An abnormal event pattern recognition method and system based on optical fiber vibration signals

Abstract

The application relates to the technical field of optical fiber sensing, and provides an abnormal event pattern recognition method and system based on an optical fiber vibration signal, which comprises the following steps: acquiring a resource state of a resource-limited device, optimizing, according to the resource state of the resource-limited device, a recognition logic used for abnormal event recognition to obtain a device-end lightweight recognition logic; based on vibration event description information, initial position information and the device-end lightweight recognition logic, performing an event recognition task and an event positioning task to obtain an abnormal event recognition result and abnormal event positioning information; acquiring a matching degree of the vibration event description information and a known event pattern, generating a reliability evaluation result according to the matching degree, the abnormal event positioning information, environment information and a historical false alarm record, and triggering a hierarchical alarm according to the reliability evaluation result. The application has the effect of improving the accuracy and reliability of system abnormal event pattern recognition.

Description

Technical Field

[0001] This invention relates to the technical field of fiber optic sensing, and specifically to a method and system for abnormal event pattern recognition based on fiber optic vibration signals. Background Technology

[0002] In modern society, the rapid development of critical infrastructure such as communications, power, oil and gas transmission, and rail transportation poses a severe challenge to the safe and stable operation of optical cables carrying critical functions. For example, optical cable outages caused by external forces such as excavation and drilling during municipal construction are becoming increasingly frequent. Traditional maintenance methods rely on manual inspections, which have drawbacks such as blind spots, slow response times, and inaccurate accident location.

[0003] To address these challenges, Distributed Acoustic Sensing (DAS) technology has emerged and is gradually becoming a core technology in the field of infrastructure security. This technology utilizes optical fibers themselves as sensing elements, based on the φ-OTDR principle, to capture minute vibration signals along the fiber optic cable, providing a feasible approach for long-distance, all-weather optical cable status monitoring. However, despite the promising future of DAS technology, it still faces several challenges in practical engineering applications. Existing systems often focus on only a few dimensions when extracting vibration signal features and rely excessively on a single analysis method, resulting in insufficient adaptability to complex and changing working environments and making it easy for abnormal events to be "ignored" by the system.

[0004] Furthermore, existing systems mostly process signals serially, resulting in a time delay of 200 milliseconds or even longer from the occurrence of vibration to the issuance of an early warning, which is insufficient for real-time early warning scenarios. Regarding anomaly location, existing systems typically have an accuracy of around ±10 meters, making it difficult for on-site personnel to accurately pinpoint the problem and take appropriate action. Traditional solutions often rely on high-performance computing devices, such as high-power graphics processing units (GPUs), for complex calculations, resulting in large-scale system analysis programs that are difficult to install directly on resource-constrained, small-sized edge devices, leading to high maintenance costs for large-scale deployments.

[0005] The existing system is inadequate in distinguishing between human-caused damage and interference from the natural environment, resulting in frequent false alarms—a persistently high false alarm rate. A large number of false alarms not only increases the workload of maintenance personnel but also significantly reduces public confidence in the system's reliability.

[0006] To address the aforementioned issues, existing technologies urgently need improvement. Summary of the Invention

[0007] This application discloses an abnormal event pattern recognition method and system based on fiber optic vibration signals, aiming to solve the technical problems faced by existing distributed acoustic sensing (DAS) technology in practical engineering applications, such as poor adaptability to complex environments, insufficient real-time performance, low positioning accuracy, high deployment cost, and high false alarm rate.

[0008] The technical solution of this application is as follows:

[0009] In a first aspect, this application discloses an abnormal event pattern recognition method based on optical fiber vibration signals, specifically including:

[0010] The vibration signal, environmental information, and historical false alarm records of the optical fiber along the route are acquired, and the initial location information of the vibration is determined based on the vibration signal; the initial location information indicates the initial location of the vibration event on the optical fiber along the route or the corresponding section information along the route.

[0011] The vibration signal is converted into audio processing data, and a correspondence is established between the audio processing data and the initial position information;

[0012] Based on the correspondence, the optical cable laying medium of the corresponding section along the line is determined, and the audio processing data is subjected to medium compensation and multi-dimensional feature extraction according to the optical cable laying medium. The signal features of multiple dimensions are fused to generate vibration event description information. Medium compensation means that the audio processing data is processed by transmission attenuation, frequency response or phase offset correction according to the vibration transmission characteristics of the optical cable laying medium.

[0013] The resource status of the resource-constrained device is obtained. Based on the resource status of the resource-constrained device, the identification logic used for abnormal event identification is optimized to obtain the device-side lightweight identification logic. The device-side lightweight identification logic represents the lightweight identification logic suitable for local operation on the resource-constrained device. It is the identification logic formed after feature compression, branch pruning, parameter reduction or operation frequency adjustment of the identification logic used for abnormal event identification.

[0014] Based on vibration event description information, initial position information, and lightweight identification logic on the device, the event identification task and event localization task are executed in parallel to obtain abnormal event identification results and abnormal event localization information.

[0015] The system obtains the degree of matching between vibration event description information and known event patterns. Based on the degree of matching, abnormal event location information, environmental information, and historical false alarm records, it performs a reliability assessment on the abnormal event identification results, generates a reliability assessment result, and triggers graded alarms based on the reliability assessment result.

[0016] This technical solution can effectively solve the problems of insufficient adaptability, poor real-time performance, low positioning accuracy, high deployment cost, and high false alarm rate of existing DAS technology in complex and ever-changing working environments, and significantly improve the accuracy, real-time performance, and reliability of abnormal event identification.

[0017] Secondly, this application also discloses an abnormal event pattern recognition system based on fiber optic vibration signals, used to perform abnormal event pattern recognition based on fiber optic vibration signals, specifically including:

[0018] The initial position determination module is used to acquire vibration signals, environmental information, and historical false alarm records of the optical fiber along the line, and to determine the initial position information of the vibration occurrence based on the vibration signals.

[0019] The correspondence establishment module is used to convert vibration signals into audio processing data and establish a correspondence between the audio processing data and the initial position information;

[0020] The vibration event description module is used to determine the optical cable laying medium of the corresponding section along the line based on the correspondence relationship, and to perform medium compensation and multi-dimensional feature extraction on the audio processing data according to the optical cable laying medium, and to fuse signal features of multiple dimensions to generate vibration event description information.

[0021] The identification logic generation module is used to obtain the resource status of resource-constrained devices, and to perform lightweight optimization on the identification logic used for abnormal event identification based on the resource status of resource-constrained devices, so as to obtain the lightweight identification logic on the device side.

[0022] The parallel task execution module is used to execute event identification and event localization tasks in parallel based on vibration event description information, initial position information and lightweight identification logic on the device, to obtain abnormal event identification results and abnormal event localization information.

[0023] The graded alarm triggering module is used to obtain the degree of matching between vibration event description information and known event patterns. Based on the degree of matching, abnormal event location information, environmental information and historical false alarm records, it performs a reliability assessment on the abnormal event identification results, generates a reliability assessment result, and triggers graded alarms based on the reliability assessment result.

[0024] This technical solution provides a system for implementing the above-mentioned abnormal event pattern recognition method. Through modular design, it ensures the coordinated operation of all parts of the system, thereby achieving comprehensive and efficient processing of fiber optic vibration signals and accurate identification and alarm of abnormal events.

[0025] Beneficial Effects: This application discloses an abnormal event pattern recognition method based on optical fiber vibration signals. By acquiring vibration signals, environmental information, and historical false alarm records along the optical fiber, and determining the initial location information of the vibration, it lays the foundation for subsequent processing. The vibration signals are converted into audio processing data and a corresponding relationship is established, making signal processing more flexible. Based on this relationship, the optical cable laying medium is determined, and medium compensation and multi-dimensional feature extraction are performed. Multi-dimensional signal features are fused to generate vibration event description information, effectively solving the problems of limited feature extraction dimensions and insufficient adaptability due to single analysis methods in existing technologies, significantly improving adaptability to complex and changing working environments. Simultaneously, the resource status of resource-constrained devices is acquired, and the recognition logic is optimized for lightweight design, resulting in a lightweight recognition logic at the device end. This overcomes the problems of traditional solutions relying on high-performance computing equipment, large program size, and difficulty in deploying to resource-constrained edge devices, reducing the operation and maintenance costs of large-scale deployments. More importantly, based on vibration event description information, initial location information, and lightweight identification logic on the device side, the event identification and event localization tasks are executed in parallel to obtain abnormal event identification results and abnormal event localization information. This effectively solves the problem of insufficient real-time performance caused by serial processing in existing systems and improves the localization accuracy from the traditional ±10 meters to a higher level. Finally, by obtaining the degree of matching between vibration event description information and known event patterns, combined with abnormal event localization information, environmental information, and historical false alarm records, the reliability of abnormal event identification results is assessed. A reliability assessment result is generated and a graded alarm is triggered, effectively distinguishing between human-caused sabotage and natural environmental interference, significantly reducing the false alarm rate and improving system reliability. Attached Figure Description

[0026] Figure 1 This is a flowchart of an abnormal event pattern recognition method based on optical fiber vibration signals in one embodiment of the present invention.

