City-level building solar radiation potential prediction method and system based on machine learning

Abstract

The invention belongs to but is not limited to the technical field of photovoltaic building integration, and discloses a city-level building solar radiation potential prediction method and system based on machine learning, and the method comprises the steps: carrying out the grid division of a building surface, constructing a sun position matrix, and building a refined building plane-level physical simulation model through combining with a ray tracing method; based on the urban block-level building data set, rapid simulation of urban building solar radiation is realized; constructing a block building multi-dimensional feature set by using meteorological, building geometry and surrounding environment features, and training to obtain a machine learning prediction model; the performance of the model is evaluated by adopting SMAPE and R2 indexes, and interpretability analysis is carried out, so that high-precision and high-efficiency urban building surface solar radiation potential prediction is realized. According to the method, the influence degree of different features on the solar radiation potential of the building planes with different orientations is revealed by training the machine learning model, and support is provided for rapid evaluation and large-scale application of the urban building integrated photovoltaic potential.

Description

Technical Field

[0001] This invention belongs to, but is not limited to, the field of building-integrated photovoltaics (BIPV), and particularly relates to a method and system for predicting the solar radiation potential of urban buildings based on machine learning. Background Technology

[0002] Global climate change has become a major challenge facing human society, making the transformation of energy structures and the achievement of carbon neutrality an urgent priority. As the primary energy consumers, cities' low-carbon development is crucial for addressing climate change. Against this backdrop, distributed photovoltaics (PV) is considered a key technological path for urban energy system transformation due to its clean and renewable characteristics. Building-integrated photovoltaics (BIPV), in particular, effectively utilizes building rooftops and facades to generate electricity on-site without requiring additional land, and has become a major trend in urban distributed PV development, receiving policy support from numerous cities worldwide.

[0003] Accurately assessing the solar radiation potential of urban buildings is a fundamental prerequisite for the scientific planning and deployment of building-integrated photovoltaic (BIPV) systems, and it has significant guiding significance for investment decisions, grid integration, and carbon emission reduction benefit assessment. Currently, assessment methods in this field mainly rely on physics-based simulations. Although these methods offer high accuracy, simulating large-scale urban buildings presents a significant computational bottleneck due to the need for detailed calculations of the shading relationships of each building at different times, leading to an exponential increase in computational complexity. Current purely physical simulation methods are insufficient to support high spatial resolution, rapid, and efficient potential assessments of entire cities, limiting their application in large-scale planning.

[0004] Based on the above analysis, the urgent technical problem to be solved by the existing technology is: how to break through the computational bottleneck of traditional physical simulation methods in the assessment of solar radiation potential of urban buildings, and achieve large-scale solar radiation potential assessment with low data dependence, high accuracy, speed and efficiency, so as to meet the needs of large-scale planning and development of urban BIPV. Summary of the Invention

[0005] To address the problems existing in the prior art, this invention provides a machine learning-based method for predicting the solar radiation potential of urban buildings. By using a data-driven model to approximate a complex physical simulation process, it achieves an order-of-magnitude improvement in computational efficiency while ensuring accuracy. This provides an efficient and reliable technical support solution for large-scale, high-spatial-precision assessment of the photovoltaic potential of urban buildings.

[0006] This invention is implemented as follows: a machine learning-based method for predicting the solar radiation potential of city-level buildings, comprising the following steps:

[0007] S1, based on building footprint and meteorological data, performs detailed regional building solar radiation physical simulation;

[0008] S2 integrates area data of different building orientations, meteorological observation data, and regional building distribution data to construct an N*M dimension feature variable matrix;

[0009] S3 uses default hyperparameters and trains a basic model on the training set, combined with R... 2 SMAPE measures model goodness of fit and prediction accuracy;

[0010] S4 saves the trained model and uses regional building information data from different geographical locations as input to predict regional solar radiation by city and building plan.

[0011] Preferably, in step S1, the building is affected by surrounding objects (such as shadows), resulting in a reduction in the actual solar radiation received. When the solar rays on the building surface are affected by obstructions (other buildings, trees, etc.), the direct sunlight will be blocked, and the building can only obtain solar radiation through scattered radiation and diffuse reflection.

