Multi-level alarm identification method, medium and equipment based on distributed optical fiber sensing

Abstract

The present invention provides a multi-level alarm identification method, medium, and device based on distributed fiber optic sensing, relating to the field of distributed fiber optic sensing technology. The method acquires light intensity signal data from multiple data points in real time; dynamically calculates the sample kurtosis of the multiple data points based on the light intensity signal data of each data point; compares the kurtosis value with a preset threshold, and proceeds to the next step if it exceeds the preset threshold; uses a cumulative strategy model to perform cumulative calculations based on the comparison results of the sample kurtosis of each data point with the preset threshold; when the cumulative results continue to accumulate and meet the preset conditions, proceeds to the next step to perform event identification on the alarm point; intercepts data blocks, maps the data blocks into grayscale images, and inputs them into a trained event recognition model based on a deep learning network for identification. The above method can enhance the signal-to-noise ratio, improve positioning accuracy, eliminate interference from accidental events, and ultimately improve the alarm event recognition rate.

Description

Technical Field

[0001] The present invention relates to the technical field of distributed optical fiber sensing, and in particular to a multi-level alarm recognition method, medium and equipment based on distributed optical fiber sensing. Background Art

[0002] Distributed fiber optic sensing technology, due to its wide monitoring range, high precision, and rapid response, is used in many fields such as energy pipelines, border security, rail transit, and structural monitoring. It uses the optical signal in the optical fiber as the information carrier, and obtains the actual alarm location and event information through photoelectric conversion and signal processing, thereby restoring the original alarm scene.

[0003] In existing technologies, distributed fiber optic sensing system alarms mostly use deep learning networks to identify alarm times. However, due to the varying conditions around the monitoring lines during deployment, they are often affected by ambient noise, resulting in a large number of false alarms, reducing system accuracy and increasing the workload of reviewers. Current technical solutions to these problems mainly focus on the following two aspects:

[0004] (1) Alarm location. Event location is the first step in screening alarm events. The accuracy of this step determines the sample base for the subsequent recognition step. If the accuracy is low, a large amount of erroneous and redundant data will be generated, thereby increasing the difficulty of subsequent recognition. Some current technologies aim to improve event recognition accuracy by simply improving positioning accuracy and reducing data false alarms.

[0005] (2) Event identification. After event location, a deep learning network is used to perform secondary identification of event data using a trained model library to clarify the type of alarm event. This technology uses data from the previous alarm location. Because the location data is accidental and random, high-intensity random events / accidental events can still pass through screening and trigger event location, resulting in reduced accuracy of deep learning sample training and event identification errors. Summary of the Invention

[0006] The purpose of this invention is to provide a multi-level alarm recognition method, medium, and device based on distributed fiber optic sensing, aiming to enhance the signal-to-noise ratio, improve positioning accuracy, eliminate interference from accidental events, and ultimately improve the alarm event recognition rate. The specific technical solution is as follows:

[0007] A multi-level alarm identification method based on distributed optical fiber sensing, the method comprising the following steps:

[0008] S100, acquiring light intensity signal data of multiple data points on a sensing optical fiber collected by a fiber optic sensing system in real time;

[0009] S200, dynamically calculating the sample kurtosis of multiple data points based on the light intensity signal data of each data point acquired at different times;

[0010] S300, comparing the calculated kurtosis value of each data point with a preset threshold, and proceeding to the next step if the kurtosis value exceeds the preset threshold;

[0011] S400: Using the accumulation strategy model, cumulative calculation is performed based on the comparison result of the sample kurtosis of each data point with the preset threshold. When the cumulative result continues to accumulate and reaches the preset condition, the data point at the corresponding moment is an alarm point, and the next step is to perform event identification on the alarm point;

[0012] S500, intercepting the sample kurtosis between the adjacent data point positions centered on the alarm point and the adjacent alarm moments as a data block, mapping the data block into a grayscale image and inputting it into a trained event recognition model based on a deep learning network for recognition, and obtaining the alarm event category.