[0027] Figure 2 This is a flowchart of an abnormal event pattern recognition method based on optical fiber vibration signals according to another embodiment of the present invention.

[0028] Figure 3 This is a system block diagram of an abnormal event pattern recognition system based on optical fiber vibration signals according to another embodiment of the present invention;

[0029] Explanation of reference numerals in the attached figures:

[0030] 1. Abnormal event pattern recognition system based on fiber optic vibration signals; 11. Initial position determination module; 12. Correspondence establishment module; 13. Vibration event description module; 14. Recognition logic generation module; 15. Task parallel execution module; 16. Hierarchical alarm triggering module. Detailed Implementation

[0031] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0032] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0033] This application proposes an abnormal event pattern recognition method based on fiber optic vibration signals, combined with... Figure 1 As shown, it includes:

[0034] S1, acquire vibration signals, environmental information and historical false alarm records of the optical fiber along the line, and determine the initial location information of the vibration occurrence based on the vibration signal; the initial location information indicates the initial location of the vibration event on the optical fiber along the line or the corresponding section information along the line;

[0035] S2 converts the vibration signal into audio processing data and establishes the correspondence between the audio processing data and the initial position information;

[0036] S3, based on the correspondence, determines the optical cable laying medium of the corresponding section along the line for audio processing data, and performs medium compensation and multi-dimensional feature extraction on the audio processing data according to the optical cable laying medium, and integrates signal features of multiple dimensions to generate vibration event description information; medium compensation means processing the audio processing data by transmission attenuation, frequency response or phase offset correction according to the vibration transmission characteristics of the optical cable laying medium.

[0037] S4. Obtain the resource status of the resource-constrained device. Based on the resource status of the resource-constrained device, perform lightweight optimization on the identification logic used for abnormal event identification to obtain the device-side lightweight identification logic. The device-side lightweight identification logic represents the lightweight identification logic suitable for local operation on the resource-constrained device. It is the identification logic formed after feature compression, branch pruning, parameter reduction, or operation frequency adjustment of the identification logic used for abnormal event identification.

[0038] S5, based on vibration event description information, initial position information and lightweight identification logic on the device, executes event identification task and event localization task in parallel to obtain abnormal event identification result and abnormal event localization information;

[0039] S6. Obtain the degree of matching between the vibration event description information and the known event pattern. Based on the degree of matching, abnormal event location information, environmental information and historical false alarm records, perform a reliability assessment on the abnormal event identification results, generate a reliability assessment result, and trigger a graded alarm based on the reliability assessment result.

[0040] "Vibration signal" refers to the acoustic or mechanical vibration information obtained from the optical fiber along the fiber optic cable using distributed acoustic sensing technology. This vibration signal carries the signal characteristics of the physical events occurring along the cable. "Initial location information" refers to the preliminary location of the vibration event on the optical fiber or the corresponding section information along the cable, which can be initially determined using the principle of optical time-domain reflectometry, phase-sensitive optical time-domain reflectometry analysis, or related positioning algorithms. "Audio processing data" refers to data suitable for audio analysis and signal feature extraction formed after sampling, quantization, format conversion, or spectral processing of the original vibration signal. "Optical cable laying medium" refers to the external environment in which the optical cable is located, such as soil, pipelines, or overhead environments. Different optical cable laying media will have different effects on the propagation attenuation, frequency response, and phase change of the vibration signal.

[0041] "Media compensation" refers to processing audio processing data by correcting transmission attenuation, frequency response, or phase shift based on the vibration transmission characteristics of the optical cable laying medium. This reduces the interference of the optical cable laying medium on the vibration signal, making the compensated audio processing data closer to the characteristics of the actual vibration event. "Multi-dimensional feature extraction" refers to extracting signal features from multiple dimensions, such as the time domain, frequency domain, and time-frequency domain, from the compensated audio processing data to describe the vibration event from different perspectives. "Vibration event description information" refers to event description data formed by fusing multiple dimensions of signal features and initial location information, used to comprehensively express the intensity, spectral structure, duration, location, and variation patterns of the vibration event.

[0042] "Resource-constrained devices" refer to edge devices with limited computing power, storage space, power consumption, or communication capabilities, such as IoT terminals, small servers, and edge computing gateways. "Identification logic" refers to the algorithm model, rule set, or judgment logic formed by combining a model and rules for anomaly event identification. "Lightweight optimization" refers to reducing the computational complexity, storage footprint, and power consumption of the identification logic through methods such as feature compression, branch pruning, parameter reduction, or operating frequency adjustment, making it suitable for the local operating environment of resource-constrained devices. "Device-side lightweight identification logic" refers to identification logic that, after lightweight optimization, can be deployed and run locally on resource-constrained devices.

[0043] "Event Recognition Task" refers to the process of determining the event type and nature based on vibration event description information and lightweight recognition logic on the device side. "Event Location Task" refers to the process of further determining the location of the event based on vibration event description information and initial location information. "Abnormal Event Recognition Result" is the output of the event recognition task, used to represent the identified abnormal event type, such as excavation, drilling, vehicle passage, knocking, or other abnormal disturbance events. "Abnormal Event Location Information" is the output of the event location task, used to represent the precise location of the abnormal event or the corresponding section along the route. "Known Event Patterns" refers to pre-stored templates of various abnormal events with clear characteristics. "Matching Degree" refers to the degree of similarity between the vibration event description information and known event patterns. "Environmental Information" refers to external environmental parameters such as temperature, humidity, wind speed, and rainfall at the time of the event. "Historical False Alarm Records" refers to records of past events that were determined to be false alarms after manual confirmation or system review. "Reliability Assessment Result" is the result obtained after quantifying or grading the reliability of the abnormal event recognition results. "Graded Alarms" refers to the mechanism that triggers different levels of alarm responses based on the reliability assessment results.

[0044] In this embodiment, vibration signals, environmental information, and historical false alarm records along the optical fiber are first acquired, and the initial location information of the vibration is determined based on the vibration signals. Specifically, the vibration signals can be acquired in real time using distributed optical fiber sensing devices. For example, phase-sensitive optical time-domain reflectometry (PTZ) technology can be used to inject laser pulses into the optical fiber, and vibration signals at different locations along the fiber can be obtained by analyzing the phase or intensity changes of the Rayleigh scattering signal. Environmental information can be obtained using environmental sensors deployed along the optical fiber, such as temperature sensors, humidity sensors, wind speed sensors, or rainfall sensors. Historical false alarm records can be obtained from maintenance databases, historical alarm systems, or manual review records, and may include the location of the false alarm, the time of the false alarm, the environmental conditions of the false alarm, the characteristics of the false alarm signal, and the cause of the false alarm. The initial location information can be determined by analyzing the optical fiber echo time, phase change location, signal intensity peak location, or time difference of arrival corresponding to the vibration signal, thereby initially obtaining the location or segment of the vibration event along the optical fiber.

[0045] Subsequently, the vibration signal is converted into audio processing data, and a correspondence is established between the audio processing data and the initial location information. Specifically, the original analog vibration signal can be converted into a digital signal using an analog-to-digital converter, and the digital signal can be processed by adjusting the sampling rate, quantizing, framing, normalizing, or filtering to form a data format suitable for audio analysis. As another implementation method, Fourier transform, short-time Fourier transform, or other spectral analysis can be performed on the vibration signal to extract the corresponding spectral data or time-frequency data, which can then be used as audio processing data. To ensure that the subsequent identification and localization processes can trace the location of the event, the system establishes a correspondence between the audio processing data and the initial location information. For example, each segment of audio processing data can be bound and stored with its corresponding initial location information through associative arrays, timestamp indexes, segment identifiers, or data mapping tables, so that subsequent media compensation, feature extraction, event identification, and event localization can all call upon the corresponding location data.

[0046] Next, based on the correspondence between the audio processing data and the initial location information, the optical cable laying medium of the corresponding section along the route is determined. Then, medium compensation and multi-dimensional feature extraction are performed on the audio processing data according to the optical cable laying medium, fusing signal features from multiple dimensions to generate vibration event description information. In specific implementation, the optical cable laying medium of the corresponding section along the route can be determined by querying geographic information system data, construction records, maintenance ledgers, or on-site survey data based on the initial location information. For example, one section of optical cable may be buried in soil, another section may be laid in a duct, or it may be in an overhead environment. Different optical cable laying media have different transmission characteristics for vibration signals. For example, soil media may cause more significant attenuation of high-frequency components, duct media may produce resonance responses in specific frequency bands, and overhead environments may be more sensitive to wind vibration and low-frequency disturbances. Therefore, the system can perform medium compensation on the audio processing data based on the transmission model, compensation parameters, or historical calibration data corresponding to the optical cable laying medium. Medium compensation can include designing inverse filters to offset transmission attenuation, correcting the frequency response of a specific frequency band, compensating for phase shift, or using adaptive filtering algorithms to adjust compensation parameters according to real-time signal changes.

