Distributed photovoltaic power prediction method and system fusing dynamic area and space-time

Abstract

The present application relates to the technical field of photovoltaic power prediction, and provides a distributed photovoltaic power prediction method and system fusing dynamic regions and space-time. The method comprises the following steps: according to historical power data and corresponding key meteorological characteristics representing power stations, a time model is used to extract time series trend characteristics, and preliminary time prediction values and time prediction errors of sub-regions of the power stations are obtained; the power stations are taken as nodes to construct a graph structure, and a space model is used to extract space correlation characteristics therein, and preliminary space prediction values and space prediction errors of the sub-regions of the power stations are obtained; the contribution degrees of the time model and the space model are dynamically adjusted by using an information entropy method, and combined prediction powers of the sub-regions of the power stations are obtained; data quality evaluation scores of the sub-regions of the power stations are determined, a scale-up weight is obtained through a proportion of the data quality evaluation scores, and the combined prediction powers of the sub-regions of the power stations are weighted and summed, and a total power of a region is predicted, so that the super-short-term prediction accuracy and robustness are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic power prediction technology, and in particular to a distributed photovoltaic power prediction method and system that integrates dynamic regional and spatiotemporal factors. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Distributed photovoltaic (PV) power generation has become an important direction for renewable energy development due to its advantages such as flexible deployment and environmental friendliness. However, the output of distributed PV is affected by multiple factors such as meteorological conditions and spatial location, exhibiting significant spatiotemporal fluctuations, which poses a huge challenge to grid dispatch and efficient consumption. Existing PV power prediction methods have many limitations: traditional single models cannot simultaneously and accurately capture both temporal variation patterns and spatial correlation characteristics, resulting in insufficient prediction accuracy and robustness; spatial correlation modeling often relies on static partitioning or simple neighborhood matching, failing to dynamically analyze the deep consistency between power stations; and the data redundancy problem of massive distributed power stations increases the risk of model overfitting, further restricting prediction performance. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a distributed photovoltaic power prediction method and system that integrates dynamic regional and spatiotemporal factors, which can effectively improve the accuracy and robustness of ultra-short-term prediction.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: The first aspect of the present invention provides a distributed photovoltaic power prediction method that integrates dynamic regional and spatiotemporal factors.

[0006] In one or more embodiments, a distributed photovoltaic power prediction method integrating dynamic regional and spatiotemporal factors is provided, including: Based on the correlation with photovoltaic power, key meteorological characteristics are screened; based on the inherent consistency of power plant output, power plant sub-regions are dynamically divided and representative power plants are screened. Based on the historical power data of representative power plants and their corresponding key meteorological characteristics, time series trend features are extracted using a time model to obtain preliminary time prediction values ​​and time prediction errors for each power plant sub-region. By constructing a graph structure with representative power stations as nodes, and using a spatial model to extract spatial correlation features, preliminary spatial prediction values ​​and spatial prediction errors for each power station sub-region are obtained. Based on the time prediction error and the spatial prediction error, the contribution of the time model and the spatial model is dynamically adjusted using the information entropy method, and used as the weight of the preliminary time prediction value and the preliminary spatial prediction value to obtain the combined predicted power of each power station sub-region. Based on the data integrity, data similarity, and rated capacity of distributed photovoltaic power stations, the data quality evaluation score of each sub-region of the power station is determined. Then, the upscaling weight is obtained by the proportion of the data quality evaluation score. Finally, the combined predicted power of each sub-region of the power station is weighted and summed to predict the total power of the region.

[0007] As one implementation method, the process of dynamically adjusting the contributions of the time model and the space model using the information entropy method is as follows: Based on the time prediction error and the spatial prediction error, calculate the first... The first model for the first proportion of each sample ; Based on sample proportion Calculate the first The entropy of the relative error of each model prediction is used to calculate the coefficient of variation of the relative error of the j-th prediction model by subtracting 1 from the entropy. Based on the above coefficient of variation, calculate the first... The contribution of each prediction model; among which... , representing the time model and the space model respectively; the sum of the contributions of the time model and the space model is 1.

[0008] As one implementation method, the expression for the data quality evaluation score of each power station sub-region is: The expression for the data quality evaluation score of each power station sub-region is as follows:

[0009] In the formula, sub-region of the power station The data quality evaluation score, i.e., the total capacity of the power plants; the total capacity of the internal power plants; The Spearman coefficients and cosine similarities for representative power plants and sub-regions are respectively; For power station The capacity is M; M is the number of power stations in the area. For power station Number of days of absence due to lack of effort.

[0010] As one implementation method, the expression for the upscaling weight is:

[0011] in, sub-region of the power station The data quality evaluation score, i.e., the total capacity of the power plant; sub-region of the power station Upscaling weights; This represents the number of sub-regions of the power station.

[0012] As one implementation method, based on the proximity propagation clustering algorithm, the initial power station area and its corresponding power station cluster label are divided by combining the geographical location of the power station and the similarity of its historical power output. The power output prediction model is used to calculate the prediction error of each power station in each power station area on the previous day, and then the power station area and its corresponding power station cluster label are updated in a rolling manner. The power station with the strongest correlation with the total power of each power station area is selected as the representative power station.

[0013] As one implementation method, Spearman correlation coefficient is used to screen key meteorological features based on their correlation with photovoltaic power.

[0014] As one implementation method, the time model uses the SSA-XGBoost model; the spatial model uses the GCNN model.

[0015] A second aspect of the present invention provides a distributed photovoltaic power prediction system that integrates dynamic regional and spatiotemporal factors.

