A method for predicting suitable regions for photovoltaic site selection based on a MaxEnt model

Abstract

The application discloses a photovoltaic site selection suitable area prediction method based on a MaxEnt model, selects nine types of environmental factors such as elevation and solar radiation and is subjected to correlation screening, extracts photovoltaic core sample points through a 300m grid and a 50% coverage threshold and is subjected to sparsification, optimizes RM and FC parameters of MaxEnt by using an ENMeval package, and takes the minimum AICc as the optimal standard, and outputs suitability results by relying on optimal parameter modeling, and is divided into five-level suitable areas by using an ArcGIS natural breakpoint method. The method does not need to manually construct negative samples, avoids subjective weighting defects of traditional site selection, is stable under a small sample, and is suitable for regional photovoltaic construction suitability evaluation.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of new energy and geographic information data analysis, specifically involving a method for predicting suitable photovoltaic site selection areas based on the MaxEnt model. Background Technology

[0002] Energy is a crucial pillar of the modern economy. Since industrialization, the energy structure dominated by fossil fuels such as coal, oil, and natural gas has not only led to the depletion of resource reserves but has also triggered severe global climate change problems. Against this backdrop, solar energy, with its advantages of wide resource distribution, clean and environmentally friendly development and utilization, and safe and reliable operation, is considered one of the best alternatives to traditional fossil fuels.

[0003] The primary task in constructing a photovoltaic (PV) power plant is site selection. Whether the site selection is scientific and reasonable directly affects the construction cost, power generation efficiency, and surrounding environment of the power plant. By assessing the solar power generation potential of different regions and conducting site suitability analysis, project costs can be effectively reduced and construction risks mitigated. Among domestic and international research on PV power plant site selection, the integration of Geographic Information System (GIS) and Multi-Criteria Decision Analysis (MCDA) has become the mainstream methodological paradigm. Among these, the Analytic Hierarchy Process (AHP) is widely used due to its clear theory and ease of operation. Combined with traditional GIS, it forms the GIS-AHP framework, which dominates in site selection practice.

[0004] However, the evaluation systems of the aforementioned methods mostly adopt models based on expert scoring and prior knowledge, assigning weights to various influencing factors, which involves a certain degree of subjectivity. To improve the scientific nature of photovoltaic power plant site selection, machine learning methods have been gradually introduced into photovoltaic site selection research because they can characterize the complex nonlinear relationship between environmental factors and site suitability, and overcome the shortcomings of traditional methods in terms of subjective weighting. However, machine learning methods often construct the site selection problem as a binary classification task of "photovoltaic sites - non-photovoltaic sites," and their modeling accuracy is highly dependent on the quantity and quality of negative samples. In practice, it is difficult to accurately obtain negative samples—the absence of a power plant in a region does not necessarily mean it is unsuitable for construction, and random selection or buffer sampling may introduce misjudgments.

[0005] The Maximum Entropy Model (MaxEnt), a niche model based on point-based data, requires only known information about the location of photovoltaic power plants and environmental variables to build a model. It eliminates the need for artificially constructed negative samples and offers advantages such as robustness under small sample conditions, the ability to characterize complex nonlinear relationships, and the output of factor contribution rates and response curves. It has been widely used in species habitat prediction and has recently been introduced into land suitability assessment, providing a new technical approach for evaluating the suitability of photovoltaic power plant sites. However, in the default settings of the MaxEnt model, the regularization multiplier (RM) and feature class (FC) parameters often lead to overfitting of the model to the training data, reducing the model's predictive ability and transferability to new environments. Therefore, systematic parameter optimization is necessary. Summary of the Invention

[0006] To address the shortcomings mentioned in the background art, this invention provides a method for predicting suitable photovoltaic site selection areas based on the MaxEnt model.