[0012] This paper utilizes the Ladybug platform of Rhino to achieve high-precision calculation of street-level solar radiation. Specifically, when considering the shading between buildings, all buildings surrounding the target building are added to the 3D model as shading elements. Ray tracing is used to calculate whether the direct solar radiation rays from each building plane in the target building geometrically intersect with the shading elements, thus determining whether direct solar radiation exists on that plane at time t. If a ray is blocked, its radiation contribution is 0; otherwise, it is 1. The intersection detection results can be represented as follows:

[0013]

[0014] Therefore, after considering shading, the total solar radiation received by each plane can be modified as follows:

[0015]

[0016] in, It is natural direct radiation (W / m 2 ); It is horizontal scattered radiation (W / m 2 ); It is the total natural solar radiation (W / m²) 2 ), which is equal to the sum of direct radiation and diffuse radiation; , and These are the correction factors for direct, scattered, and ground-reflected radiation, respectively.

[0017] Preferably, in step S2, the meteorological data only considers three annual total solar radiation indicators, including total radiation (GHI), direct radiation (DNI), and diffuse radiation (DHI), and directly sums the radiation data of 8760 hours throughout the year. The building area parameters include five building plans for the roof, south wall, north wall, east wall, and west wall, represented by TAR, TAS, TAN, TAE, and TAW, respectively. Specifically, the plan orientation is calculated based on normal vector coordinates, setting due north as 0 degrees, due east as 90 degrees, due south as 180 degrees, and due west as 270 degrees. The regional building distribution parameters were statistically analyzed using Python's Geopandas library for each study area, comprising 16 core indicators: total building footprint (ZBA), total perimeter of all buildings (ZBL), total volume of all buildings (ZBV), total floor area of ​​all buildings (BPA), volume-to-area ratio (VAR), floor area-to-area ratio (FAR), average height of all buildings (ABH), maximum height of all buildings (MBH), total volume-to-footprint ratio (BCR), building footprint-to-area ratio (ZD), average number of floors in the area (ZAF), difference between the tallest building and the average building height (ZSD), ratio of open space area to total area (OSR), ratio of open space area to total floor area (OSG), number of high-rise buildings (≥10 floors) in the area (mHRP), and total number of buildings in the area (NSB).

[0018] Preferably, step S3 uses R 2 R is used to measure the goodness of fit between model predictions and actual values, and its value ranges from [0, 1]. 2 The closer the value is to 1, the stronger the model's explanatory power and the higher the degree of agreement between the predicted results and the actual values. (Using R) 2 The ability of the evaluation model to capture the overall trend of annual total solar radiation can be expressed as:

[0019]

[0020] in, This is the actual value. These are model predictions. It refers to the number of samples.

[0021] Because the simulated solar radiation results for some building facades are close to zero (i.e., the low solar radiation in certain areas due to a small number of buildings), the traditional Mean Absolute Percentage Error (MAPE) tends to infinity near zero, causing the evaluation to fail. Therefore, this invention uses SMAPE as a supplementary indicator, calculated using the following formula:

[0022]

[0023] Another object of the present invention is to provide a city-level building solar radiation potential prediction system based on machine learning, comprising:

[0024] The solar radiation simulation module performs detailed regional physical simulations of building solar radiation based on building footprint and meteorological data; it also acquires data on building planar solar radiation and its influencing factors.

[0025] The feature variable matrix construction module integrates area data of different building orientations, meteorological observation data, and regional building distribution data to construct an N*M dimension feature variable matrix; it also identifies the model input variables and divides the training and test sets.

[0026] The model training and performance evaluation module uses default hyperparameters and performs basic model training on the training set, combined with R... 2 SMAPE measures model goodness of fit and prediction accuracy;

[0027] The solar radiation prediction module saves the trained model, takes regional building information data from different geographical locations as input, performs regional solar radiation prediction by city and building plan, and outputs the model prediction value.

[0028] Another object of the present invention is to provide a computer device including a memory and a processor, the memory storing a computer program, which, when executed by the processor, causes the processor to perform the steps of the machine learning-based method for predicting the solar radiation potential of urban buildings.

[0029] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the machine learning-based method for predicting the solar radiation potential of urban buildings.

[0030] Another objective of this invention is to provide an information data processing terminal, which includes the aforementioned machine learning-based urban building solar radiation potential prediction system.