[0013] Furthermore, in step S100, the light intensity signal data obtained is based on the light intensity amplitude data calculated after photoelectric conversion.

[0014] Furthermore, in step S200, the calculation method of the sample kurtosis is as follows: for the data matrix of M data points and N frames of data for each data point, a frame number window is selected. ω , each data point is adjacent ω The data at each moment is regarded as a processing unit. ω After the frame data is calculated, the kurtosis data of one frame is obtained. The fourth-order sample central moment and sample variance of the data in the window are calculated for each data point. The sample kurtosis is calculated based on the value of the fourth-order sample central moment divided by the square of the sample variance.

[0015] Furthermore, the sample kurtosis calculation formula is as follows:

[0016]

[0017] in, represents the sample kurtosis value, ω Represents the frame window length, x is the original light intensity signal, l is the data point, l = 1, 2, 3... M , n For data frames, n=ω , 2 ω , 3 ω ... N .

[0018] Furthermore, in step S300, the preset threshold is set when the distributed fiber optic sensing system completes the initial deployment. The specific process is as follows: data is collected when the monitoring fiber optic line is in a quiet state, and the collection time is approximately one hour; during the collection process, the maximum value of each data point is calculated in real time; and the maximum value of each data point in the quiet state is saved locally as the initial threshold setting, that is, the preset threshold.

[0019] Furthermore, the step S400 is specifically as follows: using the cumulative strategy model to perform cumulative calculations based on the sample kurtosis value of each data point and the comparison result of the sample kurtosis with the preset threshold value; when the cumulative result continues to accumulate and reaches the preset condition, proceed to the next step for event identification; the cumulative calculation rules are as follows: if the sample kurtosis is greater than the preset threshold value, the corresponding mapping factor is used according to the sample kurtosis value to obtain the mapping value, and the mapping value is multiplied by the corresponding rising factor to obtain the superposition value for superposition calculation; if the sample kurtosis is less than the preset threshold value, the corresponding mapping factor is used according to the sample kurtosis value to obtain the mapping value, and the mapping value is multiplied by the corresponding falling factor to obtain the subtraction value for subtraction calculation.

[0020] Furthermore, the up-factor, down-factor, and mapping factor of each data point in the accumulation strategy model can be adjusted.

[0021] Furthermore, in step S500, the event recognition model is a ResNet network; the training method of the event recognition model is as follows: before training, the filtered kurtosis data is collected and processed, and the features of various events are manually extracted. There are four types of events: manual construction, mechanical excavation, subway, and car driving. The extracted event data samples are all converted into 256-level grayscale images. The grayscale image is used to train the event recognition model, and the loss and accuracy of each round are recorded during the training process to adjust the model parameters. After multiple rounds of training, the model with the best test effect is selected as the trained event recognition model.

[0022] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the multi-level alarm identification method based on distributed optical fiber sensing as described above.

[0023] The present invention also provides a computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of the multi-level alarm identification method based on distributed optical fiber sensing as described above are implemented.

[0024] The present invention provides a multi-level alarm recognition method based on distributed optical fiber sensing, which has the following beneficial effects:

[0025] The present invention provides a multi-level alarm recognition method based on distributed fiber optic sensing. By using sample kurtosis values ​​to locate event positions, the signal-to-noise ratio of event signals is improved, thereby reducing missed events and enhancing the perception of events with weak intensity, thereby ensuring the timeliness of event warnings. By introducing a cumulative strategy model and adding judgment rules for the time dimension of data points, the interference of accidental events, such as accidental touch by personnel and passing of heavy machinery, is eliminated, the false alarm rate of events is reduced, and the events output by the model are ensured to be in a continuous triggering state, thereby reducing the amount of data processed in subsequent steps. A deep learning network is used for event recognition, and kurtosis data with a high signal-to-noise ratio is collected to ensure high resolution of event information in the grayscale image. Manual feature extraction and classification are performed for specific classified events, thereby ensuring obvious sample features and high recognition, thereby improving the accuracy of event recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 A schematic diagram of a flow chart of a multi-level alarm identification method based on distributed optical fiber sensing provided by the present invention;