[0007] To solve the above-mentioned technical problems, the technical solution provided by this invention is: a method for predicting suitable photovoltaic site selection areas based on the MaxEnt model, comprising the following steps: S1. Environmental Data Acquisition and Processing: Collect nine types of environmental data within the study area, including elevation, slope, aspect, land cover, precipitation, solar horizontal radiation, temperature, distance from roads, and distance from power facilities. Perform spatial reference system unification and resampling, spatiotemporal scale matching, and data fusion on the nine types of environmental data in sequence to form nine types of environmental data factors, and convert the environmental data factors into ASCII format. S2. Photovoltaic panel distribution sample point data acquisition and processing: The distribution range of photovoltaic panels within the study area is acquired through remote sensing image data. A regular grid is constructed based on the boundary of the study area. The regular grid is spatially overlaid with the distribution range of photovoltaic panels. Initial photovoltaic panel sample points are obtained by screening according to the coverage area ratio of photovoltaic panels in each grid unit. The initial photovoltaic panel sample points are then spatially sparsified. S3. MaxEnt Model Parameter Optimization: The regularization multiplier RM and feature combination FC of the MaxEnt model are systematically optimized using the ENMeval package of R language. A candidate model matrix is ​​constructed, and the minimum value of the modified Akaike Information Criterion (AICc) is used as the optimal model selection criterion to obtain the optimal RM parameters and the optimal FC parameters. S4. Model Construction and Suitability Classification: The environmental data factors obtained in step S1 and the photovoltaic panel sample points obtained in step S2 are imported into the MaxEnt model. The optimal RM parameters and optimal FC parameters obtained in step S3 are used to construct the model and output the prediction results. The prediction results are imported into ArcGIS software, and the natural discontinuity classification method is used to classify the prediction results into five levels: high suitability, relatively suitable, generally suitable, low suitability, and unsuitable, to obtain the suitability distribution map of photovoltaic power station construction in the study area.

[0008] Furthermore, in step S1, the nine types of environmental data specifically include: elevation data, slope data, aspect data, land cover data, precipitation data, solar horizontal radiation data, temperature data, distance from roads, and distance from power facilities.

[0009] Furthermore, in step S1, after forming the nine types of environmental data factors, the method further includes using the R language package to perform Pearson correlation analysis on the nine types of environmental data factors and deleting environmental data elements with an absolute value of correlation coefficient greater than 0.8.

[0010] Furthermore, in step S2, the regular grid is a 300m×300m regular grid covering the study area, and the regular grid is used as the basic unit for spatial analysis.

[0011] Furthermore, in step S2, the initial photovoltaic panel sample points are obtained by screening based on the proportion of photovoltaic panel area coverage in each grid unit. Specifically, the proportion of photovoltaic panel area coverage in each grid unit to the total area of ​​the grid is calculated, and only grid units with a photovoltaic panel area proportion greater than or equal to 50% are retained. The geometric center point of the grid unit is used as the initial photovoltaic panel sample point.

[0012] Furthermore, in step S2, the spatial sparsity processing is used to eliminate the bias caused by repeated sampling, so that the sample points tend to be evenly distributed in the study area.

[0013] Furthermore, in step S3, the value of RM ranges from 0.5 to 4, with a step size of 0.5.

[0014] Furthermore, in step S3, the FC combination includes six combinations: L, LQ, H, LQH, LQHP, and LQHPT.

[0015] Furthermore, in step S3, a candidate model matrix is ​​constructed by setting the range of RM values ​​and the FC combination. For each RM and FC combination in the candidate model matrix, a MaxEnt model is trained, and the AICc value of each model is calculated. The RM parameters and FC parameters corresponding to the model with the smallest AICc value are selected as the optimal parameters.

[0016] Furthermore, in step S4, the natural discontinuity classification method is implemented using ArcGIS software, which divides the continuous prediction probability value output by the MaxEnt model into five level intervals according to the natural discontinuities, corresponding to five suitability levels: high suitability, relatively suitable, moderately suitable, low suitability, and unsuitable.

