A method for evaluating rooftop photovoltaic potential based on machine learning

Abstract

The present invention discloses a method for evaluating the potential of rooftop photovoltaic power generation based on machine learning. In order to overcome the problem that the existing rooftop identification method is single and prone to errors and interference; the present invention includes the following steps: S1: obtaining stereoscopic remote sensing image data from the top of the building through satellite or aerial photography; S2: inputting a set of data into a trained recognition model to identify the rooftop area; performing matrix segmentation of a set of data according to the rated length and width; S3: performing boundary identification on each matrix unit data respectively, obtaining a closed boundary curve after restoration and combination, and determining the rooftop area; S4: calculating the rooftop area based on the coupling of the rooftop area identified by the recognition model and the rooftop area determined by the closed boundary curve, and evaluating the potential of rooftop photovoltaic power generation in combination with the energy consumption and environmental information of the station area where the building is located. By combining the recognition model with the segmentation recognition, the rooftop area is determined by combining the recognition results of the two methods, the recognition method is diverse, and the accuracy of the recognition result is improved.

Description

Technical Field

[0001] The present invention relates to the field of photovoltaic potential assessment, and in particular to a rooftop photovoltaic potential assessment method based on machine learning. Background Art

[0002] With the increasing global energy demand, photovoltaic power generation has attracted more and more attention due to its advantages such as cleanliness, convenience, safety, and suitability for distributed networking, and has become one of the most promising renewable energy power generation methods. Quantitative evaluation of solar energy resource potential is the basis for energy planning, providing basic data and engineering construction guidance for the development and utilization of regional solar energy resources. The existing methods have a single rooftop identification method, which is prone to interference and errors.

[0003] For example, a "distributed photovoltaic resource integration method, system, device and storage medium" disclosed in Chinese patent documents, with announcement number CN113076855A, includes: obtaining original pictures with roof information through satellites; inputting the obtained original pictures into a pre-trained artificial intelligence model, and filtering out the first pictures that can install photovoltaics from the original pictures; identifying the coordinate information of the first picture, matching the coordinate information with different customers through a big data algorithm, and obtaining a second picture associated with the customer; counting the coordinate information of the second picture into a database, and analyzing the potential data of the second picture in combination with meteorological data and industry data; integrating photovoltaic resources according to the obtained potential data. This solution has a single roof recognition method, which is prone to interference and errors, and has poor anti-interference ability. Summary of the invention

[0004] The present invention mainly solves the problem that the existing roof recognition method is single and prone to errors and interference; it provides a roof photovoltaic potential assessment method based on machine learning, which combines the recognition model with the segmentation recognition, and determines the roof area from the combination of the recognition results of the two methods. The recognition method is diverse, which improves the accuracy of the recognition result.

[0005] The above technical problems of the present invention are mainly solved by the following technical solutions:

[0006] A method for evaluating rooftop photovoltaic potential based on machine learning, comprising the following steps:

[0007] S1: Obtain stereo remote sensing image data from the top of the building through satellite or aerial photography;

[0008] S2: Copy the pre-processed stereo remote sensing image data, one copy of the data is input into the trained recognition model to identify the roof area; the other copy of the data is matrix segmented according to the rated length and width dimensions;

[0009] S3: performing boundary recognition on each matrix unit data after matrix segmentation, obtaining a closed boundary curve after restoration and combination, and determining the roof area;

[0010] S4: The roof area is calculated by coupling the roof area identified by the recognition model and the roof area determined by the closed boundary curve, and the rooftop photovoltaic potential is evaluated in combination with the energy consumption and environmental information of the station area where the building is located.

[0011] By combining the recognition model with segmentation recognition, the roof area is determined by combining the recognition results of the two methods. The recognition methods are diverse and the accuracy of the recognition results is improved.

[0012] Preferably, the stereo remote sensing image data is preprocessed by a UNB pan-sharpening method that includes both the MS band and the Pan band. This method preserves the color of the object image.

[0013] Preferably, the trained recognition model adopts U-net network recognition. Through machine learning, the model is continuously trained to recognize various forms of roofs and remove the influence of surrounding roads and vegetation.

[0014] Preferably, the step S3 specifically comprises the following steps:

[0015] S301: For each matrix unit data, extract the RGB data of each data point;

[0016] S302: traverse each data point, compare the RGB data with the points around the data point, and extract the boundary curve;

[0017] S303: according to the matrix segmentation positions, restore and splice the matrix units, connect the boundary curves, and form a closed area;

[0018] S304: the closed area is matched with each roof shape in the database; if the match is successful, the closed area is defined as a roof area; otherwise, it is determined to be an interference area and the closed area is deleted.

