Pipeline high-consequence area grade identification method

Abstract

The invention discloses a pipeline high-consequence area grade identification method, which specifically comprises the following steps of: 1, acquiring a high-resolution image, and making a label map and a division data set of a road and a house; 2, performing data preprocessing on the data set obtained in the step 1; step 3, constructing a convolutional layer and combining feature information; 4, pooling operation is carried out, and then layer-by-layer stacking is carried out; step 5, carrying out transposition convolution on the features stacked layer by layer in the step 4; and 6, if the features belong to roads, carrying out category judgment on the roads and judging the grade of the high-consequence area of the pipeline, and if the features belong to houses, carrying out connected area judgment on the house features, and judging the grade of the high-consequence area of the pipeline according to the number of the houses. According to the pipeline high-consequence area grade identification method, the problem that existing pipeline high-consequence area grade identification depends on manual judgment and cannot automatically judge is solved, and the automation degree of grade judgment is improved.

Description

Technical Field

[0001] The invention belongs to the field of oil pipeline safety and relates to a pipeline high consequence area grade identification method. Background Art

[0002] The high consequence area of ​​a pipeline refers to the area affected by a pipeline leak, such as residential areas and highways, and the high consequence area section refers to the section of the pipeline within this range. The impact of a pipeline leakage accident on the outside world depends on the distribution of the surrounding population and the fragility of the environment. Its hazards are reflected in three aspects: casualties, damage to infrastructure, and environmental pollution. Therefore, the attribute classification of the high consequence area of ​​an oil and gas pipeline should consider factors such as the degree of population concentration, facility use, environmental protection, and the severity of the high consequence area, and is divided into three levels: Level I, Level II, and Level III.

[0003] The development of intelligent identification technology for high-consequence areas is becoming more and more mature, but the classification of high-consequence areas still relies on manual classification based on the identification results table. The manual judgment method has a low level of automation, informatization, and intelligence, low efficiency, and poor timeliness. At the same time, the manual judgment method is highly subjective and cannot be automatically judged. Its effectiveness and integrity are difficult to guarantee, and it is difficult to promote the refined management of pipeline safety operations. Summary of the invention

[0004] The purpose of the present invention is to provide a pipeline high consequence area level identification method, which solves the problem that the existing pipeline high consequence area level identification relies on manual judgment and cannot be automatically judged.

[0005] The technical solution adopted by the present invention is a pipeline high consequence area level identification method, which specifically includes the following steps:

[0006] Step 1: Obtain high-resolution images and create label maps and partition datasets of roads and houses;

[0007] Step 2, preprocessing the data set obtained in step 1;

[0008] Step 3, construct a convolutional layer and combine feature information;

[0009] Step 4: Pool the features after combining the feature information in step 3, and then gradually abstract higher-level features by stacking them layer by layer;

[0010] Step 5: Perform transposed convolution on the features stacked layer by layer in step 4;

[0011] Step 6: If the feature belongs to a road, the road category is determined and the pipeline high consequence area level is determined. If the feature belongs to a house, the house feature is determined as a connected area and the pipeline high consequence area level is determined based on the number of houses.

[0012] The present invention is also characterized in that:

[0013] The specific process of obtaining high-resolution images in step 1 is to select high-resolution images on both sides of the pipeline centerline as the data source; the specific process of making label maps of roads and houses and dividing data sets is to first make label data for roads and houses respectively, and then randomly divide the image data and the corresponding label data into training set, validation set, and test set according to the proportions of 65%, 25%, and 10%.

[0014] The specific process of data preprocessing in step 2 is to crop, rotate, and add noise to the image data and label data of the training set, validation set, and test set, respectively. Finally, multiple image blocks of size 256×256 are obtained for training, validation, and testing of the road model.

[0015] The specific process of constructing the convolution layer in step 3 is to slide a small window on the preprocessed image, define the convolution kernel, construct a convolution layer conv, and set the shape of the input X to 1, 3, 64, 64, the number of channels of the convolution output Y is increased to 10, and a new feature value is obtained by weighted multiplication with the pixel value in the window. By using different convolution kernels at different positions, the parameters and weights of the model are continuously updated through iteration during the training process, and the training log is saved at the same time; the specific process of combining feature information in step 3 is to generate a series of feature maps after the convolution operation. These feature maps contain feature information at different positions in the image. These feature maps are combined to capture the rich feature information of the image.