[0016] In one or more embodiments, a distributed photovoltaic power prediction system integrating dynamic regional and spatiotemporal factors includes: The feature and representative power plant screening module is used to screen key meteorological features based on their correlation with photovoltaic power; and to dynamically divide power plant sub-regions and screen representative power plants based on the inherent consistency of power plant output. The time feature extraction and prediction module is used to extract time series trend features based on the historical power data of representative power plants and their corresponding key meteorological features, and to obtain the preliminary time prediction values ​​and time prediction errors of each power plant sub-region. The spatial feature extraction and prediction module is used to construct a graph structure with representative power stations as nodes, extract spatial correlation features using a spatial model, and obtain preliminary spatial prediction values ​​and spatial prediction errors for each power station sub-region. The combined predicted power calculation module is used to dynamically adjust the contribution of the time model and the spatial model based on the time prediction error and the spatial prediction error using the information entropy method, and use them as the weights of the preliminary time prediction value and the preliminary spatial prediction value to obtain the combined predicted power of each power station sub-region. The regional total power prediction module is used to determine the data quality evaluation score of each power station sub-region based on the data integrity, data similarity and the rated capacity of distributed photovoltaic power. Then, the upscaling weight is obtained by the proportion of the data quality evaluation score, and the combined predicted power of each power station sub-region is weighted and summed to predict the regional total power.

[0017] A third aspect of the present invention provides a computer-readable storage medium.

[0018] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the distributed photovoltaic power prediction method integrating dynamic regions and spatiotemporal factors as described above.

[0019] A fourth aspect of the present invention provides an electronic device.

[0020] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the distributed photovoltaic power prediction method integrating dynamic regional and spatiotemporal factors as described above.

[0021] Compared with the prior art, the beneficial effects of the present invention are: This invention utilizes historical power data of representative power plants and their corresponding key meteorological characteristics. It extracts time-series trend features using a time model to obtain preliminary time prediction values ​​and time prediction errors for each power plant sub-region. It then extracts spatial correlation features using a spatial model to obtain preliminary spatial prediction values ​​and spatial prediction errors for each power plant sub-region. Furthermore, it employs the information entropy method to weight and fuse the prediction results from the time and spatial dimensions, thereby reducing the relative error of the model prediction and improving its stability. Finally, it obtains upscaling weights based on the proportion of data quality evaluation scores, achieving regional total power prediction that adapts to dynamic changes in spatiotemporal characteristics and significantly improves the accuracy and robustness of distributed photovoltaic power prediction. Attached Figure Description

[0022] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0023] Figure 1 This is a flowchart of the distributed photovoltaic power prediction method that integrates dynamic region and spatiotemporal aspects according to an embodiment of the present invention; Figure 2 This is a flowchart of the dynamic clustering process according to an embodiment of the present invention; Figure 3 This is a flowchart of the SSA-XGBoost time series photovoltaic power generation prediction process according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the GCNN structure according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of the distributed photovoltaic power prediction system that integrates dynamic region and spatiotemporal aspects according to an embodiment of the present invention; Figure 6 This is a schematic diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0024] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0025] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0026] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0027] Figure 1 A schematic diagram of the distributed photovoltaic power prediction method integrating dynamic region and spatiotemporal factors according to an embodiment of the present invention is provided. Figure 1 The distributed photovoltaic power prediction method integrating dynamic region and spatiotemporal factors in this embodiment may include the following steps S101 to S105.

[0028] The specific implementation process of steps S101 to S105 is as follows: Step S101: Based on the correlation with photovoltaic power, key meteorological characteristics are screened; based on the inherent consistency of power plant output, sub-regions of power plants are dynamically divided and representative power plants are screened.

[0029] Distributed photovoltaic power is affected by meteorological factors such as total irradiance and temperature. This step improves the quality of model input by integrating multi-source data and filtering features.

[0030] Data input includes historical power data and meteorological data such as total irradiance, temperature, air pressure, and relative humidity of distributed photovoltaic power stations in a certain region, as well as the geographical location information of the power stations; Data processing: Outliers for air pressure were detected using the Z-Score method, and outliers for temperature were detected using the quartile method. Outliers were filled with the previous valid value. Missing data were completed using linear interpolation. Data standardization: Z-score standardization is used to transform the data into a distribution with mean = 0 and standard deviation = 1, which speeds up model convergence. The formula is as follows: (1) In the formula, The power data is normalized. This is actual photovoltaic power data; The actual photovoltaic power at time j on day i; Let n be the standard deviation of photovoltaic power on day i; n is the total number of days. Feature selection: Spearman correlation coefficient was used to screen meteorological features that are strongly correlated with photovoltaic power. The results showed that total irradiance and temperature are the key influencing factors. Spearman correlation coefficient was also used to analyze the output correlation between power plants, providing a basis for subsequent graph structure construction.

[0031] Traditional static region partitioning is ill-suited to the short-term, time-varying characteristics of photovoltaic power output. Dynamic clustering is used to partition sub-regions and select representative power plants to reduce the number of modeling objects. The dynamic clustering process is as follows: Figure 2 As shown, by leveraging the feature extraction capabilities of deep learning models and analyzing the power prediction performance of each power station, the evolution patterns of power generation characteristics can be captured to measure the deep consistency patterns of power stations within a cluster, thereby enabling more effective dynamic partitioning of regional clusters.