[0027] Figure 2 is an overall flow chart of an embodiment of the present invention;

[0028] Figure 3 A flow chart of sample kurtosis calculation according to an embodiment of the present invention;

[0029] Figure 4 A schematic diagram of an accumulation strategy model according to an embodiment of the present invention;

[0030] Figure 5 A flowchart of event recognition according to an embodiment of the present invention;

[0031] Figure 6 4 is a structural block diagram of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0032] The following will be combined with the accompanying drawings provided by the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. The advantages and features of the present invention will become more apparent from the following description. It should be noted that the drawings are all in a very simplified form and are not in exact proportions. They are only used to facilitate and clearly illustrate the purpose of the embodiments of the present invention.

[0033] Example 1: This example provides a multi-level alarm identification method based on distributed optical fiber sensing. Figure 1 、 2 As shown, the method includes the following steps:

[0034] S100 , acquiring light intensity signal data of multiple data points on a sensing optical fiber collected by a fiber optic sensing system in real time.

[0035] In a preferred embodiment, the acquired light intensity signal data is based on light intensity amplitude data calculated after photoelectric conversion, and the data accurately reflects the signal strength of the monitoring line.

[0036] Specifically, the number of data points collected is set according to the length of the optical fiber laid, and each data point corresponds to a different position of the optical fiber.

[0037] In one embodiment of the present invention, the optical fiber length is 9 km, the sampling frequency is 3000 kHz, that is, 3000 frames of data are received per second, and the number of data points is set to 9000, so the data matrix size per second is 3000*9000.

[0038] S200 , dynamically calculating the sample kurtosis of multiple data points based on the light intensity signal data of each data point acquired at different moments.

[0039] The sample kurtosis reflects the sharpness of the optical signal fluctuation peak and can effectively distinguish the event triggering location. The location with a larger kurtosis value is the alarm point.

[0040] Specifically, the calculation method of sample kurtosis is as follows Figure 3 As shown, for the data matrix of M data points and N frames of data for each data point, the frame window length is selected ω , each data point is adjacent ω The data at each moment is regarded as a processing unit. ω After the frame data is calculated, the kurtosis data of one frame is obtained. The fourth-order sample central moment and sample variance of the data in the window are calculated for each data point. The sample kurtosis is calculated based on the value of the fourth-order sample central moment divided by the square of the sample variance. After the sample kurtosis data calculation is completed, the data matrix size is reduced to (N / ω )*M, reducing the amount of data processed in subsequent steps.

[0041] Furthermore, the sample kurtosis calculation formula is as follows:

[0042]

[0043] in, represents the sample kurtosis value, ω represents the window length, x is the original light intensity signal, l is the data point, l= 1, 2, 3... M , n For data frames, n=ω , 2 ω , 3 ω ... N .

[0044] In one embodiment of the present invention, a window size of 100 frames is selected, and M and N correspond to 9000 and 3000 respectively. After the kurtosis data calculation is completed, the data matrix size is reduced to 30*9000, reducing the amount of data processed in subsequent steps.

[0045] S300: Compare the calculated kurtosis value of each data point with a preset threshold value, and proceed to the next step if the kurtosis value exceeds the preset threshold value.

[0046] The preset threshold is set based on data collected during a quiet state after the distributed fiber optic sensing system is deployed. When a data point reaches an event trigger state, its kurtosis value will exceed the preset threshold, thus passing the screening and allowing it to proceed to the next step.

[0047] After the sample kurtosis is calculated, it is compared with a preset threshold as the first level of screening for event location. When an event is triggered on the monitored fiber, if the kurtosis value of a data point in a frame exceeds the preset threshold, the next level of processing is entered into the accumulation strategy model; otherwise, no further processing is performed. The preset threshold is set during the initial deployment of the distributed fiber optic sensing system. The specific process is as follows: (a) Data is collected when the monitored fiber line is in a quiet state, which takes approximately one hour; (b) During the collection process, the maximum value of each data point is calculated in real time; (c) The maximum value of each data point in the quiet state is saved locally as the initial threshold setting, i.e., the preset threshold.