[0047] After completing the medium compensation, the system extracts multi-dimensional features from the compensated audio processing data. Time-domain features may include root mean square (RMS), peak value, peak-to-peak value, energy, duration, zero-crossing rate, and number of impulses; frequency-domain features may include dominant frequency, spectral energy, band energy distribution, spectral centroid, bandwidth, and harmonic structure; time-frequency domain features may include wavelet coefficients, short-time Fourier energy distribution, and Mel-frequency cepstral coefficients. Time-domain features can be obtained by directly calculating the compensated signal waveform, frequency-domain features can be obtained through Fast Fourier Transform (FFT), and time-frequency domain features can be obtained through wavelet transform, short-time Fourier transform, or other time-frequency analysis methods. The system can use feature vector concatenation, principal component analysis, feature selection models, or deep learning feature fusion models to fuse multi-dimensional signal features and combine them with initial location information to generate vibration event description information. This vibration event description information can simultaneously reflect the signal morphology, spectral structure, time-frequency evolution characteristics, and initial location of the vibration event.

[0048] Then, the resource status of the resource-constrained device is acquired. Based on this status, the identification logic used for anomaly event recognition is optimized for lightweight operation, resulting in lightweight device-side identification logic. Specifically, the resource status of the resource-constrained device can include computational resource status, storage resource status, and power consumption status. Computational resource status can include CPU utilization, processor frequency, memory usage, and task queue length; storage resource status can include available storage space, model storage usage, and cache capacity; power consumption status can include battery level, real-time power consumption, power supply mode, or temperature status. These resource statuses can be obtained through the device's built-in monitoring module, operating system interface, or edge management program. The system selects the corresponding lightweight optimization strategy based on the resource status. For example, when computational resources are scarce, the input feature size can be reduced by feature dimension compression, or smaller contributing branches in the identification logic can be removed by pruning the discriminant branches; when storage space is insufficient, model storage usage can be reduced by reducing the number of parameters, model quantization, parameter sharing, or knowledge distillation; when power consumption is limited, the continuous operation burden can be reduced by adjusting the running frequency, event-triggered operation, or low-power identification mode. After the above processing, lightweight device-side identification logic suitable for local operation on resource-constrained devices is formed.

[0049] Next, based on the vibration event description information, initial location information, and lightweight identification logic on the device side, the event identification task and the event location task are executed in parallel to obtain the abnormal event identification result and abnormal event location information. Specifically, multi-threading, multi-processing, asynchronous task scheduling, or edge task queues can be used to achieve the parallel execution of the event identification task and the event location task. The event identification task calls the lightweight identification logic on the device side to classify and judge the vibration event description information to determine whether the vibration event belongs to a known abnormal event pattern and the corresponding abnormal event type. For example, it can determine whether the vibration event is excavation, drilling, vehicle passage, knocking, animal activity, or environmental disturbance. The event location task combines the initial location information and the vibration event description information to further refine the location of the event. For example, the initial location information can be corrected according to the energy attenuation law, arrival time difference, phase change, or propagation model of the vibration signal at different locations, or a machine learning regression model can be used to predict more accurate location coordinates from the vibration event description information. By executing the event identification task and the event location task in parallel, the processing time from the occurrence of the vibration event to obtaining the abnormal event identification result and abnormal event location information can be shortened, meeting the needs of real-time monitoring and rapid alarm along the line.

[0050] Finally, the matching degree between the vibration event description information and known event patterns is obtained. Based on the matching degree, abnormal event location information, environmental information, and historical false alarm records, the reliability of the abnormal event identification results is assessed, a reliability assessment result is generated, and a tiered alarm is triggered based on the reliability assessment result. In specific implementation, the matching degree can be determined by calculating the similarity between the vibration event description information and the known event pattern template. For example, cosine similarity, Euclidean distance, dynamic time warping distance, or model output probability can be used to characterize the matching degree. The reliability assessment can use the matching degree as a basis and combine it with the accuracy of the abnormal event location information, environmental information, and historical false alarm records for correction. For example, in environments with strong winds, heavy rainfall, or high noise, certain environmental disturbances may produce vibration signals similar to abnormal events. In this case, the reliability of the abnormal event identification results can be reduced based on environmental information. Furthermore, if a certain section along the route experiences frequent historical false alarms under specific environmental conditions, the system can correct the credibility of similar identification results for that section based on historical false alarm records. The system can use Bayesian networks, fuzzy logic models, rule-weighted models, or machine learning evaluation models to comprehensively consider matching degree, abnormal event location information, environmental information, and historical false alarm records to output reliability assessment results.

[0051] Based on the reliability assessment results, the system triggers corresponding level-based alarms. For example, when the reliability assessment results indicate that the credibility of the abnormal event identification is high and the abnormal event location information is relatively accurate, a Level 1 alarm can be triggered, immediately notifying maintenance personnel, security personnel, or the monitoring center; when the reliability assessment results are at a medium level, a Level 2 alarm can be triggered, sending an alarm prompt and suggesting further verification; when the reliability assessment results are low, only logs can be recorded, a low-level alert can be triggered, or a continuous observation state can be entered. Through the above processing, this application can, in distributed fiber optic sensing scenarios, combine the fiber optic cable laying medium, resource-constrained equipment status, abnormal event identification results, abnormal event location information, environmental information, and historical false alarm records to more accurately, in real-time, and reliably identify and alarm for abnormal events along the line.

[0052] Optional, combined Figure 2 As shown, the steps for performing medium compensation and multi-dimensional feature extraction on audio processing data, and fusing signal features from multiple dimensions to generate vibration event description information include:

[0053] A1, Obtain the vibration transmission characteristics of the optical cable laying medium corresponding to the initial position information;

[0054] A2, Construct an inverse filter based on vibration transmission characteristics;

[0055] A3 uses an inverse filter to perform medium compensation on the audio processing data to obtain compensated vibration audio data.

[0056] A4, extract time-domain features, frequency-domain features, and time-frequency-domain features from the compensated vibration audio data;

[0057] A5 integrates time-domain features, frequency-domain features, time-frequency-domain features, and initial position information to form vibration event description information.

[0058] Among these, obtaining the vibration transmission characteristics of the optical cable laying medium corresponding to the initial location information refers to querying or measuring the impact of the laying medium (such as soil, pipeline, overhead, etc.) on the transmission of vibration signals based on the initial location of the vibration event or the corresponding section information along the line. Specifically, different laying media have different attenuation, frequency response, and phase shift characteristics for vibration signals, and accurately obtaining these characteristics is the basis for subsequent compensation.

[0059] Furthermore, constructing an inverse filter based on vibration transmission characteristics refers to designing a digital filter that can cancel or weaken these transmission effects based on the obtained vibration transmission characteristics of the laying medium. This inverse filter aims to restore the original form of the vibration signal distorted by the medium during transmission, making it closer to the true signal emitted by the vibration source.

[0060] Therefore, using an inverse filter to perform medium compensation on audio processing data to obtain compensated vibration audio data refers to inputting the original audio processing data into the constructed inverse filter for processing. Through medium compensation, the transmission attenuation, frequency response distortion, or phase shift caused by the fiber optic cable laying medium to the vibration signal can be effectively eliminated or reduced, thereby obtaining purer and more realistic vibration audio data, providing high-quality input for subsequent feature extraction.

[0061] Subsequently, extracting time-domain, frequency-domain, and time-frequency-domain features from the compensated vibration audio data refers to performing multi-dimensional analysis on the medium-compensated vibration audio data. Specifically, time-domain features can include signal amplitude, energy, zero-crossing rate, etc.; frequency-domain features can include the centroid, bandwidth, and dominant frequency of the spectrum, etc.; and time-frequency-domain features can be extracted using methods such as wavelet transform and short-time Fourier transform to extract the joint distribution characteristics of the signal in time and frequency, such as energy distribution and instantaneous frequency. These multi-dimensional features can comprehensively characterize the essential attributes of vibration events.

[0062] Finally, the fusion of time-domain features, frequency-domain features, time-frequency-domain features, and initial location information to form vibration event description information refers to integrating signal features extracted from different dimensions with the initial location information of the vibration event. This fusion can employ various methods such as feature vector concatenation and weighted fusion, aiming to generate a comprehensive description that includes the intrinsic characteristics of the vibration signal and spatial location information, providing comprehensive and accurate input for subsequent abnormal event identification.

[0063] Optionally, after optimizing the identification logic used for abnormal event identification based on the resource status of the resource-constrained device to obtain the lightweight identification logic on the device side, the following steps are also included:

[0064] Receive remote update packets, perform integrity verification and digital signature verification on the remote update packets, and obtain the update verification result;

[0065] When the updated verification result indicates that the verification failed, the current lightweight recognition logic on the device side is retained.

[0066] When the update verification result indicates that it has passed, the update timing is determined based on the resource status of the resource-constrained device, and the lightweight identification logic on the current device is backed up when the update timing arrives.

[0067] Extract the candidate device lightweight identification logic from the remote update package and write the candidate device lightweight identification logic into the backup storage area;

[0068] In the backup storage area, the lightweight identification logic of the candidate device is self-tested to obtain the updated self-test results;

[0069] When the self-test result indicates that the update passes, the current lightweight identification logic of the device is switched to the candidate lightweight identification logic of the device, and the candidate lightweight identification logic of the device is used as the updated lightweight identification logic of the device.

