A distributed, adaptive threshold adjustment method and a DAS system using the same

Abstract

This invention discloses a distributed, adaptive threshold adjustment method and a DAS system applying this method. The method includes the following steps: determining and initializing the capacity of the precision score array, recall score array, and confidence threshold array at system startup; user confirmation and submission of alarm feedback information; indexing and retrieving the previous precision and recall scores of the alarm location from the precision score array and recall score array; using the retrieved previous precision and recall scores, and based on the alarm feedback information, updating the precision and recall scores of the alarm location and each location within its influence range, and updating the confidence threshold within the alarm influence range; storing the updated precision, recall, and confidence thresholds on the DAS system's disk. This invention can dynamically update the precision, recall, and confidence thresholds of each location based on the actual situation of the alarm location reported by the user, in order to balance the precision and recall of the system and optimize the overall system performance.

Description

Technical Field

[0001] This invention relates to the field of distributed fiber optic vibration sensor pattern recognition technology, and in particular to a distributed, adaptive threshold adjustment method and a DAS system using this method. Background Technology

[0002] The basic principle of a DAS (Distributed Optical Array) system is to emit light pulses into an optical fiber and then receive the Rayleigh scattering signals along the fiber's path. External vibrations cause fluctuations in the reflected signals, and by detecting these reflected signals, vibrations along the fiber's path can be detected. DAS systems are currently widely used in perimeter security, pipeline early warning, and other fields. DAS system software identifies and detects potentially destructive activities such as excavation and construction based on vibration signals, and issues alarms to alert users if these events are detected.

[0003] Based on the above principles, the effect of external vibration on optical fibers is very complex. In addition, in pipeline security applications, optical cables are usually buried underground. The vibration signal between the signal to be detected (such as an excavator) and the optical cable, as well as the propagation process of the vibration signal in the soil, make it very difficult to directly model from the signal to reconstruct the target characteristics.

[0004] Currently, deep learning technology has been applied to DAS systems for pattern recognition. Deep Neural Networks (DNNs) have a strong ability to extract target signal features from large datasets. Algorithms such as Convolutional Neural Networks, Recurrent Neural Networks, and YOLO (You Only Look Once) can all be applied to DAS systems and achieve better alarm accuracy than traditional algorithms.

[0005] The final output of a DNN is usually obtained by using a sigmoid or softmax function to obtain a value in the (0,1) interval, representing the concepts of probability and confidence (this article uses confidence). If a clear alarm is required after applying a DNN to signal recognition in a DAS, a confidence threshold needs to be set. When the output confidence exceeds the set threshold, it indicates that the target has been detected and an alarm is generated. However, the confidence output of the neural network has no actual physical meaning, so the threshold setting is relatively subjective. Furthermore, the threshold is usually a globally fixed value or a partitioned fixed value, not a threshold for every location. This makes it impossible to properly set the threshold for each location on the optical fiber. Summary of the Invention

[0006] In view of this, the present invention provides a distributed, adaptive threshold adjustment method and a DAS system applying the method to solve the problems existing in the background art.

[0007] A distributed, adaptive threshold adjustment method includes the following steps:

[0008] S1. Determine and initialize the capacity of the accuracy score array, recall score array, and confidence threshold array when the system starts up, based on the spatial resolution of the optical fiber used in the DAS system and the total length of the optical fiber line.

[0009] S2, The user confirms the alarm issued by the DAS system based on the deep neural network model and submits alarm feedback information;

[0010] S3, based on the location coordinates of the alarm location, retrieve the previous precision score and recall score corresponding to the location coordinate index from the precision score array and recall score array, respectively;

[0011] S4. Using the previously obtained precision score and recall score, and based on the alarm feedback information and the set update method, update the precision score and recall score of each location within the alarm location and its influence range, and update the confidence threshold within the alarm influence range.

[0012] S5 stores the updated precision score, recall score, and confidence threshold on the DAS system's disk.

[0013] Preferably, the alarm feedback information includes correct alarms, false alarms, and missed alarms.