[0048] S400. Use the cumulative strategy model to perform cumulative calculations based on the comparison results of the sample kurtosis of each data point with the preset threshold. When the cumulative results continue to accumulate and reach the preset conditions, the data point at the corresponding moment is the alarm point, and enter the next step to perform event identification on the alarm point.

[0049] Specifically, the cumulative strategy model performs cumulative calculations based on the number of times the kurtosis value exceeds a preset threshold. The cumulative result is stored in the computer cache and serves as the baseline value for the next moment's data. If the data at the next moment exceeds the preset threshold, the cumulative result is superimposed and saved for use. Otherwise, the cumulative result will gradually decrease from the current value. When it decreases to zero, the cumulative result saved in the cumulative strategy model cache is cleared and restored to its initial state. When the cumulative result continues to accumulate and reaches the preset condition, the next step of event identification will be entered.

[0050] The accumulation strategy model has been proposed to address certain random events, often of high intensity, such as accidental human contact or the passing of heavy machinery, which can be screened out by a preset kurtosis threshold. In this model, the event data point is continuously observed over a specified period of time. If the intensity of the data point consistently or repeatedly exceeds the preset kurtosis threshold, the cumulative calculation result continues to accumulate until the preset condition is met. Conversely, if the data point exceeds the kurtosis threshold only once or a few times within a specified period of time, the cumulative result is discontinued or even reduced, ensuring that random events cannot pass the model's screening.

[0051] In a preferred embodiment, a cumulative strategy model is used to perform cumulative calculations based on the sample kurtosis value of each data point and the comparison result of the sample kurtosis with a preset threshold. When the cumulative results continue to accumulate and reach the preset conditions, the next step is entered for event identification. The cumulative calculation rules are as follows: if the sample kurtosis is greater than the preset threshold, the corresponding mapping factor is used according to the sample kurtosis value to obtain the mapping value, and the mapping value is multiplied by the corresponding rising factor to obtain the superposition value for superposition calculation; if the sample kurtosis is less than the preset threshold, the corresponding mapping factor is used according to the sample kurtosis value to obtain the mapping value, and the mapping value is multiplied by the corresponding falling factor to obtain the subtraction value for subtraction calculation.

[0052] In an exemplary embodiment, the structure diagram of the accumulation strategy model is as follows: Figure 4 As shown, the XYZ axes correspond to data points, time, and cumulative results respectively. During the model processing, the size of the data point kurtosis value and the time when the point enters the model will be taken into consideration. The specific process is as follows: (a) When the data point kurtosis data enters the model, it is first mapped according to the size of the data point kurtosis value. For example, if the kurtosis value of data point 1 is 1200, it is mapped to 1 in the model. If the kurtosis value exceeds the preset threshold and the rising factor is 1, the superposition value is 1, and the cumulative result increases by 1, which is saved as the benchmark for the next frame of data for the data point; (b) Secondly, in order to ensure that accidental events can be excluded, the time when the data point enters the cumulative strategy model is recorded, and the data point is continuously observed within a specific time to see if there are still events occurring; the cumulative results and events of the data point within this time If no event occurs, the cumulative result will not change during this time, and will gradually decrease until it reaches zero if the time exceeds. If an event is observed to occur during this time, the cumulative result will increase according to the corresponding factor. (c) According to steps (a) and (b), based on the above two reference quantities, each frame of data entering the model can be accumulated and calculated. If the triggering event is a continuous event, its cumulative result will continue to increase within a specific time, as shown in data points 1 and 2, and continue to accumulate until the preset condition is reached (such as cumulative result = 10). If a data point event is triggered only once or a few times within a specific time, its cumulative result will gradually decrease to 0 after the event disappears, such as data point 5.