[0070] After switching, monitor the running status of the updated lightweight identification logic on the device side;

[0071] When the update self-test result indicates failure or abnormal running status, a rollback is performed based on the backed-up current device lightweight identification logic to obtain the restored device lightweight identification logic.

[0072] The updated or restored lightweight device-side recognition logic is used to perform event recognition tasks.

[0073] Specifically, a remote update package can be understood as a data packet containing the lightweight identification logic of the new version of the device. Its content typically includes new model parameters, algorithm code, or configuration information, and is distributed by a remote management platform. Integrity verification refers to performing data integrity checks on the received remote update package. This is done, for example, by calculating a checksum (such as CRC, MD5, SHA-256) and comparing it with the expected checksum value to ensure that the update package has not been tampered with or corrupted during transmission. Digital signature verification uses public-key cryptography to verify the trustworthiness of the remote update package's source and confirm that its content has not been modified by an unauthorized third party. This typically involves signing the update package using the sender's private key, and the receiver using the corresponding public key for verification.

[0074] In practical applications, the update timing refers to the optimal time to perform the update operation. Its determination can comprehensively consider factors such as the current load of resource-constrained devices, network conditions, system idle time, and whether any critical tasks are running, to minimize the impact on normal business operations. Backing up the current lightweight identification logic on the device side means saving a complete copy of the currently running identification logic on the device to a secure storage location before updating, so that it can be restored to a previous stable state in case of update failure. The backup storage area refers to the storage space reserved within the device for temporarily storing and testing candidate lightweight identification logic on the device side, isolated from the storage area occupied by the currently running logic. Self-testing refers to performing a series of preset tests on the candidate lightweight identification logic on the device side in the backup storage area to verify its basic functions, compatibility, performance, and whether there are any potential errors or abnormal behaviors. This can include simulated data testing, interface testing, etc. Operational status monitoring refers to continuously collecting and analyzing key indicators during its operation after switching to the updated lightweight identification logic on the device side, such as CPU utilization, memory usage, identification accuracy, response time, and error logs, to determine whether it is working properly. Rollback refers to the automatic or manual restoration of the lightweight identification logic on the device to the stable version backed up before the update when the self-check fails during the update process or the logic after the update runs abnormally, in order to ensure the continuity and reliability of the system.

[0075] Optionally, the steps for performing an event recognition task and obtaining abnormal event recognition results based on vibration event description information, initial position information, and lightweight recognition logic on the device side include:

[0076] Read vibration event description information, initial position information, and lightweight identification logic from the device side;

[0077] The corresponding sections along the line are determined based on the initial location information, and the known event pattern library and historical false alarm baselines for the sections along the line are obtained.

[0078] Based on the lightweight identification logic on the device side, the degree of matching between the vibration event description information and the known event patterns in the known event pattern library is calculated;

[0079] By combining the matching degree and the historical false alarm baseline, the initial event identification results are obtained;

[0080] Based on the degree of matching and the initial event identification results, determine whether there are new abnormal events or event pattern evolutions;

[0081] When there are no new abnormal events and no event pattern evolution, the initial event identification result is taken as the abnormal event identification result;

[0082] When new abnormal events or event patterns evolve, the location of the vibration event description information in the feature space is determined, and the adjustment amount of the discrimination parameters is generated based on the distance between the location and the boundary of the known event pattern.

[0083] Based on the preset stability constraints, the adjustment amount of the discrimination parameter is restricted to obtain the restricted adjustment amount of the discrimination parameter;

[0084] Based on the limited discrimination parameter adjustment amount, the discrimination parameters of the lightweight identification logic on the device side are adjusted to obtain the candidate lightweight identification logic on the device side.

[0085] Based on the lightweight identification logic of the candidate device, the vibration event description information is used to identify the candidate event identification results.

[0086] Based on the changes in the confidence level of the candidate event identification results, the degree of change in the event type, and the changes in historical false alarms, determine whether the identification status of the lightweight identification logic on the candidate device side is abnormal.

[0087] When the identification status is not abnormal, the candidate event identification result is taken as the abnormal event identification result, and the candidate device-side lightweight identification logic is taken as the updated device-side lightweight identification logic.

[0088] When an abnormal status is identified, the system traces back to the lightweight identification logic on the device side before the judgment parameters were adjusted, and uses the initial event identification result as the abnormal event identification result.

[0089] Specifically, when performing an event recognition task, the first step is to read the vibration event description information, initial location information, and lightweight recognition logic on the device side. These are the basic inputs for event recognition. The vibration event description information is obtained by fusing the data after medium compensation and multi-dimensional feature extraction, and it contains the key features of the vibration event. The initial location information provides the geographical or regional context of the event. The lightweight recognition logic on the device side is the core algorithm model used to efficiently perform recognition tasks on resource-constrained devices.

[0090] Furthermore, based on the initial location information, the specific section along the line where the vibration event occurred can be determined. Each section may have its unique environmental characteristics and historical event patterns, therefore it is necessary to obtain the known event pattern library and historical false alarm baseline corresponding to that section. The known event pattern library contains the characteristic patterns of various known abnormal events within the section, while the historical false alarm baseline provides statistical information on past false alarm events within the section, which helps in correction during the identification process.

[0091] Based on this, lightweight recognition logic on the device side is used to calculate the degree of matching between the vibration event description information and each known event pattern in the known event pattern library. This degree of matching can be a similarity score, probability value, or distance metric, reflecting the similarity between the current vibration event and known patterns. Combining this degree of matching with the historical false alarm baseline, an initial event recognition result can be obtained. The introduction of the historical false alarm baseline helps to filter out some known high false alarm rate patterns in the early stages of recognition, improving the accuracy of the initial recognition.

[0092] Optionally, based on the resource status of the resource-constrained device, the steps for lightweighting and optimizing the identification logic used for abnormal event identification to obtain the lightweight identification logic on the device side include:

[0093] Based on the resource status of the resource-constrained device, obtain the real-time resource status information of the resource-constrained device, which includes computing resource status, storage resource status and power consumption status.

[0094] Based on real-time resource status information, assess the current optimization level of the identification logic used for anomaly event identification;

[0095] Based on the current level of optimization, determine the optimization strategy for the recognition logic. The optimization strategy includes feature dimension compression, discrimination branch pruning, parameter reduction, or operation frequency adjustment.

[0096] When determining the optimization strategy, a priority for identifying key events is introduced;

[0097] The recognition logic is optimized for lightweighting according to the optimization strategy, resulting in lightweight recognition logic on the device side.

[0098] Specifically, real-time resource status information refers to instantaneous data on the computing power, storage capacity, and energy consumption of resource-constrained devices at a specific point in time. Computing resource status can be reflected in CPU utilization, processing speed, or the number of available processing cores; storage resource status can be reflected in memory utilization, available disk space, or cache hit rate; and power consumption status can be reflected in battery level, wattage, or power input voltage. This information can be periodically collected through built-in sensors, operating system application programming interfaces (APIs), or dedicated monitoring modules, with the aim of providing real-time environmental data for optimizing the identification logic.

[0099] The current optimization level refers to the efficiency and resource consumption of the lightweight recognition logic on resource-constrained devices. The evaluation can be based on the comparison between preset performance indicators and actual operating data. For example, the recognition accuracy, latency, and resource consumption can be compared with preset thresholds. Alternatively, machine learning models can be used to predict the complexity and resource requirements of the recognition logic. The purpose is to quantify the resource adaptability of the current recognition logic.

[0100] In practical applications, optimization strategies include feature dimension compression, discriminant branch pruning, parameter reduction, and runtime adjustment. Specifically, feature dimension compression refers to reducing the number of features in the vibration event description information through principal component analysis (PCA), linear discriminant analysis (LDA), or other feature selection or extraction algorithms, thereby reducing the input complexity of the recognition logic. Discriminant branch pruning refers to removing decision paths or network layers that contribute little to the recognition result or are unnecessary under the current resource conditions in the recognition logic (e.g., decision trees or neural networks) to simplify the model structure. Parameter reduction refers to reducing the number of trainable parameters in the recognition logic model through model quantization (e.g., converting floating-point parameters to fixed-point parameters), parameter sharing, or knowledge distillation, thereby reducing storage and computational requirements. Runtime adjustment refers to dynamically adjusting the execution frequency or sampling rate of the recognition logic based on real-time resource conditions; for example, reducing the recognition frequency when resources are scarce and increasing it when resources are abundant. The aim is to provide multiple feasible optimization methods to adapt to different resource constraints and performance requirements.

[0101] Furthermore, when determining the optimization strategy, a critical event identification priority is introduced. Critical event identification priority refers to assigning different importance levels to different types of abnormal events. For example, intrusion and sabotage events may have high priority, while minor disturbance events may have low priority. When resources are limited, the optimization strategy will prioritize ensuring the accuracy and timeliness of high-priority event identification, even if this means greater simplification or delay in identifying low-priority events. The aim is to ensure that critical information is prioritized and protected under limited resources.

