Regional distributed power supply power upscaling prediction method and system

Abstract

The invention discloses an upscaling prediction method and system for the power of a regional distributed power supply, and belongs to the technical field of energy prediction. Actual wind power plant data are acquired, regional division is performed, and the number of sub-regions is determined; carrying out continuous grading probability scoring evaluation on the random sub-regions, obtaining an aggregation power probability prediction result of the evaluated region, gradually increasing the number of the sub-regions until the prediction result is not reduced any more, and taking the corresponding number of the sub-regions at the moment as the number of reference stations; sorting the number of the sub-regions, determining the rank corresponding to the wind power plant in each sub-region, and by determining the comprehensive rank of the wind power plants in the sub-regions, taking the wind power plant with the minimum comprehensive rank as a reference station of the sub-regions; and inputting the power generation power probability distribution parameters and the spatial correlation characteristics of the reference station into the constructed probability upscaling model with the spatial correlation for prediction, and obtaining a power supply power prediction result of the sub-region. The method can realize accurate prediction of regional new energy output.

Description

Technical Field

[0001] The present invention relates to the technical field of energy prediction, and more specifically to a method and system for predicting power upscaling of regional distributed power sources. Background Art

[0002] With the construction of new power systems, the installed capacity of distributed power sources is getting larger and larger, and its power generation output has affected the safe operation of the power grid. For regional new energy power forecasting, there will be many new energy sites in a region. Especially when some sites cannot produce forecast results, how to accurately predict the total power of new energy sites will become a difficulty in power grid dispatching. Therefore, it is necessary to finely predict the output of distributed power sources to better serve power grid dispatching.

[0003] For grid dispatching, the impact of a single distributed power source is relatively small, and accurate prediction of regional renewable energy is required. Since the geographical location of wind farms cannot truly describe the similarity of wind farm power generation, the traditional idea of ​​predicting each renewable energy source and then adding them up cannot meet the prediction needs of distributed power sources in the grid.

[0004] Due to the wide installed capacity of distributed power sources, the power grid cannot obtain the actual measurement data of distributed power sources in real time as historical measured data, and cannot accurately select reference stations, resulting in the inability to accurately predict the output of a single distributed power source. Summary of the invention

[0005] In response to the problems existing in the above-mentioned fields, the present invention proposes a method and system for upscaling prediction of regional distributed power generation power, divides the region, determines the number of benchmark stations and benchmark stations in the sub-region, selects representative new energy benchmark stations from the region, predicts the power of the benchmark stations and expands the prediction results to the regional level. This method can improve the accuracy of distributed power generation prediction.

[0006] In order to solve the above technical problems, the present invention discloses a method for predicting power upscaling of regional distributed power sources, comprising the following steps:

[0007] Obtain the actual wind farm data of regional distributed power sources and divide the regions to determine the number of sub-regions;

[0008] Conduct continuous hierarchical probability scoring assessment on random sub-areas, obtain the aggregate power probability prediction results of the assessed areas, and gradually increase the number of sub-areas until the prediction results no longer decrease. The number of sub-areas corresponding to the time when the prediction results no longer decrease is used as the number of benchmark sites.

[0009] According to the number of benchmark stations, the number of sub-regions is sorted according to the aggregated power probability prediction results and the correlation between wind farm power and regional power to determine the corresponding ranking of wind farms in each sub-region; the smaller the ranking, the higher the power correlation; according to the corresponding ranking of wind farms in each sub-region, the comprehensive ranking of wind farms in each sub-region is determined, and the wind farm with the smallest comprehensive ranking is used as the benchmark station of the sub-region;

[0010] Determine the power generation probability distribution parameters of the reference station according to the reference station in the sub-region;

[0011] By extracting the spatial correlation characteristics of the benchmark station, the probability distribution parameters and spatial correlation characteristics of the power generation of the benchmark station are input into the constructed probability upscaling model considering spatial correlation for prediction, and the power prediction results of the sub-region are obtained.