[0017] After adopting the above method, the present invention has the following advantages: 1. The present invention uses the MaxEnt model to predict the suitability of photovoltaic site selection. It only needs to know the point data of photovoltaic power station and environmental variables to build the model. There is no need to manually construct negative samples, which overcomes the problem that negative samples are difficult to obtain accurately in traditional machine learning binary classification methods. 2. This invention uses the ENMeval package in R language to systematically optimize the RM and FC parameters of the MaxEnt model, and uses the minimum AICc value as the optimal model selection criterion, which effectively reduces the overfitting of the model to the training data and improves the model's predictive ability and transferability to new environments. 3. This invention uses a 300m×300m regular grid and a 50% area threshold to screen photovoltaic panel sample points and performs spatial sparsification processing, which effectively eliminates edge-mixed pixels and reduces the impact of boundary error and resampling bias on model training. 4. This invention uses Pearson correlation analysis to remove environmental data elements with an absolute correlation coefficient greater than 0.8, thereby reducing the interference of multicollinearity among environmental factors on the model results. 5. This invention uses the natural discontinuity classification method to divide the prediction results into five suitability levels, which can intuitively output a distribution map of the suitability of photovoltaic power station construction in the study area, providing a scientific basis for photovoltaic power station site selection decisions. Detailed Implementation

[0018] The present invention will be further described in detail below through embodiments.

[0019] This embodiment takes Hunan Province as the research area to provide a detailed description of the photovoltaic site selection suitability prediction method based on the MaxEnt model described in this invention.

[0020] Step S1: Environmental Data Acquisition and Processing (1) Data Acquisition: Nine types of environmental data were collected within Hunan Province, specifically including: elevation data, slope data, aspect data, land cover data, precipitation data, solar horizontal radiation data, temperature data, distance to roads data, and distance to power facilities data. Among them, elevation, slope, and aspect data can be derived from digital elevation models (DEM); land cover data can be obtained from land use / land cover remote sensing products; precipitation and temperature data can be obtained from meteorological station observation data through spatial interpolation; solar horizontal radiation data can be obtained from radiation observation products or reanalysis data; and distance to roads and distance to power facilities data can be obtained from road vector data and power facility vector data through Euclidean distance analysis.

[0021] (2) Spatial reference system unification and resampling: The above 9 types of environmental data are unified under the same spatial reference coordinate system, and the resampling operation is used to make each environmental data have a consistent spatial resolution, so as to ensure the accuracy of subsequent spatial overlay analysis.

[0022] (3) Spatiotemporal scale matching and data fusion: Time scale matching is performed on environmental data with different time resolutions (such as precipitation, temperature, solar horizontal radiation, etc.), representative time statistics related to photovoltaic power station construction (such as multi-year average) are selected, and various types of data are fused into a unified environmental data layer to form 9 types of environmental data factors.

[0023] (4) Format conversion: The processed 9 types of environmental data factors are uniformly converted into asc format to meet the requirements of MaxEnt model for input data format.

[0024] (5) Pearson correlation analysis: Pearson correlation analysis was performed on the nine environmental data factors using the R language package. The Pearson correlation coefficient between any two environmental factors was calculated, and environmental data elements with an absolute correlation coefficient greater than 0.8 were deleted to reduce the impact of multicollinearity among environmental factors on the modeling results. The environmental factors retained after correlation analysis were used for subsequent MaxEnt modeling.

[0025] Step S2: Data Acquisition and Processing of Photovoltaic Panel Distribution Sample Points (1) Photovoltaic panel distribution range collection: Using high-resolution remote sensing image data, visual interpretation or remote sensing image classification methods are used to collect the distribution range of photovoltaic panels in Hunan Province and form photovoltaic panel area vector data.

[0026] (2) Construction of Regular Grid and Extraction of Sample Points: A 300m×300m regular grid covering the entire Hunan Province was constructed as the basic unit for spatial analysis. This regular grid was spatially overlaid with the vector data of the photovoltaic panel area to calculate the proportion of the photovoltaic panel area within each grid cell to the total area of ​​the grid. Only grid cells with a photovoltaic panel area proportion greater than or equal to 50% were retained, and the geometric center point of these grid cells was used as the initial photovoltaic panel sample point. Using a 50% area threshold effectively eliminated edge-mixed pixels, ensuring that the sample points fell within the core area of ​​the photovoltaic panel area, thus reducing the impact of boundary errors on model training.