[0102] Optionally, the steps of performing a reliability assessment on the abnormal event identification results based on the matching degree, abnormal event location information, environmental information, and historical false alarm records, generating a reliability assessment result, and triggering a graded alarm based on the reliability assessment result include:

[0103] Read the matching degree corresponding to the abnormal event identification results;

[0104] Based on the degree of matching, determine the basic reliability assessment results of the abnormal event identification results;

[0105] Temperature and humidity signals are obtained from environmental information, and vibration signals are read.

[0106] Preprocessing was performed on the vibration signal, temperature signal, and humidity signal respectively to obtain preprocessed modal data;

[0107] Based on the modal data, extract the corresponding modal signal features;

[0108] Based on the degree of influence of each modal signal characteristic on reliability assessment, modal fusion weights are assigned;

[0109] Based on the modal fusion weights, the features of each modal signal are fused to generate environmental enhancement description information;

[0110] Based on the environmental reinforcement description information, abnormal event location information, and historical false alarm records, the basic reliability assessment results are corrected to obtain the reliability assessment results, and graded alarms are triggered based on the reliability assessment results.

[0111] Specifically, the matching degree corresponding to the abnormal event identification results refers to the similarity or confidence score between the vibration event description information output by the event identification task and the known event pattern. This matching degree is an important basis for initially judging the nature of the event. The basic reliability assessment result of the abnormal event identification results determined based on the matching degree can be understood as quantifying the initial reliability of the event based on a preset threshold or mapping function of the matching degree. For example, the higher the matching degree, the higher the basic reliability score. In practical applications, obtaining temperature and humidity signals from environmental information and reading vibration signals refers to acquiring real-time temperature and humidity data through environmental sensors deployed along the optical fiber and obtaining raw vibration signals from the optical fiber vibration monitoring system. These environmental parameters and raw vibration signals are the basis for subsequent modal analysis.

[0112] Furthermore, preprocessing is performed on the vibration, temperature, and humidity signals respectively to obtain preprocessed modal data. Preprocessing may include operations such as denoising, filtering, normalization, and sampling rate adjustment to eliminate interference in the signals and unify the data format. Modal data refers to the data representation that, after preprocessing, can reflect the characteristics of different physical modes (such as vibration modes, temperature modes, and humidity modes). Specifically, extracting corresponding modal signal features based on each modal data refers to extracting time-domain, frequency-domain, or time-frequency-domain features from the preprocessed vibration modal data, features such as temperature change rate and average temperature from the temperature modal data, and features such as humidity change trend and average humidity from the humidity modal data. These features can more precisely describe the signal characteristics under different modes.

[0113] As a preferred implementation, modal fusion weights are assigned based on the degree of influence of each modal signal characteristic on the reliability assessment. This means that different environmental factors and vibration signal characteristics may contribute differently to the final reliability assessment. For example, in some scenarios, temperature changes may have a more significant impact on fiber optic vibration signals than humidity, and therefore can be assigned a higher weight. These weights can be dynamically determined based on historical data analysis, expert experience, or machine learning models. Thus, the modal signal characteristics are fused according to the modal fusion weights to generate environmental reinforcement description information. This environmental reinforcement description information supplements and enhances the original vibration event description information in the environmental dimension; it integrates the combined influence of multiple environmental factors on the vibration signal, providing a more comprehensive description of the event background.

[0114] Finally, based on the environmental reinforcement description information, abnormal event location information, and historical false alarm records, the basic reliability assessment results are revised to obtain the final reliability assessment result, which triggers tiered alarms. The revision process can be a multi-factor weighted summation, fuzzy logic reasoning, or machine learning model. It incorporates environmental reinforcement description information, precise event location information, and historical false alarm patterns to fine-tune the initial basic reliability assessment results, thereby obtaining a more accurate and reliable final assessment result, and triggering alarms of the corresponding level based on this result.

[0115] Optionally, based on vibration event description information, initial location information, and lightweight identification logic on the device side, the steps of executing event identification and event localization tasks in parallel to obtain abnormal event identification results and abnormal event localization information include:

[0116] Based on the vibration event description information and the lightweight recognition logic on the device, an event recognition task is generated.

[0117] Based on the vibration event description and initial location information, an event localization task is generated;

[0118] Deploy a resource isolation and scheduling module, and allocate computing resource quotas and memory space to the event identification task and the event location task respectively through the resource isolation and scheduling module;

[0119] Obtain task priority and historical resource usage;

[0120] Adjust the computing resource quotas and memory space corresponding to the event identification task and the event location task based on task priority and historical resource usage.

[0121] With the adjusted computing resource quota and memory space, the event identification task and the event location task are executed in parallel.

[0122] The event recognition task outputs the abnormal event recognition results, and the event location task outputs the abnormal event location information.

[0123] The event recognition task involves constructing a computational task specifically for determining the event type based on existing vibration event description information and lightweight, optimized device-side recognition logic. The event localization task involves constructing a computational task specifically for accurately determining the exact location of the event based on the vibration event description information and initial location information.

[0124] Furthermore, the resource isolation and scheduling module can be understood as a software or hardware component designed for fine-grained management and allocation of computing resources (such as CPU cores and GPU computing power) and memory space on a device. This module ensures that different tasks have independent resource environments during execution, avoiding mutual interference. In practical applications, computing resource quotas refer to the share of processor time or computing power reserved for each task, while memory space refers to the size of the storage area allocated to a task.

[0125] In addition, acquiring task priorities and historical resource usage information means that the system dynamically collects the importance level of currently pending tasks and the resource consumption patterns of these or similar tasks in the past. For example, task priorities can be set based on the urgency of events or preset business rules, while historical resource usage information can be obtained by monitoring task execution data over a long period of time.

[0126] Therefore, by adjusting the computing resource quotas and memory space corresponding to the aforementioned event identification and event location tasks based on task priorities and historical resource usage, the aim is to achieve dynamic optimization of resource allocation. For example, when an event identification task is assigned a higher priority, its computing resource quota and memory space may be appropriately increased to ensure its rapid response; simultaneously, if historical data shows that a task has low resource requirements within a specific time period, its quota may be temporarily reduced to free up resources for other high-priority tasks.

[0127] Ultimately, with adjusted computing resource quotas and memory space, the aforementioned event identification and event location tasks were executed in parallel, ensuring that the tasks ran simultaneously under optimal resource configuration. The event identification task outputs the abnormal event identification results, namely the event type and confidence level; the event location task outputs the abnormal event location information, namely the precise geographic coordinates or fiber optic segment where the event occurred.

[0128] Optionally, the steps of performing a reliability assessment on the abnormal event identification results based on the matching degree, abnormal event location information, environmental information, and historical false alarm records, generating a reliability assessment result, and triggering a graded alarm based on the reliability assessment result include:

[0129] Read the abnormal event identification results, matching degree, abnormal event location information, environmental information, and historical false alarm records;

[0130] Based on the degree of matching, a basic reliability score is determined for the abnormal event identification results;

[0131] Based on the location information of abnormal events within a continuous time period, the spatial distribution relationship of abnormal events is identified, and based on the description information of vibration events within a continuous time period, the temporal evolution relationship of abnormal events is identified.

[0132] Based on spatial distribution and temporal evolution, construct causal chains between anomalous events;

[0133] Based on the causal chain, deduce the source path of the abnormal event;

[0134] Based on environmental information, identify environmental disturbance factors within the area corresponding to the source tracing path;

[0135] Based on historical false alarm records, determine whether there are false alarm records of the same event type in the area corresponding to the tracing path;

[0136] Based on the source tracing path, environmental interference factors, and false alarm records, the basic reliability score is corrected to obtain the reliability assessment result, and a graded alarm is triggered based on the reliability assessment result.

[0137] Specifically, after obtaining the abnormal event identification results, matching degree, abnormal event location information, environmental information, and historical false alarm records, the basic reliability score of the abnormal event identification result is initially determined based on the degree of matching between the vibration event description information and known event patterns. This basic reliability score can be understood as the initial confidence level of the event identification model in the current event.

[0138] Furthermore, to more comprehensively assess the reliability of events, the system analyzes the location information of abnormal events over a continuous time period to identify the spatial distribution patterns of these events, such as whether they are concentrated in a specific area, or whether they exhibit a linear or clustered distribution. Simultaneously, it also analyzes the descriptive information of vibration events over a continuous time period to identify the temporal evolution trends of the events, such as the frequency of occurrence, intensity changes, and duration of the events.

[0139] Spatial distribution relationships refer to the patterns of clustering, diffusion, or migration of anomalous events in geographical locations, which can be identified through methods such as cluster analysis and density analysis. Temporal evolution relationships refer to the regularity, periodicity, or suddenness of anomalous events over time series, which can be identified through methods such as time series analysis and trend analysis.

[0140] Based on the identified spatial distribution and temporal evolution relationships, the system attempts to construct causal chains between anomalous events. This means that by analyzing the spatiotemporal correlation of events, it infers which events might be causes and which are consequences, thus forming an event chain. For example, a series of vibration events moving from far to near might indicate a moving source of disturbance.

