CN-121999320-A - Distributed roof photovoltaic shadow recognition method based on multi-level prediction

CN121999320ACN 121999320 ACN121999320 ACN 121999320ACN-121999320-A

Abstract

The invention discloses a distributed roof photovoltaic shadow recognition method based on multi-level prediction, and belongs to the technical field of distributed roof photovoltaic system design and image processing. The method comprises the steps of building a roof and shelter three-dimensional model based on a CAD drawing or an unmanned aerial vehicle aerial flight picture, enabling white materials to eliminate reflection interference, calculating solar altitude and azimuth angles from winter to day 9:00-15:00 per hour according to project latitude and longitude, simulating solar rays to generate a shadow range top view, adopting CSPDARKNET structures to extract features, training a multi-level YOLO model through adapting anchor frame matching strategies of different shadows after K-means clustering and dividing the dimensions, identifying the roof shadow picture to be detected, and outputting a full-period shadow closing range through union processing. The invention solves the problems of low recognition precision, poor adaptability and low efficiency of the traditional method, the recognition error is less than or equal to 3 percent, the processing time is shortened by more than 70 percent, and the invention is suitable for recognizing various distributed roof photovoltaic shadows.

Inventors

Yang Yinting
YANG BO
SUN ZHEN
SUN XIAOYANG
WU QIONG
CHENG WENJI

Assignees

华能国际工程技术有限公司
中国华能集团香港有限公司
西安热工研究院有限公司

Dates

Publication Date: 20260508
Application Date: 20260211

Claims (10)

1. A distributed roof photovoltaic shadow recognition method based on multi-level prediction is characterized by comprising the following steps: s1, building a three-dimensional model of a roof and a shelter through three-dimensional modeling, calculating a solar elevation angle and an azimuth angle in winter based on project land longitude and latitude information, simulating solar rays to irradiate the three-dimensional model, and generating a shadow range top view of each hour; S2, performing shadow recognition on the top view by adopting a multi-level prediction YOLO model, dividing the multi-level prediction YOLO model into three classes of small scale, medium scale and large scale based on different scale feature maps, respectively corresponding to the small area shadow, the medium area shadow and the large area shadow, performing union processing on the recognized shadow ranges, and outputting a winter-to-day full-time shadow recognition result.
2. The method for identifying photovoltaic shadows of a distributed roof based on multi-level prediction according to claim 1, wherein in step S1, the establishment of the three-dimensional model comprises: aiming at a training stage, a three-dimensional model comprising a roof, a wall and a shelter is established according to CAD drawings of different types of items, and all colors of the three-dimensional model are endowed with white materials; Aiming at the identification stage, generating a three-dimensional model of the roof to be detected by adopting the unmanned aerial vehicle picture.
3. The method of claim 2, wherein the RGB values of the white material are (255 ).
4. The method for recognizing photovoltaic shadows on a distributed roof based on multi-level prediction according to claim 1, wherein in the step S1, the calculation range of solar altitude and azimuth angle in winter is 9:00-15:00, calculated once per hour, and the shadow range top view corresponds to the solar ray irradiation simulation result, and one shadow is generated and stored per hour.
5. The method for identifying photovoltaic shadows of a distributed roof based on multi-level prediction according to claim 1, wherein in step S2, the training process of the multi-level prediction YOLO model includes: And dividing the feature map into a small-scale feature map, a medium-scale feature map and a large-scale feature map by K-means clustering according to the resolution of the multi-level feature map.
6. The distributed roof photovoltaic shadow recognition method based on multi-level prediction according to claim 5, wherein the clustered feature maps are respectively matched with corresponding anchor frame structures: The large-scale feature images with low resolution are distributed with large-size anchor frames, the medium-scale feature images with medium resolution are distributed with medium-size anchor frames, the small-scale feature images with high resolution are distributed with small-size anchor frames, and specific shadow positions are predicted through anchor frame matching.
7. The method for identifying photovoltaic shadows of a distributed roof based on multi-level prediction according to claim 6, wherein the predicted shadow positions are marked in the model training process to complete the model training.
8. The method for identifying photovoltaic shadows of a distributed roof based on multi-level prediction according to claim 1, wherein the solar altitude is The calculation is as follows: Azimuth angle of sun The calculation is as follows: Wherein, the Is the latitude of the sun with red color, The geographical latitude of the location is indicated, Representing the local solar time angle.
9. The distributed roof photovoltaic shadow recognition method based on multi-level prediction according to claim 1, wherein in step S2, the union process of the shadow ranges is automatically performed by a computer program, all shadow recognition closed ranges in the period from winter to 9:00-15:00 are drawn, and the recognition result is output in an on-line display manner.
10. A distributed roof photovoltaic shadow recognition system based on multi-level prediction, comprising: The three-dimensional module is used for establishing a three-dimensional model of the roof and the shielding object and comprises a training model establishment unit based on a CAD drawing and a model establishment unit to be tested based on an unmanned aerial vehicle aerial photo; the parameter module is used for acquiring project latitude and longitude information and calculating solar altitude and azimuth angles of 9:00-15:00 per hour from winter to day; the simulation module is used for simulating solar rays to irradiate the three-dimensional model, and generating and storing a shadow range top view in each hour; The training module is used for training a multi-level prediction YOLO model and comprises a feature extraction unit for extracting features by adopting a CSPDARKNET structure, a clustering unit for dividing the scale of a feature map through K-means clustering, an anchor frame matching unit for distributing corresponding anchor frames for feature maps of different scales, and a marking unit for marking shadow positions through X-AnyLabeling marking software, wherein the anchor frames configured by the anchor frame matching unit comprise large-size anchor frames adapting to large-area shadows, medium-size anchor frames adapting to medium-area shadows and small-size anchor frames adapting to small-area shadows; The recognition module is used for calling the trained multi-level prediction YOLO model and carrying out shadow recognition on the shadow range top view; the merging module is used for carrying out union processing on the identified shadow ranges of each period and drawing a shadow identification closed range; and the output module is used for displaying the shadow recognition result of the winter full time period on line.

