Building photovoltaic resource evaluation method and system based on neural radiation field

The construction of a three-dimensional model through neural radiation fields solves the problem of lack of three-dimensional information in photovoltaic resource evaluation, and achieves rapid and accurate evaluation of rooftop photovoltaic resources, which is suitable for large-scale evaluation needs.

CN119005801APending Publication Date: 2024-11-22WUHAN UNIV +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411103232.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-13
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

The lack of three-dimensional information in the evaluation of photovoltaic resource in the prior art has led to inaccurate measurement of roof geometry, area and inclination angle, affecting the evaluation accuracy, and inefficient large-scale evaluation.

Method used

Using a neural radiation field-based method, through neural implicit surface representation learning, a three-dimensional model is constructed using aerial images and light information, the location of photovoltaic modules is marked, and the installation area is calculated to achieve fast and accurate photovoltaic resource evaluation.

Benefits of technology

Improves the accuracy and efficiency of photovoltaic resource evaluation, enables rapid extraction of roof area, is suitable for large-scale roof resource evaluation, and does not require specific processing or labeling of input data.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119005801A_ABST
    Figure CN119005801A_ABST
Patent Text Reader

Abstract

The invention discloses a building photovoltaic resource evaluation method and system based on a neural radiation field. The method comprises the following steps: collecting illumination related information of a measurement and calculation area; preparing surface reconstruction data; constructing a surface reconstruction model based on a neural radiation field, inputting the processed aerial image into a neural network through neural implicit surface representation learning, carrying out constraint according to collected illumination information, and generating a three-dimensional model of all building parts which are slightly influenced by shielding in the region; marking a position where a photovoltaic module can be installed on the three-dimensional model; and calculating the total area in which the photovoltaic modules can be installed in the measurement and calculation area by adopting a method for estimating the installation area of the photovoltaic modules. According to the method, a small number of control points are measured in a measurement and calculation area, a building is modeled by using a surface reconstruction model based on a neural radiation field, and finally the installable position and the installable total area of a photovoltaic module on a roof in the measurement and calculation area are obtained by using a photovoltaic evaluation algorithm.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of renewable energy evaluation, and in particular to a photovoltaic resource evaluation method and system based on implicit three-dimensional expression modeling. Background Art

[0002] The construction scale of ultra-low energy consumption and near-zero energy consumption buildings continues to grow. For example, the energy saving rate of newly built residential buildings in extremely cold and cold regions has reached 75%, the cumulative construction area of ​​ultra-low energy consumption and near-zero energy consumption buildings has reached nearly 10 million square meters, and the energy-saving renovation area of ​​existing residential buildings has reached 514 million square meters. However, the development of green buildings still faces some challenges, such as unbalanced regional development and insufficient technological innovation.

[0003] At the same time, compared with a large number of foreign studies, my country's research on photovoltaic potential still has some shortcomings. Most of the current research focuses on the extraction of two-dimensional distribution of buildings based on remote sensing images, and there is relatively little research on three-dimensional information of buildings. The three-dimensional model can accurately measure the geometry, area and inclination angle of the roof, avoiding the possible errors in two-dimensional estimation, thereby significantly improving the evaluation accuracy. As an emerging real-life three-dimensional reconstruction technology, neural radiation field has the characteristics of high quality and high fidelity of the generated three-dimensional model and does not require specific processing or marking of the input. Three-dimensional modeling based on neural radiation field can quickly complete the surface reconstruction of the building. Applying it to photovoltaic evaluation can ensure the quality of full use of three-dimensional information for evaluation while taking into account efficiency. Summary of the invention

[0004] In view of the deficiencies in the prior art, the purpose of the present invention is to provide a building photovoltaic resource assessment method based on neural radiation field, which has the effect of measuring and marking the installation area of ​​rooftop photovoltaic resources.

[0005] The technical solution adopted by the present invention to solve the technical problem is: a building photovoltaic resource assessment method based on neural radiation field, comprising the following steps:

[0006] Step 1: Collect and measure the regional lighting related information.