[0141] Once the causal chain is established, the path to the source of the abnormal event can be deduced, that is, tracing the source or propagation trajectory of the event. This helps to understand the root cause and development process of the event.

[0142] Based on this, the system will combine environmental information to identify environmental disturbance factors within the area corresponding to the tracing path. For example, if the tracing path passes through a construction area, construction activities may be a significant environmental disturbance factor. Environmental disturbance factors may include, but are not limited to, temperature, humidity, wind force, rainfall, soil type, and surrounding construction activities.

[0143] Simultaneously, the system will review historical false alarm records to determine if there are similar false alarm records of the same event type within the area corresponding to the tracing path. If a certain type of false alarm has frequently occurred in that area historically, the reliability of the current event may need to be re-examined.

[0144] Ultimately, the system will comprehensively consider the source tracing path, environmental interference factors, and historical false alarm records to revise the previously determined basic reliability score. This revision can be achieved through weighted averaging, Bayesian updates, or other fusion algorithms, aiming to incorporate this deep-seated contextual information into the reliability assessment, thereby obtaining a more accurate and comprehensive reliability evaluation result. Based on the revised reliability assessment result, the system will trigger corresponding tiered alarms to ensure the accuracy and effectiveness of the alarms.

[0145] Optionally, the steps for triggering graded alarms based on reliability assessment results include:

[0146] Receive event information from edge devices in different regions. The event information includes abnormal event identification results, abnormal event location information, and corresponding reliability assessment results.

[0147] Specifically, event information refers to data packets generated by each edge device after identifying and locating abnormal events, combined with its own reliability assessment results. These edge devices are typically deployed in different geographical areas along the fiber optic cable, responsible for the preliminary processing and analysis of fiber optic vibration signals in their respective regions. Receiving this event information aims to aggregate event reports scattered across different regions, providing a data foundation for subsequent global analysis.

[0148] The event information is calibrated with timestamps and transformed with spatial coordinates to unify the reference benchmark for events in different regions, thus obtaining preprocessed event information;

[0149] Timestamp calibration refers to synchronizing the timestamps in the received event information to eliminate time inconsistencies caused by clock deviations between different devices or network transmission delays, ensuring that all events are analyzed on a unified timeline. Spatial coordinate transformation involves converting the event location information reported by different edge devices from their respective local coordinate systems to a unified global coordinate system, such as a geographic coordinate system or a unified linear distance coordinate system, to accurately identify the geographical location and spatial relationships of events. These preprocessing steps eliminate data heterogeneity, laying the foundation for subsequent correlation analysis.

[0150] Obtain the computing and storage resources of the currently resource-constrained device;

[0151] In practical applications, a resource-constrained device refers to a central processing unit or aggregation node responsible for receiving, processing, and analyzing event information from multiple edge devices. Acquiring its computing resources, such as CPU utilization and processing power, as well as storage resources, such as memory usage and available storage space, is crucial for understanding the device's load and processing capacity in real time. This allows for dynamic adjustments to subsequent event processing strategies, preventing resource overload.

[0152] Based on the computing and storage resources of the current resource-constrained equipment, adjust and determine the event information filtering threshold, compression ratio, and number of linkage chain levels of the preprocessed event information;

[0153] Furthermore, the event information filtering threshold refers to the criteria used to filter out unimportant or low-reliability events. For example, events with reliability assessment results below a certain preset value may be temporarily ignored or downgraded. The compression ratio refers to the degree to which event information is compressed to reduce storage and transmission overhead. The number of linkage chain levels refers to the depth of association that can be traced or expanded when building event relationships. For example, it can be set to consider only directly related events, or further consider indirectly related events. These parameters are dynamically adjusted to maximize processing efficiency and alarm effectiveness under limited resource conditions.

[0154] Based on the event information filtering threshold and compression ratio, the preprocessed event information is filtered and compressed to obtain the processed event information;

[0155] Specifically, the filtering process uses preset thresholds to eliminate event information that does not meet importance or reliability standards; for example, events with reliability scores below a certain value are filtered out. The compression process then compresses the remaining event information according to a determined compression ratio. This can be achieved by removing redundant fields, using a more compact data format, or performing lossless / lossy compression to reduce the data volume. This effectively reduces the computational and storage burden of subsequent processing.

[0156] Based on preset linkage rules, identify potential correlations between events in different regions from the processed event information;

[0157] Among these, pre-defined linkage rules can be understood as a series of logical conditions or patterns that define the correlation between events. For example, within a specific time window, multiple small-scale events in geographically close proximity may indicate the occurrence of a large-scale event; or, the continuous occurrence of a certain type of event in different regions may foreshadow a systemic problem. Identifying potential correlations aims to discover the deep connections that may exist behind seemingly independent events.

[0158] Based on the number of potential connections and linkage levels, construct linkage chains between events in different regions;

[0159] In practical applications, a chain of events refers to a sequence or network structure that connects multiple events through potential associations. The number of levels in a chain determines the depth and breadth of the chain. For example, if there is only one level, only direct associations are constructed; if there are two levels, other events related to the directly associated events are further considered. The purpose of constructing a chain of events is to visualize and analyze the mutual influence and evolution paths between events.

[0160] Based on the linkage chain between events in different regions, determine whether there are abnormal events involving multiple regions;

[0161] Therefore, by analyzing the constructed linkage chains, it is possible to determine whether there are anomalous events spanning multiple geographical regions. For example, if a linkage chain connects events from three different regions, it can be determined that there is an anomalous event involving multiple regions. This judgment is crucial for identifying large-scale, complex, or spreading anomalous events.

[0162] When there are abnormal events involving multiple regions, a linkage hierarchical alarm command is generated based on the abnormal events involving multiple regions and the reliability assessment results.

[0163] When there are no abnormal events involving multiple regions, a local hierarchical alarm instruction is generated based on the reliability assessment result corresponding to an abnormal event involving a single region.

[0164] Among them, the coordinated hierarchical alarm command refers to the alarm generated for abnormal events involving multiple regions. It typically has higher priority and a wider notification scope, and may require coordination among multiple departments or regions to respond. The local hierarchical alarm command, on the other hand, is an alarm generated for abnormal events involving only a single region. Its priority and notification scope may be relatively lower, and it is usually handled by local management personnel. Generating different types of alarm commands aims to provide differentiated alarm responses based on the nature and scope of the event.

[0165] Triage alarms are triggered based on linkage or local triage alarm commands.

[0166] Ultimately, based on the generated linked or local tiered alarm commands, the system will trigger the corresponding tiered alarms. Tiered alarms refer to the alarm level, such as urgent, important, or general, as well as the notification method, such as SMS, email, audible and visual alarms, or system pop-ups. These distinctions are made according to the severity and scope of the event, ensuring that critical information is delivered to relevant personnel in a timely and accurate manner.

[0167] This application also discloses a pattern recognition system for abnormal events based on optical fiber vibration signals, used to perform pattern recognition of abnormal events based on optical fiber vibration signals. The pattern recognition system 1 for abnormal events based on optical fiber vibration signals includes:

[0168] The initial position determination module 11 is used to acquire vibration signals, environmental information and historical false alarm records of the optical fiber along the line, and determine the initial position information of the vibration occurrence based on the vibration signal; the initial position information indicates the preliminary occurrence position of the vibration event on the optical fiber along the line or the corresponding section information along the line.

[0169] The correspondence establishment module 12 is used to convert the vibration signal into audio processing data and establish the correspondence between the audio processing data and the initial position information;

[0170] The vibration event description module 13 is used to determine the optical cable laying medium of the corresponding section along the line based on the correspondence relationship, and to perform medium compensation and multi-dimensional feature extraction on the audio processing data according to the optical cable laying medium, and to fuse signal features of multiple dimensions to generate vibration event description information; medium compensation means to perform transmission attenuation, frequency response or phase offset correction on the audio processing data according to the vibration transmission characteristics of the optical cable laying medium.

[0171] The identification logic generation module 14 is used to obtain the resource status of the resource-constrained device and optimize the identification logic used for abnormal event identification based on the resource status of the resource-constrained device to obtain the device-side lightweight identification logic. The device-side lightweight identification logic represents the lightweight identification logic suitable for local operation of the resource-constrained device. It is the identification logic formed after feature compression, branch pruning, parameter reduction or operation frequency adjustment of the identification logic used for abnormal event identification.

[0172] The parallel task execution module 15 is used to execute the event identification task and the event localization task in parallel based on the vibration event description information, initial position information and lightweight identification logic of the device, so as to obtain the abnormal event identification result and the abnormal event localization information.

[0173] The graded alarm triggering module 16 is used to obtain the degree of matching between the vibration event description information and the known event pattern. Based on the degree of matching, abnormal event location information, environmental information and historical false alarm records, it performs a reliability assessment on the abnormal event identification results, generates a reliability assessment result, and triggers graded alarms based on the reliability assessment result.

