A universal repeated alarm suppression method for transmission line monitoring system based on time series sample analysis

Abstract

The application discloses a power transmission line monitoring system general repeated alarm suppression method based on time sequence sample analysis, which can analyze and judge the change of the abnormal area by combining feature map analysis and time sequence sample modeling of a deep learning algorithm, and then determine whether to continue the alarm. The application can effectively suppress repeated alarms by comparing time series analysis and heat maps. The method of the application uses the gradient map of neural network back propagation and the output layer feature map for weighted combination to generate an activated response heat map, so that the area with the largest contribution to the abnormal output result in the image can be accurately identified. Through global average pooling and ReLU activation function, the system can accurately locate the area where the abnormality occurs. This technical means improves the accuracy of abnormal area identification and provides a reliable basis for subsequent alarm suppression.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of monitoring and alarming, and particularly relates to a general repeated alarm suppression method for a power transmission line monitoring system based on time sequence sample analysis. BACKGROUND

[0002] With the intelligent development of the power system, the monitoring and maintenance of the power transmission line have become an important link to ensure the safety of power supply. The traditional power transmission line monitoring relies on manual inspection and real-time data acquisition of equipment, which improves the safety of the power transmission line to a certain extent. In recent years, with the development of AI technology, more and more visual intelligent recognition technologies have been applied to the monitoring of the power transmission line. However, due to the large number of monitoring devices, large amount of data and wide distribution, the processing and analysis of information are also increasing in difficulty.

[0003] In the current power transmission line monitoring system, the alarm system often issues an alarm signal according to the abnormality judged by the AI visual algorithm. Most of these alarm signals are triggered by threshold setting or preset rules. However, AI visual analysis may introduce many repeated alarms and false alarms. When there is no obvious change in the target monitoring environment, the alarm will continue to occur, and these repeated alarms will bring a heavy workload to the staff.

[0004] At the same time, there are some technologies currently used to reduce false alarms, but most of these technologies focus on a single data analysis method, such as false alarm suppression based on statistical methods. These technologies depend on the specific form of the monitoring target, such as floating object intrusion on the power transmission line, tower crane construction intrusion on the line, and plant growth intrusion below the line. The system needs to develop corresponding repeated alarm suppression strategies for different monitoring targets, which means that the alarm suppression strategy for each monitoring content is independent, which obviously increases the additional monitoring and operation cost. SUMMARY

[0005] To solve the problems existing in the prior art, the present application provides a general repeated alarm suppression method for a power transmission line monitoring system based on time sequence sample analysis.

[0006] The technical scheme of the present application is as follows:

[0007] A general repeated alarm suppression method for a power transmission line monitoring system based on time sequence sample analysis, the power transmission line monitoring system is provided with a monitoring mode, the monitoring mode adopts a deep learning network to output a feature map group of the power transmission channel, and the method comprises the following steps:

[0008] Step 1) setting an alarm mode in the power transmission line monitoring system, triggering the alarm mode and performing a first alarm when an abnormality is detected in the monitoring mode;

[0009] Step 2) In the alarm mode, a feature map group is obtained, which represents that the confidence of a channel representing an abnormal category in the deep learning network is greater than the confidence of a channel representing a normal category;

[0010] Step 3) Based on the feature map group, an activation response heat map is constructed;

[0011] Step 4) A plurality of activation response heat maps at different time points are collected to form a time sequence sample;

[0012] Step 5) The time sequence sample is analyzed to obtain a difference value between adjacent activation response heat maps in the time sequence sample, and the difference value is combined with a given temperature threshold and a region area threshold to suppress repeated alarm or continuous alarm;

[0013] Step 6) When the anomaly disappears and exceeds a time threshold, the alarm mode is automatically exited and the monitoring mode is restored.

[0014] Further, the specific method of step 2) is as follows:

[0015] Step 2.1) The deep learning network includes a classification and target detection network and a segmentation network, the output channel of the classification and target detection network is a one-dimensional vector V (1 x c), and c in the vector V represents a corresponding category; the output channel of the segmentation network is a three-dimensional matrix graph A (n x m x c), and c in the matrix represents a corresponding category confidence;

[0016] Step 2.2) A feature map group output by the deep learning network is obtained , wherein w represents the width of the feature map, h represents the height of the feature map, and c represents the number of channels of the feature map;

[0017] Step 2.3) A channel with the maximum confidence in the category channel c in step 2.1) is selected as a category classification, and the category is the abnormal category when the alarm is triggered .