[0019] Divide a large image into several small images to improve computational efficiency and result accuracy.

[0020] Preferably, the step S303 further includes the following process:

[0021] The area of ​​the closed area is calculated to determine whether the area of ​​the closed area is greater than the building threshold; if so, it is determined that the closed area has the ability to build photovoltaic power and the closed area is retained; otherwise, the closed area is deleted.

[0022] Remove roof areas that do not meet building standards in advance.

[0023] Preferably, the roof area calculation process in step S4 is as follows:

[0024] S401: overlapping the roof area identified by the corresponding recognition model and the roof area determined by the closed boundary curve according to the corresponding relationship of the coordinates;

[0025] S402: For the same corresponding roof, the overlapping portion of the two roof regions is determined as the building roof;

[0026] S403: Calculate the area of ​​the non-overlapping parts of the two roof areas corresponding to the same roof; if the area of ​​the non-overlapping parts is less than or equal to the error area threshold, determine that the area is a non-building roof and delete the area; otherwise, proceed to the next step of determination;

[0027] S404: Determine whether the non-overlapping area is a custom regular shape; if so, determine that the area is a building roof, otherwise delete the area;

[0028] S405: Calculate the roof areas of the roofs determined to be building roofs respectively.

[0029] Two ways to determine the building roof make the result more accurate.

[0030] Preferably, the calculation process of the rooftop photovoltaic potential in step S4 is as follows:

[0031] S406: Obtain the average daily energy consumption of the area where the building is located in each quarter according to the satellite positioning information;

[0032] S407: According to the statistical information of the meteorological bureau, the average daily light intensity of the region in each season is obtained;

[0033] S408: Evaluate the daily average energy consumption ratio by quarter;

[0034]

[0035] Among them, P is the average daily energy consumption proportion in a quarter;

[0036] ρ is the photovoltaic conversion efficiency per unit area per unit time;

[0037] L is the average daily light intensity of the current quarter;

[0038] S na is the roof area of ​​the nth building roof;

[0039] S nht is the blocked roof area of ​​the nth building roof at time t;

[0040] T is the average daily sunshine time in the current quarter;

[0041] t is the shading time of the building roof;

[0042] t1 is the occlusion start time;

[0043] t2 is the occlusion start time;

[0044] C d is the average daily energy consumption of the area in the current quarter;

[0045] S409: Establish a rooftop photovoltaic potential assessment table, with different daily average energy consumption ratios corresponding to different potential levels;

[0046] S410: Take the weighted average of the daily energy consumption proportions of the four quarters and compare it with the roof potential assessment table to determine the roof potential level.

[0047] The proportion of electricity provided by rooftop photovoltaics to power the substation is used to assess its photovoltaic potential and determine whether it is worth building.

[0048] The beneficial effects of the present invention are:

[0049] 1. By combining the recognition model with segmentation recognition, the roof area is determined by combining the recognition results of the two methods. The recognition methods are diverse and the accuracy of the recognition results is improved.

[0050] 2. Divide the large image into several small images to improve the calculation efficiency and accuracy of the results. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a flow chart of the rooftop photovoltaic potential assessment method of the present invention. DETAILED DESCRIPTION

[0052] The technical solution of the present invention is further specifically described below through embodiments and in conjunction with the accompanying drawings.

[0053] Example:

[0054] A method for evaluating rooftop photovoltaic potential based on machine learning in this embodiment is as follows: Figure 1 As shown, the following steps are included:

[0055] S1: Obtain stereo remote sensing image data from the top of the building through satellite or aerial photography.

[0056] Using stereo remote sensing data from the RPCs associated with it, a digital surface model of the area was generated.

[0057] S2: Copy the pre-processed stereo remote sensing image data, one copy of the data is input into the trained recognition model to identify the roof area; and one copy of the data is matrix-segmented according to the rated length and width dimensions.

[0058] More than 70% of optical earth observation satellites and many modern aerial digital cameras can simultaneously capture low-resolution multispectral (MS) images and high-resolution panchromatic (Pan) images. The number of MS bands and the spectral bandwidth of Pan images often vary from sensor to sensor. In addition, seasonal and regional differences can also lead to differences in image information.

