Space-time big data and multi-dimensional environmental index driven wind energy suitability evaluation method

Abstract

The present application belongs to the field of wisdom energy geographic intelligence, and relates to a wind energy suitability evaluation method driven by time-space big data and multi-dimensional environmental indicators, comprising: 1, data acquisition and preprocessing; 2, variable screening and model building; 3, driving factor contribution effect analysis; 4, wind farm suitability mapping and verification; the present application combines multi-source time-space data fusion, interpretable nonlinear modeling and high-resolution spatial suitability mapping organically, not only improves the precision, reliability and interpretability of wind farm suitability evaluation, but also realizes the technical upgrading from single index evaluation to natural-social coupling decision analysis, significantly enhances the scientific support ability of regional wind farm site selection optimization and wind energy planning configuration.

Description

Technical Field

[0001] This invention belongs to the field of smart energy geospatial intelligence technology, specifically involving a wind energy suitability evaluation method driven by spatiotemporal big data and multi-dimensional environmental indicators. Background Technology

[0002] For evaluating the spatial layout and utilization potential of renewable energy, especially wind energy, at the regional scale, existing evaluation methods include: methods based on resource endowment and GIS / multi-criteria decision-making, methods based on statistical regression or linear weighting, predictive model methods based on machine learning, and methods based on micro-engineering optimization or planning simulation.

[0003] Existing methods for evaluating the spatial layout and utilization potential of wind energy have the following drawbacks: 1. Insufficient characterization of the coupling mechanism between natural resources and the socio-environmental system. Most existing wind farm suitability assessment methods still primarily focus on meteorological resource endowment, topographical conditions, or fixed spatial constraints, typically treating socio-economic factors such as grid connection feasibility, transportation accessibility, and policy constraints as secondary conditions. This fails to systematically characterize the deeper coupling mechanism between natural resources and human systems. Given the increasing importance of land resource constraints and development costs, existing methods do not adequately reveal the combined effects, synergistic interactions, and mutual constraints between resource endowment and socio-economic constraints. This can easily lead to discrepancies between model evaluation results and actual wind farm deployment, hindering the scientific development of wind energy.

[0004] 2. Insufficient ability to identify nonlinear response characteristics and thresholds. Existing methods typically assume a linear relationship between influencing factors and wind farm suitability, or use fixed weights for comprehensive evaluation (such as the analytic hierarchy process, entropy weight method, etc.). This makes it difficult to accurately characterize the complex nonlinear interactions among multiple factors, and also to identify the marginal effect variations and threshold ranges of key driving factors across different value ranges. Consequently, while the evaluation results may reflect certain spatial differences, they are insufficient to further reveal the strength, range, and coupled response relationships of the dominant factors in wind farm site selection.

[0005] 3. Insufficient model interpretability hinders planning and policy applications. While machine learning-based predictive models improve the accuracy of suitability identification and classification to some extent, their internal decision-making processes are often opaque. It is difficult to quantify the positive and negative effects, marginal contributions, intensity of influence, interaction relationships, and effect curves of influencing factors across different value ranges. This hinders understanding the intrinsic mechanisms of wind farm layout formation and makes it difficult to provide clear and well-founded decision support for regional energy planning, wind farm site optimization, and policy formulation. Therefore, the practicality of existing methods in engineering applications, urban scientific research, and macro-strategic levels remains limited.

[0006] 4. Insufficient high-resolution spatial mapping and integrated decision support capabilities. Existing related studies are mostly conducted at the regional, municipal, or low-resolution grid scales. Their results often focus on target outcomes such as development potential, installed capacity, or overall suitability levels, offering limited high-precision spatial representation capabilities for wind farm layouts. This makes it difficult to meet the high spatial resolution analysis requirements of the discrete deployment of wind power facilities. Furthermore, existing technologies typically lack a complete technical chain from multi-source data processing, variable selection, model building to driving factor interpretation, threshold extraction, and suitability mapping. This makes it difficult to further transform model analysis results into verifiable, mappable, and directly applicable spatial decision-making results.