[0027] (3) Spatial sparsification: The initial photovoltaic panel sample points are spatially sparsified by setting an appropriate spatial distance threshold so that the distance between any two sample points is not less than the threshold, thereby eliminating the bias caused by repeated sampling and making the sample points more uniformly distributed in the study area. The photovoltaic panel sample points after sparsification are used as the existence point input data of the MaxEnt model.

[0028] Step S3: MaxEnt model parameter optimization In the default settings of the MaxEnt model, the regularization multiplier (RM) and feature combination (FC) parameters often lead to overfitting of the model to the training data, reducing the model's predictive ability and transferability to new environments. This embodiment uses the ENMeval package in R language for systematic parameter optimization, and the specific process is as follows: (1) Parameter range setting: Set the value range of RM to 0.5 to 4, with a step size of 0.5. That is, the value of RM includes 8 values: 0.5, 1.0, 1.5, 2.0, 2.5, 3.0, 3.5, and 4.0. Set the FC combination to include 6 combinations: L (linear), LQ (linear + quadratic), H (hinge), LQH (linear + quadratic + hinge), LQHP (linear + quadratic + hinge + product), and LQHPT (linear + quadratic + hinge + product + threshold).

[0029] (2) Candidate model matrix construction: The above RM values ​​and FC combinations are fully permuted to construct the candidate model matrix, with a total of 8×6=48 candidate parameter combinations.

[0030] (3) Model training and evaluation: For each RM and FC combination in the candidate model matrix, the MaxEnt model is trained using the environmental data factors obtained in step S1 and the photovoltaic panel sample points obtained in step S2, and the AICc (correction Akaike information criterion) value of each model is calculated.

[0031] (4) Optimal parameter selection: The minimum value of AICc is used as the criterion for optimal model selection. The RM parameters and FC parameters corresponding to the model with the minimum AICc value are selected as the optimal model parameters for subsequent final model construction.

[0032] Step S4: Model Construction and Suitability Classification (1) Model Construction: The environmental data factors retained after correlation analysis in step S1 and the photovoltaic panel sample points obtained in step S2 are imported into the MaxEnt model. The optimal RM parameters and optimal FC parameters obtained in step S3 are selected, and the MaxEnt model is run to perform modeling. The predicted probability values ​​of photovoltaic power station construction suitability for each grid cell in Hunan Province are output (the value range is 0 to 1). At the same time, the MaxEnt model can output the contribution rate of each environmental factor to the model and the response curve of the environmental factor, which can be used to analyze the impact of each environmental factor on the construction suitability of photovoltaic power stations.

[0033] (2) Suitability Classification: The prediction results (suitability probability raster) output by the MaxEnt model were imported into ArcGIS software. Using the Jenks Natural Breaks classification method, the continuous prediction probability values ​​were divided into five level intervals, corresponding to five suitability levels: high suitability, relatively suitable, moderately suitable, low suitability, and unsuitable. This resulted in a suitability distribution map for photovoltaic power station construction in Hunan Province. This map visually displays the suitability levels for photovoltaic power station construction in different areas of Hunan Province, providing a scientific basis for photovoltaic power station site selection planning and decision-making.

[0034] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0035] The present invention and its embodiments have been described above. This description is not restrictive, and the actual structure is not limited thereto. In conclusion, if those skilled in the art, inspired by this description, design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the scope of protection of this invention.

