Method for estimating and predicting distributed photovoltaic power based on spatiotemporal mapping area irradiance

Abstract

The application discloses a kind of based on space-time mapping area irradiance estimation and distributed photovoltaic power prediction method, including obtaining the irradiance and distributed photovoltaic power station power data of selected power station;Radiation intensity and distributed photovoltaic power station power output correlation function model is constructed, the irradiance of selected power station is obtained;Extraterrestrial irradiance and ground irradiance mapping relationship model is constructed, weather mode is divided into sunny day and cloudy day, respectively, sunny day model and cloudy day model are constructed, and the irradiance estimation value under different modes is obtained;The cross-correlation coefficient of irradiance sequence is calculated, and the associated power station is selected;Using extraterrestrial irradiance and ground irradiance mapping relationship model, the irradiance estimation value of selected associated power station position is calculated;Radiation intensity and distributed photovoltaic power station power output correlation function model constructed is trained, and the power estimation value is calculated;Space-time double-scale network is constructed, and photovoltaic power prediction in future set time is aimed at;The prediction accuracy of the application is improved, and practical applicability is enhanced.

Description

Technical Field

[0001] This invention belongs to the field of electrical automation technology, specifically relating to a method for estimating regional irradiance and predicting distributed photovoltaic power based on spatiotemporal mapping. Background Technology

[0002] Currently, while large-scale distributed photovoltaic grid connection has alleviated the relative power shortage in some remote areas, it has also brought many adverse effects on the safe and stable operation of the power system.

[0003] Net load forecasting plays a crucial role in maintaining power system stability. Numerous studies have shown that forecasting accuracy can be improved when power consumption and power output in the power grid are decomposed and forecasted separately. However, currently, distributed photovoltaic (PV) systems do not require dedicated power forecasting systems, and due to their small individual capacity, dispersed installation, and low investment, they are not equipped with dedicated meteorological or cloud imagery observation facilities. Consequently, distributed PV power forecasting generally relies on its own power data and does not meet the conditions for directly applying centralized PV power forecasting methods.

[0004] In summary, current methods for predicting the power of distributed photovoltaic systems based on irradiance estimation of spatiotemporal mapping regions suffer from low accuracy and limited practical applicability. Summary of the Invention

[0005] The purpose of this invention is to provide a method for estimating regional irradiance and predicting distributed photovoltaic power based on spatiotemporal mapping, which improves prediction accuracy and enhances practical applicability.

[0006] The method for estimating regional irradiance and predicting distributed photovoltaic power based on spatiotemporal mapping provided by this invention includes the following steps:

[0007] S1. Obtain the irradiance and power data of the selected power station and the distributed photovoltaic power station;

[0008] S2. Using the data obtained in step S1, construct a mapping model between irradiance and the power output of a distributed photovoltaic power station, and obtain the irradiance of the selected power station;

[0009] S3. Construct a mapping model between extraterrestrial irradiance and surface irradiance in the selected area. At the same time, divide the weather patterns into sunny and cloudy days. Construct a mapping model for sunny days and a mapping model for cloudy days according to the divided weather patterns. Obtain the estimated irradiance values ​​of the set locations under different patterns through the mapping model.

[0010] S4. Calculate the cross-correlation coefficient of the irradiance sequence, and select the associated power plants within the set area that meet the set conditions;

[0011] S5. Using the mapping relationship model between extraterrestrial irradiance and surface irradiance constructed in step S3, calculate the estimated irradiance value at the selected associated power station location;

[0012] S6. Using the data obtained in step S1, train the irradiance-distributed photovoltaic power station power output mapping model constructed in step S2, and calculate the power estimate;

[0013] S7. Using the power plant selected in step S4, the irradiance estimate calculated in step S5, and the power estimate calculated in step S6, a spatiotemporal dual-scale network is constructed to predict the photovoltaic power within a set future time period.