[0007] Therefore, a method for evaluating the spatial layout and utilization potential of wind energy is needed to overcome the problems of strong reliance on fixed weights, insufficient characterization of natural-social coupling mechanisms, limited nonlinear threshold identification capabilities, and insufficient high-resolution spatial decision support in existing technologies. Summary of the Invention

[0008] This invention provides the following technical solution: a wind energy suitability evaluation method driven by spatiotemporal big data and multidimensional environmental indicators, comprising: Step S1, Data Acquisition and Preprocessing: Cleaning, spatial registration, interpolation, density calculation, rasterization and standardization are performed on multi-source heterogeneous data of meteorology, remote sensing, topography, land use, transportation, power infrastructure, end-use energy facilities, policies and wind farm samples to construct wind farm sample data, natural environmental factor data, socio-economic factor data and policy factor data under a unified spatial reference system and spatial resolution. Step S2, Variable Selection and Model Building: Based on the multidimensional dataset formed in Step S1, bivariate analysis and multicollinearity diagnosis are performed on the candidate variables to identify the key driving factors affecting the suitability of wind farms. Through machine learning model comparison, hyperparameter optimization and accuracy evaluation, a wind farm suitability identification model is built. Step S3, Analysis of the Contribution Effect of Driving Factors: Based on the optimal model obtained in Step S2, interpretability analysis is carried out on the key driving factors of wind farm suitability, quantifying the global contribution, marginal effect curve, threshold range and interaction of each factor, and forming the driving factor contribution function and scoring criteria. Step S4, Wind Farm Suitability Mapping and Verification: Based on the factor status quo-contribution scoring relationship formed in Step S3, spatial values ​​are assigned and comprehensive calculations are performed on each pixel in the study area to generate a high-resolution wind farm suitability map. Spatial decision-making results that can be directly used for planning and site selection are output through external verification of new wind power projects, hotspot identification, and administrative unit aggregation statistics.

[0009] The four steps are executed sequentially to form a complete technical system of "data acquisition and preprocessing → variable screening and model building → analysis of the contribution effect of driving factors → wind farm suitability mapping and verification".

[0010] This invention presents a wind energy suitability evaluation method based on multi-source spatiotemporal data fusion, interpretable nonlinear modeling, and geospatial suitability mapping. By integrating multi-source spatiotemporal data including meteorological, topographical, land use, infrastructure, policy, and wind farm samples, a natural-social coupling assessment framework for wind farm site selection is constructed. Based on this, a wind farm suitability identification model is established using machine learning methods, and interpretable analysis methods are further introduced to quantitatively reveal the marginal contribution characteristics, threshold ranges, and interaction mechanisms of key driving factors. Subsequently, the contribution functions of driving factors are coupled with spatial pixel information to generate high spatial resolution wind farm suitability distribution results. Through external verification of new wind power projects, spatial statistics, and key area identification, the model analysis results are transformed into spatial decision support results. This method effectively overcomes the problems of strong fixed weight dependence, insufficient characterization of natural-social coupling mechanisms, limited nonlinear threshold identification capabilities, and insufficient high-resolution spatial decision support in existing technologies, thereby significantly improving the scientific rigor, objectivity, and feasibility of wind farm site selection assessment and regional wind energy planning decisions.

[0011] The beneficial effects of this invention are: 1. This invention organically combines multi-source spatiotemporal data fusion, interpretable nonlinear modeling, and high-resolution spatial suitability mapping, which not only improves the accuracy, reliability, and interpretability of wind farm suitability assessment, but also realizes the technical upgrade from single-index evaluation to nature-society coupled decision analysis, significantly enhancing the scientific support capability for regional wind farm site selection optimization and wind energy planning and configuration.

[0012] 2. This invention enables the synergistic integrated analysis of natural environment, socio-economic conditions, and policy constraints, improving the systematicness and objectivity of wind farm suitability assessment: By integrating multi-source spatiotemporal data such as meteorology, topography, land use, transportation, power infrastructure, end-user energy facilities, and policy factors, this invention constructs a natural-social coupled assessment system for wind farm site selection. This achieves a unified expression and synergistic analysis of multi-dimensional driving factors, thereby more comprehensively depicting the combined influence and synergistic effect between resource endowment and socio-economic policy constraints, significantly improving the systematicness, objectivity, and real-world relevance of suitability assessment results.

[0013] 3. This invention can break free from fixed weight dependence and realize a nonlinear, threshold-based wind farm suitability identification mechanism: This invention introduces a method combining machine learning and interpretability analysis to establish a wind farm suitability identification model in a high-dimensional sample space, and further extracts the marginal effect curves, threshold intervals, and stage response characteristics of key driving factors. It realizes the automatic identification of the nonlinear effects of key factors such as wind speed, altitude, distance from urban areas, infrastructure density, and policy incentives, thereby avoiding the evaluation bias caused by fixed weight assumptions, improving the accuracy and stability of suitability identification results, and helping to analyze the impact of multidimensional indicators on the suitability of wind energy layout from a mechanism perspective.