[0032] The specific process of dynamically dividing power station sub-regions and selecting representative power stations is as follows: Based on the proximity propagation clustering algorithm, the initial power plant areas and their corresponding power plant cluster labels are divided by combining the geographical location of the power plant with the similarity of historical power output. The power output prediction model is used to calculate the prediction error of each power plant in each power plant area on the previous day. Then, the power plant areas and their corresponding power plant cluster labels are updated in a rolling manner, and the power plant with the strongest correlation with the total power of each power plant area is selected as the representative power plant.

[0033] Specifically, cluster center initialization: Based on the nearest neighbor propagation clustering algorithm, the initial cluster center is selected by combining the power station's geographical location and historical power output similarity; Intrinsic consistency representation: Based on the strong time-series modeling capability of LSTM models, K LSTM prediction models are trained using historical power data from K cluster centers, and the prediction error of each power station on the previous day is calculated. The smaller the error, the more consistent the power output characteristics of the power station and the cluster core are, as shown in the following formula: (2) In the formula, , Let T be the measured and predicted values ​​of model k at time i on day d-1 of power plant n, and T be the prediction time length. Regional division update: Update power plant cluster labels based on prediction errors. (3) In the formula, This is the cluster label for the power station on day d.

[0034] The average power output of each cluster of power plants is used to update the cluster center, and this process is repeated to achieve dynamic partitioning. Representative power plant selection: The power plant with the strongest correlation to the total power of the sub-region and the highest data integrity is selected as the representative power plant.

[0035] This embodiment uses Spearman correlation coefficient in two ways, which not only achieves accurate screening of meteorological features, but also ensures strong correlation with reference power stations, laying the foundation for two-dimensional modeling.

[0036] Step S102: Based on the historical power data of representative power plants and their corresponding key meteorological characteristics, extract time series trend features using a time model to obtain preliminary time prediction values ​​and time prediction errors for each power plant sub-region.

[0037] In this embodiment, the time model adopts the SSA-XGBoost model.

[0038] It should be noted that in other embodiments, the time model may also be implemented using other existing models, which will not be described in detail here.

[0039] The XGBoost model possesses strong nonlinear fitting capabilities, but its hyperparameter sensitivity affects prediction stability. SSA (Self-Optimized Algorithm) is employed to optimize XGBoost hyperparameters during model training to obtain the optimal solution and improve the accuracy of time series feature extraction. The overall optimization framework combining SSA and XGBoost is as follows: Figure 3 As shown.

[0040] SSA-XGBoost model inputs: historical power data representing the power plant and filtered meteorological characteristics such as total irradiance and temperature; SSA Hyperparameter Optimization: SSA is a swarm intelligence optimization algorithm that simulates the foraging behavior of sparrows. It includes discoverers and joiners, who work together to perform global and local searches, enabling it to quickly converge to the optimal solution and avoid getting trapped in local optima. It uses the "discoverer-joiner-watcher" mechanism of SSA for global optimization, avoiding local optima. Examples of optimized parameters include: number of iterations, tree depth, and learning rate. The optimization process is as follows: (4) In the formula: This represents the maximum number of iterations. is a random number in the range [0,1]; Q is a random number following a normal distribution; L is a 1×d matrix; Let represent the j-th dimension position information of the i-th sparrow in the t-th iteration; C is a constant in [0.5, 1], representing the safety value; The value is a random number in the range [0, 1], representing the warning value; i only iterates through the discoverers. Through this strategy, SSA can continuously adjust the particle positions during the search process, thereby improving the efficiency of hyperparameter search and avoiding the model getting trapped in local optima.

[0041] Then, SSA evaluates the hyperparameters using the fitness function and updates the position of the best individual, with the following update rules: (5) In the formula: The worst individual in t iterations; This is the current position of the best discoverer; n represents the number of participants. After an individual's location is updated, SSA trains the model using XGBoost and calculates the loss function to optimize the hyperparameter configuration.

[0042] Time Feature Output: Extract time series trend features using the optimized XGBoost model and output preliminary time prediction values. With prediction error .

[0043] This embodiment utilizes the SSA-XGBoost model to globally optimize hyperparameters, making it easier to capture temporal patterns than the traditional XGBoost, and significantly improving the accuracy of time series prediction.

[0044] Step S103: Construct a graph structure by using representative power stations as nodes, extract spatial correlation features using a spatial model, and obtain preliminary spatial prediction values ​​and spatial prediction errors for each power station sub-region.

[0045] In this embodiment, the spatial model uses the GCNN model.

[0046] It should be noted that in other embodiments, the spatial model may also be implemented using other existing models, which will not be described in detail here.

[0047] GCNN can effectively analyze the correlation information of power plants in non-Euclidean space by extracting spatial correlation features between power plants through graph structure construction. A schematic diagram of the GCNN structure is shown below. Figure 4 As shown.

[0048] Graph structure construction: Representative power plants are used as nodes, and the node characteristics are the historical power data of the first 96 sampling points. (6) In the formula, The power output of power station n during the period from t-96 to t is represented by the Spearman correlation coefficient between power stations. Spatial feature extraction: Spatial correlation features are extracted using the GCNN operator, as shown in the following formula: (7) In the formula, For spatial management and association information between power stations, Let W be the node degree matrix, W be the graph adjacency matrix, I be the identity matrix, and θ be the learnable parameter matrix. Spatial feature output: Outputs spatial feature vectors and preliminary spatial prediction values. Prediction error .

[0049] This embodiment utilizes the GCNN model combined with a dynamic graph structure, eliminating the need for predefined clustering, adapting to dynamic changes in spatial correlation, and making spatial feature extraction more flexible.

