Urban charging pile configuration demand evaluation method and device based on remote sensing image

Abstract

The application discloses a kind of urban charging pile configuration demand evaluation method and device based on remote sensing image, by obtaining the remote sensing image data of the building of target urban area;Learning offset vector model is extracted to the building footprint of the remote sensing image data using pre-constructed;Wherein, the learning offset vector model is based on U-Net model and SegNet model fusion and is constructed to obtain;The vector shape area of the building footprint is calculated based on Graham scanning method's convex hull algorithm, obtain building floor area, and calculate building volume according to the building floor area;Building environment independent variable and charging pile flexible resource are carried out regression analysis, to calculate the charging pile scale of the target urban area.Using the present application, building footprint can be identified and floor area can be calculated using multispectral remote sensing image, and then correlation analysis of regional charging pile flexible resource capacity is realized, so as to accurately predict the charging pile demand of urban area.

Description

Technical Field

[0001] The present invention relates to the technical field of charging pile configuration, and in particular to a method and device for evaluating charging pile configuration needs in urban areas based on remote sensing images. Background Art

[0002] With the improvement of environmental protection awareness and the increasingly severe fossil energy crisis, the development and application of electric vehicles have become the focus of attention in the world today. All countries are actively promoting the development of the electric vehicle industry. Charging piles are essential infrastructure for the development of electric vehicles. A reasonable number of charging piles is conducive to reducing resource waste and improving the operational efficiency of the charging piles themselves. It can also provide convenience for electric vehicle users and enhance the public's recognition of electric vehicles. It is the key to the further promotion and development of electric vehicles.

[0003] However, in the process of implementing the present invention, the inventors found that the existing technology has at least the following problems: it is difficult for the existing technology to comprehensively consider the building floor area and building environment factors to evaluate the demand for charging piles in urban areas, and it is difficult to accurately predict the demand for charging piles in urban areas. Summary of the Invention

[0004] The purpose of the embodiments of the present invention is to provide a method and device for evaluating the configuration demand of charging piles in urban areas based on remote sensing images, which can use multispectral remote sensing images to identify building footprints and calculate the occupied area, and then realize the correlation analysis of the flexible resource capacity of regional charging piles, so as to accurately predict the demand for charging piles in urban areas.

[0005] To achieve the above objectives, an embodiment of the present invention provides a method for evaluating charging pile configuration requirements in urban areas based on remote sensing images, comprising:

[0006] Acquire remote sensing image data containing buildings in the target urban area;

[0007] Extracting building footprints from the remote sensing image data using a pre-built learning offset vector model; wherein the learning offset vector model is constructed based on the fusion of a U-Net model and a SegNet model;

[0008] Calculating the vector shape area of ​​the building footprint using a convex hull algorithm based on a Graham scan method to obtain the building footprint, and calculating the building volume based on the building footprint;

[0009] A regression analysis is performed on the building environment independent variables and the charging pile flexible resources to calculate the charging pile scale of the target urban area; wherein the building environment independent variables include the building floor area and the building volume.

[0010] As an improvement to the above solution, the learning offset vector model is constructed by the following steps:

[0011] The U-Net model and the SegNet model are fused to obtain a LOFT network model system;

[0012] In the LOFT network model system, end-to-end identification of buildings is realized, and network layer transformation is performed according to actual effects, and a learning offset vector model fusing U-net and SegNet is obtained:

[0013] L = L U-Net + α 1 L SegNet + α 2 L Mask + α 3 L Offset ;

[0014] Wherein, L U-Net , L SegNet , L Mask are the losses of the U-Ne model, the SegNet model and the mask first layer respectively; L Offset is the loss of the offset head, and α1, α2 and α3 are loss weights.

[0015] As an improvement of the above scheme, after the learning offset vector model constructed in advance is used to extract the building footprint of the remote sensing image data, and before the convex hull algorithm based on the Graham scan method is used to calculate the vector shape area of the building footprint, the method further comprises:

[0016] The polygon feature mapping of the building footprint is enhanced by using a functional level offset enhancement method to obtain an enhanced building footprint.

[0017] As an improvement of the above scheme, the convex hull algorithm based on the Graham scan method calculates the vector shape area of the building footprint, specifically comprising:

[0018] Obtain the vectorized data of the building footprint, and continuously reduce the perimeter of the vector of the building footprint until it approaches the edge perimeter containing all vector points;

[0019] Take the point at the lower left corner as the pole to perform polar angle sorting, and connect the points sorted by polar angle in sequence to obtain a polygon surrounding all points;

[0020] The vector is continuously rotated counterclockwise to form a convex polygon, and the process of popping the top of the stack replaces the two edges of the triangle with the third edge while ensuring that the original vertex is included;

[0021] Calculate the polar angle of the vector between the convex hull points; wherein the convex hull is the intersection of all convex polygons that can contain all points;

[0022] Take the point at the lower left corner as the pole to perform polar angle sorting, and check whether the rotation directions of the two vectors formed by the three points are counterclockwise;

[0023] According to the clockwise given all point coordinates constituting the convex hull, the polygon area is calculated based on the cross product calculation of Graham scan method, and the vector shape area of the building footprint is obtained.