[0014] 4. This invention significantly enhances the interpretability and planning decision support capabilities of the model: By conducting contribution decomposition, marginal effect analysis, and interaction analysis of driving factors, this invention enables the model to quantitatively reveal the positive and negative impacts, intensity, and coupling relationships of different natural environmental factors, socio-economic factors, and policy factors on wind farm suitability, overcoming the shortcomings of insufficient interpretability in traditional prediction models. This not only identifies which regions are more suitable for wind power development but also further explains the dominant reasons, key thresholds, and constraints for suitability formation, thus providing interpretable, traceable, and quantifiable decision-making basis for wind farm site selection optimization, regional energy planning, and policy formulation.

[0015] 5. This invention enables high spatial resolution wind farm suitability mapping, enhancing the feasibility of spatial decision-making outcomes: Based on the coupled calculation of driving factor contribution functions and spatial image data, this invention generates high spatial resolution wind farm suitability distribution results. Combined with external verification of new wind power projects, spatial statistics, hotspot identification, and administrative unit aggregation analysis, it transforms model analysis results into spatial decision-making outcomes. This technology can not only identify highly suitable areas, low suitable areas, and key development areas, but also output a list of key county-level administrative units with practical guiding significance, thereby significantly enhancing the application value of wind farm suitability assessment results in actual planning, site selection, and regional energy allocation. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the technical framework of the wind energy suitability evaluation method driven by spatiotemporal big data and multidimensional environmental indicators of the present invention. Figure 2 This is a statistical analysis diagram of representative natural environmental driving factors in wind field areas and non-wind field areas. Figure 3 This is a statistical analysis diagram of representative social environmental driving factors and policy proxy variables in wind field areas and non-wind field areas. Figure 4 This is a graph showing the contribution of each individual factor in this invention to the wind farm layout. Figure 5 This is a diagram illustrating the marginal effects of natural environmental factors on the suitability of wind farms according to the present invention. Figure 6 This is a diagram illustrating the marginal effects of social environmental factors on the suitability of wind farms according to the present invention. Figure 7 This is a map of priority construction areas for wind energy development based on the wind energy development suitability map constructed according to this invention; Figure 8 This is a schematic diagram of the method steps of the present invention. Detailed Implementation

[0017] The relevant technologies of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0018] like Figures 1-8 As shown, the steps of the wind energy suitability evaluation method in this embodiment include: Step S1, Data Acquisition and Preprocessing Module: Cleaning, spatial registration, interpolation, density calculation, rasterization and standardization of multi-source heterogeneous data such as meteorology, remote sensing, topography, land use, transportation, power infrastructure, end-use energy facilities, policies and wind farm samples, to construct a unified spatial reference system and a database of wind farm sample data, natural environmental factor data, socio-economic factor data and policy factor data under spatial resolution.

[0019] Step S2, Variable Selection and Model Building Module: Based on the multidimensional dataset formed in Step S1, bivariate analysis and multicollinearity diagnosis are performed on the candidate variables to identify the key driving factors affecting the suitability of wind farms. Through comparison of various machine learning models, hyperparameter optimization and accuracy evaluation, a wind farm suitability identification model is constructed.

[0020] Step S3, Driving Factor Contribution Effect Analysis Module: Based on the optimal model obtained in Step S2, interpretability analysis is conducted on the key driving factors of wind farm suitability, quantifying the global contribution, marginal effect curve, threshold range and interaction of each factor, and forming the driving factor contribution function and scoring criteria.

[0021] Step S4, Wind Farm Suitability Mapping and Verification Module: Based on the factor status quo-contribution scoring relationship formed in Step S3, spatial values ​​are assigned and comprehensive calculations are performed on each pixel in the study area to generate a high-resolution wind farm suitability map. The spatial decision-making results that can directly serve planning and site selection are output through external verification of new wind power projects, hotspot identification, and administrative unit aggregation statistics.

[0022] Each step of this implementation method is as follows: Step S1, Data Acquisition and Preprocessing: S1-1, Physical Environment Modeling: Based on monitoring and station data, remote sensing images, and digital elevation models, interpolation, band processing, preprocessing, and derived calculations are performed on natural environmental indicators such as wind speed, air pressure, sunshine duration, relative humidity, altitude, and slope to construct meteorological and physical spatial information and form a natural environmental dataset that reflects the resource endowment, meteorological conditions, and topographic constraints of wind farms.

[0023] Missing values ​​in the panel data were handled using interpolation methods. For variables where observations at both time points were available, linear interpolation was used; while for sequences with consecutive missing values, an Autoregressive Integrated Moving Average (ARIMA) model was used for estimation.