[0016] The specific process of the pooling operation in step 4 is to take the maximum value or average value in a specific area, thereby downsampling the feature map and streamlining the feature information.

[0017] The specific process of transposed convolution in step 5 is to create Conv2DTranspose to construct the transposed convolution layer conv, set the convolution kernel shape, padding and stride of conv to be the same as those in conv, and set the number of output channels to 3. When the input is the output Y of the convolution layer conv, the height and width of the transposed convolution layer output are the same as the height and width of the convolution layer input. The transposed convolution layer enlarges the height and width of the feature map by 2 times respectively.

[0018] In step 6, the specific process of determining the road category and determining the level of the pipeline high consequence area is to calculate the extracted road results according to the Cartesian product topological operation and geometric features, that is, the width. The national highway width is ≥13m, the expressway width is ≥22.5m, and the provincial road width is ≥9m. If it meets the characteristics of any type of highway, expressway, or provincial road and is within 50m on both sides of the pipeline, it is determined to be a pipeline high consequence area level I.

[0019] In step 6, the specific process of determining the connected area of ​​the house characteristics and the level of the pipeline high consequence area is to count the number of closed polygons, that is, the number of houses, and use the Green's formula. Let D be a plane area. If the part enclosed by any closed curve in D belongs to D, then D is called a plane connected area closed. Using the Green's formula: By calculating the extracted closed figures one by one, the number of connected areas in the drawing can be counted, and the Cartesian product topological relationship operation can be used to identify two or more polygons connected by a wall in the middle as one household; isolated connected areas are identified as one household, and any 400m×2000m rectangular box with ≥50 households is identified as a pipeline high consequence area II; <50 households is identified as a pipeline high consequence area III.

[0020] The beneficial effects of the present invention are as follows:

[0021] The pipeline high consequence area grade identification method of the present invention performs graded judgment on the high consequence area, selects high-resolution images on both sides of the pipeline centerline as data sources, improves the judgment accuracy and efficiency through deep learning neural network technology, realizes automatic judgment of the pipeline high consequence area grade, improves the automation and intelligence of graded judgment, and timely understands the different impacts of high consequence area classification on population, environment, and facilities; and provides a scientific decision-making basis for pipeline safe operation and management. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 is a flow chart of highway extraction and high consequence area determination of the present invention;

[0023] Figure 2 is a flow chart of residential area extraction and high consequence area determination of the present invention;

[0024] Figure 3 It is a schematic diagram of obtaining road characteristic data of the present invention;

[0025] Figure 4 It is a schematic diagram of obtaining house characteristic data of the present invention. DETAILED DESCRIPTION

[0026] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0027] like Figure 1 and Figure 2 As shown, the technical solution adopted by the present invention is a pipeline high consequence area level identification method, which specifically includes the following steps:

[0028] Step 1: Obtain high-resolution images and create label maps and partition datasets of roads and houses;

[0029] The method for obtaining road features in step 1 is as follows: Figure 3As shown in the figure, specifically: select the high-resolution images on both sides of the pipeline centerline as the data source, take the pipeline centerline as the centerline, draw a 100m×2000m rectangular frame, and move arbitrarily along the axial direction on the centerline of the pipeline. The features of the graded highway (expressway, national highway, provincial highway) are extracted within the two 100m short sides of the rectangular frame; the overall idea of ​​highway extraction: according to the differences in grayscale, shape, texture, geometric features (width) and other aspects of various types of roads in high-resolution images, the extraction features of different categories of roads are obtained, and the full convolutional neural network algorithm is used to extract the results and determine the classification of highways in high-resolution images;

[0030] The method for obtaining house features in step 1 is as follows: Figure 4 As shown in the figure, specifically: select high-resolution images on both sides of the pipeline centerline as the data source, and arbitrarily divide several sections with a length of 2km and the maximum number of households within 200m on both sides of the pipeline centerline, and the densely populated area with 50 households or more. Take the pipeline centerline as the centerline, draw a 400m×2000m rectangular frame, and move arbitrarily along the axial direction on the pipeline centerline. The two 400m short sides of the rectangular frame are within the distance frame to extract the residential area (houses, walls), and count the densely populated areas with ≥50 households; Residential area (houses, walls) extraction ideas: According to the grayscale, shape, texture, vector features, topological relationship and other aspects of houses and walls on high-resolution images, the extraction features of houses and connected walls are obtained, and the connected area is formed as one household. The house and wall in the high-resolution image are extracted using the full convolutional neural network algorithm; Green's formula is used to calculate the number of connected areas, that is, the number of households in the residential area, and determine the level of high-consequence areas (level II or level III).

