Photovoltaic power generation power climbing prediction and early warning method based on multi-dimensional observation information
By performing spatiotemporal registration and accuracy correction of multidimensional observation information, and combining graph neural network analysis to identify spatial correlations between power plants, the problem of accurate power prediction for photovoltaic power generation systems under extreme weather conditions was solved, enabling precise prediction and timely early warning for large-scale photovoltaic power plant clusters.
Patent Information
- Application Number
- CN202511625190.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-07
- Publication Date
- 2026-03-20
AI Technical Summary
Existing photovoltaic power generation forecasting methods are unable to effectively integrate multi-dimensional observation information when faced with complex and ever-changing meteorological conditions. They lack the ability to cope with drastic power fluctuations under extreme weather conditions and are unable to accurately capture the mutual influence between power plants in the region.
By performing spatiotemporal registration and accuracy correction on multi-source data, a standardized multidimensional observation dataset is generated. Weather patterns are identified and a nonlinear mapping model between weather patterns and surface irradiance is constructed. Graph neural networks are used to analyze the spatial correlation between power plants, calculate the power ramp-up risk index, and trigger early warning signals.
It enables accurate prediction and timely early warning of power ramp-up risks in large-scale photovoltaic power plant clusters under extreme weather conditions, providing important decision support for grid dispatch and photovoltaic power plant management.
Smart Images
Figure CN121710167A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of information technology, and in particular to a method for predicting and warning the ramp-up of photovoltaic power generation based on multidimensional observation information. Background Technology
[0002] Photovoltaic power generation, as an important component of clean energy, plays a crucial role in the global energy transition. With the rapid growth of photovoltaic installed capacity, large-scale photovoltaic power generation systems have become an important power source for the power system, and their stability and predictability directly affect the safe operation of the power grid.
[0003] Current photovoltaic (PV) power generation forecasting methods primarily rely on single data sources and fixed forecasting models, which proves inadequate when facing complex and variable weather conditions. Traditional forecasting methods often overlook the differentiated characteristics of PV output under different weather patterns and lack the ability to effectively address drastic power fluctuations under extreme weather conditions. Existing technologies also have limitations in handling the spatial correlation of large-scale PV power plant clusters, making it difficult to accurately capture the mutual influence relationships between power plants within a region.
[0004] Under extreme weather conditions, the core challenge facing photovoltaic power generation systems stems from the effective fusion of multi-dimensional observational information. Satellite remote sensing data, numerical weather prediction, and ground-based measured data each possess different spatiotemporal resolutions and accuracy characteristics. How to organically integrate these heterogeneous data sources and construct a unified prediction framework directly determines the accuracy of the prediction model. This fusion challenge further raises technical challenges in weather model recognition and classification, because cloud distribution characteristics and irradiance variation patterns differ significantly under different weather conditions, requiring the establishment of targeted inversion models to accurately extract surface irradiance information. Based on the need for weather model classification, the complex problem of spatial correlation modeling of photovoltaic power plant clusters arises. The power output of each power plant does not change independently but is jointly influenced by the propagation path of regional weather systems. Accurately characterizing this dynamic correlation becomes a key link in achieving accurate prediction. Summary of the Invention
[0005] To address the shortcomings of existing technologies, this invention provides a photovoltaic power generation ramp-up prediction and early warning method based on multi-dimensional observation information. This method enables accurate prediction and timely early warning of power ramp-up risks in large-scale photovoltaic power plant clusters under extreme weather conditions, providing important decision support for grid dispatching and photovoltaic power plant management.
[0006] The technical solution of the present invention is as follows: A photovoltaic power generation ramp-up prediction and early warning method based on multi-dimensional observation information includes: Acquire multi-source data and perform spatiotemporal registration and accuracy correction to generate a standardized multidimensional observation dataset; Based on the standardized multidimensional observation dataset, weather patterns are identified and a weather pattern classification dataset and an extreme weather label set are generated. Based on the aforementioned weather pattern classification dataset, a mapping model between weather patterns and surface irradiance is constructed to generate a power prediction benchmark dataset. Based on the power prediction benchmark dataset and the geographical locations of each photovoltaic power station, a spatial correlation topology dataset and a dynamic weight dataset for the power station group are generated. Based on the spatial association topology dataset, dynamic weight dataset, and extreme weather label set of the power plant group, the power ramp-up risk index is calculated, and a large-scale photovoltaic power ramp-up prediction and early warning dataset based on multi-dimensional observation information under extreme weather conditions is generated.
