e.g., exceeding 80%), the system will confirm the presence of a highly suspicious abnormal target in the area, thereby triggering a high-priority alarm signal and synchronously linking the platform video surveillance subsystem to retrieve multi-angle surveillance footage of the area for subsequent event review and manual analysis.

[0037] Example 2: In order to verify the recognition capability of the present system for silent intruder targets in the actual operation environment of the high-speed railway platform, especially when the traditional visual recognition system shows a blind spot due to factors such as crowd occlusion or low target significance, it is proposed to conduct a systematic verification of its adaptability, recognition effect and system stability under different operating conditions through the combined mechanism of environmental rhythm feature baseline construction and perturbation response analysis proposed by the present system; this test was carried out in the connecting area of ​​Platforms 6 and 7 of a high-speed railway station with a comprehensive transportation hub in central China, covering the platform edge protection area, passenger waiting area and the space under the platform shield door structure. The test covered three typical platform operation conditions in the early morning, evening peak and night low peak, and the measured background noise intensity ranged from 38 dB (night) to 74 dB (during peak broadcast periods), the system deployment structure is as follows: six acoustic sensor nodes are deployed in each platform unit (installed at typical reflection / sound absorption locations such as platform columns, platform screen door tracks, and the bottom of waiting seats), and four vibration sensor nodes are also installed (installed at key nodes on the track infrastructure and platform ground). The perturbation signal is output by the platform broadcast system, modulated in a dual-frequency amplitude modulation structure with a frequency range controlled between 30 Hz and 200 Hz. The injection intensity is limited to ±3 dB of the background noise mean to ensure that the structural response differences of potential static targets to the perturbation are induced without causing interference to passengers.

[0038] Silent intrusion targets were simulated manually, primarily involving test personnel wearing sound-absorbing outerwear and slowly approaching the platform edge. Static objects wrapped in non-metallic structures, designed to resemble common objects and inconspicuous, were placed concealed near the edge of platform screen doors. Each set of simulated behaviors was conducted three times over three typical time periods. After system deployment, 72 hours of background signal data were continuously collected. A sliding window function was used to estimate the energy of the data within each 10-second window. Statistical indicators such as the mean, standard deviation, fluctuation frequency, and sudden change frequency of the energy envelope in each frequency band were extracted. These indicators were then combined with platform status labels (such as train arrival, door opening, and passengers boarding or alighting) for classification modeling, ultimately generating a set of rhythmic characteristic parameters encompassing 12 typical operating states. Based on these parameters, a baseline learning and maintenance device employed range analysis and confidence interval limit learning algorithms to extract normal fluctuation intervals and set tolerance thresholds accordingly. This process was automatically executed by the system and continuously iterated at preset intervals to ensure the dynamic adaptability of the generated environmental rhythmic characteristic baseline. The detuning event judgment standard is based on the combination of multiple rounds of initial simulation and historical data backtracking. The detuning event judgment logic adopted by each edge computing unit in the current system is as follows: the deviation of the mean energy envelope of a certain frequency band collected by any sensor from the baseline expected value exceeds 25%, and the deviation lasts for more than two sampling cycles (i.e., 3 seconds); the signal response delay between multiple sensors exceeds the average value ±15 ms setting range; within the same time window (10 seconds), if three or more spatially adjacent monitoring units simultaneously generate detuning event markers, it is classified as a localized burst event; after determining the presence of a spatially clustered detuning pattern, the system automatically triggers the perturbation injection mechanism, and the station broadcast subsystem synchronously injects a preset perturbation signal into the determination area and the reference area. The injection process is divided into three sections: 3 seconds before injection, 5 seconds during injection, and 4 seconds after injection. During this period, the edge computing unit records the response curves of the target and reference areas with millisecond sampling accuracy to generate a differential detuning response spectrum. The key parameters extracted from the differential response spectrum include frequency band structure deformation, main resonance peak drift, and cross-node response delay changes. The spectrum is then compared with the historical silent target feature database established in the system for similarity; when the confidence match of the comparison result exceeds 80%, the system determines that a highly suspicious target exists in the area, triggers a high-priority alarm, and links the video surveillance subsystem to record images and retrieve videos of the area for subsequent event verification. The test data and analysis results are as follows:

[0039] In all 18 sets of experiments, the system's recognition success rate reached 94.4% in the presence of silent targets. At the same time, there were no false alarms caused by platform background interference. The system's false alarm rate remained below 4.1%, mainly concentrated in the initial moment of broadcast frequency modulation conversion, which is within an acceptable range.