[0014] When the alarm feedback information is a correct alarm or a false alarm, the accuracy scores of the alarm location and each location within its influence range are updated according to the corresponding update method using the previously obtained accuracy scores.

[0015] When the alarm feedback information is a correct alarm or a missed alarm, the recall rate score of the alarm location and each location within its influence range is updated according to the corresponding update method, using the previously obtained recall rate score.

[0016] Preferably, the specific steps for updating the accuracy scores of the alarm location and each location within its influence range are as follows:

[0017] First, calculate the expected accuracy at the location where the alarm occurred based on the accuracy score;

[0018] Then, the accuracy score update amount for the alarm location is calculated based on the accuracy expectation, and the accuracy score for that location is updated based on the calculated accuracy score update amount.

[0019] Then, the Gaussian kernel function is used to spread the accuracy score update to each location within the influence range of the alarm event, and the accuracy score of each location within the influence range of the alarm event is updated.

[0020] Preferably, the method for calculating the expected accuracy is as follows:

[0021]

[0022] Precision score update amount at alarm location x0:

[0023] ΔP=K*(WP-EP(x0))

[0024] The method for updating the accuracy score at alarm location x0 is as follows:

[0025] P(x0)←P(x0)+K*(WP-EP(x0))

[0026] The method for updating the accuracy score of each location within the affected area of ​​the alarm event is as follows:

[0027] P(x)←P(x)+ΔP*Gk(x-x0), x∈[x0-3σ, x0+3σ]

[0028]

[0029] Where P(x0) is the previous accuracy score at the alarm location x0; EP(x0) is the expected accuracy at the alarm location x0; WP = 1 when the alarm feedback is a correct alarm and WP = 0 when the alarm feedback is a false alarm; K is the scoring coefficient; Gk is the Gaussian kernel function and σ is the variance of the Gaussian distribution.

[0030] Preferably, the specific steps for updating the recall rate score of each location within the alarm's location and its influence range are as follows:

[0031] First, calculate the expected recall rate at the location where the alarm occurred based on the recall rate score;

[0032] Then, the recall score update amount for the alarm location is calculated based on the expected recall rate, and the recall score for that location is updated based on the calculated recall score update amount.

[0033] Then, the recall score update is diffused to each position within the scope of the alarm event using a Gaussian kernel function, and the recall score at each position within the scope of the alarm event is updated.

[0034] Preferably, the expected recall rate calculation method is as follows:

[0035]

[0036] Recall score update at alarm location x0:

[0037] ΔR=K*(WR-ER(x0))

[0038] The method for updating the recall score at alarm location x0 is as follows:

[0039] R(x0)←R(x0)+K*(WR-ER(x0))

[0040] The method for updating the recall score for each location within the impact range of the alarm event is as follows:

[0041] R(x)←R(x)+ΔR*Gk(x-x0), x∈[x0-3σ, x0+3σ]

[0042]

[0043] Where R(x0) is the previous recall score at the alarm location x0; ER(x0) is the expected recall at the alarm location x0; WR = 1 when the alarm feedback is a correct alarm and WR = 0 when the alarm feedback is a false alarm; K is the scoring coefficient; σ is the variance of the Gaussian distribution.

[0044] Preferably, the method for updating the confidence threshold within the alarm's impact range in step S4 is as follows:

[0045] T(x)←T(x)+step*(ER(x)-EP(x)), x∈[x0-3σ, x0+3σ];

[0046] Where T(x) is the confidence threshold at position x, EP(x) is the expected precision at position x within the alarm's influence range, ER(x) is the expected recall at position x within the alarm's influence range, step is the threshold update coefficient used to control the update speed, and σ is the variance of the Gaussian distribution.

[0047] Preferably, the method for updating the confidence threshold within the alarm's impact range in step S4 is as follows:

[0048] T(x)←T(x)+step*(ER(x)-α*EP(x)), x∈[x0-3σ, x0+3σ]

[0049] Where EP(x) is the expected precision at position x within the alarm's influence range, ER(x) is the expected recall at position x within the alarm's influence range, step is the threshold update coefficient used to control the update speed, σ is the variance of the Gaussian distribution, and α is the balance coefficient of the DAS system when dynamically adjusting between precision and recall.