[0050] Step S104: Based on the time prediction error and spatial prediction error, dynamically adjust the contribution of the time model and the spatial model using the information entropy method, and use them as the weights of the preliminary time prediction value and the preliminary spatial prediction value to obtain the combined predicted power of each power station sub-region.

[0051] This embodiment utilizes multiple models working together to extract temporal and spatial features, providing comprehensive support for accurate prediction.

[0052] In spatiotemporal combined prediction, the prediction results of time and space dimensions each exhibit varying degrees of uncertainty. Information entropy, as an important indicator for measuring uncertainty, can be used to dynamically adjust the contribution of each dimension in the fusion process, thereby more effectively controlling uncertainty. This paper proposes a feature fusion method based on information entropy, which weights and fuses the prediction results of the time and space dimensions to reduce the relative error of the model prediction and improve its stability.

[0053] Based on the information entropy method, the weights of each model are calculated according to the errors of the individual temporal and spatial prediction models. These two weights are then combined to obtain the power of the combined spatiotemporal information prediction. The specific prediction process is as follows: Based on the time prediction error and the spatial prediction error, calculate the first... The first model for the first proportion of each sample ; (8) In the formula: Let be the relative prediction error of the j-th prediction model for the i-th sample; Based on sample proportion Calculate the first The entropy of the relative error predicted by each model Calculate the coefficient of variation of the relative prediction error of the j-th prediction model by subtracting 1 from the entropy. ; (9) (10) Based on the above coefficient of variation, calculate the first... The contribution of each prediction model; among which... , representing the time model and the space model respectively; the sum of the contributions of the time model and the space model is 1.

[0054] (11) In the formula, The contribution of the j-th model is represented by its weight. (12); The predicted power results from temporal and spatial analysis are fused according to the calculated weights to obtain the final prediction result: (13) in, sub-region of the power station Combined predicted power; and These are the contributions of the time model and the spatial model, respectively, i.e., their weights; sub-region of the power station Preliminary time forecast values; sub-region of the power station Preliminary spatial predictions.

[0055] The embodiments of the present invention utilize the above-mentioned information entropy weighted fusion strategy to dynamically adjust the contribution of the spatiotemporal model, avoiding the limitations of fixed weights and improving the robustness of prediction under complex weather conditions.

[0056] Step S105: Based on the data integrity, data similarity and rated capacity of distributed photovoltaic power stations, determine the data quality evaluation score of each sub-region of the power station, then obtain the upscaling weight through the proportion of data quality evaluation scores, and then perform a weighted summation of the combined predicted power of each sub-region of the power station to predict the total power of the region.

[0057] Accurate power forecasting relies on high-quality data; the higher the completeness of sub-regional data, the greater its usability in predicting the total regional power. Therefore, a power data quality assessment method is proposed to measure the usability of sub-regional prediction results and support the determination of upscaling weights. Furthermore, the more similar the power output of a representative power plant to that of a sub-region, the more representative its prediction results are of the sub-regional total power forecast. Based on factors such as power plant data completeness, data similarity, and the rated capacity of distributed photovoltaic power, a sub-regional data quality evaluation score is determined. The calculation formula is shown in the figure: (14) In the formula, sub-region of the power station The data quality evaluation score, i.e., the total capacity of the power plants; the total capacity of the internal power plants; The Spearman coefficients and cosine similarities for representative power plants and sub-regions are respectively; For power station The capacity is M; M is the number of power stations in the area. For power station Number of days of absence due to lack of effort.

[0058] The expression for upscaling weights is: (15) in, sub-region of the power station The data quality evaluation score, i.e., the total capacity of the power plant; sub-region of the power station Upscaling weights; This represents the number of sub-regions of the power station.

[0059] Based on the proportional relationship between the rated capacity of representative power plants and the total rated capacity of the sub-region, the power prediction results of representative power plants in the sub-region are proportionally amplified to obtain the predicted power of the sub-region. Then, the total photovoltaic output of the region is predicted by calculating the upscaling weight of the sub-regions.

[0060] (16) In the formula: This represents the total power of the region.

[0061] This invention first uses Spearman correlation coefficient to screen key meteorological features and identify strongly correlated reference power stations; second, it uses the SSA-XGBoost model to extract time series features and GCNN to analyze the spatial correlation between power stations; finally, it uses information entropy weighted fusion of multi-dimensional prediction results and upscaling to calculate the total regional power, effectively improving the accuracy and robustness of ultra-short-term prediction.

[0062] This embodiment uses historical power data of representative power plants and their corresponding key meteorological characteristics. It extracts time series trend features using a time model to obtain preliminary time prediction values ​​and time prediction errors for each power plant sub-region. It then uses a spatial model to extract spatial correlation features, obtaining preliminary spatial prediction values ​​and spatial prediction errors for each power plant sub-region. Next, it uses the information entropy method to weight and fuse the prediction results of the time and spatial dimensions, achieving the effect of reducing the relative error of the model prediction and improving the stability of the model prediction. Finally, it obtains the upscaling weight through the proportion of data quality evaluation scores, realizing the prediction of regional total power, adapting to dynamic changes in spatiotemporal characteristics, and significantly improving the accuracy and robustness of distributed photovoltaic power prediction.

[0063] like Figure 5As shown, the distributed photovoltaic power prediction system integrating dynamic regional and spatiotemporal characteristics provided in this embodiment of the invention can be implemented in software. The distributed photovoltaic power prediction system integrating dynamic regional and spatiotemporal characteristics includes the following software modules: feature and representative power plant screening module 501, time feature extraction and prediction module 502, spatial feature extraction and prediction module 503, combined predicted power calculation module 504, and regional total power prediction module 505.