[0031] Based on the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solution to be protected by this invention are as follows:

[0032] This invention provides a machine learning-based method for predicting the solar radiation potential of city-level buildings. It involves meshing the building surface, constructing a solar position matrix for 8760 hours throughout the year, and establishing a refined physical simulation model of the building planar surface using ray tracing. Based on a city block-level building dataset, it achieves rapid simulation of urban building solar radiation. A multi-dimensional feature set of block buildings is constructed using meteorological, building geometry, and surrounding environmental features, and a machine learning prediction model is trained and obtained. The method employs SMAPE and R... 2 The indicators evaluate the model performance and perform interpretability analysis, thereby achieving high-precision and high-efficiency prediction of the solar radiation potential of urban building surfaces.

[0033] This invention uses urban blocks as the basic unit to perform physical simulation of solar radiation. By training a machine learning model, it reveals the degree of influence of different features on the solar radiation potential of building planes with different orientations, providing support for the rapid assessment and large-scale application of the potential of building-integrated photovoltaics at the city level.

[0034] The machine learning model proposed in this invention achieved superior predictive performance across all building orientations (roof, south, north, east, and west facades). Error analysis revealed the difference between SMAPE and the total error, indicating that the model is extremely accurate in capturing overall radiation trends while maintaining good sensitivity to low radiation scenarios. From a data-driven perspective, this study reveals that regional building floor area is the most crucial feature influencing solar radiation potential, providing key mechanistic insights into the impact of the urban environment on solar energy resource distribution.

[0035] (1) The expected benefits and commercial value of the technical solution of this invention after transformation are as follows:

[0036] By simplifying the complex physical simulation process into an efficient machine learning prediction model, the time cost and computational resource requirements for city-level photovoltaic potential assessment can be significantly reduced, achieving an efficiency improvement from days to hours. Its commercial value lies in providing government planning departments, energy companies, and real estate developers with fast and accurate solar resource assessment services, supporting scientific decision-making and investment in large-scale distributed photovoltaic projects. This is expected to create tens to hundreds of millions of yuan in market value, while simultaneously promoting urban low-carbon transformation and the development of the renewable energy industry.

[0037] (2) The technical solution of this invention fills a technical gap in the industry both domestically and internationally:

[0038] Currently, the assessment of urban-level building photovoltaic potential both domestically and internationally mainly relies on traditional physical simulation methods, which suffer from technical bottlenecks such as low computational efficiency and difficulty in supporting large-scale applications. This invention is the first to systematically apply an interpretable machine learning method to the field of urban-level building solar radiation prediction, achieving a balance between high accuracy and high efficiency. It fills the technical gap in data-driven rapid assessment methods in this field and provides a new technical path for urban-level renewable energy planning.

[0039] (3) The technical solution of the present invention solves a technical problem that people have long wanted to solve but have never been able to solve successfully:

[0040] For a long time, the assessment of urban-level building photovoltaic potential has faced the dilemma of "difficulty in balancing accuracy and efficiency." Traditional physical simulation methods, while highly accurate, are computationally time-consuming and cannot meet the needs of large-scale applications; while simplified models, though fast, suffer significant accuracy loss. This invention successfully solves this technical problem by combining physical simulation-generated samples with machine learning for rapid prediction, while maintaining R... 2 With a high accuracy of >0.96, the evaluation efficiency is significantly improved, achieving a technological breakthrough.

[0041] (4) The technical solution of the present invention overcomes technical bias:

[0042] Traditionally, it has been believed that predicting solar radiation from urban buildings requires complex physical modeling and ray tracing calculations, and machine learning methods struggle to meet the accuracy requirements for engineering applications. This invention, through systematic feature engineering construction and interpretability analysis, demonstrates that data-driven models can achieve high-precision predictions while maintaining physical plausibility. It overcomes the technical bias that machine learning is unsuitable for predicting complex physical processes, providing new insights for technological innovation in related fields. Attached Figure Description

[0043] Figure 1 This is a flowchart of a machine learning-based method for predicting the solar radiation potential of city-level buildings, provided in an embodiment of the present invention.

[0044] Figure 2 This is a feature contribution distribution diagram of different oriented planes provided in the embodiments of the present invention.