[0018] Further, the specific method of step 3) is as follows:

[0019] The feature map group is taken as the abnormal category , and the gradient of the feature map group is calculated .

[0020] The weight of the c channels in the feature map group is obtained by global average pooling .

[0021] The feature map is calculated using the weight of the c channels .After weighting and ReLU activation, the abnormal class is obtained The activation response heat map k (w, h) on the feature map group is used to locate the region that determines the last abnormal class classification, i.e. the abnormal occurrence region , .

[0022] Further, the feature map group The weight of the c channels is calculated as follows: ; wherein, represents the size of the feature map, ; represents the value corresponding to the i-th row and j-th column on the feature map.

[0023] The specific method of step 4) includes:

[0024] Step 4.1) The power line monitoring system obtains activation response heat maps corresponding to multiple time points in alarm mode, denoted as , wherein t is the time;

[0025] Step 4.2) Sort the activation response heat maps of multiple time points in time sequence to obtain time sequence samples .

[0026] The specific method of step 5) includes:

[0027] Step 5.1) Each activation response heat map in the time sequence sample is denoted as , , , …, ; and , , , …, represent the activation response heat maps corresponding to the 1st, 2nd, 3rd, …, nth time points, respectively. Take the absolute value difference between the activation response heat maps corresponding to adjacent two time points .

[0028] Step 5.2) Given a temperature threshold T, if the value of each coordinate point is less than T, it means that the activation response heat maps of the two time points have not changed significantly, and the subsequent alarm is suppressed; if there is a coordinate point whose value is greater than T, it means that the activation response heat maps of the two time points have changed locally, and the abnormal position is changing or spreading;

[0029] ​Step 5.3) Given a region area threshold A, in step 5.2) If the value of the coordinate point is greater than T, it is determined whether the pixel area is greater than A, if greater than A, it indicates that the abnormal region is undergoing a large change, and the alarm is continued, if less than A, it indicates that the abnormal region has a slight change or no change, indicating that the abnormal situation has not spread, and subsequent alarms are inhibited.

[0030] The absolute value difference between the activation response heat map corresponding to the tth moment and the (t-1)th moment The expression is:

[0031] ;

[0032] Wherein, And The activation response heat map corresponding to the tth moment and the (t-1)th moment respectively, .

[0033] The specific method of the step 6) comprises:

[0034] Record the difference between the current moment and the last time the abnormality is found, if the difference exceeds the time threshold, automatically exit the alarm mode and restore the monitoring mode.

[0035] Compared with the prior art, the present application has the following beneficial effects:

[0036] The present application proposes a general repeated alarm suppression method for a power transmission line monitoring system based on time sequence sample analysis, which uniformly processes the AI algorithm output of different monitoring targets, avoiding designing independent suppression strategies for each monitoring target. This is different from the prior art, which usually relies on formulating suppression strategies for each type of anomaly separately, resulting in additional operating costs. The present application only needs one set of alarm suppression strategy to cope with multiple abnormal situations, and the generality of the present application significantly reduces the operating cost of the monitoring system and improves the intelligent level of the system.

[0037] The present application can analyze and judge the change of the abnormal region by combining feature map analysis and time sequence sample modeling of deep learning algorithm, and then decide whether to continue the alarm. Compared with the traditional technology based on threshold setting or rule triggering, the present application can effectively suppress repeated alarms through time sequence analysis and comparison of heat maps. This method significantly reduces invalid alarms and improves the work efficiency of the staff.

[0038] The application utilizes the gradient graph of neural network back propagation and the output layer feature graph to perform weighted combination, generates an activation response heat map, and thus accurately identifies the area in the image that contributes most to the abnormal output result. Through global average pooling and ReLU activation function, the system can accurately locate the area where the anomaly occurs. This technical means improves the accuracy of abnormal area identification and provides a reliable basis for subsequent alarm suppression.