[0053] In a further preferred embodiment, the rising factor, falling factor and mapping factor of each data point in the accumulation strategy model can be adjusted to achieve the purpose of adjusting the sensing sensitivity of the data point.

[0054] The model sets auxiliary parameters, such as mapping factor, rise factor and fall factor, to adjust the cumulative result increase value, increase speed and fall speed of the accumulation strategy model. The purpose is to take into account the different conditions around the monitoring line. The parameters of each data point can be set separately to achieve the purpose of adjusting the sensing sensitivity of each data point, ensuring that active points will not be falsely reported due to excessive sensitivity, and fatigue points will not be missed due to low perception sensitivity. The specific effect can be seen in the comparison of the cumulative results of data point 1 and data point 2 (rise factor 0.5).

[0055] S500, intercepting the sample kurtosis between the adjacent data point positions centered on the alarm point and the adjacent alarm moments as a data block, mapping the data block into a grayscale image and inputting it into a trained event recognition model based on a deep learning network for recognition, and obtaining the alarm event category.

[0056] Data that meets the preset conditions of the accumulation strategy model is clipped to a specific size, including data points before and after the alarm point location and data from the adjacent time periods before and after the alarm moment. This data is then converted into a grayscale image. During the early stages of distributed fiber optic sensing system deployment, this data is collected in large quantities and manually extracted and classified. Information is labeled based on different feature samples, and trained to generate a model file. Once the model is trained, it is put into use. Data blocks that have passed the two-level kurtosis threshold and accumulation strategy model screening are input into the model for third-level recognition. Through calculation and comparison, the alarm event category is ultimately determined.

[0057] Specifically, the specific event identification method steps are as follows Figure 5 As shown, for the data points identified by the two layers of kurtosis threshold and accumulation strategy model in step S300 and step S400, the sample kurtosis between the adjacent data point positions centered on the alarm point and the adjacent alarm moments is intercepted as a data block, and the data block is mapped to a 256-level grayscale image. An event recognition model based on the ResNet network is constructed, and the event recognition model is trained using the grayscale image of the triggering event collected in the early stage of the distributed fiber optic sensing system deployment to obtain a trained event recognition model. The grayscale image of the current triggering event is input into the trained event recognition model based on the deep learning network for recognition to obtain the alarm event category.

[0058] In a preferred embodiment, the training method of the event recognition model is as follows: before training, the filtered kurtosis data is collected and processed, and the features of various events are manually extracted. There are four types of events: manual construction, mechanical excavation, subway and car driving. The extracted event data samples are converted into 256-level grayscale images. The grayscale image is used to train the event recognition model, and the loss and accuracy of each round are recorded during the training process to adjust the model parameters. After multiple rounds of training, the model with the best test effect is selected as the trained event recognition model.

[0059] Accordingly, during the use of the model, data points that meet the preset conditions of the cumulative model will be converted into 256-level grayscale images of the same size and finally input into the trained deep learning model for event recognition.

[0060] In an exemplary embodiment, model training is mainly aimed at four types of events: manual construction, mechanical excavation, driving and subway. In order to meet the number of valid samples, it is necessary to ensure that the data collection time is not less than one day. At the initial deployment of the distributed fiber optic sensing system, after the sample kurtosis and accumulation strategy model are calculated, the data point data after two layers of recognition are continuously collected and mapped to a 256-order 100*100 grayscale image, and then the grayscale image is manually feature extracted and classified to ensure the number of samples in the training set, test set and validation set of the deep learning network model, that is, the number of samples of each event in the three sets is not less than 100. In this embodiment, the number of training rounds is set to 40 rounds. During the training process, the accuracy and loss of each round of training are recorded to modify the parameters and optimize the model in the subsequent round of training. The model with the best indicators in the test set is selected and applied to the event recognition of the third level of the present invention.

[0061] Accordingly, the basic data for event recognition is the data block that passes the kurtosis threshold and cumulative strategy model, which is converted into a 256-order 100*100 grayscale image and used as the input for event recognition. After the recognition is completed, the model will return the event type and recognition accuracy for the software platform to display the alarm information.