[0007] Step 2, surface reconstruction data preparation: Prepare the input data required for surface reconstruction including: 1) image pose and camera parameters; 2) aerial image;

[0008] Step 3: Build a surface reconstruction model based on neural radiation field. Through neural implicit surface representation learning, input the processed aerial image into the neural network, constrain it according to the illumination information collected in step 1, and generate the three-dimensional model of all the building parts in the area that are less affected by occlusion.

[0009] Step 4, marking the locations where photovoltaic modules can be installed on the 3D model;

[0010] Step 5: Calculate the total area in the measurement area where photovoltaic modules can be installed using the photovoltaic module installation area estimation method.

[0011] Furthermore, in step 1, the lighting information includes: the solar altitude angle range from 10:00 to 14:00 in the morning of each local day throughout the year, and the weather conditions from 10:00 to 14:00 recorded hourly every local day throughout the year.

[0012] Furthermore, in step 2, the position and posture of the image are solved by using the ground control points as control conditions to obtain the real position and posture at the moment of photography in the geodetic coordinate system where the control points are located. The image posture and camera parameters include: ① image posture parameters, that is, the position and direction of the camera in three-dimensional space; ② camera internal parameters, that is, focal length, image principal point position, and distortion parameters; after calculating the position and posture of the image, the aerial image needs to be removed from the imaging part outside the building, and each photo only retains the data of the building part to obtain an aerial image that only retains the building.

[0013] Furthermore, in step 3, the specific construction method of the surface reconstruction model based on the neural radiation field restricted by the lighting conditions is:

[0014] Step 3.1: Use the illumination information collected in step 1 to calculate the constraint conditions and generate the sun altitude angle α i , i=1,2,...,n-1,n, n is a constant, the horizontal direction due east is 0°, and the horizontal angle is generated at 45° intervals, denoted as β j , j=1,2,...,7,8;

[0015] Step 3.2, sampling 3D space points along the ray: First, use α i With β j 8n directions are generated Light for each pixel Launched from the origin o, along the direction Propagate, t corresponds to the position on the light, then, along 8n directions Emit rays and select only points visible in all directions for sampling to obtain the corresponding coordinate position (x, y, z) and viewing direction (θ, φ);

[0016] Step 3.3, construct the NeUS neural network F, with the coordinate position (x, y, z) and viewing direction (θ, φ) of the sampling point as input, and the corresponding color c (r, g, b) and the signed distance value SDF (p) of the sampling point as output, and the corresponding relationship is (c, SDF (p)) = F (x, y, z, θ, φ);

[0017] Step 3.4, only use the sampling points that meet the requirements in 3.2 for rendering, and record the light emitted from the pixel block on the rendered image as r(t)=o+tv d , where o represents the camera center, v d Represents the light direction of the original photo, and the volume rendering formula is Among them, C(o,v d ) represents the color of the target pixel block after rendering calculation, w(t) = T(t)σ(t), w(t) is associated with the signed distance value SDF(p), T(t) represents the light accumulation along the light, σ(t) is called the volume density, w(t) is the weight value of point r(t), c(r(t),v d ) is the spatial point r(t) in the line of sight direction v d The color value below;

[0018] Step 3.5, limit the constraints: In the loss function of the surface reconstruction model, add the constraints on the unoccluded parts along the given direction, and introduce a visibility weight ω(x,v) for each sampling point d ), where x represents the spatial coordinates of the sampling point; if a sampling point is not blocked in a given direction, it is 1, and if it is blocked, it is 0; on the basis of the original loss function, multiply it by the visibility weight ω(x,v d ), so that only the unobstructed part in a given direction will contribute to the loss function; assuming that the number of sampling points for the same ray is n and the number of rays is m, then we can get is the difference between the rendered image and the actual image, C k represents the real pixel value, is the predicted pixel value; λ is the weight coefficient of the regularization loss term; represents the regularization loss, where It's on the ray The gradient of the signed distance function of the point, where k represents different ray numbers and i represents different sampling points on the ray.