[0007] Furthermore, the acquisition of multi-source data and the performance of spatiotemporal registration and accuracy correction to generate a standardized multidimensional observation dataset include: The acquired multi-source data includes satellite remote sensing data, numerical weather prediction data, and ground-based measured data; Kriging interpolation algorithm is used to perform spatiotemporal registration of satellite remote sensing data, numerical weather prediction data and ground measured data, so as to unify the spatiotemporal resolution of multi-source data to hour-level temporal resolution and preset spatial resolution. Based on the registered data, a standardized multidimensional observation dataset containing cloud coverage, cloud movement speed, irradiance time series, temperature, and wind speed is generated. The standardized multidimensional observation dataset is subjected to accuracy correction to obtain the corrected multidimensional observation dataset.
[0008] Furthermore, the spatiotemporal registration of the satellite remote sensing data, numerical weather prediction data, and ground-based measured data using the Kriging interpolation algorithm includes: Determine the spatiotemporal resolution differences between the satellite remote sensing data, numerical weather prediction data, and ground-measured data; Based on the spatiotemporal resolution differences, the data is interpolated using the Kriging interpolation algorithm to generate a dataset with uniform spatiotemporal resolution. The temporal resolution of the dataset with the unified spatiotemporal resolution is adjusted to the hour level, and the spatial resolution is adjusted to the preset distance unit. Based on the adjusted dataset, registration data including cloud coverage, cloud movement speed, irradiance time series, temperature, and wind speed are generated.
[0009] Furthermore, the step of identifying weather patterns and generating a weather pattern classification dataset and an extreme weather label set based on the standardized multidimensional observation dataset includes: Time series of cloud cover, cloud movement speed and irradiance were extracted from the standardized multidimensional observation dataset. The K-means clustering algorithm was used to process the time series of cloud coverage, cloud movement speed and irradiance to identify sunny, cloudy and overcast weather patterns. If the cloud coverage is greater than a preset coverage threshold, the cloud movement speed is greater than a preset speed threshold, and the irradiance change gradient is greater than a preset gradient threshold, then it is marked as an extreme weather mode. Based on the recognition results, a weather pattern classification dataset and an extreme weather label set are generated.
[0010] Furthermore, the process of using the K-means clustering algorithm to process the cloud coverage, cloud movement speed, and irradiance time series to identify sunny, cloudy, and overcast weather patterns includes: The cloud coverage, cloud movement speed, and irradiance time series are used as input features; The input features are clustered using the K-means clustering algorithm to generate multiple weather pattern clusters; Based on the characteristic distribution of the weather pattern cluster, sunny, cloudy, and overcast weather patterns are determined; If the combination of cloud coverage, cloud movement speed, and irradiance change gradient satisfies the preset extreme weather conditions, then the corresponding weather pattern will be marked as an extreme weather pattern.
[0011] Furthermore, the step of constructing a mapping model between weather patterns and surface irradiance based on the weather pattern classification dataset, and generating a power prediction benchmark dataset, includes: Extract cloud cover, irradiance time series, temperature, and wind speed corresponding to the weather patterns from the weather pattern classification dataset; The cloud coverage, irradiance time series, temperature, and wind speed are converted into two-dimensional feature maps. The two-dimensional feature map is processed using a convolutional neural network algorithm to construct a nonlinear mapping model between weather patterns and surface irradiance. The theoretical power generation of each photovoltaic power station is calculated using the nonlinear mapping model to generate a power prediction benchmark dataset.
[0012] Furthermore, the step of processing the two-dimensional feature map using a convolutional neural network algorithm to construct a nonlinear mapping model between weather patterns and surface irradiance includes: The two-dimensional feature map is input into a pre-trained convolutional neural network; The spatial features of the two-dimensional feature map are extracted using the convolutional neural network. A nonlinear mapping relationship between weather patterns and surface irradiance is constructed based on the aforementioned spatial characteristics; The theoretical irradiance value of each photovoltaic power station is calculated using the aforementioned nonlinear mapping relationship; Based on the theoretical irradiance value and the conversion efficiency of the photovoltaic power station, the theoretical power generation of each photovoltaic power station is generated.
