A power transmission line anti-collision on-line monitoring system based on compressed sensing

Abstract

The application discloses a kind of based on compressive sensing's power transmission line anti-collision online monitoring system, including front-end monitoring module and background processing module, front-end monitoring module includes camera module, front-end processor and 4G communication module connected in turn;Background processing module includes background server and the early warning module connected with it, main idea is that camera module is monitored to the region in real time, when monitoring movement target, front-end processor extracts the pixel distribution characteristics of this target, and is transmitted to background server.After background server receives data, the feature database and the feature data of target are identified by compressive sensing algorithm, and then output identification result, facilitate staff to take timely disposal measures.Based on the power transmission line anti-collision online monitoring system of compressive sensing of the application, it is easy to realize, and it has higher power transmission line impact threat sample identification precision, can provide more reliable early warning information for power transmission line operation and inspection staff.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of online monitoring of power systems, and relates to a power transmission line anti-collision online early warning system, in particular to a power transmission line anti-collision online monitoring system based on compressed sensing. BACKGROUND

[0002] As the part farthest away in the power transmission process, the highest voltage level and the most complex environment, the power transmission link has been plagued by various security threats. According to statistics, among the factors causing the tripping of the power transmission line, the main ones include extreme weather, tall trees, illegal construction, etc., among which human external impact accounts for nearly one fourth. In human external impact events, nearly 80% are caused by illegal construction of large engineering machinery, and often produce permanent fault points, which not only seriously threaten the safe and stable operation of the power grid, but also greatly endanger the personal safety of construction personnel. Therefore, it is necessary to take anti-collision early warning measures for the power transmission line.

[0003] For a long time, the application of intelligent online monitoring methods in the power transmission field has been relatively weak. At present, the anti-collision of the power transmission line mainly relies on artificial on-site line patrol, artificial background video monitoring and other methods, which all need to consume a lot of human resources and have a limited monitoring coverage. SUMMARY

[0004] The purpose of the present application is to overcome the defects of the prior art, provide a power transmission line anti-collision online monitoring system based on compressed sensing, which is easy to implement and has high recognition accuracy of power transmission line impact threat samples, can provide more reliable early warning information for power transmission line operation and inspection personnel, so that the personnel can timely handle the dangerous factors around the line, thereby providing reliable protection for the safe and stable operation of the power transmission line.

[0005] The technical solution to achieve the above purpose is: a power transmission line anti-collision online monitoring system based on compressed sensing, comprising a front-end monitoring module and a background processing module, wherein:

[0006] The front-end monitoring module comprises a camera module, a front-end processor and a 4G communication module connected in sequence;

[0007] The background processing module comprises a background server and an early warning module connected thereto;

[0008] The camera module monitors the monitored area in real time, and when a moving target is monitored, sends the image of the monitored area to the front-end processor;

[0009] The front-end processor extracts the pixel distribution features of the foreign matter by using the collected image of the monitored area, and transmits the pixel distribution features of the foreign matter to the background server through the 4G communication module.

[0010] When the background server receives the pixel distribution feature of the foreign matter, the feature database and the completed compressed sensing recognition algorithm are called to identify the target feature, to determine whether the foreign matter is a threat, and to output the warning result to the warning module according to the determination result of the foreign matter. The staff executes corresponding disposal measures through the warning information of the warning module to timely handle the dangerous factors around the line.

[0011] The above-mentioned power line anti-collision online monitoring system based on compressed sensing, wherein the feature database is an offline feature database, the feature database stores historical foreign matter target feature data, the data structure of which is consistent with the pixel distribution feature of the foreign matter extracted by the front-end processor, and the determination result of the foreign matter is included. Each identification of the background server can update and learn the offline feature database to enhance the recognition ability of the compressed sensing recognition algorithm.

[0012] The above-mentioned power line anti-collision online monitoring system based on compressed sensing, wherein when the front-end processor extracts the pixel distribution feature of the foreign matter by using the collected image of the monitored area, the pixels of the image of the monitored area are converted from RGB space to HSV space, R represents red, G represents green, B represents blue, H represents hue, S represents saturation, and V represents intensity. The conversion formula between RGB and H value is:

[0013]

[0014] Through the conversion of formula (1), each pixel point corresponds to an H value, and a pixel point distribution feature map is drawn accordingly.

[0015] In order to better represent the pixel distribution feature and make the images with different total number of pixels comparable, the number of pixel points is normalized, and the number of pixel points corresponding to each H value is divided by the maximum value of the number of pixel points.