[0019] Furthermore, the specific implementation of step 3.1 is as follows:

[0020] First, organize the information collected every day throughout the year into a data set, divide the solar altitude angle into several intervals, and classify the solar altitude angle every hour of every day throughout the year according to the previous intervals. Correct the effective sunshine duration per hour according to weather conditions; accumulate the effective sunshine duration in each altitude angle interval to obtain the total illumination duration of the altitude angle interval throughout the year; compare the total illumination duration of each altitude angle interval to find the interval with a total illumination duration greater than m hours throughout the year. The altitude angle interval that meets the above requirements is the altitude angle range with the strongest illumination in a year. The illumination in these intervals is at intervals of 1°. Assume that there are n intervals, and the generated solar altitude angle is recorded as α i .

[0021] Furthermore, in step 4, the method for determining the installation position of the photovoltaic module is:

[0022] Step 4.1, generate a model of the visible roof part of the house: input the image pose and camera parameters and the processed aerial image processing into the neural radiation field-based surface reconstruction model constrained by lighting conditions for training, and output the surface model;

[0023] Step 4.2, extracting the building roof plane that can be installed;

[0024] Step 4.3, convert the coordinates of all points on the building roof plane from the three-dimensional space to a unified two-dimensional space for processing; specifically: create a new three-dimensional space coordinate in the original three-dimensional space, and record the rotation matrix R of the plane where the other roofs are located relative to the coordinate system xOy plane i , centroidalize the roof plane and record the offset of its centroid relative to the origin (X0, Y0, Z0). i ,Y i ,Z i ) according to the following formula:

[0025]

[0026] Translate and rotate to transform all roofs to the same plane of the new coordinate system, thus obtaining the two-dimensional coordinates (x i ,y i );

[0027] Step 4.4, obtain the minimum covering rectangle MRB of each roof plane, that is, find the minimum rectangle that can completely contain the given roof plane; first use the PCA algorithm to calculate the main direction of the polygon, according to the angle θ between the main direction and the coordinate axis i , get the rotation matrix R θi , R θi Multiply the coordinates of all points in the plane to get (x', y'), and then use the maximum and minimum horizontal coordinates (xmin ,x max ) and the maximum and minimum ordinates (y min ,y max ) to create a rectangle;

[0028] Step 4.5, calculate the installation position: for each building roof plane that can be installed, generate a small rectangle of the area where the photovoltaic modules can be installed starting from the upper left corner in the obtained MRB, and each two rows and columns of adjacent small rectangles are separated by a certain distance as a maintenance channel; thus, the four vertex coordinates (x′ i-1 ,y′ j-1 ),(x i-1 ′+2,y′ j-1 ),(x′ i-1 ,y′ j-1 +1),(x′ i-1 +2,y′ j-1 +1); for each small rectangle, determine whether each of its points is located in the original polygon. If not, delete it; all small rectangles that meet the requirements and their corresponding polygons are rotated using the inverse matrix Return to the original direction;

[0029] Then the two-dimensional coordinates can be rotated to the original direction and translated to transform the two-dimensional coordinates into three-dimensional coordinates. The three-dimensional coordinates can be converted to the location where the photovoltaic module can be installed in the three-dimensional space.

[0030] Furthermore, in step 4.2, the roof surfaces on which photovoltaic modules can be installed include sloping roofs and flat roofs; for flat roofs, D1 is used as the angle threshold between the surface normal vector and the horizontal plane normal vector to filter out horizontal surfaces; for sloping roofs, D2 is used as the angle threshold between the surface normal vector and the horizontal plane normal vector to filter out. Since the efficiency of installing a photovoltaic system on the north slope of the roof is too low, if the angle between the horizontal component of the normal vector and the north direction is less than 45°, it will be eliminated.

[0031] Furthermore, in step 4.5, the two-dimensional coordinates are rotated to the original direction and translated according to the following formula:

[0032]

[0033] Among them, since the small rectangular area where the photovoltaic components can be installed is generated in the two-dimensional polygon, the transformation of the photovoltaic component installation area from two-dimensional to three-dimensional is consistent with the corresponding two-dimensional polygon area.