[0024] (1) (2) in, Indicates at time The interpolated value; and They represent The most recent and last observed values; and For the corresponding time index; and These are the autoregressive coefficients and moving average coefficients of the ARIMA model, respectively. This is the white noise error term; and These represent the order of autoregression and the order of moving average, respectively.

[0025] The spatial distributions of meteorological variables, including WIN, SSD, and RHU, were generated using the kriging method based on station observation data and grid interpolation data.

[0026] (3) in, Indicates unsampled locations The predicted value at that location; For the site Observations at; Indicates site The Kriging weights, which are determined by the spatial covariance structure; This represents the total number of stations participating in the interpolation calculation.

[0027] S1-2, Social Environment Modeling: Based on spatial vector data, we perform geographic processing and data classification on social spatial elements such as commerce, residence, roads, and supply, and extract indicators such as traffic accessibility, distribution of end-use energy facilities, and density of energy supply infrastructure to form a social environment dataset that reflects the spatial location conditions and infrastructure support capabilities of wind farms.

[0028] The spatial intensity of point-based facilities—such as end-use energy sites and other energy supply facilities—is quantified using the kernel density estimation (KDE) method.

[0029] (4) (5) in, Indicates position Estimated density at; Point-like facilities The coordinates; Total number of facilities; Bandwidth for controlling the smoothness; Let be the kernel function, where express and The Euclidean distance between them.

[0030] The search radius for kernel density estimation is determined using Ripley's K function, which quantifies the spatial clustering characteristics of point-like facilities.

[0031] (6) in, Indicates distance The value of Ripley's K function at that location; Indicates the area of ​​the study region; The total number of points; This is an indicator function that indicates when the distance between two points is less than or equal to... The value is 1 if the condition is met, and 0 otherwise. This corresponds to a significant deviation from a completely spatially random distribution. The value is selected as the search radius for kernel density estimation.

[0032] The impact of transportation infrastructure such as roads and railways on wind farm site selection is assessed using Euclidean distance to characterize the accessibility and feasibility of equipment transportation.

[0033] (7) in, Indicates position The shortest Euclidean distance to the nearest transportation element; The coordinates of the target location; For transportation elements The coordinates; the minimum value is in all This was obtained from the various traffic elements.

[0034] S1-3, Policy and Panel Data Spatialization: Based on remote sensing / GEE data, panel data, planning policy texts, and related auxiliary information, this study constructs policy and socioeconomic proxy variables such as land use / cover (LULC), feed-in tariff (FiT), and GDP through supervised classification, data cleaning, corpus coding and construction, and spatialization processing, forming a policy proxy variable dataset.

[0035] Remote sensing imagery is mainly used in two aspects: first, to extract vegetation and thermal environment features; and second, to identify urban boundaries and photovoltaic power stations using convolutional neural networks (CNNs). The calculation of the Normalized Difference Vegetation Index (NDVI), Land Surface Temperature (LST), and the CNN-based feature extraction process are shown below.

[0036] (8) (9) (10) in, and These represent the near-infrared band and the red band, respectively. Indicates spectral radiance; and This is the calibration constant; CNN indicates that... Feature map of the layer; The input feature map comes from the previous layer; and These represent the convolution kernel and the bias term, respectively. This represents the convolution operation; This represents the activation function.

[0037] Land use types are further subdivided to reflect the development cost characteristics of different land types. Based on the kernel density estimation results of built-up area boundaries and point facilities, the roulette wheel selection method is used to assign values ​​to specific urban functions.

[0038] (11) in, Indicates the function The probability assigned to a given grid cell; Representation and Function The corresponding kernel density value; This represents the total number of candidate city functions considered for this grid cell. This selection process ensures that grid cells with higher suitability have a proportionally higher probability of being assigned the corresponding function.

[0039] Digital elevation models are used to calculate slopes to reflect the impact of terrain on construction costs and project feasibility, as shown in the equation.

[0040] (12) in, and They represent elevations respectively. about coordinates and The partial derivatives of coordinates are used to characterize the rate of change of terrain in the corresponding direction.

[0041] S1-4, Spatial mapping of wind farms: Based on wind power project data such as global energy monitoring, the location information, statistical attributes and installed capacity of wind power projects are cleaned and spatialized to construct a nationwide wind farm sample dataset.

[0042] The existence of wind farms is represented by a binary variable, where 1 indicates the presence of a wind farm and 0 indicates a non-wind farm area. First, samples are extracted from both wind farm and non-wind farm areas to construct an analysis dataset. To identify candidate driving factors with significant differences between the two classes of samples, this invention employs nonparametric statistical tests for analysis.