[0059] The stereo remote sensing image data is preprocessed by UNB pan-sharpening method that includes both MS band or Pan band. The color of the object image is retained, and the least square method is used to find the best fitting information between MS and Pan bands. Then this information is used to adjust the image processing parameters to achieve the best effect of Pan sharpening. During the fusion process, each group of Pan and MS effects are processed separately according to their image information. Regardless of the differences in sensors, seasons and regions, it can continuously and stably obtain good fusion effects.

[0060] The trained recognition model is identified by the U-net network. Through machine learning, the model is continuously trained to identify various forms of roofs and remove the influence of surrounding roads and vegetation. The collected stereo remote sensing data is divided into a training set and a test set. The U-net network is continuously trained through the training set containing various roofs, and finally a trained recognition model is obtained through the test set test.

[0061] S3: Perform boundary recognition on each matrix unit data after matrix segmentation, obtain a closed boundary curve after restoration and combination, and determine the roof area.

[0062] S301: For each matrix unit data, extract the RGB data of each data point.

[0063] S302: Traverse each data point, compare the RGB data with the points around the data point, and extract the boundary curve.

[0064] S303: According to the matrix segmentation positions, the matrix units are restored and spliced, and the boundary curves are connected, and the closed boundary curves form a closed area.

[0065] It is determined whether the spliced ​​boundary curves are all connected to form a continuous closed curve. If so, the process ends and proceeds to step S304; otherwise, the process proceeds to the next step of determination.

[0066] Determine the distance between the endpoint of the boundary curve at the edge of the matrix image unit and the nearest endpoint of the boundary curve at the edge of the adjacent matrix image unit. If the distance is less than the error threshold, take the middle point of the boundary curve endpoints at the connection of the two matrix image units as the correction point, move the boundary curve endpoints to the correction point, and fit them into a curve; if the distance is greater than the error threshold, return to step S302 to identify the boundary curve again.

[0067] The area of ​​the closed area is calculated to determine whether the area of ​​the closed area is greater than the building threshold; if so, it is determined that the closed area has the ability to build photovoltaic power and the closed area is retained; otherwise, the closed area is deleted.

[0068] S304: the closed area is matched with each roof shape in the database; if the match is successful, the closed area is defined as a roof area; otherwise, it is determined to be an interference area and the closed area is deleted.

[0069] S4: The roof area is calculated by coupling the roof area identified by the recognition model and the roof area determined by the closed boundary curve, and the rooftop photovoltaic potential is evaluated in combination with the energy consumption and environmental information of the station area where the building is located.

[0070] S401: According to the corresponding relationship of the coordinates, the roof area identified by the corresponding recognition model and the roof area determined by the closed boundary curve are overlapped.

[0071] S402: For the same corresponding roof, the overlapping portion of the two roof areas is determined as the building roof.

[0072] S403: Calculate the area of ​​the non-overlapping parts of the two roof areas corresponding to the same roof; if the area of ​​the non-overlapping parts is less than or equal to the error area threshold, determine that the part of the area is not a building roof and delete the part of the area; otherwise, proceed to the next step of judgment.

[0073] S404: Determine whether the non-overlapping area is a custom regular shape; if so, determine that the area is a building roof, otherwise delete the area.

[0074] S405: Calculate the roof areas of the roofs determined to be building roofs respectively.

[0075] S406: Obtain the average daily energy consumption of the area where the building is located in each quarter according to the satellite positioning information.

[0076] S407: According to the statistical information of the Meteorological Bureau, the average daily light intensity of the region in each season is obtained.

[0077] S408: Evaluate the daily average energy consumption ratio by quarter;

[0078]

[0079] Among them, P is the average daily energy consumption proportion in a quarter;

[0080] ρ is the photovoltaic conversion efficiency per unit area per unit time;

[0081] L is the average daily light intensity of the current quarter;

[0082] S nais the roof area of ​​the nth building roof;

[0083] S nht is the blocked roof area of ​​the nth building roof at time t;

[0084] T is the average daily sunshine time in the current quarter;

[0085] t is the shading time of the building roof;

[0086] t1 is the occlusion start time;

[0087] t2 is the occlusion start time;

[0088] C d is the average daily energy consumption of the area in the current quarter;

[0089] S409: Establish a rooftop photovoltaic potential assessment table, with different daily average energy consumption ratios in different ranges corresponding to different potential levels.

[0090] S410: Take the weighted average of the daily energy consumption proportions of the four quarters and compare it with the roof potential assessment table to determine the roof potential level.