[0039] According to the change of the abnormal area, the application dynamically determines whether to continue alarming by setting a temperature threshold and a region threshold. Compared with the traditional fixed threshold alarm system, this flexible alarm suppression mechanism can better adapt to different monitoring scenes, improves the response accuracy and reliability of the system, and avoids the frequent occurrence of invalid alarms.

[0040] The application divides the monitoring system into a monitoring mode and an alarm mode. The alarm mode can be applied to any monitoring system as a plug-and-play technical module. The separation of the monitoring mode and the alarm mode enables the system to be unaffected by the alarm suppression technology during normal operation and quickly switches to the alarm mode when an anomaly is detected. This technical means enhances the flexibility and scalability of the system, facilitating subsequent function expansion and maintenance.

[0041] The application realizes automatic exit of the alarm mode by continuously analyzing the change of the abnormal area and combining time threshold judgment. When the abnormal situation is handled or disappears, the system automatically returns to the monitoring mode, avoiding unnecessary alarm persistence. This technical means optimizes the alarm management process and improves the automation level and operation efficiency of the system. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 A principle block diagram of a general repeated alarm suppression method for a power transmission line monitoring system based on time sequence sample analysis;

[0043] Figure 2 A flowchart of a general repeated alarm suppression method for a power transmission line monitoring system based on time sequence sample analysis. DETAILED DESCRIPTION

[0044] The application will be further illustrated below in conjunction with the drawings and specific embodiments, and it should be understood that these embodiments are only used to illustrate the application and not to limit the scope of the application. After reading the application, those skilled in the art can make various modifications to the application, and all such modifications fall within the scope defined by the appended claims.

[0045] Embodiment:

[0046] The application discloses a power transmission line monitoring system general repeated alarm suppression method based on time sequence sample analysis. Figure 1 and Figure 2 The method comprises the following steps:

[0047] Step 1) setting an alarm mode in the power transmission line monitoring system, acquiring a power transmission line to-be-inspected image in the monitoring mode, judging the to-be-inspected image based on an inspection model, if no abnormality is found, judging the to-be-inspected image as a normal image, if an abnormality is detected, judging the to-be-inspected image as an abnormal image and triggering the alarm mode and giving a first alarm to an operator;

[0048] Step 2) acquiring, in the alarm mode, a feature map group representing that the confidence of an abnormality category channel in a deep learning network is greater than the confidence of a normality category channel;

[0049] Step 3) based on the feature map group, based on back propagation extraction and weighted processing, constructing an activation response heat map for positioning an abnormality occurrence region;

[0050] Step 4) collecting activation response heat maps at multiple time points over time to form time sequence samples;

[0051] Step 5) analyzing the time sequence samples, acquiring a difference value between adjacent activation response heat maps in the time sequence samples, and based on the difference value and in combination with a given temperature threshold and a region area threshold, suppressing repeated alarms or continuous alarms;

[0052] Step 6) when the abnormality disappears and exceeds a time threshold, automatically exiting the alarm mode and restoring the monitoring mode.

[0053] The application solves the problem of repeated alarms in the power transmission line monitoring system due to no obvious change in the environment by time sequence modeling of feature maps of an AI algorithm output layer and analysis of activation response heat maps. Specifically, the system analyzes activation response heat maps of time sequences before and after to judge whether an abnormality region changes. If the response region does not change greatly, repeated alarms are suppressed; if the response region changes obviously, continuous alarms are given. This technical means effectively reduces the frequency of invalid alarms and reduces the work burden of workers.

[0054] Further, the specific method of step 2) in the example is as follows:

[0055] Step 2.1) An alarm has been triggered in step 1, and the source of the alarm is that the monitoring algorithm has found an anomaly, which at the output level of the neural network is: the confidence of the channel representing the abnormal category is greater than the confidence of the channel representing the normal category. Deep learning networks include classification and target detection networks and segmentation networks. For classification and target detection networks, the output channel is a one-dimensional vector V (1 x c), where c represents the corresponding category in the vector V; for segmentation networks, the output channel is a three-dimensional matrix graph A (n x m x c), where c represents the corresponding category confidence in the matrix;

[0056] Step 2.2) Obtain the feature map group of the final output of the deep learning network , where w represents the width of the feature map, h represents the height of the feature map, and c represents the number of channels of the feature map;

[0057] Step 2.3) Select a channel with the maximum confidence in the category channel c in step 2.1) as the category classification, and the category is the abnormal category when the alarm is triggered .