[0043] (13) (14) in, This represents the Mann–Whitney test statistic used to test the difference between two independent samples; For the statistics corresponding to the wind farm sample group; and The sample sizes are for wind farm areas and non-wind farm areas, respectively. This is the rank sum of the observations corresponding to the wind farm samples after all samples have been merged and sorted.

[0044] S1-5, Stage Output The outputs of this stage include natural environment datasets, social environment datasets, policy proxy variable datasets, and wind farm sample datasets. These datasets are standardized and organized under a unified coordinate system and spatial resolution, providing an input basis for subsequent variable selection, model building, and suitability mapping.

[0045] Step S2, Variable Selection and Model Building: S2-1, Latent Variable Identification: Based on the multidimensional factor data generated in step S1, bivariate analysis was performed on the candidate variables to identify potential driving factors that are significantly related to the suitability of wind farms, providing a basis for subsequent screening.

[0046] S2-2, Independent Variable Screening: Further screening of potential driving factors was conducted, with a focus on analyzing the causal inference results of multicollinearity, significance, autocorrelation, and influencing mechanisms. Redundant variables were eliminated, and key variables that can effectively characterize the suitability of wind farm site selection, including natural environment, socio-economic conditions, and policy constraints, were retained.

[0047] Multicollinearity among the remaining variables was assessed using the variance inflation factor (VIF).

[0048] (15) in, Representing variables The variance inflation factor; Indicates the variable The coefficients of determination obtained by regressing all other explanatory variables. To ensure the stability of the model, variables with high multicollinearity are removed.

[0049] Based on statistical tests of differences, multicollinearity diagnosis, and mechanistic analysis, the set of driving factors affecting the suitability of wind farms was finally determined.

[0050] S2-3, Sampling: Training and validation sample sets are constructed based on wind farm samples and background non-wind farm samples to provide input data for subsequent model comparison, training and validation.

[0051] To quantitatively characterize the relationship between various driving factors and wind farm suitability, this invention introduces several machine learning algorithms, including Extreme Gradient Boosting (XGBoost), Random Forest, Extra Trees, Light Gradient Boosting Machine (LightGBM), Gradient Boosting, AdaBoost, and K-Nearest Neighbors (KNN). Based on their learning mechanisms, these models can be categorized into ensemble tree regression models, boosting-type models, and distance-based learning models.

[0052] For ensemble tree regression methods such as random forests and extreme random trees, the prediction results are obtained by averaging the outputs of multiple decision trees constructed based on random samples or random features.

[0053] (16) in, Indicates sample The predictive suitability score; Indicates the first The prediction results of each decision tree; This represents the total number of decision trees in the ensemble model. In Random Forest, each tree is trained using Bootstrap sampling and random feature selection; while Extremely Random Tree further enhances the randomness of the model by randomly selecting the split threshold.

[0054] For Boosting-type regression models, including Gradient Boosting, XGBoost, LightGBM, and AdaBoost, the final prediction result is obtained by additively combining a series of sequentially trained weak learners.

[0055] (17) in, Indicates the first A weak learner; Indicates assignment to the first The weights of each learner; This represents the total number of iterations in the Boosting process. Within this framework, each new learner focuses on fitting the residual error generated by the previous ensemble model to progressively improve overall prediction performance.

[0056] For distance-based learning methods, the KNN model estimates fitness by using the response values ​​of the nearest neighbor samples in the feature space: (18) in, In the feature space, it represents the relationship with Recent A set consisting of neighbors; For the neighbors The observation suitability value; The preset number of neighbors used for prediction.

[0057] S2-4, Multi-model comparison, model training and validation: Based on the sampling results, multi-model comparison, hyperparameter optimization, model training and validation were carried out, and the accuracy, precision, robustness and sensitivity of the models were analyzed. Finally, Model 1: Multi-dimensional Environment–Wind Model was formed.

[0058] The model with the best predictive performance is selected and further optimized through grid search to determine the optimal combination of parameters.

[0059] (19) in, This represents the optimal combination of hyperparameters; This is the preset parameter search space; Indicates the combination of parameters The prediction function of the selected machine learning model; For the suitability of observation; This represents the loss function used to evaluate model performance. Grid search, through... The system iterates through all parameter combinations to select the optimal configuration that minimizes the prediction error.

[0060] For Model 1, multiple evaluation metrics were used to assess its predictive performance, including Area Under the Receiver Operating Characteristic Curve (AUC), Precision, Recall, and F1 score.

[0061] (20) in, This refers to the True Positive Rate (TPR). This represents the false positive rate (FPR).

[0062] (twenty one) in, This represents the proportion (i.e., accuracy) of correctly predicted wind farms out of all samples predicted as wind farms. This indicates the number of True Positives (TP). This indicates the number of false positives (FP).