[0034] Furthermore, in step 5, the method for estimating the installation area of ​​the photovoltaic module is as follows: for the polygons corresponding to the installation positions on the two-dimensional plane, the area is calculated using the coordinates of their corner points. For each polygon, assuming that the polygon has N points, the calculation formula is According to this formula, we can get the area S. Assuming there are M polygons, the total area that can be installed is

[0035] The present invention also provides a building photovoltaic resource assessment system based on neural radiation fields, including a processor and a memory, the memory being used to store program instructions, and the processor being used to call the stored instructions in the memory to execute a building photovoltaic resource assessment method based on neural radiation fields as described in the above scheme.

[0036] The beneficial effects produced by the present invention are:

[0037] 1) With the increase of rooftop photovoltaic projects, the photovoltaic resource assessment method based on neural radiation field surface reconstruction can accurately and quickly extract the roof area and evaluate the photovoltaic power generation potential. Compared with the traditional roof area measurement method, which is time-consuming and labor-intensive, it can meet the needs of large-scale roof resource assessment.

[0038] 2) The data required is relatively simple, and a 3D model can be generated from any number of rooftop drone images without the need for specific processing or labeling of the input. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which:

[0040] Figure 1 is an overall flow chart of an embodiment of the present invention;

[0041] Figure 2 It is a schematic diagram of the relationship between the minimum covering matrix, the original polygon, and the small rectangles of the installable positions;

[0042] Figure 3 This is a rendering of the installation position of the photovoltaic module of the present invention and the superposition effect of the complete house model;

[0043] Figure 4 This is a rendering of an embodiment of the present invention. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0045] like Figure 1 As shown, the photovoltaic resource assessment method based on neural radiation field surface reconstruction according to an embodiment of the present invention comprises the following steps:

[0046] Step 1: Collect and calculate regional illumination information. Collect the solar altitude angle range from 10:00 to 14:00 in the morning every day throughout the year, and the weather conditions recorded hourly from 10:00 to 14:00 every day throughout the year, to prepare for the estimation of the photovoltaic module installation area and the determination of the possible installation locations;

[0047] Step 2: Prepare data for surface reconstruction. The input data required for surface reconstruction include: 1) image pose and camera parameters; 2) after calculating the position and pose of the image, the aerial image needs to be removed from the image part other than the building, and each photo only retains the data of the building part, thus obtaining an aerial image that only retains the building;

[0048] Step 3: Construct a surface reconstruction model based on the neural radiation field (the surface reconstruction model in the embodiment adopts NeUS). The specific construction method of the surface reconstruction model based on the neural radiation field limited by the illumination conditions is as follows:

[0049] Step 3.1, use the light information collected in step 1 to calculate the constraints. First, organize the information collected every day throughout the year into a data set. In order to simplify the analysis, the solar altitude angle can be divided into several intervals (for example, every 5 degrees is an interval: 0-5 degrees, 5-10 degrees, 10-15 degrees, and so on), and the solar altitude angle of each hour of each day throughout the year is classified according to the previous intervals. Correct the effective sunshine duration per hour according to the weather conditions. For example, it is calculated as 1 hour on a sunny day, 0.5 hours on a cloudy day, and 0 hours on a rainy day. Accumulate the effective sunshine duration in each altitude angle interval to obtain the total illumination duration of the altitude angle interval throughout the year. Compare the total illumination duration of each altitude angle interval and find the interval with a total illumination duration of more than 730 hours (that is, 50% of the statistical data) throughout the year. The altitude angle interval that meets this requirement is the altitude angle range with the strongest illumination in a year. The illumination in these intervals is at intervals of 1°. Assume that there are n intervals, and the generated solar altitude angle is recorded as α i (i=1,2,…,n-1,n), the horizontal direction due east is 0°, and the horizontal angle is generated at 45° intervals, denoted as β j (j=1,2,…,7,8).

[0050] Step 3.2: Sample 3D points along the ray. First, use α i With β j 8n directions are generated Light for each pixel Launched from the origin o, along the direction Propagate, t corresponds to the position on the light. Then, along 8n directions Emit rays and select only points that are visible in all directions for sampling to obtain the corresponding coordinate position (x, y, z) and viewing direction (θ, φ).

[0051] Step 3.3, construct the NeUS neural network F. The coordinate position (x, y, z) and viewing direction (θ, φ) of the sampling point are used as input, and the corresponding color c (r, g, b) and the signed distance value SDF (p) of the point are used as output to construct the neural network, and the corresponding relationship is (c, SDF (p)) = F (x, y, z, θ, φ).