[0014] Step S2, which involves using the data obtained in step S1 to construct a mapping model between irradiance and the power output of a distributed photovoltaic power station, and obtaining the irradiance of a selected power station, specifically includes:

[0015] The following formula expresses the relationship between irradiance and the power output of a distributed photovoltaic power station:

[0016]

[0017] in, This represents an estimate of the power generated by a given photovoltaic power plant; T represents the number of time steps in the data. Indicates time The vector represents the observed irradiance; α represents the vector representing the tilt angle of the photovoltaic module; β represents the vector containing the azimuth angle of the photovoltaic module; g represents the vector containing the geographical coordinates of the power station, including latitude, longitude, and altitude. Indicates time The vector of temperature observed at point P; nom A vector representing the nominal power of a photovoltaic module;

[0018] Different irradiance values ​​obtained by different functions can return the same power output. The following steps are used to process them:

[0019] (1) Estimate panel orientation from the measured power of a given photovoltaic power plant;

[0020] (2) Construct a function f using the calculated direction and solve for it;

[0021] The optimized value of irradiance is calculated using the following formula:

[0022]

[0023] Where IRRopt represents the optimized value of irradiance; P represents the observed value of photovoltaic power;

[0024] The above formula can be solved using one or more photovoltaic power values, as shown below:

[0025]

[0026] Where, n pv Indicates the total number of photovoltaic power signals; This represents the observed values ​​after normalization. This represents the power estimate after normalization.

[0027] Step S3 involves constructing a mapping model between extraterrestrial irradiance and surface irradiance in a selected area, while simultaneously classifying weather patterns into sunny and cloudy days. A sunny day mapping model and a cloudy day mapping model are constructed according to the classified weather patterns. Irradiance estimates for a given location under different weather patterns are obtained through these mapping models. Specifically, this includes:

[0028] The irradiance values ​​retrieved within a specified time period are divided into sunny or cloudy days; the division method uses five criteria: average value; maximum value; line length of the irradiance versus time curve; standard deviation of the rate of change of irradiance; and maximum difference between the irradiance change and the clear sky time series.

[0029] Construct a mapping model between extraterrestrial irradiance and surface irradiance in a selected area; select extraterrestrial irradiance, ambient temperature, solar altitude angle, and solar azimuth angle as inputs, and surface irradiance as output; and construct a clear sky mapping model and a cloudy sky mapping model according to the divided weather patterns using a BP neural network.

[0030] The external irradiance at the actual power station location is calculated using the following formula:

[0031] G = kG0 sinh = kG0cosθ z =kG0(sinδsinφ+cosδcosφcosω)

[0032] Where k represents the Earth-Sun distance correction factor; G0 represents the solar constant; θ z δ represents the zenith angle; φ represents the solar declination; φ represents the local latitude; ω represents the solar hour angle.

[0033] The calculated extraterrestrial irradiance is mapped to the estimated irradiance of the actual site, and the resulting extraterrestrial irradiance and surface irradiance mapping relationship model is trained and divided into a clear sky mapping relationship model and a cloudy sky mapping relationship model.

[0034] Step S4, which involves calculating the cross-correlation coefficient of the irradiance sequence and selecting associated power plants within a defined region that meet defined conditions, specifically includes:

[0035] Selecting associated power plants that meet the set conditions through spatiotemporal correlation analysis of irradiance:

[0036] The setting conditions include selecting a number of power plants with the highest correlation coefficients according to a set number;

[0037] An icosahedral triangular mesh is generated on a unit sphere. The irradiance sequence characteristics and spatiotemporal correlations among simulated photovoltaic power plants under different weather conditions are analyzed. The Pearson correlation coefficient between irradiance sequences is calculated to identify associated power plants that meet set conditions. The calculation includes recording irradiance values ​​sequentially within a set time period, defining the recorded values ​​as irradiance time series data. Cross-correlation coefficients are used to analyze the correlation of irradiance at different locations within the region and the lag between irradiance sequences.

[0038] Step S5, which uses the mapping model between extraterrestrial irradiance and surface irradiance constructed in step S3, calculates the estimated irradiance value for a given associated power station location, specifically including:

[0039] Using the constructed mapping model of external irradiance and surface irradiance, the estimated value of surface irradiance at a given location is mapped by calculating the external irradiance at the location of a given associated power station and the trained neural network model.

[0040] Step S6, which involves using the data obtained in step S1 to train the irradiance-distributed photovoltaic power station power output mapping model constructed in step S2 and calculating the power estimate, specifically includes:

[0041] The model is constructed using surface irradiance, ambient temperature, solar altitude angle, solar azimuth angle, and historical power as inputs and actual power as output. A BP neural network is selected.