[0050] Preferably, the capacities of the precision score array, recall score array, and confidence threshold array are:

[0051]

[0052] Where L is the total length of the fiber optic line, and spatial_resolution is the spatial resolution of the fiber optic cable used in the DAS system, in meters.

[0053] A DAS system that uses the aforementioned distributed, adaptive threshold adjustment method to dynamically update the precision, recall, and confidence thresholds of its deep neural network model.

[0054] The beneficial effects of this invention are:

[0055] 1. In the process of using a deep neural network model to process vibration signals to identify and detect potential destructive behaviors, the DAS system can use the method of this invention to evaluate the precision, recall, and confidence threshold of each location on the optical fiber. It can also dynamically update the precision score, recall score, and confidence threshold of each location based on the actual situation of the alarm location reported by the user, so as to balance precision and recall and optimize the overall system performance.

[0056] 2. When using the method of this invention, the deep neural network model of the DAS system can appropriately increase the confidence threshold for regions with high recall and low precision to obtain better precision, and conversely, decrease the confidence threshold to increase recall. This achieves distributed and adaptive thresholding, obtaining a balance between precision and recall at each position. At the same time, a balance coefficient can be set to adjust between precision and recall to optimize the overall system performance.

[0057] 3. The method of this invention is based entirely on statistical principles and does not require prior information such as on-site environment or equipment parameters. It updates warning accuracy and recall scores based on correct alarms, false alarms, and missed alarms reported by users, using only limited feedback information for evaluation and updates, without requiring complete alarm verification—that is, verifying the correctness of each alarm and identifying each missed alarm.

[0058] 4. The algorithm is lightweight, concise, and efficient, making it suitable for embedded systems and edge computing. Attached Figure Description

[0059] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0060] Figure 1 This is a diagram showing the arrays of storage precision score P, recall score R, and threshold T, as well as the update of T.

[0061] Figure 2 This is a schematic diagram of the update mechanism for P and R.

[0062] Figure 3 This is a flowchart of the method of the present invention. Detailed Implementation

[0063] To better understand the technical solution of the present invention, the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0064] It should be understood that the described embodiments are merely some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0065] The present application will now be described in further detail with reference to specific embodiments and accompanying drawings.

[0066] This invention provides a distributed, adaptive threshold adjustment method. In the process of a DAS system using a deep neural network model to process vibration signals to identify and detect potential destructive behaviors, the method of this invention can evaluate the precision, recall, and confidence thresholds at each location on the optical fiber, and dynamically update the precision score, recall score, and confidence threshold for each location to balance precision and recall and optimize the overall system performance.

[0067] The alarms issued by the DAS system are basically divided into four categories:

[0068] TP (True Positive): There are actual alarm events and alarms.

[0069] FP (False Positive): An event that did not actually trigger an alarm, but did trigger one, i.e., a false alarm.

[0070] TN (True Negative): No actual alarm events, no alarms.

[0071] FN (False Negative): An alarm event actually occurred, but no alarm was reported, i.e., a missed alarm.

[0072] Precision is defined as:

[0073]

[0074] Recall is defined as:

[0075]

[0076] Ideally, precision and recall can both reach 100%, but in reality, they are usually mutually exclusive and mutually restrictive. A combined metric (F-Measure) is typically used for evaluation, with the most common F1 score being:

[0077]

[0078] It is evident that in real-world scenarios where precision and recall are mutually exclusive, a better F1 score can be achieved when precision and recall are balanced (this lacks rigorous mathematical proof, as precision and recall are theoretically unrelated and unconstrained, but it aligns with most practical situations). According to the definitions of precision and recall above, we need to obtain the statistical values ​​of correct alerts (TP), false positives (FP), and false negatives (FN) to obtain the statistical values ​​of precision and recall (classical probability). In reality, many alerts cannot be confirmed as correct, and there are also cases where false negatives are not detected and counted, meaning that TP, FP, and FN may all be incorrect. Therefore, it is difficult to obtain precision and recall based on the above definitions in practice, and for distributed systems, the precision and recall at each location are even more difficult to calculate.