[0062] The multi-level alarm recognition method based on distributed fiber optic sensing provided by the present invention performs three levels of recognition and judgment in order, including a kurtosis threshold layer, a cumulative strategy model layer and an event recognition layer. By using the sample kurtosis value to locate the event position, the signal-to-noise ratio of the event signal is improved, thereby reducing event omissions, and enhancing the perception ability of events with weak intensity, ensuring the timeliness of event warnings; by introducing the cumulative strategy model and adding judgment rules in the time dimension of data points, the interference of accidental events, such as accidental touch by people and passing of heavy machinery, is eliminated, the false alarm rate of events is reduced, and the events output by the model are ensured to be in a continuous trigger state, reducing the amount of data processed in subsequent steps; a deep learning network is used for event recognition, and kurtosis data with a high signal-to-noise ratio is collected to ensure the high resolution of event information in the grayscale image, and artificial feature extraction and classification are performed for specific classified events, ensuring that the sample features are obvious and the recognition is high, thereby improving the accuracy of event recognition.

[0063] Embodiment 2: This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the multi-level alarm identification method based on distributed optical fiber sensing described above are implemented.

[0064] The storage medium may be a magnetic disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), a flash memory, a hard disk drive (HDD) or a solid-state drive (SSD), etc.; the storage medium may also include a combination of the above types of memory.

[0065] Example 3: This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of the multi-level alarm identification method based on distributed optical fiber sensing described above are implemented.

[0066] like Figure 6As shown, the computer device may include: at least one processor 71, such as a CPU (Central Processing Unit), at least one communication interface 73, a memory 74, and at least one communication bus 72. The communication bus 72 is used to realize the connection and communication between these components. The communication interface 73 may include a display screen (Display) and a keyboard (Keyboard), and the optional communication interface 73 may also include a standard wired interface and a wireless interface. The memory 74 may be a high-speed RAM memory (Random Access Memory, volatile random access memory) or a non-volatile memory (non-volatile memory), such as at least one disk storage. The memory 74 may optionally be at least one storage device located away from the aforementioned processor 71. The memory 74 stores application programs, and the processor 71 calls the program code stored in the memory 74 to execute any of the above method steps.

[0067] The communication bus 72 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus. The communication bus 72 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 6 Only one thick line is used in the diagram, but this does not mean that there is only one bus or one type of bus.

[0068] Among them, the memory 74 may include a volatile memory (English: volatile memory), such as a random-access memory (English: random-access memory, abbreviated: RAM); the memory may also include a non-volatile memory (English: non-volatile memory), such as a flash memory (English: flash memory), a hard disk drive (English: hard disk drive, abbreviated: HDD) or a solid-state drive (English: solid-state drive, abbreviated: SSD); the memory 74 may also include a combination of the above types of memory.

[0069] The processor 71 may be a central processing unit (CPU), a network processor (NP), or a combination of a CPU and a NP.

[0070] The processor 71 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The PLD may be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL), or any combination thereof.

[0071] Optionally, the memory 74 is further configured to store program instructions. The processor 71 can call the program instructions to implement the multi-level alarm identification method based on distributed optical fiber sensing of the present invention.

[0072] Those skilled in the art should understand that the present invention can be implemented in many other specific forms without departing from the spirit and scope of the present invention. Based on the embodiments of the present invention, any changes and modifications made by ordinary technicians in the field of the present invention in accordance with the above disclosure are within the scope of protection of the claims.

Claims

1. A multi-level alarm identification method based on distributed optical fiber sensing, characterized in that: The method comprises the following steps: S100, acquiring light intensity signal data of multiple data points on a sensing optical fiber collected by a fiber optic sensing system in real time; S200, dynamically calculating sample kurtosis values ​​of multiple data points based on light intensity signal data of each data point acquired at different times; S300, comparing the calculated sample kurtosis value of each data point with a preset threshold, and proceeding to the next step if the value exceeds the preset threshold; S400: Using the accumulation strategy model, a cumulative calculation is performed based on the comparison result of the sample kurtosis value of each data point with a preset threshold. When the cumulative result continues to accumulate and reaches a preset condition, the data point at the corresponding moment is an alarm point, and the next step is to perform event identification on the alarm point. S500, intercepting the sample kurtosis values ​​between the adjacent data point positions centered on the alarm point and the adjacent alarm moments as a data block, mapping the data block into a grayscale image and inputting the trained event recognition model based on the deep learning network for recognition to obtain the alarm event category.