[0024] As an improvement of the above scheme, the regression analysis is performed on the building environment independent variable and the flexible resource of charging pile to calculate the charging pile scale of the target urban area, and specifically comprises:

[0025] The regression analysis is performed on the building environment independent variable and the flexible resource of charging pile by using the Spearman rank correlation coefficient calculation formula;

[0026] Based on the regression analysis result, the charging pile scale is calculated in combination with the multiple correlation coefficient and the influence index factor.

[0027] As an improvement of the above scheme, the Spearman rank correlation coefficient calculation formula is:

[0028]

[0029] Wherein, V i is the building volume of the i th building in the target urban area, n i is the number of charging piles in the i th building volume, n is the total number of charging piles in the target urban area, R(V i ) and R(n i ) are the ranks of V i and n i , respectively, and represent the average rank, respectively.

[0030] As an improvement of the above scheme, the calculation formula of the charging pile scale is:

[0031]

[0032] Wherein, R p is the building environment influence coefficient, R q is the influence coefficient of other factors, R p , R q ∈(0,1), p<q; m represents the market potential of electric vehicles, f(t) represents the probability of first purchase at t, and n(t) represents the cumulative number of charging piles at t.

[0033] As an improvement of the above scheme, after the remote sensing image data containing the buildings in the target urban area is obtained, the method further comprises:

[0034] Image preprocessing operation is performed on the remote sensing image data; wherein, the image preprocessing operation comprises radiation calibration, atmospheric correction and orthorectification processing.

[0035] The embodiment of the present invention further provides a device for evaluating the configuration requirements of charging piles in urban areas based on remote sensing images, comprising:

[0036] A remote sensing image data acquisition module is used to acquire remote sensing image data of buildings including the target urban area;

[0037] A building footprint extraction module is used to extract building footprints from the remote sensing image data using a pre-built learning offset vector model; wherein the learning offset vector model is constructed based on the fusion of a U-Net model and a SegNet model;

[0038] a building volume calculation module, configured to calculate the vector shape area of ​​the building footprint based on a convex hull algorithm of a Graham scan method to obtain a building footprint, and calculate the building volume based on the building footprint;

[0039] A charging pile scale assessment module is used to perform regression analysis on building environment independent variables and charging pile flexible resources to calculate the charging pile scale of the target urban area; wherein the building environment independent variables include the building floor area and the building volume.

[0040] An embodiment of the present invention also provides a device for evaluating the configuration needs of urban charging piles based on remote sensing images, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the method for evaluating the configuration needs of urban charging piles based on remote sensing images as described in any one of the above.

[0041] Compared with the prior art, the method and device for evaluating the configuration demand of charging piles in urban areas based on remote sensing images disclosed in the present invention obtain remote sensing image data of buildings in the target urban area; use a pre-constructed learning offset vector model to extract the building footprint from the remote sensing image data; wherein the learning offset vector model is constructed based on the fusion of the U-Net model and the SegNet model; calculate the vector shape area of ​​the building footprint using the convex hull algorithm based on the Graham scan method to obtain the building footprint, and calculate the building volume based on the building footprint; perform regression analysis on the building environment independent variables and the charging pile flexible resources to calculate the scale of charging piles in the target urban area. By adopting the technical means of the embodiments of the present invention, it is possible to use multispectral remote sensing images to identify the building footprint and calculate the building footprint, perform correlation analysis on the flexible resource capacity of regional charging piles based on the building environment factor independent variables, complete regression analysis on multiple independent variables and the scale of charging piles, thereby accurately predicting the demand for charging piles in urban areas and providing a reference for the prediction of the configuration demand for charging piles in urban areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 is a flowchart of a method for evaluating charging pile configuration demand in an urban area based on remote sensing images according to an embodiment of the present application;

[0043] Figure 2 is a schematic diagram of a learning offset vector model construction principle in an embodiment of the present application;

[0044] Figure 3 is a schematic diagram of a vector shape area calculation principle of a building footprint in an embodiment of the present application;

[0045] Figure 4 is a structural schematic diagram of a device for evaluating charging pile configuration demand in an urban area based on remote sensing images according to an embodiment of the present application;

[0046] Figure 5 is a structural schematic diagram of another device for evaluating charging pile configuration demand in an urban area based on remote sensing images according to an embodiment of the present application. DETAILED DESCRIPTION