[0064] The functions of each software module in the distributed photovoltaic power prediction system that integrates dynamic regional and spatiotemporal data are described below: The feature and representative power station screening module 501 is used to screen key meteorological features based on their correlation with photovoltaic power; and to dynamically divide power station sub-regions and screen representative power stations based on the inherent consistency of power station output.

[0065] Distributed photovoltaic power is affected by meteorological factors such as total irradiance and temperature. This step improves the quality of model input by integrating multi-source data and filtering features.

[0066] Data input includes historical power data and meteorological data such as total irradiance, temperature, air pressure, and relative humidity of distributed photovoltaic power stations in a certain region, as well as the geographical location information of the power stations; Data processing: Outliers for air pressure were detected using the Z-Score method, and outliers for temperature were detected using the quartile method. Outliers were filled with the previous valid value. Missing data were completed using linear interpolation. Data standardization: Z-score standardization is used to transform the data into a distribution with mean = 0 and standard deviation = 1, which speeds up model convergence. The formula is as follows:

[0067] In the formula, The power data is normalized. This is actual photovoltaic power data; The actual photovoltaic power at time j on day i; Let n be the standard deviation of photovoltaic power on day i; n is the total number of days. Feature selection: Spearman correlation coefficient was used to screen meteorological features that are strongly correlated with photovoltaic power. The results showed that total irradiance and temperature are the key influencing factors. Spearman correlation coefficient was also used to analyze the output correlation between power plants, providing a basis for subsequent graph structure construction.

[0068] Traditional static region partitioning is ill-suited to the short-term, time-varying characteristics of photovoltaic power output. Dynamic clustering is used to partition sub-regions and select representative power plants to reduce the number of modeling objects. The dynamic clustering process is as follows: Figure 2As shown, by leveraging the feature extraction capabilities of deep learning models and analyzing the power prediction performance of each power station, the evolution patterns of power generation characteristics can be captured to measure the deep consistency patterns of power stations within a cluster, thereby enabling more effective dynamic partitioning of regional clusters.

[0069] The specific process of dynamically dividing power station sub-regions and selecting representative power stations is as follows: Based on the proximity propagation clustering algorithm, the initial power plant areas and their corresponding power plant cluster labels are divided by combining the geographical location of the power plant with the similarity of historical power output. The power output prediction model is used to calculate the prediction error of each power plant in each power plant area on the previous day. Then, the power plant areas and their corresponding power plant cluster labels are updated in a rolling manner, and the power plant with the strongest correlation with the total power of each power plant area is selected as the representative power plant.

[0070] Specifically, cluster center initialization: Based on the nearest neighbor propagation clustering algorithm, the initial cluster center is selected by combining the power station's geographical location and historical power output similarity; Intrinsic consistency representation: Based on the strong time-series modeling capability of LSTM models, K LSTM prediction models are trained using historical power data from K cluster centers, and the prediction error of each power station on the previous day is calculated. The smaller the error, the more consistent the power output characteristics of the power station and the cluster core are, as shown in the following formula:

[0071] In the formula, , Let T be the measured and predicted values ​​of model k at time i on day d-1 of power plant n, and T be the prediction time length. Regional division update: Update power plant cluster labels based on prediction errors.

[0072] In the formula, This is the cluster label for the power station on day d.

[0073] The average power output of each cluster of power plants is used to update the cluster center, and this process is repeated to achieve dynamic partitioning. Representative power plant selection: The power plant with the strongest correlation to the total power of the sub-region and the highest data integrity is selected as the representative power plant.

[0074] This embodiment uses Spearman correlation coefficient in two ways, which not only achieves accurate screening of meteorological features, but also ensures strong correlation with reference power stations, laying the foundation for two-dimensional modeling.

[0075] The time feature extraction and prediction module 502 is used to extract time series trend features based on the historical power data of representative power plants and their corresponding key meteorological features, and to obtain the preliminary time prediction values ​​and time prediction errors of each power plant sub-region.

[0076] In this embodiment, the time model adopts the SSA-XGBoost model.

[0077] It should be noted that in other embodiments, the time model may also be implemented using other existing models, which will not be described in detail here.

[0078] The XGBoost model possesses strong nonlinear fitting capabilities, but its hyperparameter sensitivity affects prediction stability. SSA (Self-Optimized Algorithm) is employed to optimize XGBoost hyperparameters during model training to obtain the optimal solution and improve the accuracy of time series feature extraction. The overall optimization framework combining SSA and XGBoost is as follows: Figure 3 As shown.

[0079] SSA-XGBoost model inputs: historical power data representing the power plant and filtered meteorological characteristics such as total irradiance and temperature; SSA Hyperparameter Optimization: SSA is a swarm intelligence optimization algorithm that simulates the foraging behavior of sparrows. It includes discoverers and joiners, who work together to perform global and local searches, enabling it to quickly converge to the optimal solution and avoid getting trapped in local optima. It uses the "discoverer-joiner-watcher" mechanism of SSA for global optimization, avoiding local optima. Examples of optimized parameters include: number of iterations, tree depth, and learning rate. The optimization process is as follows:

[0080] In the formula: This represents the maximum number of iterations. is a random number in the range [0,1]; Q is a random number following a normal distribution; L is a 1×d matrix; Let represent the j-th dimension position information of the i-th sparrow in the t-th iteration; C is a constant in [0.5, 1], representing the safety value; The value is a random number in the range [0, 1], representing the warning value; i only iterates through the discoverers. Through this strategy, SSA can continuously adjust the particle positions during the search process, thereby improving the efficiency of hyperparameter search and avoiding the model getting trapped in local optima.