[0042] Step S7 involves constructing a spatiotemporal dual-scale network using the power plant selected in step S4, the irradiance estimate calculated in step S5, and the power estimate calculated in step S6. This network is used to predict photovoltaic power within a specified future timeframe. Specifically, this includes:

[0043] (7-1) Construct an undirected graph:

[0044] Using the distributed photovoltaic power station selected in step S4, an undirected graph consisting of multiple nodes is constructed, represented by the following formula:

[0045] G t =(P t ,ε,W)

[0046] Among them, G t The graph representing time t; P t Let represent the set of nodes, which is the power set of each distributed power station at time t; ε represents the set of edges between nodes; W represents the weight of the adjacency matrix.

[0047] (7-2) Convolution extracts spatial features of data:

[0048] Spatial features of the data are extracted by graph convolution through Chebyshev polynomial approximation. The Chebyshev graph convolution formula is shown below:

[0049]

[0050] Where Θ represents an arbitrary parameter, which is initialized and adjusted using backpropagation; G represents the graph structure; x represents the input feature; Θ(L) represents the convolution kernel; K represents the graph convolution kernel size; θ k Represents the polynomial coefficients; This represents the polynomial expansion approximation of the Laplacian;

[0051] (7-3) Temporal characteristics of data extraction by gated convolution:

[0052] Gated convolutions are used to capture temporal features in the data; the convolution width is set to K. t Used to extract K t-1 Temporal characteristics;

[0053] For each node in graph G, the temporal convolution continuously fills the search for the K of the input element. t The neighborhood allows the time series length to be shortened by K each time. t-1 ;

[0054] Given that the length of the input to each node is M, and C is... i A temporal convolution sequence with channels is equivalent to: Y is mapped to a single output.

[0055] The above processing procedure can be represented by the following formula:

[0056]

[0057] Where Γ represents the convolution kernel; P represents the first half of the input of the causal convolution; ⊙ represents the Hadamard product; σ(·) represents the input P that controls the current state and is related to the composition structure and dynamic variance of the time series; Q represents the second half of the output of the causal convolution.

[0058] (7-4) Time series of spatiotemporal convolutional block joint processing graph structure:

[0059] A spatiotemporal convolutional block joint processing method is used to process time series graph structures, fusing spatial and temporal features;

[0060] The input and output of the spatiotemporal convolution block are set as three-dimensional tensors;

[0061] The input p of the l-th block is calculated using the following formula. l Output p l+1 :

[0062]

[0063] in, Represents the lower-time kernel of the l-th block; ReLU(·) represents the true streamlined unit function; Θ l This represents the kernel of graph convolution; This represents the upper temporal kernel of the l-th block;

[0064] The loss function for spatiotemporal graph convolution is calculated using the following formula:

[0065]

[0066] in, W represents the predicted value. θ p represents the training parameters of the model. t+1 p represents the true value. j This represents the output mapping of the spacetime block, j = t+1, ..., t-M+1;

[0067] (7-5) Predict the photovoltaic power output within a specified future timeframe:

[0068] Based on the given irradiance and power data of the associated power stations obtained above, the spatiotemporal graph convolutional neural network constructed above is trained, and each distributed power station is used as a node. The distance between nodes is calculated using latitude and longitude, and the photovoltaic power within a set time period in the future is predicted.

[0069] This invention provides a method for estimating regional irradiance and predicting distributed photovoltaic power based on spatiotemporal mapping. Using an unsupervised approach, it estimates irradiance using power generation measurements from a single distributed photovoltaic power station and establishes a mapping model based on external irradiance to obtain regional-level high spatiotemporal resolution irradiance. The retrieved irradiance is not only used in irradiance / PV power prediction but also facilitates the separation of PV power generation from electrical load, improving the accuracy of net load prediction in low-voltage power grids. By mapping the power data of a given PV power station, a spatiotemporal dual-scale network is established based on deep learning to deeply explore the spatiotemporal correlations among distributed power station clusters, thereby improving the accuracy of power generation prediction. The prediction accuracy and practical applicability of this invention are enhanced. Attached Figure Description

[0070] Figure 1 This is a schematic diagram of the method flow of the present invention.