[0063] (twenty two) in, The harmonic mean of precision and recall (i.e., the F1 score) is used to provide a balanced evaluation of model performance while considering both false positives and false negatives.

[0064] S2-5, Stage Output: The outputs of this stage include a set of key driving factors, a sample dataset, multi-model comparison results, model training and validation results, and Model 1, which are used to support subsequent analysis of the contribution effects of driving factors.

[0065] Step S3: Marginal effect analysis of driving factors on wind power layout: S3-1, Model Interpretability Analysis: Based on Model 1 formed in step S2, interpretability analysis methods such as SHAP are used to interpret the wind farm suitability results, extract historical contributions, marginal effect characteristics and interaction effect results, and construct Model 2: Contribution Effect Model of Multidimensional Indicators on Wind Farm Layout Suitability.

[0066] The SHapley Additive exPlanations (SHAP) method is introduced to estimate the marginal contributions of each factor and their interaction effects. (twenty three) in, Representation of features The Shapley value is used to characterize the marginal contribution of the feature to the model's predictions; For the complete set of features; Indicates that it does not contain features Any subset of features; For subset The number of elements in; and These respectively represent the inclusion and exclusion of features. Model prediction results under the following circumstances Indicates the difference between features All feature sets other than Indicates the inclusion of features The input vector corresponding to the feature subset, Indicates that only subsets are included. The input vector of the features.

[0067] S3-2, Threshold Extraction: Based on the contribution results of driving factors, correlation analysis, threshold effect identification and lookup table construction were carried out on key factors to extract key threshold ranges and response characteristics affecting the suitability of wind farms.

[0068] Based on the marginal contribution curve obtained from the SHAP value, this invention constructs a piecewise or nonlinear relationship between feature values ​​and their contribution scores through point aggregation and curve fitting methods.

[0069] (twenty four) in, Indicates the feature value The corresponding contribution score; This represents the function obtained by fitting the marginal contribution curve of the aggregated SHAP.

[0070] S3-3, Threshold Cross-Validation: Based on the threshold extraction results, threshold cross-validation is carried out, including robustness analysis, multi-region case analysis and parameter calibration, and the threshold results are verified and compared in combination with regional spatial representation.

[0071] To evaluate the robustness of the relationships obtained under different sampling strategies, this invention calculates the Mean Functional Distance (D) to measure the difference between different fitting contribution functions.

[0072] (25) in, This represents the average functional distance between two fitted functions; and At point, respectively under different sampling strategies The predicted contribution score obtained at the location; This represents the number of evaluation points sampled.

[0073] Furthermore, the Pearson correlation coefficient was used to further evaluate the consistency of results under different sampling strategies.

[0074] (26) in, and These represent the contribution scores obtained under two different sampling strategies; and It is its corresponding average value.

[0075] S3-4, Stage Output: The output of this stage includes the historical contribution results of driving factors, marginal effect characteristics, interaction effect results, threshold extraction results, and threshold cross-validation results, which are used to support subsequent wind farm suitability mapping.

[0076] Step S4: Wind farm suitability mapping: S4-1, Geographically Weighted Analysis: Based on the factor contribution relationship and threshold rule formed in step S3, nonlinear geooverlay analysis is carried out to calculate the wind farm suitability score of each pixel in the study area.

[0077] Model 3 aims to generate a spatial suitability distribution for wind farm deployment. Based on the factor-contribution relationship obtained in Model 2, the contribution scores of all driving factors are aggregated to calculate the final suitability score for each cell.

[0078] (27) in, Indicates spatial location The final suitability score for the location; Indicates factor-based The contribution function obtained by fitting the marginal contribution curve; For position Factors The value of ; This represents the total number of driving factors considered in the model.

[0079] S4-2 Verification of Newly Built Wind Farms: By leveraging the spatial consistency between newly constructed wind power projects and suitability results, the suitability results of wind farms are validated, and the reliability and real-world adaptability of the model results are evaluated.

[0080] To evaluate the predictive reliability of the suitability map, this invention selected wind farm locations newly acquired in 2025 but not included in the training dataset as external validation samples. The spatial consistency between these validation samples and the top 10% of suitable regions was calculated and evaluated using Cohen's Kappa coefficient.

[0081] (28) in, This represents the Kappa coefficient. To verify the actual consistency between the wind farm location and the predicted high suitability area; This represents the expected consistency under random conditions.

[0082] S4-3, Suitability Map Generation: Based on the results of geographic weighted analysis and validation, a wind farm suitability map is generated, and further output map representation, spatial statistical results and key area identification results are generated to form Model 3: Wind Farm Suitability Mapping Model.