[0040] Example 3: This example combines Figures 1 to 3 , this paper describes the implementation of a high-speed railway platform area intrusion detection and early warning alarm system. Figure 1 As shown in the figure, this figure shows a comparison of differential detuned response spectra, which is used to demonstrate the difference in frequency distribution between the response spectrum of the silent target area and the response spectrum of the normal area. The silent target area response spectrum is drawn with a solid line, representing the acoustic response energy spectrum collected by the system after injecting a perturbation signal in the high-risk area; the normal area response spectrum is drawn with a dotted line, indicating the response characteristics of the reference area where no abnormality is identified in the same time period and frequency range. The vertical axis in the figure is the normalized energy spectrum amplitude, reflecting the energy intensity of each frequency point, and the horizontal axis is the frequency (Hz), ranging from 20Hz to 220Hz, covering the main frequency bands involved in the perturbation injection signal of this system; the 87.5Hz and 94.1Hz marked in the figure represent the key frequency points in the Q-value drop region of the detuned response spectrum, respectively. They are located between the Q-value drop sections of the gray-marked area, corresponding to the frequency range where the quality factor (Q value) in the response spectrum is significantly reduced, revealing the influence characteristics of the silent target on the propagation characteristics of the acoustic signal.

[0041] like Figure 2 As shown in the figure, the system first uses the acoustic and vibration sensor network to realize real-time collection of environmental signals and perform multimodal data fusion, and transmits the original signal to the edge computing unit, which is responsible for real-time feature extraction, dynamic baseline comparison and generation of detuning event markers. The generated detuning event markers are sent to the detuning pattern analysis and early warning device for performing spatiotemporal correlation analysis, active perturbation injection and differential response analysis. After confirming the high risk, the alarm signal linkage module is further triggered, including video review and graded early warning operations. At the same time, the baseline learning and maintenance device in the system is responsible for constructing the environmental rhythm feature baseline and dynamically updating the tolerance threshold. According to the environmental rhythm feature baseline matching results returned from the edge computing unit, elastic deformation feedback is provided to achieve adaptive adjustment. In addition, the perturbation signal control path is specially marked in the figure, which clearly points out the perturbation injection control instructions issued by the detuning pattern analysis and early warning device to the edge computing unit. The closed loop supports the dynamic response and active identification mechanism of the entire system.

[0042] like Figure 3As shown in the figure, the system completes environmental signal collection through a multimodal sensor array (acoustic / vibration), and the collected signals are transmitted to the edge intelligent node (real-time feature extraction) for preliminary processing. After extracting key statistical features, a detuning event stream is formed, which is handed over to the intelligent analysis center (space-time pattern recognition) for analysis and identification. During the identification process, the system, on the one hand, performs matching evaluation based on the baseline comparison mechanism and the dynamic rhythm baseline (multimodal modeling). On the other hand, it starts perturbation signal injection for potential abnormal situations, and the active perturbation system (differential response analysis) confirms the abnormal area positioning results, and constructs an elastic deformation model to assist in judgment. Subsequently, the system transmits the identified risk level to the graded warning platform (video linkage review) through the risk level determination path, triggers the corresponding warning response level, and feeds back optimization suggestions to the dynamic rhythm baseline (multimodal modeling) to achieve closed-loop update.