[0071] Figure 2 This is a schematic diagram of the cross-correlation coefficients between the photovoltaic power plant ground irradiance sequences selected in the method of the present invention. Detailed Implementation

[0072] like Figure 1 The diagram shown illustrates the method flow of this invention: The method for estimating regional irradiance and predicting distributed photovoltaic power based on spatiotemporal mapping provided by this invention includes the following steps:

[0073] S1. Obtain the irradiance and power data of the selected power station and the distributed photovoltaic power station;

[0074] S2. Using the data obtained in step S1, construct a correlation function model between irradiance and the power output of a distributed photovoltaic power station to obtain the irradiance of the selected power station; specifically including:

[0075] The following formula expresses the relationship between irradiance and the power output of a distributed photovoltaic power station:

[0076]

[0077] in, This represents an estimate of the power generated by a given photovoltaic power plant; T represents the number of time steps in the data. Indicates time The vector of observed irradiance; α represents the vector containing the tilt of the module; β represents the vector containing the azimuth angle of the photovoltaic module; g represents the vector containing the geographical coordinates of the power station, including latitude, longitude, and altitude; Indicates time The vector of temperature observed at point P; nom A vector representing the nominal power of a photovoltaic module;

[0078] Different irradiance values ​​obtained by different functions can return the same power output. The following steps are used to process them:

[0079] (1) Estimate panel orientation from the measured power of a given photovoltaic power plant;

[0080] (2) Construct a function f using the calculated direction and solve for it;

[0081] The optimized value of irradiance is calculated using the following formula:

[0082]

[0083] Where IRRopt represents the optimized value of irradiance; P represents the observed value of photovoltaic power;

[0084] The above formula can be solved using one or more photovoltaic power values, as shown below:

[0085]

[0086] Where, n pvIndicates the total number of photovoltaic power signals; This represents the observed values ​​after normalization. This represents the estimated power value after normalization.

[0087] S3. Construct a mapping model between extraterrestrial irradiance and surface irradiance for the selected area, while classifying the weather patterns as sunny and cloudy. Construct separate mapping models for sunny and cloudy weather according to the defined weather patterns, and obtain irradiance estimates for the selected locations under different patterns using these mapping models. Specifically, this includes:

[0088] Construct a mapping model between extraterrestrial irradiance and surface irradiance in a selected area; select extraterrestrial irradiance, ambient temperature, solar altitude angle, and solar azimuth angle as inputs, and surface irradiance as output; and construct a clear sky mapping model and a cloudy sky mapping model according to the divided weather patterns using a BP neural network.

[0089] The external irradiance at the actual power station location is calculated using the following formula:

[0090] G = kG0 sinh = kG0cosθ z =kG0(sinδsinφ+cosδcosφcosω)

[0091] Where k represents the Earth-Sun distance correction factor; G0 represents the solar constant, which is selected as a fixed value of 1386 W / m in the method of this invention. 2 ;θ z To represent the zenith angle, θ is selected in the method of this invention. z =90° / -h; δ represents the solar declination; φ represents the local latitude; ω represents the solar hour angle;

[0092] The calculated extraterrestrial irradiance is mapped to the estimated irradiance of the actual site, and the resulting extraterrestrial irradiance and surface irradiance mapping relationship model is trained and divided into a clear sky mapping relationship model and a cloudy sky mapping relationship model.

[0093] S4. Calculate the cross-correlation coefficient of the irradiance sequence, and select associated power plants within a set area that meet the set conditions; specifically including:

[0094] Selecting associated power plants that meet the set conditions through spatiotemporal correlation analysis of irradiance:

[0095] The setting conditions include selecting a number of power plants with the highest correlation coefficients according to a set number; in the method of this invention, the number of power plants selected is 10.

[0096] An icosahedral triangular mesh is generated on a unit sphere. The irradiance sequence characteristics and spatiotemporal correlations among simulated photovoltaic power plants under different weather conditions are analyzed. The Pearson correlation coefficient between irradiance sequences is calculated to identify associated power plants that meet set conditions. The calculation includes recording irradiance values ​​sequentially within a set time period, defining the recorded values ​​as irradiance time series data. Cross-correlation coefficients are used to analyze the correlation of irradiance at different locations within the region and the lag between irradiance sequences.

[0097] like Figure 2 The diagram illustrates the cross-correlation coefficients between the irradiance sequences of the selected photovoltaic power plants in the method of this invention. In this method, three photovoltaic power plants are selected to analyze the irradiance sequence characteristics and spatiotemporal correlations under different weather conditions during cloudy weather. The power plants are designated as A, B, and C. Figure 2 It can be seen that the correlation coefficient between the data of location A and location B is close to 1, indicating that the similarity between the series is high and the series have a strong ability to express each other; the cross-correlation coefficient between the irradiance of location A and location C reaches its maximum value after a delay of about 15 minutes, indicating that the two time series are most correlated at this time.