[0016] The above-mentioned power line anti-collision online monitoring system based on compressed sensing, wherein when the background server identifies the target feature, the compressed sensing recognition algorithm is used to identify the target feature in the feature database that is most similar to the pixel distribution feature of the foreign matter this time, including two cases:

[0017] (1) The identified target type is not a threat, so the result of this target feature identification is not a threat, and the staff does not need to confirm and handle it;

[0018] (2) if the identified target type is a threat, the similarity of the target is set as a threshold for secondary judgment, if the threshold is exceeded, the target feature recognition result of this time is a threat, and the staff needs to handle it; if the threshold is not exceeded, it is determined that there is no threat.

[0019] The power line anti-collision online monitoring system based on compressed sensing, wherein the compressed sensing recognition algorithm uses the function of reconstructing data by CS algorithm to realize high-precision recognition of target features, and the modeling process of the compressed sensing recognition algorithm is as follows:

[0020] For a target recognition process, a 300-dimensional feature vector composed of pixel distribution characteristics is collected, as shown in the following formula:

[0021]

[0022] In formula (2), x j represents the feature vector of the jth target, represents the pixel distribution data in the pixel distribution characteristics;

[0023] Then, the target feature vectors in the feature database form a database matrix Ψ:

[0024]

[0025] In formula (3), N represents the number of targets in the offline database;

[0026] Let the feature data extraction vector of the jth target be S j :

[0027] S j = [0,...,0,1,0,...,0] T (4)

[0028] In formula (4), the jth element is 1;

[0029] According to formulas (2)-(4), we have:

[0030] x j = ΨS j (5)

[0031] In the actual recognition process, the known quantities are x and Ψ, S is reconstructed by CS algorithm, the target feature with the highest similarity in the feature database is found through the data characteristics of S, in order to realize this process, the standard form of CS algorithm is constructed on the basis of formula (5), and the m*300-dimensional observation matrix Φ is multiplied on both sides of formula (5) to obtain:

[0032] Φx j = ΦΨS j (6)

[0033] Let T = Φx, Θ = ΦΨ, then:

[0034] T = ΘS j (7)

[0035] In formula (7), Θ is called a sensing matrix, and the dimension is m*N, in order to embody the compression of the CS algorithm, m is much smaller than 300, and Φ is selected as a Gaussian random matrix;

[0036] Formula (7) is a standard form of the CS algorithm, and S can be reconstructed by knowing T and Θ.

[0037] Formula (7) is an underdetermined equation, and the equation cannot be directly solved, and the reconstructed vector S is a vector with a sparsity of 1, and the reconstruction problem is converted into a minimum l1 norm optimization problem, and the solving model is:

[0038] min||S||1 s.t.T = ΘS (8)

[0039] The optimization solving process is the principle of gradually approaching the original data vector through multiple iterations, and the selected solving algorithm is the orthogonal matching pursuit algorithm, and the reconstructed vector S is counted as S':

[0040] S' = argmin||S||1 s.t.T = ΘS (9)

[0041] Due to the error in the reconstruction process, the sparsity of S' is not 1, but a vector composed of a value close to 1 and a plurality of small non-zero values, and the element position corresponding to the maximum value in the vector is selected as the most similar target feature vector in the feature database, and the similarity between the value and 1 is selected as the similarity between the targets.

[0042] The power line anti-collision online monitoring system based on compressed sensing of the application has relatively small requirements for software and hardware, is easy to implement, has high recognition accuracy of power line impact threat samples, can provide reliable early warning information for power line operation and inspection personnel, so that the personnel can timely handle the dangerous factors around the line, thereby providing reliable protection for the safe and stable operation of the power line. BRIEF DESCRIPTION OF DRAWINGS

[0043] Figure 1 It is a block diagram of the power line anti-collision online monitoring system based on compressed sensing of the application;

[0044] Figure 2 It is a schematic view of an image of a crane to be grabbed;

[0045] Figure 3 It is a pixel distribution feature map of a crane grabbed in soil;

[0046] Figure 4The result graph of the test accuracy of the compressed sensing algorithm in the experiment is verified. DETAILED DESCRIPTION

[0047] In order to enable the person skilled in the art to better understand the technical solutions of the present application, the specific embodiments thereof will be described in detail below with reference to the accompanying drawings:

[0048] Please refer to Figure 1 , Figure 2 and Figure 3 , the best embodiment of the present application, a power line anti-collision online monitoring system based on compressed sensing, comprising a front-end monitoring module 1 and a background processing module 2, the front-end monitoring module 1 comprising a camera module 11, a front-end processor 12 and a 4G communication module 13 connected in sequence; the background processing module 2 comprises a background server 21 and a pre-warning module 22 connected thereto; the camera module 11 monitors the monitored area in real time, and sends the image of the monitored area to the front-end processor 12 when a moving target is monitored; the front-end processor 12 extracts the pixel distribution characteristics of the foreign matter by using the collected image of the monitored area, and transmits the pixel distribution characteristics of the foreign matter to the background server 21 through the 4G communication module 13; when the background server 21 receives the pixel distribution characteristics of the foreign matter, the feature database and the completed compressed sensing recognition algorithm are called to identify the target characteristics, to judge whether the foreign matter is a threat, and to output the pre-warning result to the pre-warning module 22 according to the judgment result of the foreign matter, and the staff executes the corresponding disposal measures through the pre-warning information of the pre-warning module 22 to timely handle the dangerous factors around the line.

[0049] The feature database is an offline feature database, and the historical collected foreign matter target feature data is stored in the feature database, the data structure of which is consistent with the pixel distribution characteristics of the foreign matter extracted by the front-end processor 12, and the judgment result of the foreign matter is contained, and each identification of the background server 21 can update and learn the data of the offline feature database to enhance the recognition ability of the compressed sensing recognition algorithm.

[0050] Pixel distribution feature extraction:

[0051] The color in the image is usually described by RGB (red, green, blue) three primary color pixels, and each primary color has a value range of 0-255. In order to represent the color characteristics of the pixel with a value, spatial conversion is needed.

[0052] When the front-end processor 12 extracts the pixel distribution characteristics of the foreign matter by using the collected image of the monitored area, the pixels of the image of the monitored area are converted from RGB space to HSV space, R refers to red, G refers to green, B refers to blue, H refers to hue, S refers to saturation, and V refers to intensity, and the hue H is taken, and the conversion formula between RGB and H value is:

[0053]

[0054] Through the conversion of formula (1), each pixel point corresponds to an H value, thereby drawing a pixel point distribution feature map; in order to better represent the pixel distribution feature, and make the images with different total numbers of pixels comparable, the number of pixel points is normalized, and the number of pixel points corresponding to each H value is divided by the maximum value of the number of pixel points.

[0055] Please refer to Figure 2 and Figure 3 , the crane is taken as an example, after the conversion of formula (1), the pixel distribution characteristics are as shown in Figure 2 . Figure 3

[0056] Target feature recognition based on compressed sensing:

[0057] When the background server 21 performs target feature recognition, the target feature recognition algorithm based on compressed sensing is used to identify the target feature in the feature database that is most similar to the pixel distribution feature of the foreign matter this time, including two cases:

[0058] (1) the identified target type is non-threatening, so the result of this target feature recognition is non-threatening, and the staff does not need to confirm and process it;

[0059] (2) the identified target type is threatening, then the similarity of the target is set to a threshold for secondary judgment, if it exceeds the threshold, the result of this target feature recognition is threatening, and the staff needs to process it; if it does not exceed the threshold, it is determined to be non-threatening.

[0060] The above is the overall logical architecture of the target feature recognition algorithm of the online monitoring system of the application. The most core part is the identification of the offline feature database, and the accuracy of this step directly affects the judgment of the final result. The compressed sensing (CS) theory points out that if the signal can meet the sparsity requirement, then the partial data collected below the Nyquist sampling frequency is used to restore the complete data of the signal. The power transmission line anti-collision online monitoring system based on compressed sensing of the application redesigns the CS algorithm, and uses the data reconstruction function to realize high-precision identification of the target feature.

[0061] For a target recognition process, a 300-dimensional feature vector composed of pixel distribution features is collected, as shown in the following formula:

[0062]

[0063] In formula (2), x j represents the feature vector of the jth target, ​pixel distribution data in pixel distribution features;

[0064] Then, the target feature vector in the feature database constitutes a database matrix Ψ:

[0065]

[0066] In formula (3), N represents the number of targets in the offline database;

[0067] Let the feature data extraction vector of the jth target be S j :

[0068] S j = [0,...,0,1,0,...,0] T (4)

[0069] In formula (4), the jth element is 1;

[0070] Comprehensive formula (2) to formula (4) is:

[0071] x j = ΨS j (5)