[0052] Step 3.4: Rendering. Only the sampling points that meet the requirements in 3.2 are used for rendering, and the light emitted from the pixel block on the rendering image is recorded as r(t)=o+tv d , where o represents the camera center, which is different from 3.2 Represents the simulated sunlight direction, where v d Represents the light direction of the original photo, and the volume rendering formula is Among them, C(o,v d ) represents the color of the target pixel block after rendering calculation, w(t) = T(t)σ(t), and w(t) is associated with the signed distance value SDF(p). T(t) represents the light accumulation along the light, σ(t) is called the volume density, and w(t) is the weight value of point r(t). c(r(t),v d ) is the spatial point r(t) in the line of sight direction v d The color value below.

[0053] Step 3.5: Constraints. In the loss function, add constraints on the unobstructed parts along a given direction. Introduce a visibility weight ω(x,v d ), where x represents the spatial coordinate of the sampling point. If a point is not blocked in a given direction, it is 1, and if it is blocked, it is 0. Based on the original loss function, multiply it by the visibility weight ω(x,v d ) In this way, only the unobstructed part in a given direction will contribute to the loss function. Assuming that the number of sampling points of the same ray is n and the number of rays is m, we can get is the difference between the rendered image and the actual image, C k represents the real pixel value, is the predicted pixel value; λ is the weight coefficient of the regularization loss term;

[0054] represents the regularization loss, where It's on the ray The gradient of the signed distance function (SDF) of a point. SDF is used here to represent the implicit surface of a 3D geometric shape. Since the SDF function has the property that the modulus of the point normal on the zero level set is equal to 1, this loss term is set here to constrain the SDF network to make it close to the real SDF function. Here k represents different ray numbers, and i represents different sampling points on the ray.

[0055] Step 4: The method for determining the installation location of the photovoltaic modules is as follows:

[0056] Step 4.1: Generate a model of the visible roof part of the house. Input the image pose, camera parameters and processed aerial image processing into the surface reconstruction model based on neural radiation field under lighting conditions for training, and output the surface model.

[0057] Step 4.2, extract the building roof plane that can be installed. For the roof surface where photovoltaic modules can be installed, two cases are considered: sloping roof and flat roof. For flat roof, use D1 as the angle threshold between the surface normal vector and the horizontal plane normal vector (can be set to 15°) to filter almost horizontal surfaces. For sloping roof, use D2 as the angle threshold between the surface normal vector and the horizontal plane normal vector (can be set to 10° to 35°) for screening. Since the efficiency of installing photovoltaic systems on the north slope of the roof is too low, if the angle between the horizontal component of the normal vector and the due north direction is less than 45°, it will be eliminated.

[0058] Step 4.3, from 3D projection to 2D projection. All point coordinates are converted from 3D space to a unified 2D space for processing. Specifically: create a new 3D space coordinate in the original 3D space, and record the rotation matrix R of the plane where the other roofs are located relative to the xOy plane of the coordinate system. i , centroidalize the roof plane and record the offset of its centroid relative to the origin (X0, Y0, Z0). i ,Y i ,Z i ) according to the following formula:

[0059]

[0060] Translate and rotate all roofs to the same plane of the new coordinate system (near the origin), thus obtaining the two-dimensional coordinates (x i ,y i ).

[0061] Step 4.4, obtain the minimum covering rectangle (MBR) of each roof plane. That is, find the smallest rectangle that can completely contain the given roof plane. First, use the PCA algorithm to calculate the main direction of the polygon, and then use the angle θ between the main direction and the coordinate axis to calculate the minimum covering rectangle (MBR).i , get the rotation matrix R θi , R θi Multiply the coordinates of all points in the plane to get (x', y'). Then use the maximum and minimum horizontal coordinates (x min ,x max ) and the maximum and minimum ordinates (y min ,y max ) to create a rectangle.