[0047] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0048] Referring to Figure 1 is a flowchart of a method for evaluating charging pile configuration demand in an urban area based on remote sensing images according to an embodiment of the present application. The present application provides a method for evaluating charging pile configuration demand in an urban area based on remote sensing images, which specifically includes the following steps S11 to S14:

[0049] S11, acquiring remote sensing image data containing buildings in a target urban area;

[0050] S12, extracting a building footprint from the remote sensing image data by using a pre-constructed learning offset vector model, wherein the learning offset vector model is constructed based on a fusion of a U-Net model and a SegNet model;

[0051] S13, calculating a vector shape area of the building footprint based on a convex hull algorithm of the Graham scan method to obtain a building floor area, and calculating a building volume according to the building floor area;

[0052] S14, performing regression analysis on building environment independent variables and charging pile flexible resources to calculate a charging pile scale of the target urban area, wherein the building environment independent variables include the building floor area and the building volume.

[0053] In the embodiment of the present application, the charging pile configuration demand in the target urban area is evaluated, first, remote sensing image data containing buildings in the target urban area is obtained through an image sensor, and the building environment independent variable in the target urban area is calculated through analysis of the remote sensing image data.

[0054] As a preferred embodiment, after step S11, the method further comprises step S11':

[0055] S11', image preprocessing operation is performed on the remote sensing image data; wherein the image preprocessing operation comprises radiation calibration, atmospheric correction and ortho correction processing.

[0056] Specifically, the operation of radiation calibration is to transform the measurement value (DN value) obtained by the image sensor into a relative value related to physical quantities such as surface reflectivity and surface temperature, eliminate the error generated by the sensor itself, obtain the calibration coefficient of the multispectral remote sensing image, and calibrate the image. The calculation formula of radiation calibration is as follows:

[0057] L e (λ e )=Gain*DN+Offset;

[0058] Wherein, L e (λ e ) is the converted radiance, DN is the observation value received by the sensor, Gain is the calibration gain value, and Offset is the calibration offset value.

[0059] The purpose of atmospheric correction is to eliminate the influence of factors such as atmosphere and light on the reflection of ground objects, and the process of obtaining the true reflectivity of ground objects after atmospheric correction facilitates the accuracy of construction land feature extraction. The ENVI5.3 fast atmospheric correction method is used for correction. The fast atmospheric correction automatically collects spectral information of different substances from the image to obtain empirical values to complete the fast atmospheric correction of hyperspectral and multispectral, and the accuracy of the result is approximately + / - 15% of FLAASH or other radiation transmission model-based. The multispectral and hyperspectral spectral range it supports is 0.4-2.5um.

[0060] The operation of ortho correction is to select some ground control points on the image, use the digital elevation model DEM data of the image range that has been obtained, correct the tilt and projection difference of the image at the same time, and resample the image into an ortho image through the rational polynomial model ortho correction (RPC ortho correction method). The RPC correction polynomial equation is as follows:

[0061] x i =a0+(a1X i +a2Y i )+(a3X i2 +a4X i Y i +a5Y i 2 )+(a6X i 3 +a7X i 2 Y i +a8X i Y i 2 +a9Y i 3 )

[0062] y i =b0+(b1X i +b2Y i )+(b3X i 2 +b4X i Y i +b5Y i 2 )+(b6X i 3 +b7X i 2 Y i +b8X i Y i 2 +b9Y i 3 );

[0063] where (x, y) is the original image coordinates, (X, Y) is the ground or map coordinates of the homonymic point, (x, y) is the original image coordinates, (X, Y) is the ground or map coordinates of the homonymic point.

[0064] Further, the learning offset vector model is constructed based on the fusion of the U-Net model and the SegNet model in advance, and then the learning offset vector model is used to extract the building footprint of the remote sensing image data after the remote sensing image data is preprocessed.

[0065] Preferably, the learning offset vector model is constructed by the following steps S21 to S22:

[0066] S21, fuse the U-Net model and the SegNet model to obtain a LOFT network model system;

[0067] S22, implement end-to-end identification of buildings in the LOFT network model system, and perform network layer transformation according to the actual effect, to obtain a learning offset vector model fused with U-net and SegNet:

[0068] L = L U-Net + a1L SegNet + a2L Mask + a3L Offset ;

[0069] Wherein, L U-Net , L SegNet , L Mask are the loss of U-Ne model, SegNet model and mask first layer respectively; L Offset is the loss of offset head, and a1, a2 and a3 are loss weights.

[0070] Specifically, referring to Figure 2 , the construction principle diagram of learning offset vector model in the embodiment of the application, U-Net is pre-trained on the image data set, and SegNet is used as the backbone. The learning offset vector model is constructed by using the fusion of U-Net and SegNet model, and the data training of the overall structure system is carried out.