[0058] Further, the specific method of step 3) in this example is as follows:

[0059] with the abnormal category , the gradient of the feature map group is calculated , which is consistent in dimension with the feature map group . This gradient reflects the contribution of the feature layer to the final result: ;

[0060] The weights of the c channels in the feature map group are obtained by global average pooling: ;

[0061] wherein represents the size of the feature map and serves as a weight scaling function, ; represents the value corresponding to the i-th row and j-th column on the feature map;

[0062] The feature map is weighted using the weights of the c channels, and then activated by a ReLU activation function to obtain the activation response heat map k (w, h) of the abnormal category on the feature map group. The activation response heat map k (w, h) is used to locate the area that determines the final abnormal category classification, i.e., the abnormal occurrence area , .

[0063] Further, the specific method of step 4) in this example includes:

[0064] Step 4.1) As time goes by, the transmission line monitoring system uses the method of step 2) and step 3) in the alarm mode to obtain the activation response heat map corresponding to multiple moments, which is recorded as , where t is the time; these activation response heat maps reflect the changes in the spatial position of the abnormal areas considered by the algorithm;

[0065] Step 4.2) Heatmap of activation responses at multiple moments in time Sort and get time series samples .

[0066] Furthermore, the specific method of step 5) in this example includes:

[0067] Obtain the difference between adjacent activation response heat maps in the time series sample, and dynamically determine whether to suppress repeated alarms in combination with the temperature threshold and the area threshold: If the value of each coordinate point in the difference value is less than the temperature threshold, suppress the subsequent alarm; otherwise, compare the abnormal area area of ​​all coordinate points in the difference value that are greater than or equal to the temperature threshold with the area threshold. If the area of ​​the abnormal area exceeds the area threshold, a continuous alarm is issued.

[0068] Step 5.1) Separate the time series samples Each activation response heat map is recorded as 、 、 ,..., ; 、 、 ,..., Represent the activation response heatmaps corresponding to the 1st, 2nd, 3rd, ..., nth moments respectively; take the absolute value difference of the activation response heatmaps corresponding to two adjacent moments, and the absolute value difference between the activation response heatmaps corresponding to the tth moment and the t-1th moment The expression is: ,in, and are the activation response heat maps corresponding to the t-th moment and the t-1-th moment, ;

[0069] Step 5.2) Given a temperature threshold T, if If the value of each coordinate point is less than T, it means that the activation response heat map at the two moments has not changed significantly, and the subsequent alarm is suppressed; if If there is a point with a coordinate value greater than T, it means that the activation response heat map at two moments has changed locally, and the abnormal location is changing or spreading; it can be expressed as: , the alarm is suppressed; , the abnormal region is expanding.

[0070] Step 5.3) Given a region area threshold A, in step 5.2) If there is a coordinate point whose value is greater than T, then determine whether the pixel area of the abnormal region surrounded by the coordinate points greater than T is greater than A. If greater than A, it indicates that the abnormal region is undergoing a large change, and the alarm is continued. If less than A, it indicates that the abnormal region has a slight change or no change, indicating that the abnormal situation has not spread, and subsequent alarms are suppressed. This can be expressed as: Then the abnormality is rapidly expanding. Then the abnormality is slowly expanding, and the alarm is suppressed.

[0071] Further, the specific method of step 6) in this example includes:

[0072] Record the difference between the current time and the last time the abnormality was found. If the difference exceeds the time threshold, it indicates that the abnormality has disappeared, and the alarm mode is automatically exited and the monitoring mode is resumed.

[0073] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or replacements within the technical scope disclosed by the present application can be easily thought of by those skilled in the art, and should be covered within the protection scope of the present application.