[0079] Therefore, the distributed, adaptive threshold adjustment method of this invention applies the Elo rating system, using a precision score P (hereinafter referred to as P-score) and a recall score R (hereinafter referred to as R-score) to evaluate the precision and recall of the DAS system, and updates and determines the confidence threshold based on the scoring results. The DAS system of this invention uses three arrays to store and represent these distributed values: P, R, and the threshold T. These three arrays are loaded with data during system initialization and maintained during system operation.

[0080] Specifically, the distributed, adaptive threshold adjustment method of the present invention includes the following steps:

[0081] S1. Determine and initialize the capacities of the P, R, and T arrays during system startup based on the spatial resolution and total length of the fiber optic lines used in the DAS system. The capacity calculation is as follows:

[0082]

[0083] Where L is the total length of the fiber optic line, and spatial_resolution is the spatial resolution of the fiber optic cable used in the DAS system, in meters. The index of the x-position is:

[0084]

[0085] Here, idx is the index at position x, and the value of the array at position x can be found through the index. The array capacity of the three arrays will be adjusted accordingly depending on the spatial resolution of the optical fiber. For DAS systems, if different spatial resolutions are used, only the array capacity and index calculation need to be adjusted, which will not increase the complexity of this method. The following assumes a spatial resolution of 1 meter for simplification.

[0086] S2, the user confirms the alarm issued by the DAS system based on the deep neural network model and submits alarm feedback information.

[0087] The alarm feedback information includes correct alarms, false alarms, and missed alarms. That is, the user confirms whether the alarm is a correct alarm, a false alarm, or a missed alarm, and submits the confirmation result as alarm feedback information.

[0088] S3, based on the location coordinates of the alarm location, retrieve the previous P score and R score corresponding to the location coordinate index from the P and R arrays respectively.

[0089] S4. Using the previously obtained P score and R score, and based on the alarm feedback information and the set update method, update the P and R scores of each location within the alarm occurrence location and its influence range, and update the confidence threshold within the alarm influence range.

[0090] User confirmation of an alarm's correctness or error will trigger an update to the P score, while confirmation of an alarm's correctness or a missed alarm will trigger an update to the R score. That is, when the alarm feedback information is a correct alarm or a false alarm, the previously obtained P score is used to update the P scores at the alarm's location and its affected area according to the corresponding update method. The specific update steps are as follows:

[0091] The calculation method for the expected accuracy obtained from P is as follows:

[0092]

[0093] Where P(x) is the previous P score at position x, and EP(x) is the accuracy expectation at x, which will affect both the threshold update and the P score update.

[0094] Then, simply substitute the previous P(x0) at the alarm location x0 into the above formula (1) to calculate the expected accuracy EP(x0) at that location. P(x0) is the value corresponding to the location coordinate obtained from the P array based on its location coordinate x0.

[0095] Then, based on the expected accuracy, the P update amount ΔP at the alarm location x0 is calculated:

[0096] ΔP=K*(WP-EP(x0)) (2)

[0097] The update method for P at position x0 is as follows:

[0098] P(x0)←P(x0)+K*(WP-EP(x0)) (3)

[0099] WP determines its value based on the alarm feedback information confirmed by the user. That is, when the alarm feedback information is a correct alarm, WP=1, and when the alarm feedback information is a false alarm, WP=0. K is the scoring coefficient. When the score value is higher, the K value will be reduced to avoid losing too many points due to a single error in the high score range.

[0100] The above update only updates the P score at position x0. Then, a Gaussian kernel function is needed to spread the precision update ΔP to all positions within the influence range of the alarm event. The update method is as follows:

[0101] P(x)←P(x)+ΔP*Gk(x-x0),x∈[x0-3σ,x0+3σ] (4)

[0102]

[0103] Where Gk is the Gaussian kernel function, σ is the variance of the Gaussian distribution, and σ is the actual range of influence of the alarm event. For example, σ is taken as 30 meters for a mechanical excavation event.