[0174] The system proposed in this application integrates functions such as fiber optic vibration signal acquisition, processing, feature extraction, identification logic optimization, parallel task execution, reliability assessment, and alarm triggering through a modular design. This system aims to address the shortcomings of traditional distributed acoustic sensing technology, including insufficient feature extraction dimensions, poor real-time performance, low positioning accuracy, and high false alarm rates. Through the collaborative work of its various functional modules, the system can achieve accurate and real-time identification and location of abnormal events along the route, and issue tiered alarms based on the reliability of the events, thereby improving the security and operational efficiency of infrastructure.

[0175] The above embodiments have already described how the initial position determination module acquires vibration signals, environmental information, and historical false alarm records along the optical fiber, and determines the initial position information of the vibration occurrence based on the vibration signals; these details will not be repeated here. It is important to emphasize that the initial position determination module can be configured to achieve its function in various ways. For example, the vibration signal can be acquired by a dedicated optical fiber sensing unit, which converts the raw optical signal into an electrical signal and performs preliminary digitization. Environmental information can be provided by a separate sensor array and transmitted to the module via wired or wireless means. Historical false alarm records can be read from local memory or a remote database. The determination of the initial position information, besides being based on signal arrival time difference or intensity attenuation, can also be achieved through simple threshold detection. That is, when the vibration signal intensity exceeds a preset threshold, the center position of the current optical fiber segment is taken as the initial position information, the purpose of which is to provide a preliminary indication of the event occurrence area.

[0176] The above embodiments have already described how the correspondence establishment module converts vibration signals into audio processing data and establishes a correspondence between the audio processing data and the initial position information, which will not be repeated here. It is important to emphasize that the correspondence establishment module can be designed to directly perform simple sampling and quantization of the original vibration signal to form basic digital audio data. This module can establish a correspondence between each audio data block and its corresponding initial position information by simply associating them with a label, for example, by adding a position identifier to the header of the data packet. The purpose is to ensure that subsequent processing can trace back to the initial occurrence area of ​​the event.

[0177] The above embodiments have already described how the vibration event description module determines the optical cable laying medium of the corresponding section along the route based on the correspondence relationship, performs medium compensation and multi-dimensional feature extraction on the audio processing data according to the optical cable laying medium, and fuses signal features from multiple dimensions to generate vibration event description information, which will not be repeated here. It is important to emphasize that the vibration event description module can be configured to perform a simple medium type search based on the initial location information using a preset list of medium types. Medium compensation can use a fixed-parameter equalization filter to correct the audio processing data according to an empirically set medium attenuation model. Multi-dimensional feature extraction can extract only a few basic time-domain features, such as the average amplitude and energy of the signal, and simple frequency-domain features, such as the highest energy frequency point obtained through Fast Fourier Transform. These features can be fused through a simple splicing method to form preliminary vibration event description information, the purpose of which is to provide a basic overview of the event's characteristics.

[0178] The above embodiments have already described how the recognition logic generation module obtains the resource status of the resource-constrained device, and based on the resource status, performs lightweight optimization on the recognition logic used for abnormal event identification to obtain the content of the lightweight recognition logic on the device side, which will not be repeated here. It should be emphasized that the recognition logic generation module can be designed to obtain real-time CPU and memory usage as resource status by periodically reading the resource monitoring interface of the device's operating system. Lightweight optimization can adopt a preset simplified rule set. For example, when the CPU utilization exceeds a certain threshold, a pre-trained recognition model version with fewer parameters can be automatically selected, or the running frequency of the recognition logic can be directly reduced. The purpose is to ensure that the recognition function can run stably on resource-constrained devices.

[0179] The above embodiments have already described how the parallel execution module, based on vibration event description information, initial position information, and lightweight identification logic on the device side, executes event identification and event localization tasks in parallel to obtain abnormal event identification results and abnormal event localization information, which will not be repeated here. It is important to emphasize that the parallel execution module can be configured to allocate the event identification and event localization tasks to different processor cores or threads via a simple task scheduler, allowing them to start executing simultaneously. The event identification task can be a rule-based classifier that matches key features in the vibration event description information. The event localization task can be a simple calculator based on initial position information and a signal strength attenuation model, aiming to simultaneously obtain the event type and approximate location to improve response speed.

[0180] The above embodiments have already described how the graded alarm triggering module obtains the matching degree between the vibration event description information and the known event pattern. Based on the matching degree, abnormal event location information, environmental information, and historical false alarm records, it performs a reliability assessment on the abnormal event identification results, generates a reliability assessment result, and triggers graded alarms based on the reliability assessment result. This will not be elaborated further here. It is important to emphasize that the graded alarm triggering module can be designed to determine the matching degree through a simple threshold comparison. The reliability assessment can be a linear model based on a weighted sum of the matching degree, location information accuracy, environmental interference level, and historical false alarm frequency using preset weights. Based on the final reliability assessment result, the module can trigger a level 1 alarm based on preset alarm level thresholds. For example, a reliability score higher than 80% triggers a level 2 alarm, 60%-80% triggers a level 2 alarm, and below 60% only logs the results. The purpose is to provide differentiated responses based on the urgency and credibility of the event.

[0181] The abnormal event pattern recognition system based on fiber optic vibration signals proposed in this application effectively solves many problems existing in the prior art through its modular and collaborative design. Traditional systems have limited dimensions in feature extraction and rely on single analysis, which easily leads to missed detections. This system performs medium compensation and multi-dimensional feature extraction through a vibration event description module, and integrates signal features from multiple dimensions to generate more comprehensive and robust vibration event description information, significantly improving the system's ability to identify various abnormal events.

[0182] The above are merely embodiments of this application and are not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for identifying abnormal event patterns based on fiber optic vibration signals, characterized in that, include: The vibration signal, environmental information, and historical false alarm records of the optical fiber along the line are acquired, and the initial location information of the vibration is determined based on the vibration signal. The initial location information indicates the preliminary location of the vibration event along the optical fiber or the corresponding section information along the fiber. The vibration signal is converted into audio processing data, and a correspondence is established between the audio processing data and the initial position information; Based on the correspondence, the optical cable laying medium of the corresponding section along the line is determined, and the audio processing data is subjected to medium compensation and multi-dimensional feature extraction according to the optical cable laying medium. The signal features of multiple dimensions are fused to generate vibration event description information. The medium compensation means that the audio processing data is processed by transmission attenuation, frequency response or phase offset correction according to the vibration transmission characteristics of the optical cable laying medium. The resource status of the resource-constrained device is obtained. Based on the resource status of the resource-constrained device, the identification logic used for abnormal event identification is optimized to obtain the device-side lightweight identification logic. The device-side lightweight identification logic represents a lightweight identification logic suitable for local operation on the resource-constrained device. It is the identification logic formed after feature compression, branch pruning, parameter reduction, or operation frequency adjustment of the identification logic used for abnormal event identification. Based on the vibration event description information, the initial position information, and the lightweight identification logic of the device, the event identification task and the event localization task are executed in parallel to obtain the abnormal event identification result and the abnormal event localization information. The degree of matching between the vibration event description information and the known event pattern is obtained. Based on the degree of matching, the abnormal event location information, the environmental information and the historical false alarm records, the reliability of the abnormal event identification result is evaluated, a reliability evaluation result is generated, and a graded alarm is triggered based on the reliability evaluation result.

2. The method for abnormal event pattern recognition based on fiber optic vibration signals according to claim 1, characterized in that, The steps of performing medium compensation and multi-dimensional feature extraction on the audio processing data, and fusing signal features from multiple dimensions to generate vibration event description information include: Obtain the vibration transmission characteristics of the optical cable laying medium corresponding to the initial position information; Construct an inverse filter based on the vibration transmission characteristics; The inverse filter is used to perform medium compensation on the audio processing data to obtain compensated vibration audio data. Extract time-domain features, frequency-domain features, and time-frequency-domain features from the compensated vibration audio data; The vibration event description information is formed by fusing the time-domain features, the frequency-domain features, the time-frequency-domain features, and the initial position information.

3. The abnormal event pattern recognition method based on fiber optic vibration signals according to claim 1, characterized in that, After optimizing the identification logic for abnormal event identification based on the resource status of the resource-constrained device to obtain the lightweight identification logic on the device side, the method further includes: Receive the remote update packet, and perform integrity verification and digital signature verification on the remote update packet to obtain the update verification result; When the updated verification result indicates that it has failed, the current lightweight identification logic on the device side is retained; When the update verification result indicates that it passes, the update timing is determined according to the resource status of the resource-constrained device, and the lightweight identification logic of the current device is backed up when the update timing arrives. Extract the candidate device lightweight identification logic from the remote update package and write the candidate device lightweight identification logic into the backup storage area; The lightweight identification logic of the candidate device is self-tested in the backup storage area to obtain an updated self-test result; When the update self-test result indicates that it passes, the current device lightweight identification logic is switched to the candidate device lightweight identification logic, and the candidate device lightweight identification logic is used as the updated device lightweight identification logic. After switching, monitor the running status of the updated lightweight identification logic on the device side; When the update self-test result indicates failure, or the running status is abnormal, a rollback is performed based on the backed-up current device lightweight identification logic to obtain the restored device lightweight identification logic. The updated or restored lightweight device-side recognition logic is used to perform event recognition tasks.