[0083] Furthermore, highly suitable areas are aggregated at the basic administrative unit scale to quantify their spatial distribution characteristics. Specifically, the area of ​​the top 10% of suitable pixels within each administrative unit is calculated.

[0084] (29) in, Indicates administrative unit The total area of ​​the top 10% of pixels in terms of interior suitability; This represents a single pixel within the unit; As an indicator function, when a cell The value is 1 if the candidate belongs to the top 10% suitability category, and 0 otherwise.

[0085] Furthermore, based on the fitted factor-contribution curves, statistical summaries were conducted for different suitability levels to analyze their distribution characteristics. Simultaneously, Moran's I index was used to assess the spatial autocorrelation of the final suitability map.

[0086] (30) in, This represents Moran's I index; This refers to the number of spatial units; and Representing spatial units and Suitability score; The average of the suitability scores; Indicates position and Spatial weights between them; This is the sum of spatial weights.

[0087] S4-4, Stage Output: The outputs of this stage include wind farm suitability maps, spatial statistics, key area identification results, and verification results of newly built wind farms, which are used to support regional wind power planning, site selection optimization, and spatial decision-making.

[0088] In summary, this invention constructs a complete technical framework for wind energy suitability evaluation by integrating physical environment, social environment, policy and panel data, and wind farm spatial information. This framework first standardizes and spatializes multi-source data through a data acquisition and preprocessing module, laying the foundation for subsequent analysis. Then, using a variable selection and model building module, it introduces various machine learning algorithms and optimizes them through multi-model comparison to form Model 1, which accurately depicts the relationship between multi-dimensional environment and wind energy suitability. Subsequently, using a model interpretability analysis, threshold extraction, and cross-validation module, it deeply analyzes the marginal effects and key thresholds of driving factors, constructing Model 2. Finally, through a geographic weighted analysis, new wind farm verification, and suitability map generation module, it completes the spatial mapping of wind farm suitability in the study area, forming Model 3. The entire methodology closely integrates spatiotemporal big data and multi-dimensional environmental indicators, achieving full-chain technical support from data processing to model building, effect analysis, and spatial mapping. This provides a scientific and accurate decision-making basis for regional wind power planning, effectively improving the rationality and efficiency of wind energy resource development and utilization.

[0089] It should be emphasized that the above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any way. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention shall still fall within the scope of the technical solution of the present invention.

Claims

1. A wind energy suitability evaluation method driven by spatiotemporal big data and multidimensional environmental indicators, characterized in that, include: Step S1, Data Acquisition and Preprocessing: Cleaning, spatial registration, interpolation, density calculation, rasterization and standardization are performed on multi-source heterogeneous data of meteorology, remote sensing, topography, land use, transportation, power infrastructure, end-use energy facilities, policies and wind farm samples to construct wind farm sample data, natural environmental factor data, socio-economic factor data and policy factor data under a unified spatial reference system and spatial resolution. Step S2, Variable Selection and Model Building: Based on the multidimensional dataset formed in Step S1, bivariate analysis and multicollinearity diagnosis are performed on the candidate variables to identify the key driving factors affecting the suitability of wind farms. Through machine learning model comparison, hyperparameter optimization and accuracy evaluation, a wind farm suitability identification model is built. Step S3, Analysis of the Contribution Effect of Driving Factors: Based on the optimal model obtained in Step S2, interpretability analysis is carried out on the key driving factors of wind farm suitability, quantifying the global contribution, marginal effect curve, threshold range and interaction of each factor, and forming the driving factor contribution function and scoring criteria. Step S4, Wind Farm Suitability Mapping and Verification: Based on the factor status quo-contribution scoring relationship formed in Step S3, spatial values ​​are assigned and comprehensive calculations are performed on each pixel in the study area to generate a high-resolution wind farm suitability map. Spatial decision-making results that can be directly used for planning and site selection are output through external verification of new wind power projects, hotspot identification, and administrative unit aggregation statistics.