[0104] In this way, the alarm feedback information from the user at position x0 updates the P score of the area [x0-3σ, x0+3σ] near x0.

[0105] When the alarm feedback information indicates a correct alarm or a missed alarm, the previously obtained R score is used to update the R score at the alarm location and within its affected area according to the corresponding update method. The specific update steps are as follows:

[0106] The method for calculating the expected recall rate based on R is as follows:

[0107]

[0108] Where R(x) is the previous R score at position x, and ER(x) is the expected recall at position x, which will affect both the threshold update and the R score update.

[0109] Then, simply substitute the previous R(x0) at the alarm location x0 into the above formula (6) to calculate the expected recall ER(x0) at that location. R(x0) is the value corresponding to the location coordinate obtained from the R array based on its location coordinate x0.

[0110] Then, based on the expected recall rate, the R update amount ΔR at the alarm location x0 is calculated.

[0111] ΔR=K*(WR-ER(x0)) (7)

[0112] The method for updating R at position x0 is as follows:

[0113] R(x0)←R(x0)+K*(WR-ER(x0)) (8)

[0114] WR determines its value based on the alarm feedback information confirmed by the user. That is, when the alarm feedback information is a correct alarm, WR=1, and when the alarm feedback information is a missed alarm, WR=0. K is the scoring coefficient. When the score is higher, the K value will be reduced to avoid losing too many points due to a single error in the high score range.

[0115] The above update only updates R at position x0. Then, a Gaussian kernel function is needed to spread the update amount ΔR to all positions within the influence range of the alarm event. The update method is as follows:

[0116] R(x)←R(x)+ΔR*Gk(x-x0),x∈[x0-3σ,x0+3σ] (9)

[0117]

[0118] Where σ is the variance of the Gaussian distribution described above.

[0119] In this way, the alarm feedback information from the user at position x0 updates the R score of the area [x0-3σ, x0+3σ] near x0.

[0120] The confidence threshold T(x) for the affected area of ​​the aforementioned alarm event is updated based on the expected precision and recall rates, and the update method is as follows:

[0121] T(x)←T(x)+step*(ER(x)-EP(x)) (11)

[0122] Where EP(x) is the expected precision of position x within the alarm's influence range, ER(x) is the expected recall of position x within the alarm's influence range, and step is the threshold update coefficient used to control the update speed.

[0123] When updating the confidence threshold according to formula (11), the system will dynamically adjust towards the direction where recall equals precision. However, sometimes users believe that recall and precision have different priorities, for example, they believe that recall is more important than alarm precision. In this case, the update method of the confidence threshold T(x) is modified as follows:

[0124] T(x)←T(x)+step*(ER(x)-α*EP(x)) (12)

[0125] Here, α is the balance coefficient of the DAS system when dynamically adjusting between precision and recall. The system adjusts towards this balance; if α > 1, recall is considered to have higher priority than precision. Using this method, the system dynamically updates the threshold for the area near the alarm location based on user-submitted alarm feedback information during runtime, thereby obtaining a distributed, adaptive confidence threshold setting across the entire line.

[0126] S5 stores the updated P score, R score, and confidence threshold T on the DAS system's disk.

[0127] The following examples illustrate the distributed, adaptive threshold adjustment method of the present invention in detail.

[0128] Assuming the fiber optic line used in the DAS system is 50km long and the spatial resolution is 1m, the capacity of the P array, R array and confidence threshold T array is initialized to 50,000. The P(x) and R(x) values ​​at position x can be obtained by indexing from the corresponding arrays through their position coordinates x.

[0129] When the DAS system starts, the P, R, and confidence threshold array T load data stored on disk from previous system use to restore the previous state.

[0130] If the DAS system is used for the first time, it needs to be initialized, in which P and R are all initialized to 0, and T is all initialized to the traditional fixed threshold. In this method, T is initialized to 0.5.