[0045] Figure 3 This is a distribution map of solar radiation data for building planes with different orientations at the street level, provided in an embodiment of the present invention. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0047] like Figure 1 As shown, this embodiment of the invention provides a method for predicting the solar radiation potential of city-level buildings based on machine learning, including the following steps:

[0048] S1: Based on building footprint and meteorological data, the building surface is meshed and a solar position matrix for 8760 hours throughout the year is constructed. A refined planar physical simulation model of the building is established using ray tracing, and regional physical simulations are conducted. Based on a city block-level building dataset, rapid simulation of urban building solar radiation is achieved. Specific steps are as follows:

[0049] In complex urban environments, buildings are affected by surrounding objects (such as shadows), resulting in a reduction in the actual solar radiation received. When the solar rays on a building's surface are blocked by obstructions (other buildings, trees, etc.), its direct sunlight will be blocked, and it can only receive solar radiation through scattered radiation and diffuse reflection.

[0050] This paper utilizes the Ladybug platform of Rhino to achieve high-precision calculation of street-level solar radiation. Specifically, when considering the shading between buildings, all buildings surrounding the target building are added to the 3D model as shading elements. Ray tracing is used to calculate whether the direct solar radiation rays from each building plane in the target building geometrically intersect with the shading elements, thus determining whether direct solar radiation exists on that plane at time t. If a ray is blocked, its radiation contribution is 0; otherwise, it is 1. The intersection detection results can be represented as follows:

[0051]

[0052] Therefore, after considering shading, the total solar radiation received by each plane can be modified as follows:

[0053]

[0054] in, It is natural direct radiation (W / m 2 ); It is horizontal scattered radiation (W / m 2 ); It is the total natural solar radiation (W / m²) 2 ), which is equal to the sum of direct radiation and diffuse radiation; , and These are the correction factors for direct, scattered, and ground-reflected radiation, respectively.

[0055] S2: Integrate area data of building plans with different orientations, meteorological observation data, and regional building distribution data to construct an N*M dimension feature variable matrix. Meteorological data only considers three annual total solar radiation indicators: total solar radiation (GHI), direct solar radiation (DNI), and diffuse solar radiation (DHI), directly summing the radiation data from 8760 hours throughout the year. Building area parameters include five building plans for the roof, south wall, north wall, east wall, and west wall, represented by TAR, TAS, TAN, TAE, and TAW, respectively. Specifically, plan orientation is calculated based on normal vector coordinates, setting due north as 0 degrees, due east as 90 degrees, due south as 180 degrees, and due west as 270 degrees. The regional building distribution parameters were statistically analyzed using Python's Geopandas library for each study area, comprising 16 core indicators: total building footprint (ZBA), total perimeter of all buildings (ZBL), total volume of all buildings (ZBV), total floor area of ​​all buildings (BPA), volume-to-area ratio (VAR), floor area-to-area ratio (FAR), average height of all buildings (ABH), maximum height of all buildings (MBH), total volume-to-footprint ratio (BCR), building footprint-to-area ratio (ZD), average number of floors in the area (ZAF), difference between the tallest building and the average building height (ZSD), ratio of open space area to total area (OSR), ratio of open space area to total floor area (OSG), number of high-rise buildings (≥10 floors) in the area (mHRP), and total number of buildings in the area (NSB).

[0056] S3: Use default hyperparameters and train the basic model on the training set. Then, use R... 2 R is used to measure the goodness of fit between model predictions and actual values, and its value ranges from [0, 1]. 2 The closer the value is to 1, the stronger the model's explanatory power and the higher the degree of agreement between the predicted results and the actual values. (Using R) 2 The ability of the evaluation model to capture the overall trend of annual total solar radiation can be expressed as:

[0057]

[0058] in, This is the actual value. These are model predictions. It refers to the number of samples.

[0059] Because the simulated solar radiation results for some building facades are close to zero (i.e., the low solar radiation in certain areas due to a small number of buildings), the traditional Mean Absolute Percentage Error (MAPE) tends to infinity near zero, causing the assessment to fail. Therefore, this study uses SMAPE as a supplementary indicator, calculated using the following formula:

[0060]

[0061] S4: Save the trained model and further use regional building information data from different geographical locations as input to conduct regional solar radiation predictions by city and building plan. In the future, it can also be combined with photovoltaic power generation models and economic assessment models to quickly obtain techno-economic multi-scale predictions of building rooftop and facade photovoltaic potential at the block, city, and national levels.