Claims

1. A general repeated alarm suppression method for a transmission line monitoring system based on time series sample analysis, wherein the transmission line monitoring system is provided with a monitoring mode, wherein the monitoring mode uses a deep learning network to output a feature map group of the transmission channel, characterized in that: The method comprises: Step 1) setting an alarm mode in the transmission line monitoring system, and triggering the alarm mode and issuing a first alarm when an abnormality is detected in the monitoring mode; Step 2) In the alarm mode, a feature map group is obtained for indicating that the confidence level of the channel representing the abnormal category in the deep learning network is greater than the confidence level of the channel representing the normal category. The specific method of step 2) is as follows: Step 2.1) The deep learning network includes a classification and object detection network and a segmentation network. The output channel of the classification and object detection network is a one-dimensional vector V (1×c), where c in vector V represents the corresponding category. The output channel of the segmentation network is a three-dimensional matrix A (n×m×c), where c in the matrix represents the corresponding category confidence. Step 2.2) Obtain the feature map group of the final output of the deep learning network , where w represents the width of the feature map, h represents the height of the feature map, and c represents the number of channels of the feature map; Step 2.3) Select the channel with the highest confidence in the category channel c in step 2.1) as the category classification. When an alarm is issued, this category is considered an abnormal category. ; Step 3) constructing an activation response heat map based on the feature map group; Step 4) collecting activation response heat maps at multiple moments to form time series samples; the specific method of step 4) includes: Step 4.1) The transmission line monitoring system obtains activation response heat maps corresponding to multiple moments in the alarm mode, which are recorded as , where t is the time; Step 4.2) Heatmap of activation responses at multiple moments in time Sort and get time series samples ; Step 5) analyzing the time series samples to obtain difference values ​​between adjacent activation response heat maps in the time series samples, and suppressing repeated alarms or continuous alarms based on the difference values ​​and in combination with given temperature thresholds and area thresholds; the specific method of step 5) includes: Step 5.1) Separate the time series samples Each activation response heat map is recorded as 、 、 ,..., ; 、 、 ,..., Represent the activation response heat maps corresponding to the 1st, 2nd, 3rd, ..., nth moments respectively; take the activation response heat maps corresponding to two adjacent moments as the absolute value difference ; Step 5.2) Given a temperature threshold T, if If the value of each coordinate point is less than T, it means that the activation response heat map at the two moments has not changed significantly, and the subsequent alarm is suppressed; if If there is a point with a coordinate value greater than T, it means that the activation response heat map at the two moments has changed locally, and the abnormal location is changing or spreading; Step 5.3) Given a region area threshold A, in step 5.2) If the value of a coordinate point is greater than T, the pixel area is determined to be greater than A. If it is greater than A, it indicates that the abnormal area is undergoing significant changes, and a continuous alarm is issued. If it is less than A, it indicates that the abnormal area has slightly changed or has not changed, indicating that the abnormal situation has not spread, and subsequent alarms are suppressed. Step 6) When the anomaly disappears and exceeds the time threshold, the system automatically exits the alarm mode and resumes the monitoring mode.

2. The universal repeated alarm suppression method for a power transmission line monitoring system based on time series sample analysis according to claim 1 is characterized in that: The specific method of step 3) is as follows: The exception category For feature map group Finding the gradient : ; Get the feature map group through global average pooling The weights of the c channels in ; Using the weights of c channels , for the feature map Weighted, and then through the ReLU activation function, get the abnormal category The activation response heat map k(w, h) on the feature map group is used to locate the area that plays a decisive role in the final abnormality classification, that is, the abnormality occurrence area , .

3. The universal repeated alarm suppression method for a power transmission line monitoring system based on time series sample analysis according to claim 2, characterized in that: The feature map group The weights of the c channels in The calculation formula is as follows: ;in, represents the size of the feature map, ; Represents the value corresponding to the i-th row and j-th column on the feature map.

4. The universal repeated alarm suppression method for a power transmission line monitoring system based on time series sample analysis according to claim 3 is characterized in that: The absolute value difference between the activation response heatmaps corresponding to time t and time t-1 The expression is: ； in, and are the activation response heat maps corresponding to the t-th moment and the t-1-th moment, .

5. The universal repeated alarm suppression method for a power transmission line monitoring system based on time series sample analysis according to claim 1, characterized in that: The specific method of step 6) includes: Record the difference between the current time and the time when the last abnormality was discovered. If the difference exceeds the time threshold, the system will automatically exit the alarm mode and resume the monitoring mode.

Images