2. The multi-level alarm identification method based on distributed optical fiber sensing according to claim 1 is characterized in that: In step S100 , the light intensity signal data obtained is based on the light intensity amplitude data calculated after photoelectric conversion.

3. The multi-level alarm identification method based on distributed optical fiber sensing according to claim 1 is characterized in that: In step S200, the calculation method of the sample kurtosis value is as follows: for the data matrix of M data points and N frames of data for each data point, a frame number window is selected. ω , each data point is adjacent ω The data at each moment is regarded as a processing unit. ω After the frame data is calculated, the kurtosis data of one frame is obtained. The fourth-order sample central moment and sample variance of the data in the window are calculated for each data point. The sample kurtosis value is calculated based on the value of the fourth-order sample central moment divided by the square of the sample variance.

4. The multi-level alarm identification method based on distributed optical fiber sensing according to claim 3 is characterized in that: The formula for calculating the sample kurtosis value is as follows: ； in, represents the sample kurtosis value, ω Represents the frame window length, x is the original light intensity signal, l is the data point, l= 1, 2, 3... M , n For data frames, n=ω , 2 ω , 3 ω ... N .

5. The multi-level alarm identification method based on distributed optical fiber sensing according to claim 1 is characterized in that: In step S300, the preset threshold is set when the distributed fiber optic sensing system is initially deployed. The specific process is as follows: data is collected when the monitored fiber optic line is in a quiet state, and the collection time is one hour; during the collection process, the maximum value of each data point is calculated in real time; and the maximum value of each data point in the quiet state is saved locally as the initial threshold setting, i.e., the preset threshold.

6. The multi-level alarm identification method based on distributed optical fiber sensing according to claim 1 is characterized in that: The step S400 is specifically as follows: using the cumulative strategy model to perform cumulative calculations based on the sample kurtosis value of each data point and the comparison result of the sample kurtosis value with the preset threshold value; when the cumulative result continues to accumulate and reaches the preset condition, proceed to the next step for event identification; the cumulative calculation rules are as follows: if the sample kurtosis value is greater than the preset threshold value, a corresponding mapping factor is used according to the sample kurtosis value to obtain a mapping value, and the mapping value is multiplied by the corresponding rising factor to obtain a superposition value for superposition calculation; if the sample kurtosis value is less than the preset threshold value, a corresponding mapping factor is used according to the sample kurtosis value to obtain a mapping value, and the mapping value is multiplied by the corresponding falling factor to obtain a subtraction value for subtraction calculation.

7. The multi-level alarm identification method based on distributed optical fiber sensing according to claim 6 is characterized in that: The up-factor, down-factor, and mapping factor for each data point in the accumulation strategy model can be adjusted.

8. The multi-level alarm identification method based on distributed optical fiber sensing according to claim 1, characterized in that: In step S500, the event recognition model is a ResNet network; the event recognition model is trained as follows: before training, the filtered kurtosis data is collected and processed, and the features of various events are manually extracted. There are four types of events: manual construction, mechanical excavation, subway, and car driving. The extracted event data samples are all converted into 256-level grayscale images, and the event recognition model is trained using the grayscale images. During the training process, the loss and accuracy of each round are recorded to adjust the model parameters. After multiple rounds of training, the model with the best test results is selected as the trained event recognition model.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the multi-level alarm identification method based on distributed optical fiber sensing are implemented as described in any one of claims 1 to 8.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the multi-level alarm identification method based on distributed optical fiber sensing are implemented as described in any one of claims 1 to 8.

Images