[0062] Step 4.5, calculate the installation position. For each building roof plane that can be installed, generate a small rectangle of the area where the photovoltaic modules can be installed starting from the upper left corner of the obtained MRB (the long side l can be set to 2m, and the short side w can be set to 1m), and the interval between every two adjacent rows and columns of small rectangles is 1m as a maintenance channel. In this way, the four vertex coordinates (x′ i-1 ,y′ j-1 ),(x i-1 ′+2,y′ j-1 ),(x′ i-1 ,y′ j-1 +1),(x′ i-1 +2,y′ j-1 +1). For each small rectangle, determine whether each point is located in the original polygon. If not, delete it. All small rectangles that meet the requirements and their corresponding polygons are rotated using the inverse matrix Return to the original direction. The relationship between the minimum covering matrix, the original polygon, and the small rectangle of the installable position is as follows: Figure 2 As shown, the red ones are not installable and the green ones are installable.

[0063] According to the following formula, the two-dimensional coordinates can be rotated to the original direction and translated to transform the two-dimensional coordinates into three-dimensional coordinates:

[0064]

[0065] Among them, since the small rectangular area where the photovoltaic modules can be installed is generated in the two-dimensional polygon, the transformation of the photovoltaic module installation area from two-dimensional to three-dimensional is consistent with the corresponding two-dimensional polygon area. Finally, the output is converted into three-dimensional coordinates to know the location where the photovoltaic modules can be installed in three-dimensional space. The effect is as follows Figure 3 As shown, the figure shows the possible installation locations of PV panels superimposed on the complete house model.

[0066] Step 5: The method for estimating the installation area of ​​PV modules is as follows: for the polygons corresponding to the installation positions on the two-dimensional plane, the area is calculated using the coordinates of their corner points. For each polygon, assuming there are N points, the calculation formula is According to this formula, we can get the area S. Assuming there are M polygons, the total area that can be installed is

[0067] According to the specific steps of the above embodiment, taking the illumination-related information collected in the measurement area and the image with internal and external parameters of the camera as an example, a surface reconstruction model based on the neural radiation field is constructed with the help of the constraints of illumination information. A three-dimensional model of the surface of the house that is not obstructed is established through image data. Since the position and posture of the images are all in the geographic coordinate system, the constructed model is also in the geographic coordinate system and the scale is consistent with reality. It can be processed to obtain the installation position and area of ​​photovoltaic components with practical significance. Using the constructed roof model, the location where photovoltaic resources can be installed is processed, and the location is superimposed with the complete model of the house in the area. The visual effect is as follows: Figure 4 shown. Figure 4 The visual effect proves the feasibility and good effect of the present invention on building photovoltaic resource assessment.

[0068] In specific operations, the method proposed in the present invention can be implemented with the aid of computer software technology to achieve automated process execution. The system device implementing the method may include a readable storage medium storing relevant computer programs, and a computer device capable of running these computer programs. It should be emphasized that these implementation methods are within the scope of protection of the present invention and can be adopted by professionals in the technical field.

[0069] An embodiment of the present invention also provides a building photovoltaic resource assessment system based on neural radiation fields, including a processor and a memory, the memory being used to store program instructions, and the processor being used to call the stored instructions in the memory to execute a building photovoltaic resource assessment method based on neural radiation fields as described in the above scheme.

[0070] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A building photovoltaic resource assessment method based on neural radiation field, characterized in that: The following steps are involved: Step 1: Collect and measure the regional lighting related information. Step 2, surface reconstruction data preparation: Prepare the input data required for surface reconstruction including: 1) image pose and camera parameters; 2) aerial image; Step 3: Build a surface reconstruction model based on neural radiation field. Through neural implicit surface representation learning, input the processed aerial image into the neural network, constrain it according to the illumination information collected in step 1, and generate the three-dimensional model of all the building parts in the area that are less affected by occlusion. Step 4, marking the locations where photovoltaic modules can be installed on the 3D model; Step 5: Calculate the total area in the measurement area where photovoltaic modules can be installed using the photovoltaic module installation area estimation method.

2. The method for evaluating building photovoltaic resources based on neural radiation field according to claim 1, characterized in that: In step 1, the illumination information includes: the solar altitude angle range from 10:00 to 14:00 in the morning of each local day throughout the year, and the weather conditions from 10:00 to 14:00 recorded hourly on each local day throughout the year.