[0013] Furthermore, the step of generating a spatial association topology dataset and a dynamic weight dataset for the power plant cluster based on the power prediction benchmark dataset and the geographical location of the photovoltaic power plants includes: Obtain the geographical coordinates of each photovoltaic power station; Based on the aforementioned power prediction benchmark dataset, calculate the Pearson correlation coefficient matrix among the photovoltaic power plants; If the Pearson correlation coefficient is greater than a preset correlation coefficient threshold, the weight of the associated edge is calculated according to the distance-weighted formula, which is:
[0014] in As weight, Let be the distance between any two photovoltaic power stations. This is the preset smoothing factor; Based on the calculated associated edge weights, a spatial association topology dataset and a dynamic weight dataset for the power plant cluster are generated.
[0015] Furthermore, the step of calculating the weight of associated edges according to the distance-weighted formula includes: Obtain the geographical coordinates of each photovoltaic power station; Calculate the distance between any two photovoltaic power stations ; Based on the distance between any two photovoltaic power stations and preset smoothing factor The weights of the associated edges between power stations are calculated using a distance-weighted formula. If the Pearson correlation coefficient is greater than the preset correlation coefficient threshold, the corresponding associated edge weight is retained; Generate a spatial association topology dataset of power plant clusters based on the retained associated edge weights; A dynamic weighted dataset is generated based on the spatial association topology dataset of the power plant group.
[0016] Furthermore, the step of calculating the power ramp-up risk index based on the spatial association topology dataset, dynamic weight dataset, and extreme weather label set of the power plant group, and generating a large-scale photovoltaic power ramp-up prediction and early warning dataset under extreme weather conditions based on multi-dimensional observation information, includes: Extract the weights of the associated edges from the spatial association topology dataset and the dynamic weight dataset of the power plant group; The power fluctuation value is determined based on the extreme weather tag set; The power ramp-up risk index is calculated using the weighted average method. If the risk index is greater than the preset risk threshold and the duration of fluctuation in the irradiance time series is greater than the preset time threshold, a power ramp-up warning signal is generated. Based on the power ramp-up warning signal, a large-scale photovoltaic power generation power ramp-up prediction and warning dataset under extreme weather conditions is generated, based on multi-dimensional observation information.
[0017] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention first performs spatiotemporal registration and accuracy correction on multi-source data to generate a standardized multidimensional observation dataset. Then, based on cloud features and irradiance changes, it identifies extreme weather patterns and constructs a nonlinear mapping model between weather patterns and surface irradiance to calculate the theoretical power generation of each photovoltaic power station. Next, it uses a graph neural network algorithm to analyze the spatial correlation between power stations, generating a dynamic weighted dataset. Finally, it combines extreme weather labels to calculate a power ramp-up risk index, triggering an early warning signal when the risk index and fluctuation duration exceed thresholds. This invention achieves accurate prediction and timely early warning of power ramp-up risk for large-scale photovoltaic power station clusters under extreme weather conditions, providing important decision support for grid dispatching and photovoltaic power station management. Attached Figure Description
[0018] Figure 1 This is a flowchart illustrating the overall process of an embodiment of the present invention.
[0019] Figure 2 This is a schematic diagram of the specific process of step S101 in an embodiment of the present invention.
[0020] Figure 3 This is a schematic diagram of the specific process of step S102 in an embodiment of the present invention.
[0021] Figure 4 This is a schematic diagram of the specific process of step S103 in an embodiment of the present invention.
[0022] Figure 5 This is a schematic diagram of the specific process of step S104 in an embodiment of the present invention.
[0023] Figure 6 This is a schematic diagram of the specific process of step S105 in an embodiment of the present invention. Detailed Implementation
[0024] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.
[0025] refer to Figures 1 to 6 This embodiment provides a photovoltaic power generation ramp-up prediction and early warning method based on multi-dimensional observation information, which may specifically include: Step S101: Acquire multi-source data and perform spatiotemporal registration and accuracy correction to generate a standardized multidimensional observation dataset.
[0026] The acquired multi-source data includes satellite remote sensing data, numerical weather prediction data, and ground-based measured data. A timestamp alignment method is used to perform preliminary spatiotemporal registration of the multi-source data, resulting in an initially aligned multi-source dataset. Based on this initially aligned dataset, the Kriging interpolation algorithm is used to unify the spatiotemporal resolution of the multi-source data to an hourly temporal resolution and a preset spatial resolution. The spatiotemporal resolution differences between the satellite remote sensing data, numerical weather prediction data, and ground-based measured data are determined. Based on these differences, the Kriging interpolation algorithm is used to interpolate the data, generating a dataset with a unified spatiotemporal resolution. The temporal resolution of this unified spatiotemporal resolution dataset is adjusted to the hourly level, and the spatial resolution is adjusted to a preset distance unit. Based on the registered data, a standardized multidimensional observation dataset containing cloud cover, cloud movement speed, irradiance time series, temperature, and wind speed is generated. The standardized multidimensional observation dataset is then subjected to accuracy correction to obtain a corrected multidimensional observation dataset.