[0031] Step 2, preprocessing the data set obtained in step 1;

[0032] The specific process of data preprocessing in step 2 is: image cropping, rotation, and noise addition are performed on the image data and label data of the training set, validation set, and test set respectively. Finally, multiple image blocks of size 256×256 are obtained for training, validation, and testing of the road model. The input layer reads the regularized image, and each neuron in each layer takes a group of small local neighbor units in the previous layer as input.

[0033] Step 3, construct a convolutional layer and combine feature information;

[0034] The specific process of constructing the convolution layer in step 3 is as follows: slide a small window on the preprocessed image, define the convolution kernel, construct a convolution layer conv, and set the shape of the input X to 1, 3, 64, 64, the number of channels of the convolution output Y is increased to 10, and a new feature value is obtained by weighted multiplication with the pixel value in the window. By using different convolution kernels at different positions, the parameters and weights of the model are continuously updated through iteration during the training process, and the training log is saved at the same time;

[0035] The specific process of combining feature information in step 3 is as follows: after the convolution operation, a series of feature maps are generated, which contain feature information at different positions in the image. These feature maps are combined to capture the rich feature information of the image.

[0036] Step 4: Pool the features after combining the feature information in step 3, and then gradually abstract higher-level features by stacking them layer by layer;

[0037] The specific process of the pooling operation in step 4 is: taking the maximum value or average value in a specific area, thereby downsampling the feature map and streamlining the feature information.

[0038] Step 5: Perform transposed convolution on the features stacked layer by layer in step 4;

[0039] The specific process of transposed convolution in step 5 is: create Conv2DTranspose to construct the transposed convolution layer conv, set the convolution kernel shape, padding and stride of conv to be the same as those in conv, and set the number of output channels to 3. When the input is the output Y of the convolution layer conv, the height and width of the transposed convolution layer output are the same as the height and width of the convolution layer input. The transposed convolution layer enlarges the height and width of the feature map by 2 times respectively.

[0040] Step 6: If the feature belongs to a road, the road category is determined and the road high-consequence area level is determined. If the feature belongs to a house, the house feature is determined as a connected area and the house high-consequence area level is determined based on the number of houses.

[0041] In step 6, the specific process of determining the road category and determining the level of the pipeline high consequence area is as follows: the extracted road results are calculated according to the Cartesian product topological operation and geometric features, that is, the width, where the national highway width is ≥13m, the expressway width is ≥22.5m, and the provincial road width is ≥9m. If it meets the characteristics of any type of highway, expressway, or provincial road and the highway is within 50m on both sides of the pipeline, it is determined to be a pipeline high consequence area level I.

[0042] In step 6, the specific process of determining the connected area of ​​the house characteristics and the level of the pipeline high consequence area is as follows: the number of closed polygons is counted as the number of houses, and Green's formula is used. Let D be the plane area. If the part enclosed by any closed curve in D belongs to D, then D is called a closed plane connected area. Using Green's formula: By calculating the extracted closed figures one by one, the number of connected areas in the drawing can be counted, and the Cartesian product topological relationship operation can be used to identify two or more polygons connected by a wall in the middle as one household; isolated connected areas are identified as one household, and any 400m×2000m rectangular box with ≥50 households is identified as a pipeline high consequence area II; <50 households is identified as a pipeline high consequence area III.

[0043] Refer to "Oil and Gas Gathering and Transportation Pipeline and Plant Station Integrity Management Code Part 3: Identification and Risk Assessment of High Consequence Areas" (Q / SY 01039.3-2019), where the classification of high consequence areas is shown in the following table:

[0044] Classification of high consequence areas.