2. The wind energy suitability evaluation method driven by spatiotemporal big data and multidimensional environmental indicators according to claim 1, characterized in that, Step S1 includes the following sub-steps: S1-1, Physical Environment Modeling: Based on monitoring and station data, remote sensing images and digital elevation models, interpolation, band processing, preprocessing and derived calculations are performed on natural environment indicators to construct meteorological and physical spatial information and form a natural environment dataset that reflects the wind farm's resource endowment, meteorological conditions and topographic constraints. S1-2, Social Environment Modeling: Based on spatial vector data, the social spatial elements are geographically processed and classified to extract indicators such as traffic accessibility, distribution of end-user energy facilities and density of energy supply infrastructure, forming a social environment dataset that reflects the spatial location conditions and infrastructure support capabilities of wind farms. S1-3, Spatialization of Policy and Panel Data: Based on remote sensing / GEE data, panel data, planning policy texts and related auxiliary information, policy and socio-economic proxy variables are constructed through supervised classification, data cleaning, corpus coding and construction, and spatialization processing to form a policy proxy variable dataset. S1-4. Wind Farm Spatial Mapping: Based on wind power project data, the location information, statistical attributes, and installed capacity of wind power projects are cleaned and spatialized to construct a nationwide wind farm sample dataset.

3. The wind energy suitability evaluation method driven by spatiotemporal big data and multidimensional environmental indicators according to claim 1, characterized in that, Step S2 includes the following sub-steps: S2-1. Identification of latent variables: Based on the multidimensional factor data formed in step S1, bivariate analysis is performed on candidate variables to identify potential driving factors that are significantly related to the suitability of wind farms. S2-2, Independent Variable Screening: Further screening of potential driving factors, focusing on the causal inference results of multicollinearity, significance, autocorrelation and influence mechanism, eliminating redundant variables, and retaining key variables that can effectively characterize the suitability of wind farm site selection, such as natural environment, socio-economic conditions and policy constraints. S2-3. Sampling: Construct training and validation sample sets based on wind farm samples and background non-wind farm samples; S2-4, Multi-model comparison, model training and validation: Based on the sampling results, multi-model comparison, hyperparameter optimization, model training and validation are carried out respectively, and the accuracy, precision, robustness and sensitivity of the models are analyzed to form a multi-dimensional environment-wind power model.

4. The wind energy suitability evaluation method driven by spatiotemporal big data and multidimensional environmental indicators according to claim 3, characterized in that, The multidimensional environment-wind power model is as follows: （19） in, This represents the optimal combination of hyperparameters. For the preset parameter search space, Indicates the combination of parameters The prediction function of the selected machine learning model. Indicates the suitability of the observation. This represents the loss function used to evaluate model performance.

5. The wind energy suitability evaluation method driven by spatiotemporal big data and multidimensional environmental indicators according to claim 1, characterized in that, Step S3 includes the following sub-steps: S3-1, Model Interpretability Analysis: Based on the multidimensional environment-wind power model formed in step S2, interpretability analysis method is used to interpret the wind farm suitability results, extract historical contribution, marginal effect characteristics and interaction effect results, and construct a multidimensional index contribution effect model; S3-2, Threshold Extraction: Based on the contribution results of driving factors, conduct correlation analysis, threshold effect identification and lookup table construction for key factors, and extract key threshold intervals and response characteristics that affect the suitability of wind farms; S3-3, Threshold Cross-Validation: Based on the threshold extraction results, threshold cross-validation is carried out, and the threshold results are verified and compared in combination with the regional spatial representation.

6. The wind energy suitability evaluation method driven by spatiotemporal big data and multidimensional environmental indicators according to claim 5, characterized in that, The multidimensional index contribution effect model is as follows: （23） in, Representation of features Shapley value, For the complete set of features, Indicates that it does not contain features Any feature subset, For subset The number of elements in and These respectively represent the inclusion and exclusion of features. Model prediction results under the following circumstances Indicates the difference between features All feature sets other than Indicates the inclusion of features The input vector corresponding to the feature subset, Indicates that only subsets are included. The input vector of the features.

7. The wind energy suitability evaluation method driven by spatiotemporal big data and multidimensional environmental indicators according to claim 1, characterized in that, Step S4 includes the following sub-steps: S4-1, Geographic Weighted Analysis: Based on the factor contribution relationship and threshold rules formed in step S3, nonlinear geographic overlay analysis is carried out to calculate the wind farm suitability score of each pixel in the study area. S4-2, Verification of New Wind Farms: By utilizing the spatial consistency between new wind power projects and suitability results, the suitability results of wind farms are verified, and the reliability and real-world adaptability of the model results are evaluated. S4-3. Suitability Map Generation: Based on the geographically weighted analysis results and verification results, a wind farm suitability map is generated, and further output map representation, spatial statistical results and key area identification results are generated to form a wind farm suitability mapping model.

8. The wind energy suitability evaluation method driven by spatiotemporal big data and multidimensional environmental indicators according to claim 7, characterized in that, The wind farm suitability mapping model is as follows: （29） in, Indicates administrative unit The total area of ​​the top 10% of pixels in terms of interior suitability; This represents a specific cell within the unit. This is an indicator function.

Images