[0071] The network of U-Net model mainly consists of a compression path (i.e. an encoder network) for capturing context, an expansion symmetric path (i.e. a decoder network) for finding precise location and a skip connection, follows a typical convolutional network architecture, and an encoder network is established on the basis of VGG-13 for extracting features to generate a feature map, only the convolutional layer is reserved, and the fully connected layer is deleted. The decoder network is used for decoding feature information, restoring the spatial resolution and detail information of the input image, and generating a high-resolution prediction map. It should be noted that the main effective operation of this network is convolution, and the entire network has 24 convolution operation layers, including 20 convolution layers and 4 deconvolution layers.

[0072] The core of the SegNet network architecture consists of an encoder network, a decoder network and a pixel-level classification layer, and the entire network contains 10 layers, the first layer to the fifth layer is the encoder layer, the sixth layer to the tenth layer is the decoder layer, the encoder network is composed of 13 layers of 3x3 convolution, the core backbone is composed of the first 13 convolutional layers of the VGG-16 network, and the full connection layer is eliminated, so that the encoder can maintain good resolution even at the deepest network layer, while greatly reducing the network parameter amount.

[0073] In order to reduce the parameters in the functional level offset enhancement module, the parameters of the FC layer in all branches need to be shared. In the training stage, each branch will calculate the offset vector loss respectively, while in the inference stage, the multiple offset vectors predicted in the corresponding branch will be inversely rotated according to the rotation angle. The final fused offset is generated by the maximum selection strategy, and the value of the offset is often less than the ground truth.

[0074] Referring to Figure 2 , the Fo For the functional level offset enhancement, the network architecture of the functional level offset enhancement module is shown as Each branch is composed of a series of convolutional layers Conv and fully connected layers FC, whose parameters are the same as the offset head in Sec. The functional level offset enhancement has four branches, and the parameters of FC layers are shared. During the training phase, each branch rotates the input feature map F o by a given rotation angle and regresses the corresponding offset vector. During the inference phase, the offset vectors regressed by the four branches are inversely rotated and fused to form the final offset vector.

[0075] To solve the problem of serious occlusion of the facade in the boundary part of the roof projection footprint, the present application designs a new offset head to predict the "roof-to-footprint" offset vector of each building. According to the predicted offset vector, the building footprint is estimated by translating the predicted roof mask.

[0076] Therefore, it is necessary to establish building footprint labeling data. Given an aerial image with a lowest point, the building footprint estimation task is to design a model to locate the footprint and extract all the buildings' boundaries containing me. Through supervised learning of a deep building footprint extraction model data D, that is

[0077] D={(I i ,T i );i=1,...,N};

[0078] Where T i is the corresponding label of the building contained in the image I i in d. In particular, T i is composed of K footprints Each is represented by a polygon with M vertices. In addition, since the footprint of the building is partially visible, while the roof is always fully visible, the building footprint is extracted by estimating the roof associated with an offset vector of each building's footprint. Therefore, the footprint label in each dataset D corresponds to a roof label and an additional offset label which represents the offset vector from the roof in the image I i to the jth building's footprint. Therefore, the labeled data will be:

[0079]

[0080] Next, the learned offset vector model is combined with the fusion to extract the building footprint.

[0081] The learning offset vector model method LOFT is to predict the footprint of a building and its related offset vector from the lowest image, supervised dataset D, its process is to insert a new offset head learning vector to a top-down instance segmentation framework for learning building roof. The learning offset vector module is used for some other instance segmentation frameworks, including U-net and SegNet, the training of the learning offset vector model system is as follows:

[0082] Building bounding box (B-Bbox) Roof mask (R-Mask) And offset vector Each is a bounding box representation And each Polygon of vertices. This information can be obtained from the label .

[0083] With a set of labeled data D, the input image I is input into the backbone network to generate the backbone feature map B. Then, the U-net model generates the region proposal with the feature map B as input. Next, three selection layers calculate the feature map F i =(x i ,y i ,ω i ,h i )∈P and B as input b , F r , F o , p i is represented as a building proposal, F b , F r , F o are the building bounding box, roof mask, and offset vector feature map respectively. The SegNet and mask head inherit from the mask SegNet. For the offset head, the feature map F o is used as input, which consists of several convolution (Conv) layers and fully connected (FC) layers. The ground truth of the U-net and SegNet head is to establish the bounding box, because the receptive field of the network needs to cover the roof and footprint at the same time to regress the offset vector at the offset head, that is, the information of the offset vector is embedded into the visible building facade structure. Finally, the learning offset vector model LOFT network system is obtained by minimizing a joint loss function:

[0084] L=L U-Net +α1L SegNet +α2L Mask +α3L Offset ;

[0085] Where, LU-Net , L SegNet , L Mask is the loss of the mask SegNet, i.e., the loss of the U-Net model, the SegNet model and the mask first layer, respectively. Offset is the loss of the offset head, where the standard smooth L1 loss is used. Optionally, the loss weights are set to α1=1, α2=1 and α3=2.