Claims

1. A method for predicting suitable photovoltaic site selection areas based on the MaxEnt model, characterized in that, Includes the following steps: S1. Environmental Data Acquisition and Processing: Collect nine types of environmental data within the study area, including elevation, slope, aspect, land cover, precipitation, solar horizontal radiation, temperature, distance from roads, and distance from power facilities. Perform spatial reference system unification and resampling, spatiotemporal scale matching, and data fusion on the nine types of environmental data in sequence to form nine types of environmental data factors, and convert the environmental data factors into ASCII format. S2. Photovoltaic panel distribution sample point data acquisition and processing: The distribution range of photovoltaic panels within the study area is acquired through remote sensing image data. A regular grid is constructed based on the boundary of the study area. The regular grid is spatially overlaid with the distribution range of photovoltaic panels. Initial photovoltaic panel sample points are obtained by screening according to the coverage area ratio of photovoltaic panels in each grid unit. The initial photovoltaic panel sample points are then spatially sparsified. S3. MaxEnt Model Parameter Optimization: The ENMeval package in R language is used to systematically optimize the regularization multiplier RM and feature combination FC of the MaxEnt model, construct a candidate model matrix, and use the minimum value of AICc as the optimal model selection criterion to obtain the optimal RM parameters and the optimal FC parameters. S4. Model Construction and Suitability Classification: The environmental data factors obtained in step S1 and the photovoltaic panel sample points obtained in step S2 are imported into the MaxEnt model. The optimal RM parameters and optimal FC parameters obtained in step S3 are used to construct the model and output the prediction results. The prediction results are imported into ArcGIS software, and the natural discontinuity classification method is used to classify the prediction results into five levels: high suitability, relatively suitable, generally suitable, low suitability, and unsuitable, to obtain the suitability distribution map of photovoltaic power station construction in the study area.

2. The method for predicting suitable photovoltaic site selection areas based on the MaxEnt model according to claim 1, characterized in that, In step S1, the nine types of environmental data are specifically: elevation data, slope data, aspect data, land cover data, precipitation data, solar horizontal radiation data, temperature data, distance from roads, and distance from power facilities.

3. The method for predicting suitable photovoltaic site selection areas based on the MaxEnt model according to claim 1, characterized in that, In step S1, after forming the nine types of environmental data factors, the method further includes using the R language package to perform Pearson correlation analysis on the nine types of environmental data factors and deleting environmental data elements with an absolute value of correlation coefficient greater than 0.

8.

4. The method for predicting suitable photovoltaic site selection areas based on the MaxEnt model according to claim 1, characterized in that, In step S2, the regular grid is a 300m×300m regular grid covering the study area, and the regular grid is used as the basic unit for spatial analysis.

5. The method for predicting suitable photovoltaic site selection areas based on the MaxEnt model according to claim 1 or 4, characterized in that, In step S2, the initial photovoltaic panel sample points are obtained by screening based on the proportion of photovoltaic panel area coverage in each grid unit. Specifically, the proportion of photovoltaic panel area coverage in each grid unit to the total area of ​​the grid is calculated, and only grid units with a photovoltaic panel area proportion greater than or equal to 50% are retained. The geometric center point of the grid unit is used as the initial photovoltaic panel sample point.

6. The method for predicting suitable photovoltaic site selection areas based on the MaxEnt model according to claim 1, characterized in that, In step S2, the spatial sparsity processing is used to eliminate the bias caused by repeated sampling, so that the sample points tend to be evenly distributed in the study area.

7. The method for predicting suitable photovoltaic site selection areas based on the MaxEnt model according to claim 1, characterized in that, In step S3, the value of RM ranges from 0.5 to 4, with a step size of 0.

5.

8. The method for predicting suitable photovoltaic site selection areas based on the MaxEnt model according to claim 1, characterized in that, In step S3, the FC combination includes six combinations: L, LQ, H, LQH, LQHP, and LQHPT.

9. The method for predicting suitable photovoltaic site selection areas based on the MaxEnt model according to claim 1, characterized in that, In step S3, a candidate model matrix is ​​constructed by setting the range of RM values ​​and the combination of FC. For each RM and FC combination in the candidate model matrix, a MaxEnt model is trained, and the AICc value of each model is calculated. The RM parameters and FC parameters corresponding to the model with the smallest AICc value are selected as the optimal parameters.

10. The method for predicting suitable photovoltaic site selection areas based on the MaxEnt model according to claim 1, characterized in that, In step S4, the natural discontinuity classification method is implemented using ArcGIS software. The continuous prediction probability value output by the MaxEnt model is divided into five level intervals according to the natural discontinuities, corresponding to five suitability levels: high suitability, relatively suitable, moderately suitable, low suitability, and unsuitable.