3. The method for evaluating building photovoltaic resources based on neural radiation field according to claim 1, characterized in that: In step 2, the position and posture of the image are solved by using the ground control points as control conditions to obtain the real position and posture at the moment of photography in the geodetic coordinate system where the control points are located. The image posture and camera parameters include: ① image posture parameters, that is, the position and direction of the camera in three-dimensional space; ② camera internal parameters, that is, focal length, image principal point position, and distortion parameters; after calculating the position and posture of the image, the aerial image needs to be removed from the imaging part outside the building, and each photo only retains the data of the building part to obtain an aerial image that only retains the building.

4. The method for evaluating building photovoltaic resources based on neural radiation field according to claim 1, characterized in that: In step 3, the specific construction method of the surface reconstruction model based on the neural radiation field restricted by the lighting conditions is: Step 3.1: Use the illumination information collected in step 1 to calculate the constraint conditions and generate the sun altitude angle α i , i=1,2,…,n-1,n, n is a constant, the horizontal direction due east is 0°, and the horizontal angle is generated at 45° intervals, denoted as β j , j=1,2,…,7,8; Step 3.2, sampling 3D space points along the ray: First, use α i With β j 8n directions are generated k=1,2,…,8n-1,8n, each pixel corresponds to a ray Emitted from the origin o, along the direction Propagate, t corresponds to the position on the light, then, along 8n directions Emit rays and select only points visible in all directions for sampling to obtain the corresponding coordinate position (x, y, z) and viewing direction (θ, φ); Step 3.3, construct the NeUS neural network F, with the coordinate position (x, y, z) and viewing direction (θ, φ) of the sampling point as input, and the corresponding color c (r, g, b) and the signed distance value SDF (p) of the sampling point as output, and the corresponding relationship is (c, SDF (p)) = F (x, y, z, θ, φ); Step 3.4, only use the sampling points that meet the requirements in 3.2 for rendering, and record the light emitted from the pixel block on the rendered image as r(t)=o+tv d , where o represents the camera center, v d Represents the light direction of the original photo, and the volume rendering formula is Among them, C(o,v d ) represents the color of the target pixel block after rendering calculation, w(t) = T(t)σ(t), w(t) is associated with the signed distance value SDF(p), T(t) represents the light accumulation along the light, σ(t) is called the volume density, w(t) is the weight value of point r(t), c(r(t),v d ) is the spatial point r(t) in the line of sight direction v d The color value below; Step 3.5, limit the constraints: In the loss function of the surface reconstruction model, add the constraints on the unoccluded parts along the given direction, and introduce a visibility weight ω(x,v) for each sampling point d ), where x represents the spatial coordinates of the sampling point; if a sampling point is not blocked in a given direction, it is 1, and if it is blocked, it is 0; on the basis of the original loss function, multiply it by the visibility weight ω(x,v d ), so that only the unobstructed part in a given direction will contribute to the loss function; assuming that the number of sampling points for the same ray is n and the number of rays is m, then we can get is the difference between the rendered image and the actual image, C k represents the real pixel value, is the predicted pixel value; λ is the weight coefficient of the regularization loss term; represents the regularization loss, where It's on the ray The gradient of the signed distance function of the point, where k represents different ray numbers and i represents different sampling points on the ray.

5. The method for evaluating building photovoltaic resources based on neural radiation field according to claim 4, characterized in that: The specific implementation of step 3.1 is as follows: First, organize the information collected every day throughout the year into a data set, divide the solar altitude angle into several intervals, and classify the solar altitude angle for every hour of every day throughout the year according to the previous intervals. Correct the effective sunshine duration per hour according to weather conditions; The effective sunshine duration in each altitude interval is accumulated to obtain the total sunshine duration of the altitude interval throughout the year; the total sunshine duration of each altitude interval is compared to find the interval with a total sunshine duration of more than m hours throughout the year. The altitude interval that meets the above requirements is the altitude range with the strongest sunshine in a year. The sunshine in these intervals is set at an interval of 1°. Assuming there are n intervals, the generated solar altitude angle is recorded as α i .