[0098] S5. Using the mapping model between extraterrestrial irradiance and surface irradiance constructed in step S3, calculate the estimated irradiance value for the selected associated power station location; specifically including:

[0099] Using the constructed mapping model of external irradiance and surface irradiance, the estimated value of surface irradiance at a given location is mapped by calculating the external irradiance at the location of a given associated power station and the trained neural network model.

[0100] S6. Using the data obtained in step S1, train the irradiance-distributed photovoltaic power station power output mapping model constructed in step S2, and calculate the power estimate; specifically including:

[0101] The model is constructed using surface irradiance, ambient temperature, solar altitude angle, solar azimuth angle, and historical power as inputs and actual power as output. A BP neural network is selected.

[0102] S7. Using the power plant selected in step S4, the irradiance estimate calculated in step S5, and the power estimate calculated in step S6, a spatiotemporal dual-scale network is constructed to predict photovoltaic power within a set future time period; specifically including:

[0103] (7-1) Construct an undirected graph:

[0104] Using the distributed photovoltaic power station selected in step S4, an undirected graph consisting of multiple nodes is constructed, represented by the following formula:

[0105] G t =(P t ,ε,W)

[0106] Among them, G t The graph representing time t; P t Let represent the set of nodes, which is the power set of each distributed power station at time t; ε represents the set of edges between nodes; W represents the weight of the adjacency matrix.

[0107] (7-2) Convolution extracts spatial features of data:

[0108] Spatial features of the data are extracted by graph convolution through Chebyshev polynomial approximation. The Chebyshev graph convolution formula is shown below:

[0109]

[0110] Where Θ represents an arbitrary parameter, which is initialized and adjusted using backpropagation; G represents the graph structure; x represents the input feature; Θ(L) represents the convolution kernel; K represents the graph convolution kernel size; θ k Represents the polynomial coefficients; This represents the polynomial expansion approximation of the Laplacian;

[0111] (7-3) Temporal characteristics of data extraction by gated convolution:

[0112] Gated convolutions are used to capture temporal features in the data; the convolution width is set to K. t Used to extract K t-1 Temporal characteristics;

[0113] For each node in graph G, the temporal convolution continuously fills the search for the K of the input element. t The neighborhood allows the time series length to be shortened by K each time. t-1 ;

[0114] Given that the length of the input to each node is M, and C is... i A temporal convolution sequence with channels is equivalent to: Y is mapped to a single output.

[0115] The above processing procedure can be represented by the following formula:

[0116]

[0117] Where Γ represents the convolution kernel; P represents the first half of the input of the causal convolution; ⊙ represents the Hadamard product; σ(·) represents the input P that controls the current state and is related to the composition structure and dynamic variance of the time series; Q represents the second half of the output of the causal convolution.

[0118] (7-4) Time series of spatiotemporal convolutional block joint processing graph structure:

[0119] A spatiotemporal convolutional block joint processing method is used to process time series graph structures, fusing spatial and temporal features;

[0120] The input and output of the spatiotemporal convolution block are set as three-dimensional tensors;

[0121] The input p of the l-th block is calculated using the following formula. l Output p l+1 :

[0122]

[0123] in, Represents the lower-time kernel of the l-th block; ReLU(·) represents the true streamlined unit function; Θ l This represents the kernel of graph convolution; This represents the upper temporal kernel of the l-th block;

[0124] The loss function for spatiotemporal graph convolution is calculated using the following formula:

[0125]

[0126] in, W represents the predicted value. θ p represents the training parameters of the model. t+1 p represents the true value. j This represents the output mapping of the spacetime block, j = t+1, ..., t-M+1;

[0127] (7-5) Predict the photovoltaic power output within a specified future timeframe:

[0128] Based on the given irradiance and power data of the associated power stations obtained above, the spatiotemporal graph convolutional neural network constructed above is trained, and each distributed power station is used as a node. The distance between nodes is calculated using latitude and longitude, and the photovoltaic power within a set time period in the future is predicted.

