Search

CN-121997755-A - Urban building solar radiation potential prediction method and system based on machine learning

CN121997755ACN 121997755 ACN121997755 ACN 121997755ACN-121997755-A

Abstract

The invention belongs to the technical field of integrated photovoltaic building, but not limited to, discloses a method and a system for predicting solar radiation potential of urban building based on machine learning, which are used for carrying out grid division on a building surface, constructing a solar position matrix, establishing a refined building plane level physical simulation model by combining a ray tracing method, realizing rapid simulation of solar radiation of urban building based on a urban block level building dataset, constructing a block building multidimensional feature set by utilizing weather, building geometry and surrounding environment features, training and obtaining a machine learning prediction model, evaluating model performance by adopting SMAPE and R 2 indexes, carrying out interpretative analysis, and realizing high-precision and high-efficiency urban building surface solar radiation potential prediction. According to the invention, the influence degree of different characteristics on solar radiation potential of different orientation building planes is revealed through training a machine learning model, and support is provided for rapid evaluation and large-scale application of integrated photovoltaic potential of urban building.

Inventors

  • TIAN ZHIYONG
  • LONG HAI
  • CHEN XINYU
  • MA LING

Assignees

  • 华中科技大学

Dates

Publication Date
20260508
Application Date
20260128

Claims (10)

  1. 1. The urban building solar radiation potential prediction method based on machine learning is characterized by comprising the following steps of: Acquiring building footprint data and meteorological data, and constructing a three-dimensional space model comprising a target building and surrounding buildings; Based on the three-dimensional space model, performing solar radiation physical simulation considering shielding relation on each building plane of the target building to obtain building plane solar radiation data serving as a supervision tag; fusing building plane area features, meteorological radiation features and regional building morphological features to construct a feature variable matrix for machine learning model input; Training the characteristic variable matrix and the supervision labels by adopting a machine learning model to obtain a solar radiation prediction model of the urban building; Building information of different geographical location areas is input into the prediction model, and a solar radiation prediction result of the building planes of different cities is output.
  2. 2. The method of claim 1, wherein the solar radiation physics simulation comprises: emitting rays pointing to the sun direction at each moment to each building plane of the target building; judging whether the ray intersects with the geometry of the surrounding building or not; when a ray intersects any geometry, setting the direct radiation contribution corresponding to the ray to 0; When a ray does not intersect any geometry, the direct radiation contribution corresponding to that ray is set to 1.
  3. 3. The method according to claim 2, characterized in that the total solar radiation of the building plane, taking into account the shielding, is obtained by: multiplying the direct radiation without shielding by a direct correction coefficient to obtain corrected direct radiation; multiplying the scattered radiation by a scatter correction coefficient to obtain corrected scattered radiation; multiplying the ground reflected radiation by a reflection correction coefficient to obtain corrected reflected radiation; and adding the corrected direct radiation, the corrected scattered radiation and the corrected reflected radiation to obtain the total solar radiation of the building plane.
  4. 4. A city building solar radiation shielding perception and radiation correction calculation method is characterized by comprising the following steps: Constructing a three-dimensional geometric model containing a target building and surrounding buildings; Generating rays directed to the sun's azimuth at each time step for each building plane of the target building; detecting whether the ray intersects any of the peripheral building geometries; setting the direct radiation contribution of the time step to 0 when the intersection occurs; setting the direct radiation contribution of the time step to 1 when no intersection occurs; And based on the direct radiation contribution, carrying out weighted correction and accumulation on the direct radiation, the scattered radiation and the ground reflection radiation to obtain the total solar radiation of the building plane.
  5. 5. The method of claim 4, wherein the building plane comprises a roof plane and at least 4 vertical facade planes.
  6. 6. The method of claim 4, wherein the time steps are 1 hour and the number of annual time steps is 8760.
  7. 7. A machine learning-based urban building solar radiation potential prediction system, comprising: The three-dimensional radiation simulation module is used for constructing a three-dimensional building model and calculating solar radiation data of a building plane considered to be shielded; the feature construction module is used for fusing the building plane area features, the meteorological radiation features and the regional building morphological features to construct a feature variable matrix; the model training module is used for training a machine learning prediction model based on the characteristic variable matrix and the solar radiation data; The prediction module is used for receiving building data of different areas and outputting corresponding solar radiation prediction results; the data output by the three-dimensional radiation simulation module is used as the supervision tag input of the model training module, so that closed loop coordination between physical simulation and machine learning is realized.
  8. 8. The system of claim 7, wherein the feature construction module includes an area morphology analysis unit for calculating building floor space ratio, floor area ratio, building height statistic and open space ratio.
  9. 9. The system of claim 7, wherein the model training module comprises a model evaluation unit for calculating a decision coefficient and a symmetric mean absolute percentage error.
  10. 10. The system of claim 7, wherein the prediction module is configured to output a total annual solar radiation prediction for each city, each building plane.

