Tree and bamboo risk identification method for power transmission line

Abstract

The present application belongs to the technical field of power transmission line inspection, and particularly relates to a tree and bamboo risk identification method for power transmission lines. The present application comprises data acquisition, data processing, feature engineering, and tree and bamboo risk identification. In the feature part, the present application introduces vegetation index data with a short update frequency and high-resolution DEM data, extracts the statistical features of slope, slope direction, elevation, terrain position index, terrain roughness index, and vegetation index, and extracts the terrain and vegetation coverage around the tower in multiple directions and multiple scales. In the model training part, under the condition of serious imbalance between positive and negative samples, the method of equally dividing negative samples and then merging positive samples for model training proposed by the present application avoids overfitting in oversampling or loss of part of the sample information in undersampling.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of power transmission line inspection, and particularly relates to a tree and bamboo risk identification method for a power transmission line. BACKGROUND

[0002] In the operation of a power transmission line, when the distance between a conductor and a tree is small, especially less than the specified minimum safety distance, a short circuit of the line may occur, which may further cause tripping, the line cannot normally supply power, and the production and life of users are affected, and in serious cases, life safety is threatened. Therefore, the identification of tree and bamboo risks is particularly important for the normal operation of a power transmission line.

[0003] The identification of tree and bamboo risks generally adopts three ways of manual distance measurement, online device observation and model identification to identify tree and bamboo risks. However, manual field distance measurement has a high cost, and is generally measured once every few months, which is difficult to meet the demand of line normalization inspection. The online device observation has a large error in identifying tree and bamboo risks due to the limitations of the shooting angle of the device, the weather and the experience of the inspection personnel. The data of model identification generally comes from laser radar data, sensor data, satellite images and the like, but is limited by the flight frequency of the unmanned aerial vehicle, the number of installed sensors and the identification difficulty of the image, and therefore the accuracy and update frequency of model identification are not very ideal. SUMMARY

[0004] In view of the above problems, the present application provides a tree and bamboo risk identification method for a power transmission line based on the existing tree and bamboo felling records and based on vegetation index and topographic features.

[0005] The technical scheme of the present application is as follows:

[0006] A tree and bamboo risk identification method for a power transmission line, as shown in Figure 1 The method comprises the following steps:

[0007] S1, data acquisition: collecting tree and bamboo felling records, DEM elevation data and vegetation index data for all towers within a specified range, wherein the tree and bamboo felling records include tower number, longitude, latitude, discovery time, felling time and tree and bamboo risk description, the DEM elevation data includes longitude, latitude and DEM value, and the vegetation index data includes longitude, latitude, time and vegetation index; and the tower number, longitude and latitude are integrated to form tower basic data;

[0008] S2, data processing:

[0009] Tree and bamboo felling area acquisition: acquiring specific position information of tree and bamboo felling according to the tower basic information and the tree and bamboo risk description in the tree and bamboo felling records, and updating the tree and bamboo felling records based on the obtained position information;

[0010] Data space-time matching: matching the updated tree and bamboo felling record, DEM data and vegetation index of each time to obtain space-time matching data;

[0011] Division of positive and negative samples: the obtained space-time matching data is divided into positive and negative samples;

[0012] S3, feature extraction is performed according to the obtained sample, and feature selection is performed, so as to obtain sample features;

[0013] S4, the obtained sample features are used to train a recognition model, and according to the obtained recognition model, the obtained tower data, vegetation index data and DEM data are used for tree and bamboo risk identification.

[0014] The beneficial effects of the present application are:

[0015] Compared with the existing tree and bamboo risk identification method, the present application is not affected by the external environment, the obtained data is stable and continuous, and is more easily used in power transmission line inspection, which helps to improve the inspection efficiency. The present application extracts the statistical features of vegetation index and DEM data, divides the positive and negative samples based on the existing tree and bamboo felling record, and identifies the tree and bamboo risk through the supervised learning method. Among them, in the feature generation part, the terrain and vegetation coverage around the tower are extracted in multiple directions and multiple scales; in the model training part, in the case of serious imbalance of positive and negative samples, the method of dividing the negative samples and then merging the positive samples for model training is used to avoid overfitting in oversampling or loss of part of the sample information in undersampling in general cases. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 is the method flowchart of the present application. DETAILED DESCRIPTION

[0017] The technical scheme of the present application will be described in detail below.