[0012] Preferably, the determining the number of sub-areas comprises the following steps:

[0013] According to the actual wind farm data of regional distributed power sources obtained, including the geographical distribution data of the stations, the region is divided by the inter-cluster distance of hierarchical clustering to obtain the wind farm data of each cluster;

[0014] According to the wind farm data of each cluster, the maximum value of the distance between the power generation of any two wind farms in the cluster is determined and used as the inter-cluster distance;

[0015] The calculation formula for the maximum distance between the power generation of any two wind farms in the cluster is:

[0016] D i,j =max{d(S A ,S B ),S A ∈C i ,S B ∈C j}

[0017] Where D i,j Represents cluster C i and C j The distance between A ,S B ) represents cluster C i and C j Any wind farm S A and S B The distance between the generating power;

[0018] The similarity of power generation of different wind farms is measured by Pearson correlation coefficient, and the distance between the output power of two wind farms is determined;

[0019] The calculation formula of the Pearson correlation coefficient is:

[0020]

[0021] Where Y A and Y B Represents wind farm S A and S B The power generation sequence of , E(·) represents the expectation, μ and σ represent the mean and standard deviation of the power sequence respectively;

[0022] The distance between the output powers of the two wind farms is:

[0023] d(S A ,S B )=1-|ρ(Y A ,Y B )|

[0024] Distance d(S A ,S B ) is smaller, indicating that the distance between the power generation of wind farms is smaller and the similarity between the powers is higher;

[0025] Among them, hierarchical clustering includes the following steps:

[0026] Step 1: Calculate the distance d (S) between any two wind farms in the region. A ,S B );

[0027] Step 2: Initialize all wind farms into clusters C1, C2, ..., C N ;

[0028] Step 3: Calculate the distance D between any two clusters;

[0029] Step 4: Merge the nodes with the smallest D i,j Two clusters C i and C j For a cluster C ij ;

[0030] Step 5: Repeat Step 3 and Step 4 until only one cluster remains;

[0031] All wind farms are clustered into M clusters through hierarchical clustering, corresponding to M sub-regions. The wind farms included in the mth sub-region are Where N m Indicates the number of wind farms in the mth sub-region.

[0032] Preferably, the method of taking the number of the sub-region corresponding to when the prediction result no longer decreases as the number of reference stations specifically includes:

[0033] Perform continuous hierarchical probability scoring evaluation on the random sub-region M to obtain the aggregate power probability prediction result of the evaluated region;

[0034] And gradually increase the number of sub-regions. When the number of sub-regions increases five times in a row and the aggregate power probability prediction results of the regions evaluated by continuous graded probability scores no longer decrease, that is:

[0035] When the number of sub-regions M=1, the corresponding regional aggregate power probability prediction result when the number of sub-regions evaluated by continuous graded probability scoring is 1 is b1; when the number of sub-regions M=2, the corresponding regional aggregate power probability prediction result when the number of sub-regions evaluated by continuous graded probability scoring is 2 is b2; when the number of sub-regions M=3, the corresponding regional aggregate power probability prediction result when the number of sub-regions evaluated by continuous graded probability scoring is 3 is b3; when the number of sub-regions is 4, the corresponding regional aggregate power probability prediction result when the number of sub-regions evaluated by continuous graded probability scoring is M=4 is b4; when the number of sub-regions M=5, the corresponding regional aggregate power probability prediction result when the number of sub-regions evaluated by continuous graded probability scoring is 5 is b5; when the number of sub-regions M=6, the corresponding regional aggregate power probability prediction result when the number of sub-regions evaluated by continuous graded probability scoring is 6 is b6, and b6≥b1;

[0036] The number of sub-regions M=6 corresponding to the regional aggregate power probability prediction result b6 is taken as the number of reference sites.

[0037] Preferably, determining the ranking corresponding to the wind farms in each sub-region specifically includes:

[0038] Sub-region C m Wind farms in m,1 ,S m,2 ,...,S m,Nm} Sort the aggregate power probability prediction results based on the continuous hierarchical probability score evaluation to obtain the ranking of the aggregate power probability prediction results based on the continuous hierarchical probability score evaluation for:

[0039]

[0040] Sub-region C m Wind farms in According to the correlation between power generation and regional power measured by Pearson correlation coefficient, the correlation ranking between power generation and regional power measured by Pearson correlation coefficient is obtained. for:

[0041]

[0042] In the formula, r1 CRPS and r1 Pearson Represent sub-regions C m The ranking of the aggregated power probability prediction results of the first wind farm based on the continuous hierarchical probability score evaluation and the correlation ranking of the Pearson correlation coefficient; the smaller the ranking, the higher the prediction accuracy or the higher the power correlation.