[0045]

[0046] Example 1

[0047] like Figure 1 As shown, the pipeline high consequence area level identification method proposed in this embodiment specifically includes the following steps:

[0048] Step 1: Obtain high-resolution images and create label maps and partition datasets of roads and houses;

[0049] Step 2, preprocessing the data set obtained in step 1;

[0050] Step 3, construct a convolutional layer and combine feature information;

[0051] Step 4: Pool the features after combining the feature information in step 3, and then gradually abstract higher-level features by stacking them layer by layer;

[0052] Step 5: Perform transposed convolution on the features stacked layer by layer in step 4;

[0053] Step 6: If the feature belongs to a road, the road category is determined and the pipeline high consequence area level is determined. If the feature belongs to a house, the house feature is determined as a connected area and the pipeline high consequence area level is determined based on the number of houses.

[0054] Example 2

[0055] like Figure 1 and Figure 3As shown, the pipeline high consequence area level identification method proposed in this embodiment specifically includes the following steps: the specific process of obtaining the high-resolution image in step 1 is to select the high-resolution images on both sides of the pipeline centerline as the data source; the specific process of making the label map of the highway and dividing the data set is to first make the label data of the highway, and then randomly divide the image data and the corresponding label data into a training set, a verification set, and a test set according to the proportions of 65%, 25%, and 10%.

[0056] The specific process of data preprocessing in step 2 is to crop, rotate, and add noise to the image data and label data of the training set, validation set, and test set, respectively. Finally, multiple image blocks of size 256×256 are obtained for training, validation, and testing of the road model.

[0057] The specific process of constructing the convolution layer in step 3 is to slide a small window on the preprocessed image, define the convolution kernel, construct a convolution layer conv, and set the shape of the input X to 1, 3, 64, 64, the number of channels of the convolution output Y is increased to 10, and a new feature value is obtained by weighted multiplication with the pixel value in the window. By using different convolution kernels at different positions, the parameters and weights of the model are continuously updated through iteration during the training process, and the training log is saved at the same time; the specific process of combining feature information in step 3 is to generate a series of feature maps after the convolution operation. These feature maps contain feature information at different positions in the image. These feature maps are combined to capture the rich feature information of the image.

[0058] The specific process of the pooling operation in step 4 is to take the maximum value or average value in a specific area, thereby downsampling the feature map and streamlining the feature information.

[0059] The specific process of transposed convolution in step 5 is to create Conv2DTranspose to construct the transposed convolution layer conv, set the convolution kernel shape, padding and stride of conv to be the same as those in conv, and set the number of output channels to 3. When the input is the output Y of the convolution layer conv, the height and width of the transposed convolution layer output are the same as the height and width of the convolution layer input. The transposed convolution layer enlarges the height and width of the feature map by 2 times respectively.

[0060] In step 6, the specific process of determining the road category and determining the level of the pipeline high consequence area is to calculate the extracted road results according to the Cartesian product topological operation and geometric features, that is, the width. The national highway width is ≥13m, the expressway width is ≥22.5m, and the provincial road width is ≥9m. If it meets the characteristics of any type of highway, expressway, or provincial road and is within 50m on both sides of the pipeline, it is determined to be a pipeline high consequence area level I.

[0061] Example 3

[0062] like Figure 2 and Figure 4 As shown, the pipeline high consequence area level identification method proposed in this embodiment specifically includes the following steps: the specific process of obtaining the high-resolution image in step 1 is to select the high-resolution images on both sides of the pipeline centerline as the data source; the specific process of making the label map of the house and dividing the data set is to first make the label data of the house, and then randomly divide the image data and the corresponding label data into a training set, a verification set, and a test set according to the proportions of 65%, 25%, and 10%.

[0063] The specific process of data preprocessing in step 2 is to crop, rotate, and add noise to the image data and label data of the training set, validation set, and test set, respectively. Finally, multiple image blocks of size 256×256 are obtained for training, validation, and testing of the road model.

[0064] The specific process of constructing the convolution layer in step 3 is to slide a small window on the preprocessed image, define the convolution kernel, construct a convolution layer conv, and set the shape of the input X to 1, 3, 64, 64, the number of channels of the convolution output Y is increased to 10, and a new feature value is obtained by weighted multiplication with the pixel value in the window. By using different convolution kernels at different positions, the parameters and weights of the model are continuously updated through iteration during the training process, and the training log is saved at the same time; the specific process of combining feature information in step 3 is to generate a series of feature maps after the convolution operation. These feature maps contain feature information at different positions in the image. These feature maps are combined to capture the rich feature information of the image.