[0027] Step S102: Based on the standardized multidimensional observation dataset, identify weather patterns and generate a weather pattern classification dataset and an extreme weather label set.
[0028] Cloud cover, cloud movement speed, and irradiance time series are extracted from the standardized multidimensional observation dataset. K-means clustering is used to process these time series to identify sunny, cloudy, and overcast weather patterns. The cloud cover, cloud movement speed, and irradiance time series are used as input features, and K-means clustering is applied to cluster these features, generating multiple weather pattern clusters. Based on the feature distribution of these weather pattern clusters, sunny, cloudy, and overcast weather patterns are determined. If the cloud cover is greater than a preset coverage threshold, the cloud movement speed is greater than a preset speed threshold, and the irradiance change gradient is greater than a preset gradient threshold, these are labeled as extreme weather patterns. A weather pattern classification dataset and an extreme weather label set are generated based on the identification results.
[0029] Step S103: Based on the weather pattern classification dataset, construct a mapping model between weather patterns and surface irradiance to generate a power prediction benchmark dataset.
[0030] Based on weather pattern classification datasets and extreme weather label sets, a convolutional neural network (CNN) algorithm is used to process two-dimensional feature maps of cloud cover, irradiance time series, temperature, and wind speed. A CNN model is constructed based on these two-dimensional feature maps, and a nonlinear mapping relationship between weather patterns and surface irradiance is trained. The two-dimensional feature maps are then input into the pre-trained CNN to extract spatial features. Based on these spatial features, a nonlinear mapping relationship between weather patterns and surface irradiance is constructed. The theoretical irradiance value for each photovoltaic (PV) power station is calculated using this nonlinear mapping relationship. Based on the theoretical irradiance value and the conversion efficiency of the PV power station, a theoretical power generation dataset for each PV power station is generated. Finally, based on the theoretical power generation dataset and the equipment parameters and efficiency coefficients of the PV power station, a power prediction benchmark dataset is generated.
[0031] Step S104: Based on the power prediction benchmark dataset and the geographical location of the photovoltaic power station, generate a spatial association topology dataset and a dynamic weight dataset for the power station group.
[0032] Based on the power prediction benchmark dataset, the geographical coordinates of each photovoltaic power station are obtained, and the distance between any two power stations is calculated using the Euclidean distance formula. Preset smoothing factor Based on historical power data in the power prediction benchmark dataset, a graph neural network algorithm is used to calculate the Pearson correlation coefficient of the power time series between each power station, resulting in a Pearson correlation coefficient matrix. If the Pearson correlation coefficient is greater than a preset threshold, the weights of the associated edges are calculated according to a distance-weighted formula. The distance-weighted formula is as follows:
[0033] If the Pearson correlation coefficient is greater than the preset correlation coefficient threshold, the corresponding associated edge weights are retained, and a spatial association topology dataset of the power plant group is generated; based on the spatial association topology dataset of the power plant group, a dynamic weight dataset is generated.
[0034] Step S105: Calculate the power ramp-up risk index based on the power plant group spatial association topology dataset, dynamic weight dataset, and extreme weather label set, and generate a large-scale photovoltaic power ramp-up prediction and early warning dataset based on multi-dimensional observation information under extreme weather conditions.
[0035] Based on the spatial association topology dataset of the power plant cluster, the association edge weights between each photovoltaic power plant are obtained to acquire spatial association strength data between power plants. Based on the dynamic weight dataset, the real-time weight values of each photovoltaic power plant are extracted to determine the weight distribution dataset. Through an extreme weather label set, extreme weather patterns within the current time period are identified to obtain a set of weather influencing factors. A convolutional neural network algorithm is used to process the two-dimensional feature maps of irradiance time series and cloud cover rate to obtain power plant power fluctuation values. The power ramp-up risk index was calculated using the weighted average method.
[0036] in The power ramp-up risk index, For weights, The power fluctuation value of the power plant is used; if the risk index is greater than the preset threshold, the fluctuation data of the irradiance time series is extracted to determine the duration of the fluctuation; if the duration of the fluctuation is greater than the preset time threshold, the power ramp-up warning signal is triggered to generate a preliminary warning dataset; the power fluctuation value in the preliminary warning dataset is calibrated using the power prediction benchmark dataset to obtain the calibrated warning dataset; the spatial association topology dataset of the power plant group is updated using geographical location coordinates and Pearson correlation coefficient matrix to obtain a large-scale photovoltaic power ramp-up prediction and warning dataset based on multi-dimensional observation information under extreme weather conditions.