[0062] This invention provides a machine learning-based system for predicting the solar radiation potential of city-level buildings, comprising:

[0063] The solar radiation simulation module performs detailed regional physical simulations of building solar radiation based on building footprint and meteorological data; it also acquires data on building planar solar radiation and its influencing factors.

[0064] The feature variable matrix construction module integrates area data of different building orientations, meteorological observation data, and regional building distribution data to construct an N*M dimension feature variable matrix; it also identifies the model input variables and divides the training and test sets.

[0065] The model training and performance evaluation module uses default hyperparameters and performs basic model training on the training set, combined with R... 2 SMAPE measures model goodness of fit and prediction accuracy;

[0066] The solar radiation prediction module saves the trained model, takes regional building information data from different geographical locations as input, performs regional solar radiation prediction by city and building plan, and outputs the model prediction value.

[0067] A specific example: A city's block-level buildings are selected as a representative application. Data includes three categories: block geographical boundaries, meteorological data, and architectural data. Using a physical simulation model, the regional-level solar radiation distribution per unit area of ​​building plans with different orientations is obtained, such as... Figure 2 As shown in the figure. The results indicate that there are significant differences in solar radiation received by building planes with different orientations. The roof receives the most radiation, ranging from 1200 to 1500 kWh / m². 2The radiation distribution on the facade exhibits a clear gradient characteristic. In the main urban area, due to higher building density and severe shading, radiation resources are relatively weaker. For the south wall, the radiation values ​​in the main urban area are mostly concentrated between 300-600 kWh / m². 2 The range is relatively wide, while the suburbs are concentrated in the 600-900 kWh / m² range. 2 The east and west walls are mostly located at 300-600 kWh / m³. 2 The radiation received by the north wall is the lowest overall, with radiation levels below 300 kWh / m² in most areas. 2 .

[0068] The R-values ​​of the machine learning model under different building orientations were obtained through simulation. 2 The SMAPE and total error performance are shown in Table 1. The SMAPE errors for the five orientations—roof, south wall, north wall, east wall, and west wall—are 4.55%, 11.23%, 11.80%, 13.71%, and 11.65%, respectively, and R... 2 All values ​​were above 0.96, indicating that the model has extremely strong explanatory power and predictive stability. In terms of total solar radiation, the south wall had the lowest total error (only 0.01%), and the total errors for all facing facades were below 1.4%. It can be observed that there is a significant difference between SMAPE and the total error, which stems from the characteristics of the error assessment index. SMAPE focuses on relative error and is sensitive to low radiation values, while the total error reflects the overall deviation.

[0069] Table 1. Model representation of different building plans

[0070]

[0071] Through simulation analysis, the distribution of the top 12 feature contributions of the machine learning model under different building orientations can be further analyzed, such as... Figure 3 As shown in the figure, building area is a key feature influencing solar radiation prediction results, ranking among the top two in terms of contribution across all orientations. This is followed by building distribution parameters in the ZBL and ZBA regions. Meteorological parameters have a relatively small impact because meteorological data in the study area varies little.

[0072] This invention is primarily applied to the rapid assessment of distributed photovoltaic potential in urban planning and new energy fields, and is particularly suitable for efficient and accurate solar resource analysis of large-scale urban building complexes (such as blocks, urban areas, cities, and even national to global scales). Related products include, but are not limited to: city-level photovoltaic planning and assessment platforms, solar energy analysis modules in building energy management (BEMS) systems, renewable energy analysis components in smart city digital twin systems, and customized solar resource assessment reports and decision support tools for governments, energy companies, and real estate developers.

[0073] The technical effects provided by the embodiments of this invention (i.e., significantly simplifying the physical simulation modeling process and enabling rapid evaluation of large-scale building data) are supported by clear data. As shown in Table 1, in the application to urban block-level buildings, the machine learning model trained based on the method of this invention achieved extremely high accuracy (R0) in predicting all building orientations (roof, four walls). 2 All values ​​are above 0.96) and the overall bias is extremely low (total error is below 1.4%). This demonstrates that the model can reproduce complex physical simulation results with extremely high accuracy. Furthermore, this method bypasses the traditional physical simulation process of performing detailed geometric modeling and time-consuming ray tracing calculations for each building individually. Instead, it uses pre-simulated sample data to train a general prediction model. This enables a paradigm shift from "building-by-building simulation" to "batch prediction" when evaluating thousands of buildings, reducing computation time from days or even weeks to hours or less. This represents an order-of-magnitude improvement in evaluation efficiency and provides a practical technical solution for large-scale, city-level solar resource surveys and dynamic assessments.