[0086] In order to accelerate the convergence speed of the offset head, the following encoding function is used:

[0087] φ x = o x / ω P , φ y = o y / h P ;

[0088] where ω P and h P are the width and height of the matched proposal p m ∈P, [o x , o y ] T is the ground truth offset vector, [φ x , φ y ] T is the encoded offset vector for regression. The predicted offset vector is used in the inference stage to convert the predicted roof mask into a building footprint mask. Specifically, a predicted roof mask will first be represented as a polygon with M vertices by a topological structure analysis algorithm By translating the roof polygon r with the predicted offset vector , the polygon f of the building footprint is finally calculated.

[0089] With the embodiments of the present application, the full convolution semantic segmentation model U-Net and SegNet are implemented in the fused LOFT network model system to identify buildings end-to-end, and the network layers are reformed according to the actual effect to realize the optimal model performance. The building identification precision is compared and analyzed on the local data set, the functions and characteristics of each network layer are researched, the best network structure and hyperparameters are adjusted, and the segmentation precision of the existing network is improved.

[0090] As a preferred embodiment, after step S12 and before step S13, the method further comprises a step S12':

[0091] S12', the polygon feature map of the building footprint is enhanced by using a functional level offset enhancement method to obtain an enhanced building footprint.

[0092] Specifically, in the polar coordinate system, o = [ox ,o y ] T The offset vector can be converted to [o ρ ,o θ ] T 。o θ approximately uniformly distributed, because the viewing angle of the aerial imaging platform is almost arbitrary to the scene. Therefore, when learning the offset, the network needs to handle arbitrary rotation transformations. In addition, the offset head may not converge well to a limited number of offset training samples. Therefore, designing an enhanced module requires learning more robust robust offset features. One approach is to use image-level rotation augmentation. It is observed that the offset vector in the image can only be rotated by an angle of training era and the network therefore needs more time to converge. Embodiments of the present application propose a function-level offset augmentation module to rotate the offset feature and improve the robustness of the offset prediction operation in the abstract feature space.

[0093] The method extends a single forward offset head to multiple parallel offset head branches, which parallelly regress multiple rotation offset vectors at different angles. Specifically, the input feature map F o and the corresponding ground truth offset vector [o x ,o y ] T will be rotated simultaneously by a set of rotation angles Θ = {θ1, θ2,..., θ n}.

[0094] Specifically, a spatial transformer is used to rotate the feature map. First, a sampling grid, i.e. a set of points, is created using the rotation angle θ i ∈ Θ to sample the input feature map, generating the transformed feature map. More precisely, given the rotation matrix:

[0095]

[0096] The point-level transformation on the input feature map is:

[0097]

[0098] where, is the target coordinate of the regular grid in the output feature map, is the source coordinate in the input feature map. A similar approach is applied to the rotation of the offset vector, i.e. the rotated offset vector is calculated from o * = A θ o.

[0099] Next, according to the extraction result of the building footprint polygon f obtained by the learning offset vector model, the building floor area is calculated. Since the obtained building footprint extraction result is a raster type.tiff format file, it is necessary to first realize the conversion of the raster.tif format file into a vector.shp format file. The raster image in.tif format needs to be rendered in a dynamic slicing manner on the map, and the switching of multiple images needs to be done.

[0100] Each pixel is converted into a rectangle, and then similar pixels are merged, and the conversion method is as follows:

[0101]

[0102] The output file is the merged vector data, which contains the corresponding value of each pixel in the map interaction. After obtaining the vectorized data of the building footprint, the next step is to calculate the area of the building floor vector shape using the convex hull algorithm. Referring to Figure 3 , it is the vector shape area calculation principle diagram of the building footprint in the embodiment of the application, and the convex hull algorithm divides the unit vector to calculate the area of the vector shape. The steps are as follows:

[0103] Obtain the vectorized data of the building footprint, and continuously reduce the perimeter of the vector of the building footprint until the edge perimeter containing all vector points is approached;

[0104] Take the left lower corner point as the pole to perform polar angle sorting, and connect the polar angle sorted points in turn to obtain a polygon surrounding all points;

[0105] The vector is continuously rotated counterclockwise to form a convex polygon, and the process of popping the top of the stack replaces the two edges of the triangle with the third edge, while ensuring that the original vertex is included;