[0131] Figure 1 As shown, this diagram illustrates the arrays storing P, R, and the threshold T, as well as the update of T. In the left diagram, the values ​​of the P and R arrays represent the P score and R score at each position. The expected EP and ER at each position can be obtained according to the formulas mentioned above, as shown in the EP and ER curves. In the right diagram, the values ​​of the T array are the confidence thresholds for each position. The solid line in the curve represents the T value before the update, and the dashed line represents the T+dT value after the update. In the diagram, the update range of T reaches 1000 meters, but in actual implementation, the update range is usually not this large.

[0132] Figure 2 The diagram shows the update mechanism of P and R. User confirmation and feedback of correct alarms, incorrect alarms, and missed alarms will trigger the update of P and R. The G curve in the figure is the Gaussian distribution update curve of the area near the confirmation point. The solid line in the right figure is the P and R curve before the update, and the dashed line is the P and R curve after the update.

[0133] Assuming a user confirms an alarm as correct at a location of 294m, P updates the alarm as follows: First, calculate the update amount at the 294m location as described above:

[0134] ΔP=K*(WP-EP

[294] )

[0135] The correct alarm is set to WP = 1. Based on P

[294] = 41.9, the expected EP

[294] = 0.56 can be obtained. In this method, K takes the value 32 when the score is in (-1000, 1000), and K takes the value 16 when it exceeds this range. Therefore, K = 32 is substituted into the above formula to get ΔP = 14. Then the update value is spread to the nearby area, σ = 30 is taken, and the Gaussian kernel Gk is calculated according to the method described above. Figure 2 As shown in the graph at the top right, the P-values ​​within this range are updated as follows:

[0136] P(x)←P(x)+14*Gk(x-294),x∈[204,384]

[0137] Correct alerts will also cause the R value to be updated in the same way, which will not be elaborated further.

[0138] Suppose a user confirms a missed detection at a location of 294 meters. R updates for missed detections as follows: First, calculate the update amount for the 294-meter location as described above:

[0139] ΔR=K*(WR-ER

[294] )

[0140] The missed alarm is set to WR=0. Based on R

[294] =107.5, the expected ER

[294] =0.65 can be obtained. Substituting K=32 into the above formula, we get ΔR=-20.8. Applying the same Gaussian kernel Gk as above, such as... Figure 2 As shown in the graph at the bottom right, the R values ​​within this range are updated as follows:

[0141] R(x)←R(x)-20.8*Gk(x-294), x∈[204, 384]

[0142] False alarms will also cause the P value to be updated in the same way, which will not be elaborated further. After P and R are updated, the threshold T for that region also needs to be updated. This method uses an update rate step = 0.3 and a balance coefficient α = 1.1. First, the ER and EP for that region are calculated, and then T is updated:

[0143] T(x) ← T(x) + 0.3*(ER(x) - 1.1*EP(x)), x ∈ [204, 384]

[0144] After the update is complete, P, R, and T need to be saved to disk so that the state can be restored when the system restarts.