[0072] In the actual recognition process, the known quantities are x and Ψ, S is reconstructed through the CS algorithm, and the target feature with the highest similarity in the feature database is found through the data features of S. In order to realize this process, the standard form of the CS algorithm is constructed on the basis of formula (5), and the m*300-dimensional observation matrix Φ is multiplied on both sides of formula (5) to obtain:

[0073] Φx j = ΦΨS j (6)

[0074] Let T = Φx and Θ = ΦΨ, then:

[0075] T = ΘS j (7)

[0076] In formula (7), Θ is called a sensing matrix, and the dimension is m*N. In order to embody the compression of the CS algorithm, m is much smaller than 300, and Φ is selected as a Gaussian random matrix;

[0077] Formula (7) is the standard form of the CS algorithm, and S can be reconstructed when T and Θ are known;

[0078] Formula (7) is an underdetermined equation, and the equation cannot be directly solved. The reconstructed vector S is a vector with a sparsity of 1. The reconstruction problem is converted into a minimum l1 norm optimization problem, and the solution model is:

[0079] min ||S||1 s.t. T = ΘS (8)

[0080] The principle of the optimization solving process is to gradually approach the original data vector through multiple iterations, and the selected solving algorithm is an orthogonal matching pursuit algorithm (OMP), and the reconstructed vector S is denoted as S':

[0081] S' = argmin||S||1 s.t.T = ΘS (9)

[0082] Due to the error in the reconstruction process, the sparsity of S' is not 1, but a vector composed of a value close to 1 and several small non-zero values, and the element position corresponding to the maximum value in the vector is selected as the most similar target feature vector in the feature database, and the similarity between the value and 1 is used as the similarity between the targets.

[0083] Test verification:

[0084] In order to verify the effectiveness of the power line anti-collision online monitoring system based on compressed sensing of the application, first, an offline feature database for identification is established, and a test database for verifying the algorithm is established. Through field collection and processing of historical samples, an offline feature database containing 976 samples and a test database containing 273 samples are established. The database contains samples of various mechanical equipment in various environments, which can cover most types of power line large mechanical impact accidents. Among the 273 samples in the test database, there are 76 threatening samples and 197 non-threatening samples.

[0085] The compressed sensing algorithm directly outputs a similarity value, and the final result needs to be compared with the set threshold value. Therefore, different thresholds have a significant impact on the accuracy of the algorithm output. Therefore, the power line anti-collision online monitoring system based on compressed sensing of the application counts the recognition accuracy under different thresholds during testing.

[0086] Please refer to Figure 4 , in order to verify the effectiveness of the CS algorithm used in the application, BP neural network and K-Nearest Neighbor (KNN) two pattern recognition algorithms are selected for comparison and verification in the test. The recognition test results are shown in Figure 4 .

[0087] From the results in Figure 4 , it can be concluded that the recognition accuracy of the three algorithms increases first and then decreases with the increase of the threshold value. The threshold values of CS, BP and KNN algorithms to reach the maximum accuracy are 86%, 80% and 78% respectively, and the corresponding recognition accuracies are 96.34%, 91.94% and 87.55% respectively.

[0088] It can be seen from the above results that the power transmission line anti-collision online monitoring system based on compressed sensing has an advantage in the recognition effect of the CS algorithm, which not only has the largest accuracy value, but also has the largest corresponding threshold value. This indicates that the CS algorithm can output a higher similarity value when identifying the target. When the threshold value is lower, the number of "misjudgments" will increase, and when the threshold value is higher, there will be no excessive "omissions". Therefore, the CS algorithm has the highest robustness.

[0089] In summary, the power transmission line anti-collision online monitoring system based on compressed sensing is easy to implement and has high recognition accuracy of power transmission line impact threat samples. It can provide more reliable early warning information for power transmission line operation and inspection personnel, so that they can timely handle the dangerous factors around the line, thereby providing reliable protection for the safe and stable operation of the power transmission line.

[0090] Those skilled in the art of the present technology should recognize that the above embodiments are only used to illustrate the present application, and are not used as a limitation of the present application. Any changes and modifications to the above described embodiments within the spirit and principles of the present application will fall within the scope of the claims of the present application.