Claims

1. A method for estimating regional irradiance and predicting distributed photovoltaic power based on spatiotemporal mapping, comprising the following steps: S1. Obtain the irradiance and power data of the selected power station and the distributed photovoltaic power station; S2. Using the data obtained in step S1, construct a mapping model between irradiance and the power output of distributed photovoltaic power stations, and estimate the irradiance using the power generation measurement of a single distributed photovoltaic power station; S3. Construct a mapping model between extraterrestrial irradiance and surface irradiance in the selected area. At the same time, divide the weather patterns into sunny and cloudy days. Construct a mapping model for sunny days and a mapping model for cloudy days according to the divided weather patterns. Obtain the estimated irradiance values ​​of the set locations under different patterns through the mapping model. S4. Calculate the cross-correlation coefficient of the irradiance sequence and select the associated power plants within the selected area that meet the set conditions; S5. Using the mapping relationship model between extraterrestrial irradiance and surface irradiance constructed in step S3, calculate the estimated irradiance values ​​for the locations of associated power plants that meet the set conditions; S6. Using the data obtained in step S1, train the irradiance-distributed photovoltaic power station power output mapping model constructed in step S2, and calculate the power estimate of the associated power station selected in step S4. S7. Using the associated power plants selected in step S4, the irradiance estimate calculated in step S5, and the power estimate calculated in step S6, a spatiotemporal dual-scale network is constructed to predict the photovoltaic power within a set future time period.

2. The method for estimating regional irradiance and predicting distributed photovoltaic power based on spatiotemporal mapping as described in claim 1, characterized in that... Step S2 involves using the data obtained in step S1 to construct a mapping model between irradiance and the power output of a distributed photovoltaic power station, and estimating irradiance using the power generation measurement of a single distributed photovoltaic power station. Specifically, this includes: The following formula expresses the relationship between irradiance and the power output of a distributed photovoltaic power station: in, This represents an estimate of the power generated by a given photovoltaic power plant. Indicates the number of time steps in the data; Indicates time The vector of irradiance observed at a location; A vector representing the tilt angle of a photovoltaic module; This represents a vector containing the azimuth angle of the photovoltaic module; A vector representing the geographic coordinates of the power station, including latitude, longitude, and altitude; Indicates time The vector of the observed temperature at that location; A vector representing the nominal power of a photovoltaic module; Different irradiance value functions can return the same power output. The following steps are used to process them: (1) Estimating panel orientation from measured power of a given photovoltaic power plant; (2) Construct a function using the calculated direction. Solve for it; The optimized value of irradiance is calculated using the following formula: in, This represents the optimized value of irradiance. Represents observed photovoltaic power; The above formula can be solved using one or more photovoltaic power values, as shown below: in, Indicates the total number of photovoltaic power signals; This represents the observed values ​​after normalization. This represents the estimated power value after normalization.

3. The method for estimating regional irradiance and predicting distributed photovoltaic power based on spatiotemporal mapping as described in claim 2, characterized in that... Step S3 involves constructing a mapping model between extraterrestrial irradiance and surface irradiance in a selected area, while simultaneously classifying weather patterns into sunny and cloudy days. A sunny day mapping model and a cloudy day mapping model are constructed according to the classified weather patterns. Irradiance estimates for a given location under different weather patterns are obtained through these mapping models. Specifically, this includes: The irradiance values ​​retrieved within a specified time period are divided into sunny or cloudy days; the division method uses five criteria: average value; maximum value; line length of the irradiance versus time curve; standard deviation of the rate of change of irradiance; and maximum difference between the irradiance change and the clear sky time series. Construct a mapping model between extraterrestrial irradiance and surface irradiance in a selected area; select extraterrestrial irradiance, ambient temperature, solar altitude angle, and solar azimuth angle as inputs, and surface irradiance as output; and construct a clear sky mapping model and a cloudy sky mapping model according to the divided weather patterns using a BP neural network. The external irradiance at the actual power station location is calculated using the following formula: in, This represents the correction factor for the Earth-Sun distance; Represents the solar constant; Indicates the zenith angle; Indicates the solar declination; Indicates the local latitude; Indicates solar hour angle; The calculated extraterrestrial irradiance is mapped to the estimated irradiance of the actual field station. The resulting mapping model between extraterrestrial irradiance and surface irradiance is trained and divided into a clear sky mapping model and a cloudy sky mapping model.

4. The method for estimating regional irradiance and predicting distributed photovoltaic power based on spatiotemporal mapping as described in claim 3, characterized in that... Step S4, which involves calculating the cross-correlation coefficient of the irradiance sequence and selecting associated power plants within a defined region that meet defined conditions, specifically includes: Selecting associated power plants that meet the set conditions through spatiotemporal correlation analysis of irradiance: The setting conditions include selecting a number of power stations with the largest cross-correlation coefficients according to a set number; An icosahedral triangular mesh is generated on a unit sphere. The irradiance sequence characteristics and spatiotemporal correlations among simulated photovoltaic power plants under different weather conditions are analyzed. The Pearson correlation coefficient between irradiance sequences is calculated to identify associated power plants that meet set conditions. The calculation includes recording irradiance values ​​sequentially within a set time period, defining the recorded values ​​as irradiance time series data. Cross-correlation coefficients are used to analyze the correlation degree of irradiance at different locations within the region and the lag between irradiance sequences.