[0037] Based on the above, this embodiment of the method for predicting and warning of power ramp-up in large-scale photovoltaic power generation under extreme weather conditions based on multi-dimensional observation information first performs spatiotemporal registration and accuracy correction on multi-source data to generate a standardized multi-dimensional observation dataset. Then, based on cloud characteristics and irradiance changes, extreme weather patterns are identified, and a nonlinear mapping model between weather patterns and surface irradiance is constructed to calculate the theoretical power generation of each photovoltaic power station. Next, a graph neural network algorithm is used to analyze the spatial correlation between power stations, generating a dynamic weighted dataset. Finally, combined with extreme weather labels, a power ramp-up risk index is calculated, and a warning signal is triggered when the risk index and the duration of fluctuation exceed a threshold. This invention achieves accurate prediction and timely warning of power ramp-up risk for large-scale photovoltaic power station clusters under extreme weather conditions, providing important decision support for grid dispatching and photovoltaic power station management.
[0038] The above description discloses only preferred embodiments of the present invention and should not be construed as limiting the scope of the invention. Those skilled in the art will understand that implementing all or part of the above embodiments and making equivalent changes in accordance with the claims of the present invention are still within the scope of the invention.
Claims
1. A method for predicting and warning the ramp-up of photovoltaic power generation based on multi-dimensional observation information, characterized in that, include: Acquire multi-source data and perform spatiotemporal registration and accuracy correction to generate a standardized multidimensional observation dataset; Based on the standardized multidimensional observation dataset, weather patterns are identified and a weather pattern classification dataset and an extreme weather label set are generated. Based on the aforementioned weather pattern classification dataset, a mapping model between weather patterns and surface irradiance is constructed to generate a power prediction benchmark dataset. Based on the power prediction benchmark dataset and the geographical locations of each photovoltaic power station, a spatial correlation topology dataset and a dynamic weight dataset for the power station group are generated. Based on the spatial association topology dataset, dynamic weight dataset, and extreme weather label set of the power plant group, the power ramp-up risk index is calculated, and a large-scale photovoltaic power ramp-up prediction and early warning dataset based on multi-dimensional observation information under extreme weather conditions is generated.
2. The method as described in claim 1, characterized in that, The process of acquiring multi-source data and performing spatiotemporal registration and accuracy correction to generate a standardized multidimensional observation dataset includes: The acquired multi-source data includes satellite remote sensing data, numerical weather prediction data, and ground-based measured data; Kriging interpolation algorithm is used to perform spatiotemporal registration of satellite remote sensing data, numerical weather prediction data and ground measured data, so as to unify the spatiotemporal resolution of multi-source data to hour-level temporal resolution and preset spatial resolution. Based on the registered data, a standardized multidimensional observation dataset containing cloud coverage, cloud movement speed, irradiance time series, temperature, and wind speed is generated. The standardized multidimensional observation dataset is subjected to accuracy correction to obtain the corrected multidimensional observation dataset.
3. The method as described in claim 2, characterized in that, The spatiotemporal registration of the satellite remote sensing data, numerical weather prediction data, and ground-based measured data using the Kriging interpolation algorithm includes: Determine the spatiotemporal resolution differences between the satellite remote sensing data, numerical weather prediction data, and ground-measured data; Based on the spatiotemporal resolution differences, the data is interpolated using the Kriging interpolation algorithm to generate a dataset with uniform spatiotemporal resolution. The temporal resolution of the dataset with the unified spatiotemporal resolution is adjusted to the hour level, and the spatial resolution is adjusted to the preset distance unit. Based on the adjusted dataset, registration data including cloud coverage, cloud movement speed, irradiance time series, temperature, and wind speed are generated.
4. The method as described in claim 1, characterized in that, The step of identifying weather patterns and generating a weather pattern classification dataset and an extreme weather label set based on the standardized multidimensional observation dataset includes: Time series of cloud cover, cloud movement speed and irradiance were extracted from the standardized multidimensional observation dataset. The K-means clustering algorithm was used to process the time series of cloud coverage, cloud movement speed and irradiance to identify sunny, cloudy and overcast weather patterns. If the cloud coverage is greater than a preset coverage threshold, the cloud movement speed is greater than a preset speed threshold, and the irradiance change gradient is greater than a preset gradient threshold, then it is marked as an extreme weather mode. Based on the recognition results, a weather pattern classification dataset and an extreme weather label set are generated.