Claims

1. A power transmission line anti-collision online monitoring system based on compressed sensing, characterized in that, It comprises a front-end monitoring module and a background processing module, wherein: The front-end monitoring module comprises a camera module, a front-end processor and a 4G communication module connected in sequence; The background processing module comprises a background server and a pre-warning module connected with the background server; The camera module performs real-time monitoring on the monitored area, and sends the image of the monitored area to the front-end processor when a moving target is monitored; The front-end processor extracts the pixel distribution feature of the foreign matter by using the collected image of the monitored area, and transmits the pixel distribution feature of the foreign matter to the background server through the 4G communication module; When the background server receives the pixel distribution feature of the foreign matter, it calls the feature database and the modeling completed compressed sensing recognition algorithm to perform target feature recognition, judges whether the foreign matter is a threat, and outputs the pre-warning result to the pre-warning module according to the judgment result of the foreign matter, so that the staff can execute corresponding disposal measures through the pre-warning information of the pre-warning module to timely handle the dangerous factors around the line; When the front-end processor extracts the pixel distribution feature of the foreign matter by using the collected image of the monitored area, the pixels of the image of the monitored area are converted from RGB space to HSV space, R represents red, G represents green, B represents blue, H represents hue, S represents saturation, and V represents intensity, and the hue H is taken, and the conversion formula between RGB and H value is: （1） Through the conversion of formula (1), each pixel point corresponds to an H value, and a pixel distribution feature map is drawn accordingly; In order to better represent the pixel distribution feature and make the images with different total number of pixels comparable, the number of pixel points is normalized, and the number of pixel points corresponding to each H value is divided by the maximum value of the number of pixel points; The compressed sensing recognition algorithm realizes high-precision recognition of target features by using the function of reconstructing data by CS algorithm, and the modeling process of the compressed sensing recognition algorithm is as follows: For a target recognition process, a 300-dimensional feature vector composed of pixel distribution features is collected, as shown in the following formula: （2） In formula (2), x j represent a feature vector of the j th target, Then, the target feature vectors in the feature database form a database matrix Ψ: 1,j , According to formula (2)-(4), we have: 2,j ,…, s.t. T 300,j represent pixel distribution data in the pixel distribution feature; s.t. T In formula (3), N represents the number of targets in the feature database; Set the feature data extraction vector of the j S j :​ （4） In formula (4), the first element is 1. j the first element is 1. The feature database is an offline feature database, and the feature database stores historical foreign matter target feature data, the data structure of which is consistent with the pixel distribution feature of the foreign matter extracted by the front-end processor, and contains the judgment result of the foreign matter, and each identification of the background server can update and learn the offline feature database to enhance the recognition ability of the compressed sensing recognition algorithm. （5） In the actual identification process, the known quantity is x j and Ψ, reconstructing by CS algorithm S j , finding the most similar target feature in the feature database by the data characteristics of S j In order to achieve this process, the standard form of CS algorithm is constructed on the basis of formula (5), and both sides of formula (5) are multiplied by m ×300-dimensional observation matrix Φ: （6） Let T =Φ x j , Θ = ΦΨ, then we have: （7） In formula (7), Θ is called a sensing matrix, with a dimension of m × N In order to embody the compressibility of the CS algorithm, m much smaller than 300, and Φ is a Gaussian random matrix. Equation (7) is the standard form of the CS algorithm, where T and Θ, i.e., can be reconstructed S j ; Equation (7) is an underdetermined equation, which cannot be solved directly. The reconstructed vector S j is a vector with sparsity 1. The reconstruction problem is converted into a least l 1-norm optimization problem, and the solution model is min‖ S j ‖1 When the background server performs target feature recognition, it uses the compressed sensing recognition algorithm to identify the target feature in the feature database that is most similar to the pixel distribution feature of the foreign matter, including two cases: =Θ S j (8) The optimization solution process is to gradually approach the original data vector by multiple iterations. The selected solution algorithm is the orthogonal matching pursuit algorithm, and the reconstructed vector S j The count is S j ' S j = argmin ||y - Xb||2 S j = ||b||1 (1) The identified target type is not a threat, so the result of the target feature recognition is not a threat, and the staff does not need to confirm and handle it; = Θ S j (9) Since the process of reconstruction has errors, S j The sparsity is not 1, but a vector consisting of a value close to 1 and several smaller non-zero values. The element position corresponding to the maximum value in the vector is selected as the most similar target feature vector in the feature database, and the similarity between the value and 1 is used as the similarity between the targets.

2. The power line anti-collision on-line monitoring system based on compressed sensing according to claim 1, characterized in that, ​ 3. The power line anti-collision on-line monitoring system based on compressed sensing according to claim 1, characterized in that, ​ ​ (2) if the identified target type is a threat, a threshold is set for the similarity of the target for secondary judgment, and if the threshold is exceeded, the target feature recognition result of this time is a threat, and a staff needs to handle it; If the threshold is not exceeded, it is determined that there is no threat.

Images