[0043] Preferably, the taking the smallest wind farm in the comprehensive ranking as the reference station of the sub-region specifically includes:

[0044] The ranking of the aggregate power probability prediction results based on the continuous hierarchical probability score evaluation and the correlation ranking of the Pearson correlation coefficient of the same wind farm are added together to obtain the comprehensive ranking R of the wind farms in the sub-region. m :

[0045]

[0046] Select comprehensive ranking R m The smallest wind farm in m As sub-area C m The reference station of the wind farm S m The overall ranking of m =min{r∈R m}.

[0047] Preferably, the determining of the power generation probability distribution parameters of the reference station specifically includes:

[0048] The power generation of the benchmark station is predicted by the probability prediction method of the linear spline quantile function, and the prediction result of the benchmark station is expressed as P m (y m ;Ω m ), m=1,2,...,M, where Ω m =(γ m ,β m ,ξ m ) represents the parameter of the quantile function;

[0049] For the spline quantile function with K knots, the corresponding parameters Uniquely represents the reference station S m Probability prediction results;

[0050] The corresponding reference station S m The probability distribution parameters of the power generation include the cumulative distribution function and quantile function of the power generation, which are F m (y m ) and Qm (Ω m ).

[0051] Preferably, the constructed probability upscaling model considering spatial correlation specifically includes:

[0052] The probability upscaling model transforms the input matrix Λ into the probability distribution parameter Ω of regional aggregate power through the nonlinear mapping composed of MLP and projection layer, where:

[0053] The input matrix Λ includes the parameters of the probability distribution of the power generation of the reference station [Ω1, ..., Ω M ] and the spatial correlation characteristics of the benchmark stations extracted by the Gaussian Copula function Θ G ;

[0054] Flatten the input matrix Λ into a one-dimensional vector of the lower triangular matrix L, input the lower triangular matrix L into the MLP, and output the quantile function parameters Ω=(γ, β, ξ) of the regional aggregate power, where the quantile function parameters Ω=(γ, β, ξ) are estimated by minimizing the continuous hierarchical probability scoring loss function;

[0055] The nonlinear mapping includes learning the input features through MLP, and representing the learning process of MLP as an intermediate variable h:

[0056] h=ρ MLP (Λ; θ MLP )

[0057] In the formula, ρ MLP and θ MLP They represent the nonlinear mapping and its parameters learned by MLP respectively;

[0058] The nonlinear mapping also includes a projection layer composed of an affine transformation and an activation function, which transforms the intermediate variable h into a quantile function parameter Ω of the regional aggregation power:

[0059] Ω=ρ proj (h; θ proj )

[0060] In the formula, ρ proj and θ proj They represent the nonlinear mapping of the projection layer and its parameters respectively;

[0061] In the projection layer, the activation function type depends on the constraints on the quantile function parameters;

[0062] The parameter θ is adjusted by the Adam algorithm MLP and θ proj Optimize, through the optimized parameters θ MLP and θ proj, and optimize the constructed probability upscaling model taking into account spatial correlation.

[0063] Preferably, obtaining the power prediction result of the sub-region power supply specifically includes:

[0064] According to the optimized probability upscaling model considering spatial correlation, the power prediction result of the sub-region is obtained as the probability distribution parameter of the regional aggregate power.

[0065] Preferably, a regional distributed power generation power upscaling prediction system is also included, comprising:

[0066] A regional division module is used to obtain the actual wind farm data of regional distributed power sources and divide the regions to determine the number of sub-regions;

[0067] The reference station number determination module is used to perform continuous hierarchical probability scoring evaluation on random sub-areas, obtain the aggregate power probability prediction result of the evaluated area, and gradually increase the number of sub-areas until the prediction result no longer decreases, and the number of sub-areas corresponding to the prediction result no longer decreases is used as the reference station number;

[0068] The reference station selection module is used to sort the number of sub-regions according to the aggregate power probability prediction results and the correlation between wind farm power and regional power according to the number of reference stations, and determine the corresponding ranking of wind farms in each sub-region; wherein, the smaller the ranking, the higher the power correlation; according to the ranking of wind farms in each sub-region, by determining the comprehensive ranking of wind farms in each sub-region, the wind farm with the smallest comprehensive ranking is used as the reference station of the sub-region;

[0069] The regional power prediction module is used to determine the probability distribution parameters of the power generation of the benchmark station according to the benchmark station in the sub-region; by extracting the spatial correlation characteristics of the benchmark station, the probability distribution parameters of the power generation of the benchmark station and the spatial correlation characteristics are input into the constructed probability upscaling model considering spatial correlation for prediction, so as to obtain the power prediction results of the sub-region.