5. The method for estimating regional irradiance and predicting distributed photovoltaic power based on spatiotemporal mapping as described in claim 4, characterized in that... Step S5, which uses the mapping model between extraterrestrial irradiance and surface irradiance constructed in step S3, calculates the estimated irradiance values ​​for the locations of associated power plants that meet the set conditions. Specifically, this includes: Using the constructed mapping model of external irradiance and surface irradiance, the estimated value of surface irradiance at the location of the associated power station that meets the set conditions is mapped by calculating the external irradiance at the location of the associated power station that meets the set conditions and the trained neural network model.

6. The method for estimating regional irradiance and predicting distributed photovoltaic power based on spatiotemporal mapping as described in claim 5, characterized in that... Step S6 involves using the data obtained in step S1 to train the irradiance-distributed photovoltaic power station power output mapping model constructed in step S2, and calculating the power estimate of the associated power station selected in step S4. Specifically, this includes: The model is constructed using surface irradiance, ambient temperature, solar altitude angle, solar azimuth angle, and historical power as inputs and actual power as output. A backpropagation neural network is selected.

7. The method for estimating regional irradiance and predicting distributed photovoltaic power based on spatiotemporal mapping as described in claim 6, characterized in that... Step S7 involves constructing a spatiotemporal dual-scale network using the associated power plants selected in step S4, the irradiance estimate calculated in step S5, and the power estimate calculated in step S6. This network is used to predict photovoltaic power within a specified future timeframe. Specifically, this includes: (7-1) Construct an undirected graph: Using the distributed photovoltaic power station selected in step S4, an undirected graph consisting of multiple nodes is constructed, represented by the following formula: in, express A graph of moments; Represents a set of nodes, which is The power collection of each distributed power station at any given time; This represents the set of edges between nodes; Represents the weights of the adjacency matrix; (7-2) Spatial features of data extracted by graph convolution: Spatial features of the data are extracted by graph convolution through Chebyshev polynomial approximation. The Chebyshev graph convolution formula is shown below: in, This represents any parameter that is initialized and adjusted using backpropagation of error. Representing the graph structure; Represents input features; Represents the convolution kernel; Indicates the size of the graph convolution kernel; Represents the coefficients of a polynomial; This represents the polynomial expansion approximation of the Laplacian; (7-3) Temporal characteristics of data extraction using gated convolution: Gated convolutions are used to capture temporal features in the data; the convolution width is set to... Used for extraction Temporal characteristics; Regarding the diagram At each node in the process, the temporal convolution continuously fills the search for the input element. Neighborhood, which makes the time series length shorten each time. ; The length of the input to each node is... ,have A temporal convolution sequence with channels is equivalent to: , Mapped to a single output ; The above processing procedure can be represented by the following formula: in, Represents the convolution kernel; This represents the first half of the input to a causal convolution; Represents the Hadamard product; Input indicating the control of the current state It is related to the compositional structure and dynamic variance of the time series; This indicates the second half of the output of a causal convolution. (7-4) Time series of spatiotemporal convolutional block joint processing graph structure: A spatiotemporal convolutional block joint processing method is used to process time series graph structures, fusing spatial and temporal features; The input and output of the spatiotemporal convolution block are set as three-dimensional tensors; Calculate the following formula Block input Output : in, Indicates the first The lower temporal kernel of the block; Represents a true streamlined unit function; This represents the kernel of graph convolution; Indicates the first The upper-temporal kernel of the block; The loss function for spatiotemporal graph convolution is calculated using the following formula: in, Indicates the predicted value; Represents the training parameters of the model; Represents the actual value; Represents the output mapping of the spacetime block. ; (7-5) Predict the photovoltaic power output within a specified future timeframe: Based on the irradiance and power data of the associated power stations that meet the set conditions, the spatiotemporal graph convolutional neural network is trained. Each distributed power station is used as a node, and the distance between nodes is calculated using latitude and longitude to predict the photovoltaic power within a set time period in the future.

Images