[0106] Calculate the polar angle of the vector between the convex hull points; wherein the convex hull is the intersection of all convex polygons that can contain all points;

[0107] Take the left lower corner point as the pole to perform polar angle sorting, and check whether the rotation direction of the two vectors composed of the three points is counterclockwise;

[0108] According to the clockwise given all point coordinates constituting the convex hull, the cross product calculation is performed based on the Graham scan method, and the polygon area is calculated to obtain the vector shape area of the building footprint, and the specific formula is as follows:

[0109]

[0110] Therefore,

[0111] ​Then, the corresponding building volume can be obtained by using the radar data: preprocessing the radar data table to create a spatial index; preprocessing the remote sensing map table to create a spatial index; creating a new column in the remote sensing map to contain the point patch within each building footprint; describing the building elevation as the statistical data of all point elevations within the building footprint; obtaining the ground elevation by identifying the building contour and combining it with the radar data; and multiplying the building elevation and the preceding building footprint in turn to obtain the volume of each building.

[0112] Further, regression analysis is performed on the building environment independent variables and the flexible resources of charging piles to calculate the charging pile scale of the target urban area, specifically including steps S141 and S142:

[0113] S141, regression analysis is performed on the building environment independent variables and the flexible resources of charging piles using the Spearman rank correlation coefficient calculation formula;

[0114] S142, based on the regression analysis results, the charging pile scale is calculated in combination with the multiple correlation coefficient and the influence index factor.

[0115] Based on the calculation results of the above building index, correlation analysis is performed, and the building environment dependent variables of the variable elements with correlation are analyzed, and the correlation between the elements needs to exist a certain relationship or probability to perform correlation analysis.

[0116] First, the data of social, economic, and environmental factors in the implemented urban area are used to construct a geographic information system based on remote sensing images, providing a data basis for the extraction of subsequent electric vehicle traffic behavior characteristics in the area, and providing dependent variables for subsequent electric vehicle ownership prediction and charging load demand calculation in the area. Factors affecting car ownership are widely accepted, in addition to family structure, social and economic characteristics, and are related to the building environment (including diversity, building footprint density, design, destination accessibility, transit distance, and demand management, etc.). The building environment independent variables include: building footprint density in the target urban area, average building volume, total building volume, volume proportion of each functional unit of the building group (such as office area, public square, residential area, etc.), resident population density, per capita electric vehicle ownership, and total electric vehicle ownership, etc.

[0117] Let the dependent variable y be the charging pile configuration demand index, and the above building environment index and social and economic activity index be the independent variables x1, x2,..., x p , construct a linear model:

[0118] y = b0 + b1x1 +... + b p xp + ε;

[0119] Linear regression is performed on the above linear equation to estimate b0,..., b p :

[0120]

[0121] After obtaining the dependent variables above that affect the number of charging pile configurations, the respective complex correlation coefficients need to be obtained, y and x1, x2,..., x p Correlation analysis is performed on y and The general equation for determining the coefficient is calculated as follows:

[0122]

[0123] As an example, the correlation analysis of the building floor area and the number of charging piles obtained from the remote sensing image data in Tianjin is taken as an example:

[0124]

[0125]

[0126] The distribution rule of the number of charging piles is further analyzed from the relationship between the building environment factors and social economic factor indexes and the number of charging piles. By summarizing the values of the two, the rule is found and the correlation analysis is established. The Spearman rank correlation coefficient calculation formula is as follows:

[0127]

[0128] Where V i is the building volume of the i-th building in the target urban area, n i is the number of charging piles in the i-th building volume, n is the total number of charging piles in the target urban area, R(V i ) and R(n i ) are the ranks of V i and n i , respectively, and represent the average rank, respectively.

[0129] The determination coefficient γ 2 in linear regression can explain the ratio of the sum of squared deviations to the total sum of squared deviations. Through the principle of Spearman rank correlation analysis, the simple equation is shown as follows, and the autocorrelation regression of the building environment factors and the configurable degree of charging piles is performed:

[0130]

[0131] Where, d iThe difference between the bit values of the ith data pair, k is the total number of building observation samples

[0132] The determination coefficient evaluates the degree of explanation of the regression model to the change of the dependent variable y in the regression analysis, that is, the determination coefficient γ 2 is an index for evaluating the goodness of the regression model. γ 2 The value range is also 0-1, usually expressed in percentage. For example, if the γ 2 of the regression model is 0.7, it means that the explainable degree of the regression model to the prediction result is 70%. Generally, γ 2 >0.75 indicates that the model fitting degree is good and the explainable degree is high; γ 2 <0.5 indicates that the model fitting is problematic and is not suitable for regression analysis.