6. The method for evaluating building photovoltaic resources based on neural radiation field according to claim 1, characterized in that: In step 4, the method for determining the installation location of the photovoltaic module is as follows: Step 4.1, generate a model of the visible roof part of the house: input the image pose and camera parameters and the processed aerial image processing into the neural radiation field-based surface reconstruction model constrained by lighting conditions for training, and output the surface model; Step 4.2, extracting the building roof plane that can be installed; Step 4.3, convert the coordinates of all points on the building roof plane from the three-dimensional space to a unified two-dimensional space for processing; specifically: create a new three-dimensional space coordinate in the original three-dimensional space, and record the rotation matrix R of the plane where the other roofs are located relative to the coordinate system xOy plane i , centroidalize the roof plane and record the offset of its centroid relative to the origin (X0, Y0, Z0). i ,Y i ,Z i ) according to the following formula: Translate and rotate to transform all roofs to the same plane of the new coordinate system, thus obtaining the two-dimensional coordinates (x i ,y i ); Step 4.4, obtain the minimum covering rectangle MRB of each roof plane, that is, find the minimum rectangle that can completely contain the given roof plane; first use the PCA algorithm to calculate the main direction of the polygon, according to the angle θ between the main direction and the coordinate axis i , get the rotation matrix R θi , R θi Multiply the coordinates of all points in the plane to get (x', y'), and then use the maximum and minimum horizontal coordinates (x min ,x max ) and the maximum and minimum ordinates (y min ,y max ) to create a rectangle; Step 4.5, calculate the installation position: for each building roof plane that can be installed, generate a small rectangle of the area where the photovoltaic modules can be installed starting from the upper left corner in the obtained MRB, and each two rows and columns of adjacent small rectangles are separated by a certain distance as a maintenance channel; thus, the four vertex coordinates (x′ i-1 ,y′ j-1 ),(x i-1 ′+2,y′ j-1 ),(x′ i-1 ,y′ j-1 +1),(x′ i-1 +2,y′ j-1 +1); for each small rectangle, determine whether each of its points is located in the original polygon. If not, delete it; all small rectangles that meet the requirements and their corresponding polygons are rotated using the inverse matrix Return to the original direction; Then the two-dimensional coordinates can be rotated to the original direction and translated to transform the two-dimensional coordinates into three-dimensional coordinates. The three-dimensional coordinates can be converted to the location where the photovoltaic module can be installed in the three-dimensional space.

7. The method for evaluating building photovoltaic resources based on neural radiation field according to claim 6, characterized in that: In step 4.2, the roof surfaces where photovoltaic modules can be installed include sloping roofs and flat roofs. For flat roofs, D1 is used as the angle threshold between the surface normal vector and the horizontal plane normal vector to select horizontal surfaces. For sloping roofs, D2 is used as the angle threshold between the surface normal vector and the horizontal plane normal vector to select. Since the efficiency of installing a photovoltaic system on the north slope of the roof is too low, if the angle between the horizontal component of the normal vector and the north direction is less than 45°, it will be eliminated.

8. The method for evaluating building photovoltaic resources based on neural radiation field according to claim 6, characterized in that: In step 4.5, the two-dimensional coordinates are rotated to the original direction and translated according to the following formula: Among them, since the small rectangular area where the photovoltaic components can be installed is generated in the two-dimensional polygon, the transformation of the photovoltaic component installation area from two-dimensional to three-dimensional is consistent with the corresponding two-dimensional polygon area.

9. The method for evaluating building photovoltaic resources based on neural radiation field according to claim 8, characterized in that: In step 5, the method for estimating the installation area of ​​PV modules is as follows: for the polygons corresponding to the installation positions on the two-dimensional plane, the area is calculated using the coordinates of their corner points. For each polygon, assuming that the polygon has N points, the calculation formula is According to this formula, we can get the area S. Assuming there are M polygons, the total area that can be installed is 10. A building photovoltaic resource assessment system based on neural radiation field, characterized in that: It includes a processor and a memory, the memory is used to store program instructions, and the processor is used to call the stored instructions in the memory to execute a building photovoltaic resource assessment method based on neural radiation field as described in any one of claims 1 to 9.