[0074] Example 1: Construction of a Closed Loop for City-Level Physical Simulation and Machine Learning

[0075] This study selects a central urban area as the research object, collecting building footprint data, building height data, and hourly meteorological radiation data to construct a 3D spatial model including the target building and surrounding buildings. Based on this model, hourly solar radiation is calculated for the target building's roof and facade. The presence of direct solar radiation at each moment is determined by the intersection of rays with the geometry of obstructing buildings. Based on this, direct radiation, diffuse radiation, and ground-reflected radiation are corrected and accumulated to obtain the annual total solar radiation value of the building plan as a supervision label. The building plan area characteristics, meteorological radiation characteristics, and regional building morphology characteristics are fused to form a feature matrix, which is used to train a machine learning model. This achieves closed-loop synergy between data generation driven by physical mechanisms and prediction models driven by data, improving the reliability of cross-regional predictions.

[0076] Example 2: Ray Intersection Determination Based on Shadow Occlusion Perception

[0077] To address the prevalent shading problem in densely built-up areas, this method incorporates all buildings within a certain radius of the target building as shading elements when constructing the 3D model. For each plane of the target building, a ray pointing towards the sun is generated at each time step, and geometric intersection with the shading elements is checked. When the ray intersects with any shading geometry, the direct radiation contribution for that time step is set to 0; otherwise, it is set to 1. This method accurately characterizes the impact of inter-building shading relationships on solar radiation, avoiding the direct application of radiation results under unshaded conditions to the urban scale, thus fundamentally improving the realism of the model's input data.

[0078] Example 3: Cooperative Correction Mechanism for Multi-Source Radiation Components

[0079] After obtaining the shading perception results, independent correction coefficients are introduced for direct radiation, diffuse radiation, and ground reflected radiation for weighting. Specifically, unshaded direct radiation is multiplied by a direct radiation correction coefficient, diffuse radiation by a diffuse radiation correction coefficient, and ground reflected radiation by a reflection correction coefficient, and then these are summed to obtain the total radiation value of the building plan. This fractional processing method can distinguish the contribution of different radiation sources to the total radiation, avoiding the problem that a single radiation model cannot reflect the complex urban lighting environment.

[0080] Example 4: Feature Collaborative Construction for Machine Learning

[0081] The model uses the building's roof and south-, north-, east-, and west-facing facade areas as geometric features, and total annual radiation, direct radiation, and diffuse radiation as meteorological features. Simultaneously, it extracts regional building footprint ratios, floor area ratios, average heights, and open space ratios as morphological features, forming a unified feature matrix. This multi-source feature matrix complements each other at the spatial, physical, and statistical levels, enabling the model to simultaneously perceive both the building's own attributes and its surrounding urban environment, thus avoiding the insufficient generalization ability that can result from relying solely on single-category features.

[0082] Example 5: Model Training and Performance Evaluation Mechanism

[0083] The model is trained using the constructed feature matrix and radiation data generated by physical simulation. The coefficient of determination is used to evaluate the model's ability to fit the overall trend, and the symmetric mean absolute percentage error is used to evaluate the prediction accuracy. This dual-indicator constraint avoids the problem of pursuing only trend fitting while ignoring absolute error, thus ensuring that the model maintains stable performance when extended across regions.

[0084] Example 6: Cross-city prediction and engineering application verification

[0085] The trained model is applied to multiple urban areas in different geographical locations. By inputting only building footprint data, building height data, and meteorological statistics, it can quickly output the annual total solar radiation prediction results for each building plane. Compared with direct building-by-building and hourly physical simulation, this method significantly reduces computational costs while maintaining the data reliability constrained by physical mechanisms, achieving a balance between high accuracy and high efficiency. It is suitable for engineering scenarios such as city-level photovoltaic site selection, building energy conservation planning, and energy system optimization.

[0086] The above embodiments illustrate the feasibility, completeness, and technical effects of the technical solution from multiple levels, including physical simulation mechanism, occlusion perception mechanism, radiation correction mechanism, feature collaborative construction mechanism, and machine learning prediction mechanism. They fully demonstrate the significant improvement in accuracy, scalability, and engineering applicability compared to traditional single physical simulation or single data-driven methods, thereby supporting its technical creativity and feasibility.