[0145] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A distributed, adaptive threshold adjustment method, characterized in that, Specifically, the following steps are included: S1. Determine and initialize the capacity of the accuracy score array, recall score array, and confidence threshold array when the system starts up, based on the spatial resolution of the optical fiber used in the DAS system and the total length of the optical fiber line. S2, The user confirms the alarm issued by the DAS system based on the deep neural network model and submits alarm feedback information; S3, based on the location coordinates of the alarm location, retrieve the previous precision score and recall score corresponding to the location coordinate index from the precision score array and recall score array, respectively; S4. Using the previously obtained precision score and recall score, and based on the alarm feedback information and the set update method, update the precision score and recall score of each location within the alarm location and its influence range, and update the confidence threshold within the alarm influence range. The specific steps for updating the accuracy scores of the alarm location and each location within its affected area are as follows: First, the expected accuracy at the alarm location is calculated based on the accuracy score. The method for calculating the expected accuracy is as follows: Then, the accuracy score update amount for the alarm location is calculated based on the accuracy expectation, and the accuracy score for that location is updated based on the calculated accuracy score update amount: The accuracy score update amount at alarm location x0 is Δp = K*(WP-EP(x0)); The method for updating the accuracy score at the alarm location x0 is P(x0)←P(x0)+K*(WP-EP(x0)); Then, the Gaussian kernel function is used to diffuse the accuracy score update to each location within the influence range of the alarm event, and the accuracy score of each location within the influence range of the alarm event is updated: P(x)←P(x)+Δp*GK(x-x0),x∈[x0-3σ,x0+3σ], Where P(x0) is the previous accuracy score at the alarm location x0; EP(x0) is the expected accuracy at the alarm location x0; WP = 1 when the alarm feedback is a correct alarm and WP = 0 when the alarm feedback is a false alarm; K is the scoring coefficient; σ is the variance of the Gaussian distribution. The specific steps for updating the recall score for each location within the alarm's location and its affected area are as follows: First, the expected recall rate at the location where the alarm occurred is calculated based on the recall rate score. The method for calculating the expected recall rate is as follows: Then, the recall score update amount for the alarm location is calculated based on the expected recall rate, and the recall score for that location is updated based on the calculated recall score update amount. The recall score update amount at alarm location x0 is ΔR = K*(WR-ER(x0)); The method for updating the recall score at the alarm location x0 is R(x0)←R(x0)+K*(WR-ER(x0)); Then, the recall score update is propagated to all locations within the influence range of the alarm event using a Gaussian kernel function, and the recall score at each location within the influence range of the alarm event is updated: The method for updating the recall score for each location within the scope of the alarm event is R(x)←R(x)+Δp*GK(x-x0),x∈[x0-3σ,x0+3σ]. Where R(x0) is the previous recall score at alarm location x0; ER(x0) is the expected recall at alarm location x0; WR = 1 when the alarm feedback is a correct alarm and WR = 0 when the alarm feedback is a false negative; K is the scoring coefficient; σ is the variance of the Gaussian distribution. The update method for the confidence threshold within the alarm influence range is: T(x)←T(x)+step*(ER(x)-EP(x)),x∈[x0-3σ,x0+3σ], where EP(x) is the expected precision at position x within the alarm influence range, ER(x) is the expected recall at position x within the alarm influence range, step is the threshold update coefficient used to control the update speed, and σ is the variance of the Gaussian distribution; S5 stores the updated precision score, recall score, and confidence threshold on the DAS system's disk.

2. The distributed, adaptive threshold adjustment method according to claim 1, characterized in that, The alarm feedback information includes correct alarms, false alarms, and missed alarms. When the alarm feedback information is a correct alarm or a false alarm, the accuracy scores of the alarm location and each location within its influence range are updated according to the corresponding update method using the previously obtained accuracy scores. When the alarm feedback information is a correct alarm or a missed alarm, the recall rate score of the alarm location and each location within its influence range is updated according to the corresponding update method, using the previously obtained recall rate score.

3. The distributed, adaptive threshold adjustment method according to claim 1, characterized in that, The capacity of the precision score array, recall score array, and confidence threshold array L is the total length of the fiber optic line, and spatial_resolution is the spatial resolution of the fiber optic cable used in the DAS system, in meters.

4. A DAS system, characterized in that, The DAS system uses the distributed, adaptive threshold adjustment method described in any one of claims 1-3 to dynamically update the precision score, recall score, and confidence threshold of its deep neural network model.