[0043] Example 4: During nighttime operation at a typical high-speed rail station, passenger flow is low and the ambient background noise level is relatively low. Due to the sparse passenger density, frequently changing lighting conditions, and the reduced contrast of the video surveillance system under low light conditions, this period presents a high-risk window for the placement of illegal items or the lurking of static abnormal targets. This example focuses on the system's operating mechanism under these complex conditions. By combining a dynamic baseline learning mechanism, an active perturbation differential response strategy, and multimodal signal collaborative analysis capabilities, it achieves effective identification and early warning response to static hidden intrusion behaviors. Specifically, in this scenario, multiple acoustic and vibration sensor nodes deployed in the platform waiting area and the leading edge of the platform screen door have completed a basic learning cycle and constructed an environmental rhythm feature baseline covering the background noise characteristics and structural response laws under nighttime operation mode. The baseline parameters include the mean value of the energy envelope in different frequency bands, the fluctuation period, the frequency of local amplitude mutations, and the cross-node response delay distribution pattern. These parameters are dynamically extracted and updated by the baseline learning and maintenance device during a continuous 72-hour learning process to form a stable model with nighttime scene adaptability.

[0044] When the system enters the night monitoring state, the edge computing unit collects acoustic and vibration signals at a sampling frequency of twice per second, and extracts short-term statistical features in real time. The statistical features are then matched and compared with the rhythm feature baseline in the currently activated night operation mode. When the energy envelope mean value of the 30 to 100 Hz frequency band of a certain area deviates by more than 20% compared with the baseline expected value in two consecutive sampling cycles, and the deviation is accompanied by a decrease in the fluctuation frequency from twice per minute to less than once per minute, the system determines that there are signs of signal detuning in the area. The edge computing unit generates a structured detuning event marker based on this and reports it to the detuning pattern analysis module; the detuning pattern analysis module After receiving the detuning event markers from three adjacent monitoring nodes, a cluster analysis in spatial and temporal dimensions is performed based on the attached geographic tags and timestamps. If it is determined that the event sequence has the characteristics of spatial continuity and concentrated outbreak within a ten-second time window, the system will preliminarily identify it as a static abnormal risk pattern and further activate the active perturbation injection mechanism. This mechanism controls the station broadcast system to synchronously inject a specially modulated acoustic disturbance signal into the judgment area and its symmetrical reference area in an acoustic manner below the audible threshold. The signal covers the frequency band of 30 to 120 Hz and adopts a dual-frequency progressive amplitude modulation structure, aiming to stimulate the structural nonlinear response caused by potential static objects.

[0045] In the three stages before and after the perturbation signal injection (i.e., before injection, during injection, and after injection), the edge computing unit collects the acoustic response data of the target area and the reference area respectively, and constructs a differential detuning response spectrum based on the response amplitude change trend, spectrum drift characteristics, and cross-sensor phase delay pattern. The differential response analysis process does not rely on any formula expression, but summarizes the response changes in full Chinese form. For example, it is observed that the response energy of the target area near the 50 Hz frequency point is more than 30% higher than that of the reference area, and the frequency drift range of the response in this frequency band is expanded to more than 10 Hz. At the same time, the similarity between this type of response feature and the typical response template induced by non-metallic objects in the system's pre-built silent target feature database is more than 85%; when the differential detuning response spectrum is confirmed by the system's internal preset judgment logic, the system determines that there is a high-confidence static abnormal target in the area, and immediately triggers a high-priority alarm signal. It also links the video surveillance system through the communication interface to package the real-time image and historical image data of the target area and transmit it to the early warning processing terminal for subsequent manual analysis and To avoid false triggering of high-level alarms when there are fluctuations in environmental background noise, the system also simultaneously searches for the changing trends of the corresponding baseline parameters in the area over the past fifteen minutes. If the elastic deformation exceeds twice the normal standard deviation of the parameter, the alarm level is further increased to enhance the system's accuracy in identifying real threats. The frequency and intensity of the perturbation signal used in this scenario are set based on the following: the frequency range is selected based on early structural response backtest data. After actual measurement of the natural frequency of the platform concrete structure, the 30 to 120 Hz frequency range is preferred as the main excitation frequency range; the intensity of the perturbation signal is set based on the statistical distribution of the local nighttime background sound level to ensure that its injection intensity is controlled within three decibels above and below the background noise mean, thereby avoiding perceptible interference to passengers and ensuring sufficient structural response excitation capability. The judgment threshold used in the differential response spectrum is taken from the empirical distribution of response differences obtained from more than 100 sets of scenario simulation verifications. Based on this, the system sets the critical frequency band amplitude difference exceeding 25% and the phase response drift exceeding 0.2 seconds as the judgment boundaries. After multiple rounds of actual testing, the system's success rate in identifying static abnormal targets exceeded 92 percent, and no false alarms caused by disturbance signals from the broadcast system occurred.