5. The method as described in claim 4, characterized in that, The K-means clustering algorithm is used to process the time series data of cloud cover, cloud movement speed, and irradiance to identify sunny, cloudy, and overcast weather patterns, including: The cloud coverage, cloud movement speed, and irradiance time series are used as input features; The input features are clustered using the K-means clustering algorithm to generate multiple weather pattern clusters; Based on the characteristic distribution of the weather pattern cluster, sunny, cloudy, and overcast weather patterns are determined; If the combination of cloud coverage, cloud movement speed, and irradiance change gradient satisfies the preset extreme weather conditions, then the corresponding weather pattern will be marked as an extreme weather pattern.
6. The method as described in claim 1, characterized in that, The step of constructing a mapping model between weather patterns and surface irradiance based on the weather pattern classification dataset, and generating a power prediction benchmark dataset, includes: Extract cloud cover, irradiance time series, temperature, and wind speed corresponding to the weather patterns from the weather pattern classification dataset; The cloud coverage, irradiance time series, temperature, and wind speed are converted into two-dimensional feature maps. The two-dimensional feature map is processed using a convolutional neural network algorithm to construct a nonlinear mapping model between weather patterns and surface irradiance. The theoretical power generation of each photovoltaic power station is calculated using the nonlinear mapping model to generate a power prediction benchmark dataset.
7. The method as described in claim 6, characterized in that, The step of processing the two-dimensional feature map using a convolutional neural network algorithm to construct a nonlinear mapping model between weather patterns and surface irradiance includes: The two-dimensional feature map is input into a pre-trained convolutional neural network; The spatial features of the two-dimensional feature map are extracted using the convolutional neural network. A nonlinear mapping relationship between weather patterns and surface irradiance is constructed based on the aforementioned spatial characteristics; The theoretical irradiance value of each photovoltaic power station is calculated using the aforementioned nonlinear mapping relationship; Based on the theoretical irradiance value and the conversion efficiency of the photovoltaic power station, the theoretical power generation of each photovoltaic power station is generated.
8. The method as described in claim 1, characterized in that, The process of generating a spatial association topology dataset and a dynamic weight dataset for a power plant cluster based on the power prediction benchmark dataset and the geographical location of the photovoltaic power plants includes: Obtain the geographical coordinates of each photovoltaic power station; Based on the aforementioned power prediction benchmark dataset, calculate the Pearson correlation coefficient matrix among the photovoltaic power plants; If the Pearson correlation coefficient is greater than a preset correlation coefficient threshold, the weight of the associated edge is calculated according to the distance-weighted formula, which is: ； in As weight, Let be the distance between any two photovoltaic power stations. This is the preset smoothing factor; Based on the calculated associated edge weights, a spatial association topology dataset and a dynamic weight dataset for the power plant cluster are generated.
9. The method as described in claim 8, characterized in that, The calculation of the associated edge weights according to the distance-weighted formula includes: Obtain the geographical coordinates of each photovoltaic power station; Calculate the distance between any two photovoltaic power stations ; Based on the distance between any two photovoltaic power stations and preset smoothing factor The weights of the associated edges between power stations are calculated using a distance-weighted formula. If the Pearson correlation coefficient is greater than the preset correlation coefficient threshold, the corresponding associated edge weight is retained; Generate a spatial association topology dataset of power plant clusters based on the retained associated edge weights; A dynamic weighted dataset is generated based on the spatial association topology dataset of the power plant group.
10. The method as described in claim 1, characterized in that, The process involves calculating a power ramp-up risk index based on the spatial association topology dataset, dynamic weight dataset, and extreme weather label set of the power plant group, and generating a large-scale photovoltaic power ramp-up prediction and early warning dataset under extreme weather conditions based on multi-dimensional observation information, including: Extract the weights of the associated edges from the spatial association topology dataset and the dynamic weight dataset of the power plant group; The power fluctuation value is determined based on the extreme weather tag set; The power ramp-up risk index is calculated using the weighted average method. If the risk index is greater than the preset risk threshold and the duration of fluctuation in the irradiance time series is greater than the preset time threshold, a power ramp-up warning signal is generated. Based on the power ramp-up warning signal, a large-scale photovoltaic power generation power ramp-up prediction and warning dataset under extreme weather conditions is generated, based on multi-dimensional observation information.