[0133] Finally, based on the above regression analysis of the correlation between the independent variables including building environmental factors and the charging pile configuration, the charging pile scale is calculated, and the charging pile scale calculation formula is:

[0134]

[0135] Where, R p is the building environmental influence coefficient, R q is the influence coefficient of other factors, R p , R q ∈(0,1), p<q; m represents the market potential of electric vehicles, f(t) represents the probability of first purchase at t, F(t) represents the cumulative proportion of buyers at t (f=df / dt), and n(t) represents the cumulative number of charging piles at t.

[0136] Thus, from the building footprint extraction of remote sensing images to the regression analysis of the correlation between building environmental independent variables, the calculation and configuration demand statistics of the charging pile scale in the urban area are completed.

[0137] The embodiment of the present invention provides a method for evaluating the configuration demand of charging piles in an urban area based on remote sensing images, which obtains remote sensing image data of buildings in a target urban area; uses a pre-constructed learning offset vector model to extract building footprints from the remote sensing image data; wherein the learning offset vector model is constructed based on the fusion of the U-Net model and the SegNet model; calculates the vector shape area of ​​the building footprint using a convex hull algorithm based on the Graham scan method to obtain the building footprint, and calculates the building volume based on the building footprint; performs regression analysis on the building environment independent variables and the charging pile flexible resources to calculate the scale of charging piles in the target urban area. By adopting the technical means of the embodiment of the present invention, it is possible to use multispectral remote sensing images to identify the building footprint and calculate the building footprint, perform correlation analysis on the flexible resource capacity of regional charging piles based on the building environment factor independent variables, complete regression analysis on multiple independent variables and charging pile scale, thereby accurately predicting the charging pile demand in the urban area and providing a reference for the prediction of the configuration demand of charging piles in the urban area.

[0138] See also Figure 4 , is a schematic diagram of the structure of a device for evaluating the configuration requirements of charging piles in urban areas based on remote sensing images provided by an embodiment of the present invention. The embodiment of the present invention provides a device 20 for evaluating the configuration requirements of charging piles in urban areas based on remote sensing images, comprising:

[0139] The remote sensing image data acquisition module 21 is used to acquire remote sensing image data of buildings including the target urban area;

[0140] A building footprint extraction module 22 is configured to extract building footprints from the remote sensing image data using a pre-built learning offset vector model; wherein the learning offset vector model is constructed based on a fusion of a U-Net model and a SegNet model;

[0141] A building volume calculation module 23 is configured to calculate the vector shape area of ​​the building footprint based on a convex hull algorithm of a Graham scan method to obtain a building footprint, and calculate the building volume based on the building footprint;

[0142] The charging pile scale assessment module 24 is configured to perform regression analysis on building environment independent variables and charging pile flexible resources to calculate the charging pile scale of the target urban area; wherein the building environment independent variables include the building floor area and the building volume.

[0143] It should be noted that the urban charging pile configuration demand assessment device based on remote sensing images provided in an embodiment of the present invention is used to execute all the process steps of the urban charging pile configuration demand assessment method based on remote sensing images in the above embodiment. The working principles and beneficial effects of the two correspond one to one, and therefore will not be repeated here.

[0144] See also Figure 5 Fig. 30 is a structural schematic diagram of another city charging pile configuration demand evaluation device based on remote sensing images provided by an embodiment of the present application. The present embodiment also provides a city charging pile configuration demand evaluation device 30 based on remote sensing images, which comprises a processor 31, a memory 32, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the city charging pile configuration demand evaluation method based on remote sensing images is realized, as described in any one of the above embodiments.

[0145] It can be understood by those skilled in the art that all or part of the processes in the above-mentioned embodiments can be completed by a computer program instructing relevant hardware. The program can be stored in a computer readable storage medium. When the program is executed, it can include the processes of the above-mentioned embodiments. The storage medium can be a magnetic disc, an optical disc, a read-only memory (ROM), a random access memory (RAM), or the like.

[0146] The above describes the preferred embodiments of the present application. It should be noted that those of ordinary skill in the art can make several improvements and refinements without departing from the principles of the present application. These improvements and refinements are also considered within the scope of protection of the present application.