[0046] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0047] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A high-speed railway platform area intrusion detection and early warning alarm system, characterized in that: The system comprises: Multiple acoustic sensors and vibration sensors are deployed in the high-speed rail platform area to continuously collect the ambient soundscape and vibration signals in the high-speed rail platform area; A baseline learning and maintenance device is used to extract statistical features from historically collected environmental soundscape and vibration signals to generate and dynamically maintain an environmental rhythm feature baseline representing the normal operation of the high-speed railway platform area under different predetermined operating modes. The environmental rhythm feature baseline includes expected statistical feature values ​​corresponding to each predetermined operating mode and tolerance thresholds set based on the expected statistical feature values; Multiple edge computing units, each communicating with at least one acoustic sensor or vibration sensor, are configured to extract statistical features of a currently collected signal in real time, compare the features with expected statistical feature values ​​corresponding to the current operating mode of the high-speed railway platform area in the environmental rhythm feature baseline, and generate a detuning event marker when the comparison result indicates that the deviation between the statistical features of the currently collected signal and the expected statistical feature values ​​exceeds a tolerance threshold; A detuning pattern analysis and early warning device is used to receive detuning event markers from multiple edge computing units, perform spatiotemporal correlation analysis on the detuning event markers to identify detuning patterns that indicate potential intrusion or anomaly, and trigger an alarm signal when a detuning pattern is identified.

2. The high-speed railway platform area intrusion detection and early warning alarm system according to claim 1 is characterized in that: The statistical features contained in the environmental rhythm feature baseline include: the average energy envelope of one or more frequency bands pre-configured for the baseline learning and maintenance device, the fluctuating rhythmic characteristics of the signal's short-time energy, and the basic time domain correlation pattern and basic frequency domain correlation pattern between multiple sensor signals; and the baseline learning and maintenance device is also used to adaptively adjust the currently effective environmental rhythm feature baseline based on real-time platform operation information obtained from an external interface or the platform operation plan stored in the system, the real-time platform operation information or the platform operation plan including train arrival and departure time data and passenger flow level data.

3. The high-speed railway platform area intrusion detection and early warning alarm system according to claim 2 is characterized in that: The detune event marker generated by the edge computing unit is a data structure with a predetermined format, which includes a detune type code for identifying the detune type, a precise timestamp of the detune occurrence, and a detune intensity level value indicating the detune amplitude.

4. The high-speed railway platform area intrusion detection and early warning alarm system according to claim 1 is characterized in that: The detuning patterns identified by the detuning pattern analysis and early warning device include one or any combination of the following pattern types recorded in the pre-stored pattern library: the continuous diffusion pattern of detuning events in spatial positions, the synchronous outbreak pattern of detuning events at multiple different positions in time, and the continuous occurrence pattern of the same type of detuning events corresponding to the detuning type code in the detuning event mark in a single monitoring area.

5. The high-speed railway platform area intrusion detection and early warning alarm system according to claim 1 is characterized in that: The detuning pattern analysis and early warning device is further configured to, when a trigger condition preset by the detuning pattern analysis and early warning device for initiating an active identification process is met, control at least one existing acoustic device capable of outputting a wide-area acoustic signal or an existing vibration device capable of generating structural vibration deployed within the high-speed railway platform area to synchronously inject a synchronous perturbation signal having time-frequency spectrum characteristics predetermined for the detuning pattern analysis and early warning device, and having an output intensity controlled to be equivalent to the current environmental background noise, into a target monitoring area and a reference area designated by an operator; Acoustic sensors and vibration sensors deployed in the target monitoring area and the reference area are used to collect response signals during the injection of synchronous perturbation signals. A differential detuned response spectrum of the target monitoring area is generated by comparing the difference between the response signal of the target monitoring area and at least one of the following reference signals: the background signal before the injection of the synchronous perturbation signal, the response signal of the reference area, or a healthy state response model established by historical data learning or on-site calibration of the target monitoring area under normal conditions; and based on whether the differential detuned response spectrum shows identifiable characteristic changes related to the presence of a hidden silent target, it is determined whether there is a hidden silent target in the target monitoring area, and an alarm signal is generated or the level of the existing alarm signal is adjusted accordingly.