Description

Urban building solar radiation potential prediction method and system based on machine learning Technical Field The invention belongs to the technical field of photovoltaic building integration, and particularly relates to a method and a system for predicting solar radiation potential of urban building based on machine learning. Background Global climate change has become a major challenge for human society, pushing energy structure transformation, achieving carbon neutralization, and the goal is urgent. Cities are the main source of energy consumption, and their low carbonization development is critical for coping with climate change. In this context, distributed photovoltaics are considered to be one of the key technological paths for urban energy system conversion due to their clean, renewable nature. The building photovoltaic integrated system can effectively utilize building roof and facade resources, realizes on-site power generation on the premise of not occupying extra land, becomes a main trend of urban distributed photovoltaic development, and is supported by policies of numerous cities worldwide. The solar radiation potential of the urban building is accurately estimated, is a basic premise of scientific planning and deployment of a building photovoltaic system, and has important guiding significance for investment decision, power grid consumption and carbon emission reduction benefit estimation. The current assessment method in the field mainly relies on simulation based on physical principles. Although the method has higher precision, when simulating urban-level and large-scale buildings, the computing complexity is exponentially increased due to the need of finely computing the shielding relation of each building at different moments, and the method faces serious computing bottleneck. The current pure physical simulation method is difficult to support for carrying out high spatial resolution, rapid and efficient potential evaluation on the whole city, and limits the application of the method in large-scale planning. In view of the analysis, the technical problem in the prior art, which needs to be solved, is how to break through the calculation bottleneck of the traditional physical simulation method in the solar radiation potential evaluation of the urban building, realize the large-scale solar radiation potential evaluation with low data dependence, high accuracy, rapidness and high efficiency, and adapt to the requirements of urban BIPV large-scale planning and development. Disclosure of Invention Aiming at the problems existing in the prior art, the invention provides a machine learning-based urban building solar radiation potential prediction method, which approximates a complex physical simulation process through a data driving model, realizes the order-of-magnitude improvement of calculation efficiency on the premise of ensuring the precision, and provides a high-efficiency and reliable technical support scheme for urban building photovoltaic potential evaluation with large-scale and high spatial precision. The invention is realized in such a way that the urban building solar radiation potential prediction method based on machine learning comprises the following steps: s1, carrying out refined regional building solar radiation physical simulation according to building footprints and meteorological data; s2, integrating area data, meteorological observation data and regional building distribution data of different oriented planes of a building to construct a characteristic variable matrix with N-M dimensions; S3, adopting default super parameters and carrying out basic model training on a training set, and measuring model fitting goodness and prediction precision by combining R 2 and SMAPE; And S4, saving the model obtained by training, and taking regional building information data of different geographic positions as input to predict regional solar radiation according to cities and building planes. Preferably, in step S1, the building is subjected to surrounding objects (e.g. shadow masking), resulting in attenuation of the actually received solar radiation. When the solar rays on the surface of the building are affected by the shielding objects (other buildings, trees and the like), the direct solar rays are blocked, and the solar radiation can be obtained only by means of scattered radiation and diffuse reflection; And a Ladybug platform based on the Rhino is used for realizing high-precision calculation of solar radiation at the street level. Specifically, when shadow occlusion between buildings is considered, all buildings around the target building are added to the 3D model as occlusion elements. And calculating whether the direct solar radiation rays of each building plane geometrically intersect with the shielding element in the target building by adopting a ray tracing method, so as to judge whether the direct solar radiation exists in the plane at the moment t. If a ray is bloc