[0081] Then, SSA evaluates the hyperparameters using the fitness function and updates the position of the best individual, with the following update rules:

[0082] In the formula: The worst individual in t iterations; This is the current position of the best discoverer; n represents the number of participants. After an individual's location is updated, SSA trains the model using XGBoost and calculates the loss function to optimize the hyperparameter configuration.

[0083] Time Feature Output: Extract time series trend features using the optimized XGBoost model and output preliminary time prediction values. With prediction error .

[0084] This embodiment utilizes the SSA-XGBoost model to globally optimize hyperparameters, making it easier to capture temporal patterns than the traditional XGBoost, and significantly improving the accuracy of time series prediction.

[0085] The spatial feature extraction and prediction module 503 is used to construct a graph structure by taking representative power stations as nodes, extracting spatial correlation features from the spatial model, and obtaining preliminary spatial prediction values ​​and spatial prediction errors for each power station sub-region.

[0086] In this embodiment, the spatial model uses the GCNN model.

[0087] It should be noted that in other embodiments, the spatial model may also be implemented using other existing models, which will not be described in detail here.

[0088] GCNN can effectively analyze the correlation information of power plants in non-Euclidean space by extracting spatial correlation features between power plants through graph structure construction. A schematic diagram of the GCNN structure is shown below. Figure 4 As shown.

[0089] Graph structure construction: Representative power plants are used as nodes, and the node characteristics are the historical power data of the first 96 sampling points.

[0090] In the formula, The power output of power station n during the period from t-96 to t is represented by the Spearman correlation coefficient between power stations. Spatial feature extraction: Spatial correlation features are extracted using the GCNN operator, as shown in the following formula:

[0091] In the formula, For spatial management and association information between power stations, Let W be the node degree matrix, W be the graph adjacency matrix, I be the identity matrix, and θ be the learnable parameter matrix. Spatial feature output: Outputs spatial feature vectors and preliminary spatial prediction values. Prediction error .

[0092] This embodiment utilizes the GCNN model combined with a dynamic graph structure, eliminating the need for predefined clustering, adapting to dynamic changes in spatial correlation, and making spatial feature extraction more flexible.

[0093] The combined predicted power calculation module 504 is used to dynamically adjust the contribution of the time model and the spatial model according to the time prediction error and the spatial prediction error using the information entropy method, and use them as the weights of the preliminary time prediction value and the preliminary spatial prediction value to obtain the combined predicted power of each power station sub-region.

[0094] This embodiment utilizes multiple models working together to extract temporal and spatial features, providing comprehensive support for accurate prediction.

[0095] In spatiotemporal combined prediction, the prediction results of time and space dimensions each exhibit varying degrees of uncertainty. Information entropy, as an important indicator for measuring uncertainty, can be used to dynamically adjust the contribution of each dimension in the fusion process, thereby more effectively controlling uncertainty. This paper proposes a feature fusion method based on information entropy, which weights and fuses the prediction results of the time and space dimensions to reduce the relative error of the model prediction and improve its stability.

[0096] Based on the information entropy method, the weights of each model are calculated according to the errors of the individual temporal and spatial prediction models. These two weights are then combined to obtain the power of the combined spatiotemporal information prediction. The specific prediction process is as follows: Based on the time prediction error and the spatial prediction error, calculate the first... The first model for the first proportion of each sample ; In the formula: Let be the relative prediction error of the j-th prediction model for the i-th sample; Based on sample proportion Calculate the first The entropy of the relative error predicted by each model Calculate the coefficient of variation of the relative prediction error of the j-th prediction model by subtracting 1 from the entropy. ; ; ; Based on the above coefficient of variation, calculate the first... The contribution of each prediction model; among which... , representing the time model and the space model respectively; the sum of the contributions of the time model and the space model is 1.

[0097] ; In the formula, The contribution of the j-th model is represented by its weight. ; The predicted power results from temporal and spatial analysis are fused according to the calculated weights to obtain the final prediction result: ; in, sub-region of the power station Combined predicted power; and These are the contributions of the time model and the spatial model, respectively, i.e., their weights; sub-region of the power station Preliminary time forecast values; sub-region of the power station Preliminary spatial predictions.

[0098] The embodiments of the present invention utilize the above-mentioned information entropy weighted fusion strategy to dynamically adjust the contribution of the spatiotemporal model, avoiding the limitations of fixed weights and improving the robustness of prediction under complex weather conditions.

[0099] The regional total power prediction module 505 is used to determine the data quality evaluation score of each power station sub-region based on the data integrity, data similarity and the rated capacity of distributed photovoltaic power. Then, it obtains the upscaling weight through the proportion of the data quality evaluation score, and then performs a weighted summation of the combined predicted power of each power station sub-region to predict the regional total power.

[0100] Accurate power forecasting relies on high-quality data; the higher the completeness of sub-regional data, the greater its usability in predicting the total regional power. Therefore, a power data quality assessment method is proposed to measure the usability of sub-regional prediction results and support the determination of upscaling weights. Furthermore, the more similar the power output of a representative power plant to that of a sub-region, the more representative its prediction results are of the sub-regional total power forecast. Based on factors such as power plant data completeness, data similarity, and the rated capacity of distributed photovoltaic power, a sub-regional data quality evaluation score is determined. The calculation formula is shown in the figure: ; In the formula, sub-region of the power station The data quality evaluation score, i.e., the total capacity of the power plants; the total capacity of the internal power plants; The Spearman coefficients and cosine similarities for representative power plants and sub-regions are respectively; For power station The capacity is M; M is the number of power stations in the area. For power station Number of days of absence due to lack of effort.