[0065] The specific process of the pooling operation in step 4 is to take the maximum value or average value in a specific area, thereby downsampling the feature map and streamlining the feature information.

[0066] The specific process of transposed convolution in step 5 is to create Conv2DTranspose to construct the transposed convolution layer conv, set the convolution kernel shape, padding and stride of conv to be the same as those in conv, and set the number of output channels to 3. When the input is the output Y of the convolution layer conv, the height and width of the transposed convolution layer output are the same as the height and width of the convolution layer input. The transposed convolution layer enlarges the height and width of the feature map by 2 times respectively.

[0067] In step 6, the specific process of determining the connected area of ​​the house characteristics and the level of the pipeline high consequence area is to count the number of closed polygons, that is, the number of houses, and use the Green's formula. Let D be a plane area. If the part enclosed by any closed curve in D belongs to D, then D is called a plane connected area closed. Using the Green's formula: By calculating the extracted closed figures one by one, the number of connected areas in the drawing can be counted, and the Cartesian product topological relationship operation can be used to identify two or more polygons connected by a wall in the middle as one household; isolated connected areas are identified as one household, and any 400m×2000m rectangular box with ≥50 households is identified as a pipeline high consequence area II; <50 households is identified as a pipeline high consequence area III.

[0068] Example 4

[0069] like Figure 1-4 As shown, the pipeline high consequence area level identification method proposed in this embodiment specifically includes the following steps: the specific process of obtaining the high-resolution image in step 1 is to select the high-resolution images on both sides of the pipeline centerline as the data source; the specific process of making the label map of the road and the house and dividing the data set is to first make the label data of the road and the house respectively, and then randomly divide the image data and the corresponding label data into a training set, a verification set, and a test set according to the proportions of 65%, 25%, and 10%.

[0070] The specific process of data preprocessing in step 2 is to crop, rotate, and add noise to the image data and label data of the training set, validation set, and test set, respectively. Finally, multiple image blocks of size 256×256 are obtained for training, validation, and testing of the road model.

[0071] The specific process of constructing the convolution layer in step 3 is to slide a small window on the preprocessed image, define the convolution kernel, construct a convolution layer conv, and set the shape of the input X to 1, 3, 64, 64, the number of channels of the convolution output Y is increased to 10, and a new feature value is obtained by weighted multiplication with the pixel value in the window. By using different convolution kernels at different positions, the parameters and weights of the model are continuously updated through iteration during the training process, and the training log is saved at the same time; the specific process of combining feature information in step 3 is to generate a series of feature maps after the convolution operation. These feature maps contain feature information at different positions in the image. These feature maps are combined to capture the rich feature information of the image.

[0072] The specific process of the pooling operation in step 4 is to take the maximum value or average value in a specific area, thereby downsampling the feature map and streamlining the feature information.

[0073] The specific process of transposed convolution in step 5 is to create Conv2DTranspose to construct the transposed convolution layer conv, set the convolution kernel shape, padding and stride of conv to be the same as those in conv, and set the number of output channels to 3. When the input is the output Y of the convolution layer conv, the height and width of the transposed convolution layer output are the same as the height and width of the convolution layer input. The transposed convolution layer enlarges the height and width of the feature map by 2 times respectively.

[0074] In step 6, the specific process of determining the road category and determining the level of the pipeline high consequence area is to calculate the extracted road results according to the Cartesian product topological operation and geometric features, that is, the width. The national highway width is ≥13m, the expressway width is ≥22.5m, and the provincial road width is ≥9m. If it meets the characteristics of any type of highway, expressway, or provincial road and is within 50m on both sides of the pipeline, it is determined to be a pipeline high consequence area level I.

[0075] In step 6, the specific process of determining the connected area of ​​the house characteristics and the level of the pipeline high consequence area is to count the number of closed polygons, that is, the number of houses, and use the Green's formula. Let D be a plane area. If the part enclosed by any closed curve in D belongs to D, then D is called a plane connected area closed. Using the Green's formula: By calculating the extracted closed figures one by one, the number of connected areas in the drawing can be counted, and the Cartesian product topological relationship operation can be used to identify two or more polygons connected by a wall in the middle as one household; isolated connected areas are identified as one household, and any 400m×2000m rectangular box with ≥50 households is identified as a pipeline high consequence area II; <50 households is identified as a pipeline high consequence area III.