[0018] The present application provides a tree and bamboo risk identification method for power transmission lines, including data acquisition, data processing, feature engineering and tree and bamboo risk identification. In the feature part, the present application introduces vegetation index data with short update frequency and high-resolution DEM data, extracts the statistical features of slope, slope direction, elevation, terrain position index, terrain roughness index and vegetation index, and extracts the terrain and vegetation coverage around the tower in multiple directions and multiple scales. In the model training part, in the case of serious imbalance of positive and negative samples, the method of dividing the negative samples and then merging the positive samples for model training proposed by the present application avoids overfitting in oversampling or loss of part of the sample information in undersampling.

[0019] The specific method is:

[0020] 1. Data acquisition

[0021] The present application collects the tree and bamboo cutting records, DEM elevation data, and vegetation index data of all towers in a specified range within a certain time period, and then constructs a training sample set based on the tree and bamboo cutting records of each tower and the DEM elevation data and vegetation index data of the corresponding position.

[0022] The tree and bamboo cutting records include tower number, longitude, latitude, discovery time, cutting time, and tree and bamboo risk description, the DEM elevation data includes longitude, latitude, and DEM value, and the vegetation index data includes longitude, latitude, time, and vegetation index. The spatial resolution of the DEM data is 5 meters, and the spatial resolution of the vegetation index data is 10 meters with an update frequency of 5 days.

[0023] Since the position information recorded in the tree and bamboo cutting records only includes the longitude and latitude of the current tower, but the location of the tree and bamboo risk mostly lies in the channel between two towers, it is necessary to obtain the longitude and latitude of the adjacent tower of each tower in the tree and bamboo cutting records, i.e. the tower basic data, which mainly includes tower number, longitude, and latitude.

[0024] 2. Data processing

[0025] Data processing checks and eliminates error data in the data through certain methods, mainly including the following steps:

[0026] 2.1 Data repeatability check. It refers to the absence of repeated data, and the repeated data is deleted.

[0027] 2.2 Missing value processing. Since the DEM and vegetation index data are raster data, the average value near the missing point is used for filling.

[0028] 2.3 Abnormal value processing. Data with extremely small or impossible probability of occurrence in a certain area within a certain time period is removed.

[0029] 2.4 Tree and bamboo cutting area acquisition. According to the tower basic information and the tree and bamboo risk description in the tree and bamboo cutting records, the specific position information of the tree and bamboo cutting is obtained.

[0030]

[0031]

[0032] ,

[0033]

[0034] Among them , Total number of tree bamboo felling records. Discovery time, Felling time. Tree bamboo felling records are processed as follows: for example, in a tree bamboo felling record Time Tree bamboo risk description as "large side 50 meters away from tree bamboo ultra-high, felling range 3 meters", from the tower foundation data Large side adjacent tower , get the straight line to and the distance from is 50 meters, the latitude and longitude of the point is recorded as , and the felling range is recorded as =3 meters. Therefore, the processed tree bamboo felling record is:

[0035]

[0036]

[0037] ,

[0038]

[0039] 2.5 Spatio-temporal matching of data. Since DEM data has spatial attributes, tree bamboo felling records and vegetation index data have both temporal and spatial attributes, and spatio-temporal matching of the three types of data is required. For each record of the result of step 2.4 Match the DEM data and the vegetation index of each time, the time series of the vegetation index , the location of the DEM is:

[0040]

[0041]

[0042] wherein, and represent the row number and column number of the grid data DEM respectively.

[0043] The location of the vegetation index is:

[0044]

[0045]

[0046] wherein, and represent the row number and column number of the grid data vegetation index respectively.

[0047] The result of spatio-temporal matching is as follows:

[0048]

[0049] wherein,

[0050] and the following principles are satisfied:

[0051]

[0052]

[0053] wherein, the function represents the spherical distance of the coordinate .

[0054] 2.6 Division of positive and negative samples.

[0055] wherein, is the vegetation index time, is the discovery time, is the felling time, if is satisfied, the corresponding target variable is a positive sample, otherwise, it is a negative sample.

[0056] 3. Feature engineering

[0057] Feature engineering mainly includes feature generation and feature selection.

[0058] 3.1 Feature generation

[0059] Extract features in DEM, such as slope, aspect, elevation, etc. These feature matrices are consistent with the format of DEM data, so the corresponding slope, aspect, elevation, terrain position index, terrain roughness index values can be extracted according to in . Similarly, according to extract the vegetation index value. Since these features are not specific values, but a column of values, statistical features are extracted, such as the range of elevation, the variance of terrain roughness index, the average of vegetation index, etc. The finally generated feature matrix is:

[0060]

[0061] wherein, represents the feature matrix has features.