[0091] This solution combines the recognition model with segmentation recognition, and determines the roof area from the combination of the recognition results of the two methods. The recognition methods are diverse and the accuracy of the recognition results is improved.

[0092] It should be understood that the embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms also fall within the scope limited by the appended claims of the application.

Claims

1. A method for evaluating rooftop photovoltaic potential based on machine learning, characterized in that: The following steps are involved: S1: Obtain stereo remote sensing image data from the top of the building through satellite or aerial photography; S2: Copy the pre-processed stereo remote sensing image data, one copy of the data is input into the trained recognition model to identify the roof area; the other copy of the data is matrix segmented according to the rated length and width dimensions; S3: performing boundary recognition on each matrix unit data after matrix segmentation, obtaining a closed boundary curve after restoration and combination, and determining the roof area; S4: The roof area is calculated by coupling the roof area identified by the recognition model and the roof area determined by the closed boundary curve, and the rooftop photovoltaic potential is evaluated in combination with the energy consumption and environmental information of the station area where the building is located.

2. A method for evaluating rooftop photovoltaic potential based on machine learning according to claim 1, characterized in that: The stereo remote sensing image data is preprocessed by using the UNB pan-sharpening method that includes both the MS band and the Pan band.

3. A method for evaluating rooftop photovoltaic potential based on machine learning according to claim 1 or 2, characterized in that: The trained recognition model adopts U-net network recognition.

4. The method for evaluating rooftop photovoltaic potential based on machine learning according to claim 1, characterized in that: The step S3 specifically includes the following steps: S301: For each matrix unit data, extract the RGB data of each data point; S302: traverse each data point, compare the RGB data with the points around the data point, and extract the boundary curve; S303: according to the matrix segmentation position, restore and splice the matrix units, connect the boundary curves, and close the boundary curves to form a closed area; S304: the closed area is matched with each roof shape in the database; if the match is successful, the closed area is defined as a roof area; otherwise, it is determined to be an interference area and the closed area is deleted.

5. A method for evaluating rooftop photovoltaic potential based on machine learning according to claim 4, characterized in that: The step S303 also The process includes: Calculate the area of ​​the closed area and determine whether the area of ​​the closed area is greater than the construction threshold; If so, it is determined that the closed area has the capability to build photovoltaic power, and the closed area is retained; otherwise, the closed area is deleted.

6. A method for evaluating rooftop photovoltaic potential based on machine learning according to claim 1, 4 or 5, characterized in that: The roof area calculation process in step S4 is specifically as follows: S401: overlapping the roof area identified by the corresponding recognition model and the roof area determined by the closed boundary curve according to the corresponding relationship of the coordinates; S402: For the same corresponding roof, the overlapping portion of the two roof regions is determined as the building roof; S403: Calculating the area of ​​the non-overlapping parts of the two roof areas corresponding to the same roof; if the area of ​​the non-overlapping parts is less than or equal to the error area threshold, the area is determined to be a non-building roof, and the area is deleted; Otherwise, proceed to the next step; S404: Determine whether the non-overlapping area is a custom regular shape; If yes, then the area is determined to be the roof of the building, otherwise the area is deleted; S405: Calculate the roof areas of the roofs determined to be building roofs respectively.

7. A method for evaluating rooftop photovoltaic potential based on machine learning according to claim 6, characterized in that: The calculation process of the rooftop photovoltaic potential in step S4 is as follows: S406: Obtain the average daily energy consumption of the area where the building is located in each quarter according to the satellite positioning information; S407: According to the statistical information of the meteorological bureau, the average daily light intensity of the region in each season is obtained; S408: Evaluate the daily average energy consumption ratio by quarter; Among them, P is the average daily energy consumption proportion in a quarter; ρ is the photovoltaic conversion efficiency per unit area per unit time; L is the average daily light intensity of the current quarter; S na is the roof area of ​​the nth building roof; S nht is the blocked roof area of ​​the nth building roof at time t; T is the average daily sunshine time in the current quarter; t is the shading time of the building roof; t1 is the occlusion start time; t2 is the occlusion start time; C d is the average daily energy consumption of the area in the current quarter; S409: Establish a rooftop photovoltaic potential assessment table, with different daily average energy consumption ratios corresponding to different potential levels; S410: Take the weighted average of the daily energy consumption proportions of the four quarters and compare it with the roof potential assessment table to determine the roof potential level.

Images