[0101] The expression for upscaling weights is: ; in, sub-region of the power station The data quality evaluation score, i.e., the total capacity of the power plant; sub-region of the power station Upscaling weights; This represents the number of sub-regions of the power station.

[0102] Based on the proportional relationship between the rated capacity of representative power plants and the total rated capacity of the sub-region, the power prediction results of representative power plants in the sub-region are proportionally amplified to obtain the predicted power of the sub-region. Then, the total photovoltaic output of the region is predicted by calculating the upscaling weight of the sub-regions.

[0103] In the formula: This represents the total power of the region.

[0104] It should be noted that each module in the distributed photovoltaic power prediction system integrating dynamic region and spatiotemporal data in this embodiment corresponds one-to-one with each step in the distributed photovoltaic power prediction method integrating dynamic region and spatiotemporal data in the above embodiment, and their specific implementation processes are the same, so they will not be repeated here.

[0105] In this embodiment, firstly, key meteorological features are screened and strongly correlated reference power stations are identified using Spearman correlation coefficient; secondly, time series features are extracted using the SSA-XGBoost model, and spatial correlations between power stations are analyzed using GCNN; finally, the total regional power is obtained by weighted fusion of multi-dimensional prediction results using information entropy and then upscaled, effectively improving the accuracy and robustness of ultra-short-term prediction.

[0106] This embodiment uses historical power data of representative power plants and their corresponding key meteorological characteristics. It extracts time series trend features using a time model to obtain preliminary time prediction values ​​and time prediction errors for each power plant sub-region. It then uses a spatial model to extract spatial correlation features, obtaining preliminary spatial prediction values ​​and spatial prediction errors for each power plant sub-region. Next, it uses the information entropy method to weight and fuse the prediction results of the time and spatial dimensions, achieving the effect of reducing the relative error of the model prediction and improving the stability of the model prediction. Finally, it obtains the upscaling weight through the proportion of data quality evaluation scores, realizing the prediction of regional total power, adapting to dynamic changes in spatiotemporal characteristics, and significantly improving the accuracy and robustness of distributed photovoltaic power prediction.

[0107] The structure of the electronic device according to an embodiment of the present invention will be described in detail below. Figure 6 This is a schematic diagram of the composition structure of an electronic device provided in an embodiment of the present invention. It can be understood that... Figure 6 The diagram shows only an exemplary structure of the electronic device, not the entire structure. Some or all of the structures shown may be implemented as needed.

[0108] The electronic device provided in this embodiment of the invention includes: at least one processor 601, a memory 602, a user interface 603, and at least one network interface 604. Various components in a distributed photovoltaic power prediction system integrating dynamic regional and spatiotemporal data are coupled together via a bus system 605. It is understood that the bus system 605 is used to realize communication between these components. In addition to a data bus, the bus system 605 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in… Figure 6 The general designated all buses as Bus System 605.

[0109] The user interface 603 may include a monitor, keyboard, mouse, trackball, click wheel, buttons, touchpad, or touch screen.

[0110] It is understood that memory 602 can be volatile memory or non-volatile memory, or both. In this embodiment of the invention, memory 602 is capable of storing data to support the operation of the terminal. Examples of this data include any computer programs used to operate on the terminal, such as operating systems and applications. The operating system includes various system programs, such as framework layers, core library layers, driver layers, etc., used to implement various basic services and handle hardware-based tasks. Applications can include various applications.

[0111] In some embodiments, the distributed photovoltaic power prediction system integrating dynamic region and spatiotemporal dimensions provided in this invention can be implemented using a combination of hardware and software. For example, the distributed photovoltaic power prediction system integrating dynamic region and spatiotemporal dimensions provided in this invention can be a processor in the form of a hardware decoding processor, programmed to execute the distributed photovoltaic power prediction method integrating dynamic region and spatiotemporal dimensions provided in this invention. For instance, the processor in the form of a hardware decoding processor can employ one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components.

[0112] As an example, processor 601 can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., wherein the general-purpose processor can be a microprocessor or any conventional processor, etc.

[0113] As an example of the hardware implementation of the distributed photovoltaic power prediction system integrating dynamic region and spatiotemporal dimensions provided in this embodiment of the invention, the device provided in this embodiment of the invention can be directly executed by a processor 601 in the form of a hardware decoding processor. For example, it can be executed by one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components to implement the distributed photovoltaic power prediction method integrating dynamic region and spatiotemporal dimensions provided in this embodiment of the invention.

[0114] The memory 602 in this embodiment of the invention is used to store various types of data to support the operation of a distributed photovoltaic power prediction system that integrates dynamic regional and spatiotemporal data, or to store data for execution. Figure 1 The program code for the method shown. Examples of this data include: any executable instructions for operation on a distributed photovoltaic power prediction system that integrates dynamic regional and spatiotemporal dimensions, such as executable instructions that can be included in the executable instructions to implement the distributed photovoltaic power prediction method integrating dynamic regional and spatiotemporal dimensions of the present invention.

[0115] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program including functions for executing... Figure 1 The program code for the method shown. In such an embodiment, the computer program can be downloaded and installed from a network via a communication component, and / or installed from a removable medium. When the computer program is executed by the central processing unit, it performs the various functions defined in the apparatus of this application.