Description

Distributed roof photovoltaic shadow recognition method based on multi-level prediction Technical Field The invention belongs to the technical field of design and image processing of distributed roof photovoltaic systems, and particularly relates to a distributed roof photovoltaic shadow recognition method based on multi-level prediction. Background In recent years, with the increasing demands for global energy structure transformation and environmental protection, distributed roof photovoltaic systems are receiving increasing attention as a clean, renewable energy solution. The distributed roof photovoltaic not only can effectively utilize the roof space of a building and reduce the dependence on traditional fossil energy sources, but also can remarkably reduce the emission of greenhouse gases, and assist the world to meet the challenge of climate change. However, shadow shielding in the construction of distributed roof photovoltaic power plants can significantly reduce the power generation efficiency, even lead to hot spot effects, and damage photovoltaic modules. The conventional shadow detection method relies on a sensor or a simple image processing technology, and is difficult to cope with complex and changeable rooftop environments. Accurately identifying and effectively handling shadow problems is an important issue that needs to be addressed in the design phase of distributed roof photovoltaic systems. Shadows are a complex problem in image processing, and they can affect the accuracy of image recognition by changing pixel values, creating false edges and contours. Therefore, it is important to develop efficient shadow processing techniques. The current algorithm for shadow recognition of the distributed roof photovoltaic system mainly comprises the following four types: 1. a method based on image processing; 2. shadow prediction based on GIS and solar model; 3. A method of identifying shadows based on sensor monitoring illumination; 4. Shadow recognition is performed based on machine learning methods in combination with manual features (e.g., texture, color). Although various shadow recognition algorithms have been proposed, the above algorithms have the following problems in practical application: 1. In the face of complex roof environments, shadow generating factors are relatively large and scattered, and a traditional algorithm can identify shadow areas to a certain extent, but the identification accuracy of small targets and complex backgrounds is insufficient. 2. Distributed roof photovoltaics are generally advanced on the whole industrial and commercial roof or county, have large coverage, and are difficult to meet the requirements of different levels of shadow detection by a single model. Disclosure of Invention The invention aims to solve the technical problems that aiming at the defects in the prior art, a distributed roof photovoltaic shadow recognition method and a system based on multi-level prediction are provided, and the method and the system are used for solving the technical problems that in the prior art, when facing a complex roof environment comprising a chimney, a peripheral building, trees and other shielding objects, the conventional shadow recognition algorithm is insufficient in small-area shadow target recognition precision, and a single detection model is difficult to simultaneously meet the requirements of high-precision and high-robustness detection on different layers of shadows of large area, medium area, small area and the like. The invention adopts the following technical scheme: a distributed roof photovoltaic shadow recognition method based on multi-level prediction comprises the following steps: s1, building a three-dimensional model of a roof and a shelter through three-dimensional modeling, calculating a solar elevation angle and an azimuth angle in winter based on project land longitude and latitude information, simulating solar rays to irradiate the three-dimensional model, and generating a shadow range top view of each hour; S2, performing shadow recognition on the top view by adopting a multi-level prediction YOLO model, dividing the multi-level prediction YOLO model into three classes of small scale, medium scale and large scale based on different scale feature maps, respectively corresponding to the small area shadow, the medium area shadow and the large area shadow, performing union processing on the recognized shadow ranges, and outputting a winter-to-day full-time shadow recognition result. Preferably, in step S1, the building of the three-dimensional model includes: aiming at a training stage, a three-dimensional model comprising a roof, a wall and a shelter is established according to CAD drawings of different types of items, and all colors of the three-dimensional model are endowed with white materials; Aiming at the identification stage, generating a three-dimensional model of the roof to be detected by adopting the unmanned aerial vehicle picture.