[0087] It should be noted that embodiments of the present invention can be implemented in hardware, software, or a combination of both. The hardware portion can be implemented using dedicated logic; the software portion can be stored in memory and executed by a suitable instruction execution system, such as a microprocessor or dedicated-design hardware. Those skilled in the art will understand that the above-described devices and methods can be implemented using computer-executable instructions and / or included in processor control code, for example, such code provided on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuitry such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field-programmable gate arrays, programmable logic devices, etc., or by software executed by various types of processors, or by a combination of the above-described hardware circuitry and software, such as firmware.

[0088] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions, and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention, and within the spirit and principles of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for predicting the solar radiation potential of city-level buildings based on machine learning, characterized in that, Includes the following steps: Acquire building footprint data and meteorological data to construct a three-dimensional spatial model containing the target building and its surrounding buildings; Based on the aforementioned three-dimensional spatial model, a physical simulation of solar radiation is performed on each building plane of the target building, taking into account the shading relationship, to obtain solar radiation data of the building plane as a supervision label. By integrating building plan area characteristics, meteorological radiation characteristics, and regional building morphology characteristics, a feature variable matrix is ​​constructed for input to machine learning models. A machine learning model is used to train the feature variable matrix and the supervision label to obtain a city-level building solar radiation prediction model; The prediction model inputs building information from different geographical locations and outputs solar radiation prediction results for different cities and building planes.

2. The method according to claim 1, characterized in that, The solar radiation physics simulation includes: At every moment, rays pointing towards the sun are emitted from each floor plan of the target building. Determine whether the ray intersects with the geometry of the surrounding buildings; When a ray intersects any geometric object, the direct radiation contribution corresponding to that ray is set to 0; When a ray does not intersect any geometry, the direct radiation contribution corresponding to that ray is set to 1.

3. The method according to claim 2, characterized in that, The total solar radiation of the building plan, after taking shading into account, is obtained in the following way: The corrected direct radiation is obtained by multiplying the unobstructed direct radiation by the direct radiation correction factor. Multiply the scattered radiation by the scattering correction factor to obtain the corrected scattered radiation; The corrected reflected radiation is obtained by multiplying the ground reflected radiation by the reflection correction factor. The total solar radiation on the building plane is obtained by adding the corrected direct radiation, corrected scattered radiation, and corrected reflected radiation.

4. A method for sensing and calculating solar radiation shading of buildings at the city level, characterized in that, include: Construct a three-dimensional geometric model that includes the target building and its surrounding buildings; For each floor plan of the target building, generate a ray pointing towards the sun at each time step; Detect whether the ray intersects with any surrounding building geometry; When an intersection occurs, the direct radiation contribution at that time step is set to 0; When no intersection occurs, the direct radiation contribution at that time step is set to 1; Based on the direct radiation contribution, the direct radiation, scattered radiation, and ground reflected radiation are weighted, corrected, and summed to obtain the total solar radiation of the building plane.

5. The method according to claim 4, characterized in that, The building plan includes a roof plan and at least four vertical elevation plans.

6. The method according to claim 4, characterized in that, The time step is 1 hour, and the total number of time steps throughout the year is 8760.

7. A city-level building solar radiation potential prediction system based on machine learning, characterized in that, include: The 3D radiation simulation module is used to build 3D models of buildings and calculate solar radiation data of building planes considering shading. The feature construction module is used to integrate building floor area features, meteorological radiation features, and regional building morphology features to construct a feature variable matrix. The model training module is used to train a machine learning prediction model based on the feature variable matrix and the solar radiation data. The prediction module is used to receive building data from different regions and output corresponding solar radiation prediction results; The data output by the three-dimensional radiation simulation module is used as the supervisory label input for the model training module, thereby realizing closed-loop collaboration between physical simulation and machine learning.

8. The system according to claim 7, characterized in that, The feature construction module includes a regional morphology analysis unit for calculating the building footprint ratio, floor area ratio, building height statistics, and open space ratio.

9. The system according to claim 7, characterized in that, The model training module includes a model evaluation unit for calculating the coefficient of determination and the symmetric mean absolute percentage error.

10. The system according to claim 7, characterized in that, The prediction module is used to output the annual total solar radiation prediction results by city and by building plan.

Images