Claims

1. A pipeline high consequence area level identification method, characterized in that: The specific steps include: Step 1: Obtain high-resolution images and create label maps and partition datasets of roads and houses; Step 2, preprocessing the data set obtained in step 1; Step 3, construct a convolutional layer and combine feature information; Step 4: Pool the features after combining the feature information in step 3, and then gradually abstract higher-level features by stacking them layer by layer; Step 5: Perform transposed convolution on the features stacked layer by layer in step 4; Step 6: If the feature belongs to a road, the road category is determined and the pipeline high consequence area level is determined. If the feature belongs to a house, the house feature is determined as a connected area and the pipeline high consequence area level is determined based on the number of houses.

2. The pipeline high consequence area level identification method according to claim 1 is characterized in that: The specific process of obtaining the high-resolution image in step 1 is to select the high-resolution images on both sides of the pipeline centerline as the data source; The specific process of making label maps of roads and houses and dividing data sets in step 1 is to first make label data of roads and houses respectively, and then randomly divide the image data and the corresponding label data into training set, verification set and test set according to the proportions of 65%, 25% and 10%.

3. The pipeline high consequence area level identification method according to claim 2 is characterized in that: The specific process of data preprocessing in step 2 is to crop, rotate, and add noise to the image data and label data of the training set, validation set, and test set, respectively, and finally obtain multiple image blocks of size 256×256 for training, validation, and testing of the road model.

4. The pipeline high consequence area level identification method according to claim 3 is characterized in that: The specific process of constructing the convolution layer in step 3 is to slide a small window on the preprocessed image, define the convolution kernel, construct a convolution layer conv, and set the shape of the input X to 1, 3, 64, 64, the number of channels of the convolution output Y is increased to 10, and a new feature value is obtained by weighted multiplication with the pixel value in the window. By using different convolution kernels at different positions, the parameters and weights of the model are continuously updated through iteration during the training process, and the training log is saved at the same time; The specific process of combining feature information in step 3 is to generate a series of feature maps after the convolution operation, and these feature maps contain feature information at different positions in the image. These feature maps are combined to capture rich feature information of the image.

5. The pipeline high consequence area level identification method according to claim 4 is characterized in that: The specific process of the pooling operation in step 4 is to take the maximum value or the average value in a specific area, thereby downsampling the feature map and simplifying the feature information.

6. The pipeline high consequence area level identification method according to claim 5 is characterized in that: The specific process of the transposed convolution in step 5 is to create Conv2DTranspose to construct a transposed convolution layer conv, set the convolution kernel shape, padding and stride of conv to be the same as those in conv, and set the number of output channels to 3. When the input is the output Y of the convolution layer conv, the height and width of the transposed convolution layer output are the same as those of the convolution layer input, and the transposed convolution layer enlarges the height and width of the feature map by 2 times respectively.

7. The pipeline high consequence area level identification method according to claim 6 is characterized in that: In step 6, the specific process of determining the category of the road and determining the level of the pipeline high consequence area is to calculate the width of the extracted road results based on Cartesian product topological operations and geometric features, where the national highway width is ≥13m, the expressway width is ≥22.5m, and the provincial road width is ≥9m. If it meets the characteristics of any type of highway, expressway, or provincial road and the highway is within 50m on both sides of the pipeline, it is determined to be a pipeline high consequence area level I.

8. The pipeline high consequence area level identification method according to claim 6 is characterized in that: In step 6, the specific process of determining the connected area of ​​the house characteristics and the level of the pipeline high consequence area is to count the number of closed polygons, that is, the number of houses, and use the Green's formula. Let D be a plane area. If the part surrounded by any closed curve in D belongs to D, then D is called a plane connected area closed. Using the Green's formula: By calculating the extracted closed figures one by one, the number of connected areas in the drawing can be counted, and the Cartesian product topological relationship operation can be used to identify two or more polygons connected by a wall in the middle as one household; isolated connected areas are identified as one household, and any 400m×2000m rectangular box with ≥50 households is identified as a pipeline high consequence area II; <50 households is identified as a pipeline high consequence area III.

Images