[0116] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0117] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A distributed photovoltaic power prediction method integrating dynamic regional and spatiotemporal factors, characterized in that, include: Based on the correlation with photovoltaic power, key meteorological characteristics are screened; based on the inherent consistency of power plant output, power plant sub-regions are dynamically divided and representative power plants are screened. Based on the historical power data of representative power plants and their corresponding key meteorological characteristics, time series trend features are extracted using a time model to obtain preliminary time prediction values ​​and time prediction errors for each power plant sub-region. By constructing a graph structure with representative power stations as nodes, and using a spatial model to extract spatial correlation features, preliminary spatial prediction values ​​and spatial prediction errors for each power station sub-region are obtained. Based on the time prediction error and the spatial prediction error, the contribution of the time model and the spatial model is dynamically adjusted using the information entropy method, and used as the weight of the preliminary time prediction value and the preliminary spatial prediction value to obtain the combined predicted power of each power station sub-region. Based on the data integrity, data similarity, and rated capacity of distributed photovoltaic power stations, the data quality evaluation score of each sub-region of the power station is determined. Then, the upscaling weight is obtained by the proportion of the data quality evaluation score. Finally, the combined predicted power of each sub-region of the power station is weighted and summed to predict the total power of the region.

2. The distributed photovoltaic power prediction method integrating dynamic regional and spatiotemporal factors as described in claim 1, characterized in that, The process of dynamically adjusting the contributions of the time model and the space model using the information entropy method is as follows: Based on the time prediction error and the spatial prediction error, calculate the first... The first model for the first proportion of each sample ; Based on sample proportion Calculate the first The entropy of the relative error of each model prediction is used to calculate the coefficient of variation of the relative error of the j-th prediction model by subtracting 1 from the entropy. Based on the above coefficient of variation, calculate the first... The contribution of each prediction model; among which... , representing the time model and the space model respectively; the sum of the contributions of the time model and the space model is 1.

3. The distributed photovoltaic power prediction method integrating dynamic regional and spatiotemporal factors as described in claim 1, characterized in that, The expression for the data quality evaluation score of each power station sub-region is as follows: In the formula, sub-region of the power station The data quality evaluation score, i.e., the total capacity of the power plants; the total capacity of the internal power plants; The Spearman coefficients and cosine similarities for representative power plants and sub-regions are respectively; For power station The capacity is M; M is the number of power stations in the area. For power station Number of days of absence due to lack of effort.

4. The distributed photovoltaic power prediction method integrating dynamic regional and spatiotemporal factors as described in claim 1, characterized in that, The expression for upscaling weights is: in, sub-region of the power station The data quality evaluation score, i.e., the total capacity of the power plant; sub-region of the power station Upscaling weights; This represents the number of sub-regions of the power station.

5. The distributed photovoltaic power prediction method integrating dynamic regional and spatiotemporal factors as described in claim 1, characterized in that, Based on the proximity propagation clustering algorithm, the initial power plant areas and their corresponding power plant cluster labels are divided by combining the geographical location of the power plant with the similarity of historical power output. The power output prediction model is used to calculate the prediction error of each power plant in each power plant area on the previous day. Then, the power plant areas and their corresponding power plant cluster labels are updated in a rolling manner, and the power plant with the strongest correlation with the total power of each power plant area is selected as the representative power plant.

6. The distributed photovoltaic power prediction method integrating dynamic regional and spatiotemporal factors as described in claim 1, characterized in that, Based on the correlation with photovoltaic power, the Spearman correlation coefficient was used to screen key meteorological features.

7. The distributed photovoltaic power prediction method integrating dynamic regional and spatiotemporal factors as described in claim 1, characterized in that, The temporal model uses the SSA-XGBoost model; the spatial model uses the GCNN model.

8. A distributed photovoltaic power prediction system integrating dynamic regional and spatiotemporal characteristics, characterized in that, The distributed photovoltaic power prediction method based on the integration of dynamic regional and spatiotemporal factors as described in any one of claims 1-7 includes: The feature and representative power plant screening module is used to screen key meteorological features based on their correlation with photovoltaic power; and to dynamically divide power plant sub-regions and screen representative power plants based on the inherent consistency of power plant output. The time feature extraction and prediction module is used to extract time series trend features based on the historical power data of representative power plants and their corresponding key meteorological features, and to obtain the preliminary time prediction values ​​and time prediction errors of each power plant sub-region. The spatial feature extraction and prediction module is used to construct a graph structure with representative power stations as nodes, extract spatial correlation features using a spatial model, and obtain preliminary spatial prediction values ​​and spatial prediction errors for each power station sub-region. The combined predicted power calculation module is used to dynamically adjust the contribution of the time model and the spatial model based on the time prediction error and the spatial prediction error using the information entropy method, and use them as the weights of the preliminary time prediction value and the preliminary spatial prediction value to obtain the combined predicted power of each power station sub-region. The regional total power prediction module is used to determine the data quality evaluation score of each power station sub-region based on the data integrity, data similarity and the rated capacity of distributed photovoltaic power. Then, the upscaling weight is obtained by the proportion of the data quality evaluation score, and the combined predicted power of each power station sub-region is weighted and summed to predict the regional total power.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the distributed photovoltaic power prediction method that integrates dynamic regional and spatiotemporal factors as described in any one of claims 1-7.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the distributed photovoltaic power prediction method that integrates dynamic regional and spatiotemporal factors as described in any one of claims 1-7.

Images