[0062] 3.2 Feature selection

[0063] Feature matrix in 3.1 Feature selection is performed. Random Forest is selected to perform feature selection on the features in the feature matrix

[0064] 4. Tree and bamboo risk identification

[0065] 4.1 Data set division

[0066] The positive and negative samples divided in step 2.6 have a serious sample imbalance problem, that is, the negative samples are much more than the positive samples. Therefore, when performing cross-validation, the present application first divides the negative samples equally, and the number of each part is close to that of the positive samples, then sequentially combines each part of the negative samples and the positive samples to form a new data set, and divides the training set and the test set according to a 7:3 ratio according to the new data set.

[0067] 4.2 Model training

[0068] The present application adopts Logistic Regression to fit and train the training set of each cross-validation, and then performs prediction test on the test set by using the trained model, and finally selects a model with convergence, high prediction accuracy and good robustness.

[0069] After the obtained tower data, vegetation index data and DEM data are processed according to steps 2 and 3, the trained model is input, and the tree and bamboo risk identification result can be obtained.​​​

Claims

1. A power transmission line-oriented tree bamboo risk identification method, characterized in that, Comprise the following steps: S1, data acquisition: all towers in the specified range, gather tree bamboo cutting records, DEM elevation data, vegetation index data, wherein, tree bamboo cutting records contain tower number, longitude, latitude, discovery time, cutting time, tree bamboo risk description, DEM elevation data includes longitude, latitude, DEM value, vegetation index data contains longitude, latitude, time, vegetation index; The tower number, longitude, latitude are integrated to form the tower basic data; S2, data processing: Tree bamboo cutting area acquisition: according to the tower basic information and the tree bamboo risk description in the tree bamboo cutting record, the specific position information of the tree bamboo cutting is obtained, and the tree bamboo cutting record is updated based on the obtained position information; Data space-time matching: matching DEM data and vegetation index of each time for the updated tree bamboo cutting record, to obtain space-time matching data; Division of positive and negative samples: the obtained space-time matching data is divided into positive and negative samples; S3, according to the obtained sample, the feature extraction is carried out, and the feature selection is carried out, so as to obtain the sample feature; S4, the obtained sample feature is used to train the recognition model, and according to the obtained recognition model, the tower data, vegetation index data and DEM data are used for tree bamboo risk identification.

2. The method for identifying the risk of trees and bamboos towards power transmission lines according to claim 1, characterized in that, In step S2, the tree bamboo cutting record is defined as: , wherein , represents the total number of tree and bamboo felling records, is the tower number , is the longitude and latitude, is the discovery time, is the felling time, is the tree and bamboo risk description; the tree and bamboo felling record is processed as follows: according to the corresponding tree and bamboo risk description, the specific position information of the tree and bamboo felling is obtained based on the tower foundation information, and the corresponding longitude and latitude is updated to obtain the processed tree and bamboo felling record: , wherein, are the updated longitude and latitude; The resulting Each record of the , DEM data and the time series of vegetation indices , DEM data and the time series of vegetation indices wherein, and represent the row number and column number of the grid data DEM, respectively; The position of vegetation index is: wherein, and NDVIi,jdenotes the row and column number of the grid data vegetation index, respectively; The result of space-time matching is as follows: wherein and the following principles are met: where the function represents the spherical distance of the coordinates . Mid, is the vegetation index time, is the discovery time, is the felling time, if then the corresponding target variable is a positive sample, otherwise, it is a negative sample.

3. The method according to claim 2, wherein, The specific method of step S3 is: The features in the DEM are extracted, and the obtained feature matrix is consistent with the format of the DEM data. According to , the corresponding slope, slope direction, altitude, terrain position index and terrain roughness index values are extracted; according to , the vegetation index values are extracted, and the statistical feature extraction is performed on them. Finally, the generated feature matrix is:​ wherein represents a feature matrix there are features; The feature matrix obtained is Feature selection is performed on the feature matrix , and the random forest is used to perform feature selection on features in the feature matrix, and finally features are selected, .

4. The method according to claim 3, wherein, The specific method of step S4 is: Firstly, the negative samples are equally divided, and the number of each part is close to that of the positive samples, then the negative samples and the positive samples are combined into new data set in turn, and the new data set is divided into training set and test set according to the proportion of 7:3; The training set of each cross validation is fitted and trained by using logistic regression, and then the trained model is used for prediction test of the test set, and finally the model with good convergence, high prediction accuracy and good robustness is selected; After the obtained tower data, vegetation index data and DEM data are processed by steps S2 and S3, the trained model is input, and the tree bamboo risk identification result is obtained.

Images