6. The high-speed railway platform area intrusion detection and early warning alarm system according to claim 2 is characterized in that: The rhythmic characteristics of the short-term energy fluctuations of the signal specifically include the periodic fluctuation characteristics and non-periodic burst characteristics of the signal energy envelope in the time series; the basic time domain correlation pattern and the basic frequency domain correlation pattern between multiple sensor signals specifically include the specific time sequence pattern of signal strength increase or decrease between the signals collected by different sensors expected based on the known deployment positions of each sensor, as well as the predetermined delay relationship pattern between these signal energy envelopes.

7. The high-speed railway platform area intrusion detection and early warning alarm system according to claim 1 is characterized in that: The detuning mode analysis and early warning device is further used for: when the time window preset for the detuning mode analysis and early warning device is Within, their respective strengths Exceeds a saturation intensity judgment threshold preset by a detuning mode analysis and warning device The number of detuning event markers The total number of edge computing units involved in monitoring proportion Meet the conditions When A global detuning saturation event trigger ratio threshold is pre-set for a detuning pattern analysis and early warning device to determine the occurrence of a global detuning saturation event; when a global detuning saturation event occurs, the macro-statistical characteristics of the initial detuning event flow that triggers the global detuning saturation event are extracted to generate saturation portrait data, and the dynamic change quantization value of the environmental rhythm characteristic baseline under the influence of the global detuning saturation event is evaluated to characterize its baseline elastic deformation modulus; and based on the evaluation results of the saturation portrait data and the baseline elastic deformation modulus, the interpretation logic parameters or alarm priority judgment rules for detuning events or detuning patterns that occur in the context of a global detuning saturation event or within a predetermined recovery period thereafter are dynamically adjusted.

8. The high-speed railway platform area intrusion detection and early warning alarm system according to claim 1 is characterized in that: The logic of the edge computing unit to determine whether the degree of deviation between the statistical characteristics of the currently collected signal and the expected statistical characteristic values ​​exceeds the tolerance threshold includes: determining whether the value of the average energy envelope currently collected, corresponding to one or more frequency band ranges pre-configured for the baseline learning and maintenance device in the environmental rhythm feature baseline, exceeds the normal fluctuation range defined by the tolerance threshold; determining whether the fluctuation pattern of the current signal energy envelope and the difference measure between it and the expected rhythm pattern corresponding to the current operating mode recorded in the environmental rhythm feature baseline exceeds the limit determined by the tolerance threshold; and determining whether the spatiotemporal correlation pattern between multiple sensor signals and the expected spatiotemporal correlation pattern corresponding to the current operating mode recorded in the environmental rhythm feature baseline exceeds the limit determined by the tolerance threshold.

9. The high-speed railway platform area intrusion detection and early warning alarm system according to claim 4, characterized in that: When the detuning pattern analysis and early warning device determines that the identified detuning pattern has reached the pre-configured high-risk confirmation standard based on a comprehensive analysis of the identified detuning pattern, the system also links the station's existing video surveillance subsystem through a predetermined communication interface to conduct a key target video review of the corresponding geographical area where the detuning pattern occurs. The video image or video clip data obtained by the video surveillance subsystem is used as supporting data to assist in judging the nature of potential intrusions or abnormal events.

10. The high-speed railway platform area intrusion detection and early warning alarm system according to claim 2, characterized in that: When dynamically maintaining the environmental rhythm feature baseline, the baseline learning and maintenance device also uses statistical data on environmental soundscapes and vibration patterns in the system's historical operating data that have been identified and marked as originating from routine operations or known non-threatening interference sources, as well as feedback learning data obtained after correlation analysis of historical alarm events and their subsequent verification results, to continuously iteratively update the expected statistical characteristic values ​​and tolerance thresholds in the environmental rhythm feature baseline.

Images