5. A distributed, adaptive threshold adjustment method, characterized in that, Specifically, the following steps are included: S1. Determine and initialize the capacity of the accuracy score array, recall score array, and confidence threshold array when the system starts up, based on the spatial resolution of the optical fiber used in the DAS system and the total length of the optical fiber line. S2, The user confirms the alarm issued by the DAS system based on the deep neural network model and submits alarm feedback information; S3, based on the location coordinates of the alarm location, retrieve the previous precision score and recall score corresponding to the location coordinate index from the precision score array and recall score array, respectively; S4. Using the previously obtained precision score and recall score, and based on the alarm feedback information and the set update method, update the precision score and recall score of each location within the alarm location and its influence range, and update the confidence threshold within the alarm influence range. The specific steps for updating the accuracy scores of the alarm location and each location within its affected area are as follows: First, the expected accuracy at the alarm location is calculated based on the accuracy score. The method for calculating the expected accuracy is as follows: Then, the accuracy score update amount for the alarm location is calculated based on the accuracy expectation, and the accuracy score for that location is updated based on the calculated accuracy score update amount: The accuracy score update amount at alarm location x0 is Δp = K*(WP-EP(x0)); The method for updating the accuracy score at the alarm location x0 is P(x0)←P(x0)+K*(WP-EP(x0)); Then, the Gaussian kernel function is used to diffuse the accuracy score update to each location within the influence range of the alarm event, and the accuracy score of each location within the influence range of the alarm event is updated: P(x)←P(x)+Δp*GK(x-x0),x∈[x0-3σ,x0+3σ], Where P(x0) is the previous accuracy score at the alarm location x0; EP(x0) is the expected accuracy at the alarm location x0; WP = 1 when the alarm feedback is a correct alarm and WP = 0 when the alarm feedback is a false alarm; K is the scoring coefficient; σ is the variance of the Gaussian distribution. The specific steps for updating the recall score for each location within the alarm's location and its affected area are as follows: First, the expected recall rate at the location where the alarm occurred is calculated based on the recall rate score. The method for calculating the expected recall rate is as follows: Then, the recall score update amount for the alarm location is calculated based on the expected recall rate, and the recall score for that location is updated based on the calculated recall score update amount. The recall score update amount at alarm location x0 is ΔR = K*(WR-ER(x0)); The method for updating the recall score at the alarm location x0 is R(x0)←R(x0)+K*(WR-ER(x0)); Then, the recall score update is propagated to all locations within the influence range of the alarm event using a Gaussian kernel function, and the recall score at each location within the influence range of the alarm event is updated: The method for updating the recall score for each location within the scope of the alarm event is R(x)←R(x)+Δp*GK(x-x0),x∈[x0-3σ,x0+3σ]. Where R(x0) is the previous recall score at alarm location x0; ER(x0) is the expected recall at alarm location x0; WR = 1 when the alarm feedback is a correct alarm and WR = 0 when the alarm feedback is a false negative; K is the scoring coefficient; σ is the variance of the Gaussian distribution. The update method for the confidence threshold within the alarm influence range is: T(x)←T(x)+step*(ER(x)-α*EP(x)),x∈[x0-3σ,x0+3σ], where EP(x) is the precision expectation at position x within the alarm influence range, ER(x) is the recall expectation at position x within the alarm influence range, step is the threshold update coefficient used to control the update speed, σ is the variance of the Gaussian distribution, and α is the balance coefficient of the DAS system when dynamically adjusting between precision and recall. S5 stores the updated precision score, recall score, and confidence threshold on the DAS system's disk.

6. The distributed, adaptive threshold adjustment method according to claim 5, characterized in that, The alarm feedback information includes correct alarms, false alarms, and missed alarms. When the alarm feedback information is a correct alarm or a false alarm, the accuracy scores of the alarm location and each location within its influence range are updated according to the corresponding update method using the previously obtained accuracy scores. When the alarm feedback information is a correct alarm or a missed alarm, the recall rate score of the alarm location and each location within its influence range is updated according to the corresponding update method, using the previously obtained recall rate score.

7. The distributed, adaptive threshold adjustment method according to claim 5, characterized in that, The capacity of the precision score array, recall score array, and confidence threshold array L is the total length of the fiber optic line, and spatial_resolution is the spatial resolution of the fiber optic cable used in the DAS system, in meters.

8. A DAS system, characterized in that, The DAS system uses the distributed, adaptive threshold adjustment method described in any one of claims 5-7 to dynamically update the precision score, recall score, and confidence threshold of its deep neural network model.

Images