4. The abnormal event pattern recognition method based on fiber optic vibration signals according to claim 1, characterized in that, The steps for performing an event identification task and obtaining an abnormal event identification result based on the vibration event description information, the initial position information, and the lightweight identification logic of the device include: Read the vibration event description information, the initial position information, and the lightweight identification logic of the device; Based on the initial location information, the corresponding section along the line is determined, and the known event pattern library and historical false alarm baseline corresponding to the section along the line are obtained; Based on the lightweight identification logic of the device, the degree of matching between the vibration event description information and the known event patterns in the known event pattern library is calculated; By combining the matching degree and the historical false alarm baseline, the initial event identification result is obtained; Based on the matching degree and the initial event identification results, it is determined whether there are new abnormal events or event pattern evolutions. When no new abnormal events occur and no event pattern evolves, the initial event identification result is taken as the abnormal event identification result. When a new abnormal event or event pattern evolution occurs, the position of the vibration event description information in the feature space is determined, and the discrimination parameter adjustment amount is generated based on the distance between the position and the boundary of the known event pattern. Based on preset stability constraints, the adjustment amount of the discrimination parameter is restricted to obtain the restricted adjustment amount of the discrimination parameter; Based on the restricted discrimination parameter adjustment amount, the discrimination parameters of the device-side lightweight identification logic are adjusted to obtain the candidate device-side lightweight identification logic; Based on the lightweight identification logic of the candidate device, the vibration event description information is used to identify the candidate event identification result. Based on the changes in the confidence level of the candidate event identification results, the degree of change in the event type, and the changes in historical false alarms, it is determined whether the identification status of the lightweight identification logic on the candidate device side is abnormal. When the identification status is not abnormal, the candidate event identification result is taken as the abnormal event identification result, and the candidate device-side lightweight identification logic is taken as the updated device-side lightweight identification logic. When the identification state is abnormal, the process is to backtrack to the lightweight identification logic on the device side before the judgment parameter adjustment, and use the initial event identification result as the abnormal event identification result.

5. The abnormal event pattern recognition method based on fiber optic vibration signals according to claim 1, characterized in that, The step of optimizing the identification logic used for abnormal event identification based on the resource status of the resource-constrained device to obtain the lightweight identification logic on the device side includes: Based on the resource status of the resource-constrained device, obtain the real-time resource status information of the resource-constrained device, which includes computing resource status, storage resource status and power consumption status. Based on the real-time resource status information, assess the current optimization level of the identification logic used for anomaly event identification; Based on the current level of optimization, an optimization strategy for the recognition logic is determined. The optimization strategy includes feature dimension compression, discrimination branch pruning, parameter reduction, or operation frequency adjustment. When determining the optimization strategy, a key event identification priority is introduced; The recognition logic is optimized for lightweighting according to the optimization strategy to obtain lightweight recognition logic on the device side.

6. The abnormal event pattern recognition method based on fiber optic vibration signals according to claim 1, characterized in that, The steps of performing a reliability assessment on the abnormal event identification result based on the matching degree, the abnormal event location information, the environmental information, and the historical false alarm records, generating a reliability assessment result, and triggering a graded alarm based on the reliability assessment result include: Read the matching degree corresponding to the abnormal event identification result; Based on the degree of matching, the basic reliability assessment result of the abnormal event identification result is determined; Temperature and humidity signals are obtained from the environmental information, and the vibration signal is read. Preprocessing is performed on the vibration signal, the temperature signal, and the humidity signal respectively to obtain preprocessed modal data; Based on the modal data, extract the corresponding modal signal features; Based on the degree of influence of each modal signal characteristic on reliability assessment, modal fusion weights are assigned; The features of each modal signal are fused according to the modal fusion weights to generate environmental enhancement description information; Based on the environmental reinforcement description information, the abnormal event location information, and the historical false alarm records, the basic reliability assessment result is corrected to obtain the reliability assessment result, and a graded alarm is triggered based on the reliability assessment result.

7. The abnormal event pattern recognition method based on fiber optic vibration signals according to claim 1, characterized in that, The step of executing event identification and event localization tasks in parallel based on the vibration event description information, the initial position information, and the lightweight identification logic of the device, to obtain abnormal event identification results and abnormal event localization information includes: Based on the vibration event description information and the lightweight identification logic of the device, an event identification task is generated; Based on the vibration event description information and the initial position information, an event localization task is generated; Deploy a resource isolation and scheduling module, and allocate computing resource quotas and memory space to the event identification task and the event location task respectively through the resource isolation and scheduling module; Obtain task priority and historical resource usage; Adjust the computing resource quotas and memory space corresponding to the event identification task and the event location task based on task priority and historical resource usage. Under the adjusted computing resource quota and memory space, the event identification task and the event location task are executed in parallel; The event recognition task outputs the abnormal event recognition result, and the event location task outputs the abnormal event location information.

8. The method for abnormal event pattern recognition based on fiber optic vibration signals according to claim 1, characterized in that, The steps of performing a reliability assessment on the abnormal event identification result based on the matching degree, the abnormal event location information, the environmental information, and the historical false alarm records, generating a reliability assessment result, and triggering a graded alarm based on the reliability assessment result include: Read the abnormal event identification result, the matching degree, the abnormal event location information, the environmental information, and the historical false alarm records; Based on the degree of matching, a basic reliability score for the abnormal event identification result is determined; Based on the location information of abnormal events within a continuous time period, the spatial distribution relationship of abnormal events is identified, and based on the description information of vibration events within a continuous time period, the temporal evolution relationship of abnormal events is identified. Based on the spatial distribution relationship and the temporal evolution relationship, a causal chain between abnormal events is constructed; Based on the aforementioned causal chain, infer the source path of the abnormal event; Based on the environmental information, identify environmental interference factors within the area corresponding to the tracing path; Based on the historical false alarm records, determine whether there are false alarm records of the same event type in the area corresponding to the source tracing path; Based on the source tracing path, the environmental interference factors, and the false alarm records, the basic reliability score is corrected to obtain a reliability assessment result, and a graded alarm is triggered based on the reliability assessment result.

9. The method for abnormal event pattern recognition based on fiber optic vibration signals according to claim 1, characterized in that, The step of triggering a graded alarm based on the reliability assessment result includes: Receive event information from edge devices in different regions, the event information including abnormal event identification results, abnormal event location information and corresponding reliability assessment results; The event information is calibrated with timestamps and transformed with spatial coordinates to unify the reference benchmark for events in different regions, thus obtaining preprocessed event information; Obtain the computing and storage resources of the currently resource-constrained device; Based on the computing and storage resources of the current resource-constrained equipment, adjust and determine the event information filtering threshold, compression ratio, and number of linkage chain levels of the preprocessed event information; Based on the event information filtering threshold and the compression ratio, the preprocessed event information is filtered and compressed to obtain the processed event information; Based on preset linkage rules, identify potential correlations between events in different regions from the processed event information; Based on the potential associations and the number of linkage chain levels, a linkage chain is constructed between events in different regions; Based on the linkage chain between events in different regions, determine whether there are abnormal events involving multiple regions; When there are abnormal events involving multiple regions, a linkage hierarchical alarm instruction is generated based on the abnormal events involving multiple regions and the reliability assessment results. When there are no abnormal events involving multiple regions, a local hierarchical alarm instruction is generated based on the reliability assessment result corresponding to an abnormal event involving a single region. A hierarchical alarm is triggered based on the linked hierarchical alarm command or the local hierarchical alarm command.

10. An abnormal event pattern recognition system based on fiber optic vibration signals, used to perform abnormal event pattern recognition based on fiber optic vibration signals, characterized in that, include: The initial position determination module is used to acquire vibration signals, environmental information, and historical false alarm records of the optical fiber along the line, and determine the initial position information of the vibration occurrence based on the vibration signals. The correspondence establishment module is used to convert the vibration signal into audio processing data and establish a correspondence between the audio processing data and the initial position information; The vibration event description module is used to determine the optical cable laying medium of the section along the line corresponding to the audio processing data based on the correspondence, and to perform medium compensation and multi-dimensional feature extraction on the audio processing data according to the optical cable laying medium, and to fuse signal features of multiple dimensions to generate vibration event description information. The identification logic generation module is used to obtain the resource status of resource-constrained devices, and to perform lightweight optimization on the identification logic used for abnormal event identification based on the resource status of resource-constrained devices, so as to obtain the lightweight identification logic on the device side. The parallel task execution module is used to execute the event identification task and the event localization task in parallel based on the vibration event description information, the initial position information and the lightweight identification logic of the device, so as to obtain the abnormal event identification result and the abnormal event localization information. The graded alarm triggering module is used to obtain the degree of matching between the vibration event description information and the known event pattern, and to perform a reliability assessment on the abnormal event identification result based on the degree of matching, the abnormal event location information, the environmental information and the historical false alarm records, generate a reliability assessment result, and trigger a graded alarm based on the reliability assessment result.

Images