Claims

1. A method for evaluating the configuration demand of charging piles in urban areas based on remote sensing images, characterized in that: include: Acquire remote sensing image data containing buildings in the target urban area; Extracting building footprints from the remote sensing image data using a pre-built learning offset vector model; wherein the learning offset vector model is constructed based on the fusion of a U-Net model and a SegNet model; Calculating the vector shape area of ​​the building footprint using a convex hull algorithm based on a Graham scan method to obtain the building footprint, and calculating the building volume based on the building footprint; Performing a regression analysis on building environment independent variables and charging pile flexible resources to calculate the scale of charging piles in the target urban area; wherein the building environment independent variables include the building floor area and the building volume; The regression analysis of the building environment independent variables and the charging pile flexible resources is performed to calculate the scale of charging piles in the target urban area, specifically including: The Spearman rank correlation coefficient calculation formula was used to conduct regression analysis on the building environment independent variables and charging pile flexible resources; Based on the regression analysis results, the scale of charging piles is calculated by combining the complex correlation coefficient and influencing index factors; The Spearman rank correlation coefficient calculation formula is: Among them, V i is the building volume of the i-th building in the target urban area, n i is the number of charging piles in the i-th building volume, n is the total number of charging piles in the target urban area, R(V i ) and R(n i ) are V i and n i rank, and Respectively represent the average rank; The calculation formula for the charging pile scale is: Among them, R p is the building environment impact coefficient, R q is the impact coefficient of other factors, R p , R q ∈(0, 1), p < q; m represents the market potential of electric vehicles, f(t) represents the probability of the first purchase at time t, and n(t) represents the cumulative number of charging piles at time t.

2. The method for evaluating charging pile configuration requirements in urban areas based on remote sensing images according to claim 1, characterized in that: The learning offset vector model is constructed by the following steps: The U-Net model and the SegNet model are integrated to obtain the LOFT network model system; We implemented end-to-end building recognition in the LOFT network model system and modified the network layer based on the actual results, resulting in a learning offset vector model that integrates U-net and SegNet: L=L U-Net +α1L SegNet +α2L Mask +α3L Offset ; Among them, L U-Net , L SegNet , L Mask are the losses of U-Ne model, SegNet model and the first layer of mask respectively; L Offset is the loss of the offset head, and α1, α2, and α3 are the loss weights.

3. The method for evaluating charging pile configuration requirements in urban areas based on remote sensing images according to claim 1, characterized in that: After extracting the building footprint from the remote sensing image data using the pre-built learning offset vector model, and before calculating the vector shape area of ​​the building footprint using the convex hull algorithm based on the Graham scan method, the method further includes: The polygonal feature map of the building footprint is enhanced by using a function-level offset enhancement method to obtain an enhanced building footprint.

4. The method for evaluating charging pile configuration requirements in urban areas based on remote sensing images according to claim 1, wherein: The convex hull algorithm based on the Graham scan method calculates the vector shape area of ​​the building footprint, specifically including: Obtaining vectorized data of the building footprint, and continuously reducing the perimeter of the vector of the building footprint until it approaches the perimeter of the edge containing all vector points; Sort the points by polar angles with the point at the bottom left corner as the pole, and connect the points after polar angle sorting in sequence to obtain a polygon that surrounds all points; The vector rotates counterclockwise to form a convex polygon. When popping the top of the stack, the two sides of the triangle are replaced by the third side, while ensuring that the original vertex is included. Calculate the polar angle of the vector between the convex hull points; wherein the convex hull is the intersection of all convex polygons that can contain all points; Sort the polar angles with the point at the bottom left corner as the pole, and check whether the rotation direction of the two vectors composed of the three points is counterclockwise; According to the clockwise coordinates of all points constituting the convex hull, a cross product calculation is performed based on the Graham scan method to calculate the polygon area and obtain the vector shape area of ​​the building footprint.

5. The method for evaluating charging pile configuration requirements in urban areas based on remote sensing images according to claim 1, characterized in that: After acquiring the remote sensing image data containing the buildings in the target urban area, the method further includes: Perform image preprocessing operations on the remote sensing image data; wherein the image preprocessing operations include radiometric calibration, atmospheric correction and orthorectification processing.

6. A device for evaluating the configuration requirements of charging piles in urban areas based on remote sensing images, characterized in that: The method for evaluating the configuration requirements of charging piles in urban areas based on remote sensing images according to any one of claims 1 to 5 is executed, wherein the device comprises: A remote sensing image data acquisition module is used to acquire remote sensing image data of buildings including the target urban area; A building footprint extraction module is used to extract building footprints from the remote sensing image data using a pre-built learning offset vector model; wherein the learning offset vector model is constructed based on the fusion of a U-Net model and a SegNet model; a building volume calculation module, configured to calculate the vector shape area of ​​the building footprint based on a convex hull algorithm of a Graham scan method to obtain a building footprint, and calculate the building volume based on the building footprint; A charging pile scale assessment module is used to perform regression analysis on building environment independent variables and charging pile flexible resources to calculate the charging pile scale of the target urban area; wherein the building environment independent variables include the building floor area and the building volume.

7. A device for evaluating the configuration requirements of charging piles in urban areas based on remote sensing images, characterized in that: It includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and when the processor executes the computer program, it implements the urban charging pile configuration demand assessment method based on remote sensing images as described in any one of claims 1 to 5.

Images