[0070] Compared with the prior art, the present invention has the following beneficial effects:

[0071] The regional distributed power generation power upscaling prediction method proposed by the present invention determines the number of corresponding benchmark stations and benchmark stations in the sub-region by obtaining actual wind farm data and performing regional division, so as to select representative new energy benchmark stations from the region, predict the power of the benchmark station by constructing a probability upscaling model considering spatial correlation, and expand the prediction results to the regional level to achieve regional power upscaling prediction. Compared with the traditional method, this method can be applied to regional power prediction of large-scale new energy power generation clusters. In the process of determining the number of benchmark stations, by gradually increasing the number of sub-regions and using continuous hierarchical probability scoring to evaluate the aggregated power probability prediction results of the sub-regions, the regional power prediction accuracy is guaranteed, the number of benchmark stations is reduced as much as possible, and the model scale is simplified. When selecting the benchmark station, the number of wind farms in the sub-region is sorted according to the aggregated power probability prediction results evaluated by continuous hierarchical probability scoring and the correlation between the wind farm power and the regional power, and the corresponding ranking of the wind farms in each sub-region is determined. By determining the comprehensive ranking of the wind farms in each sub-region, the smallest wind farm in the comprehensive ranking is used as the benchmark station of the sub-region. The smaller the ranking, the higher the prediction accuracy or the higher the power correlation. The regional distributed power generation power upscaling prediction method proposed in the present invention taking into account spatial correlation can achieve accurate prediction of regional renewable energy output. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 A schematic diagram of the process flow of the regional distributed power generation power upscaling prediction method of the present invention

[0073] Figure 2 A diagram showing the architecture of a probability upscaling model considering spatial correlation provided in an embodiment of the present invention;

[0074] Figure 3 A schematic diagram of the principle and idea of ​​the scaling method provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0075] The following will be combined with the attached embodiment of the present invention Figure 1-Figure 3 , the technical solutions in the embodiments of the present invention are clearly and completely described. It should be understood that the terms described in the present invention are only used to describe specific implementation methods and are not used to limit the present invention.

[0076] Example

[0077] like Figure 1 As shown, a method for predicting power upscaling of regional distributed power sources proposed by the present invention comprises the following steps:

[0078] S1: Obtain the actual wind farm data of regional distributed power sources and divide the regions to determine the number of sub-regions;

[0079] S2: Conduct continuous hierarchical probability scoring evaluation on random sub-areas, obtain the aggregate power probability prediction results of the evaluated areas, and gradually increase the number of sub-areas until the prediction results no longer decrease. The number of sub-areas corresponding to the time when the prediction results no longer decrease is used as the number of benchmark sites.

[0080] S3: According to the number of benchmark stations, the number of sub-regions is sorted according to the aggregate power probability prediction results and the correlation between wind farm power and regional power to determine the corresponding ranking of wind farms in each sub-region; the smaller the ranking, the higher the power correlation; according to the corresponding ranking of wind farms in each sub-region, the comprehensive ranking of wind farms in each sub-region is determined, and the wind farm with the smallest comprehensive ranking is used as the benchmark station of the sub-region;

[0081] S4: Determine the probability distribution parameters of the power generation of the benchmark station according to the benchmark station in the sub-region; extract the spatial correlation characteristics of the benchmark station, input the probability distribution parameters and spatial correlation characteristics of the power generation of the benchmark station into the constructed probability upscaling model considering spatial correlation for prediction, and obtain the power prediction result of the sub-region.

[0082] Specifically, the method mainly includes 1) sub-region division based on hierarchical clustering and 2) selection of benchmark stations, where:

[0083] 1) Use Hierarchical Clustering Analysis (HCA) to realize the division of sub-regions

[0084] Hierarchical clustering requires the calculation of inter-cluster distances. Inter-cluster distances can be measured in the following ways: minimum distance, maximum distance, mean distance, class average distance, etc.

[0085] According to the actual wind farm data of regional distributed power sources, the region is divided by the inter-cluster distance of HCA to obtain the wind farm data of each cluster. The clustering result should avoid becoming a chain and make the number of wind farms in different clusters similar.

[0086] According to the wind farm data of each cluster, the maximum value of the distance between the power generation of any two wind farms in the cluster is determined and used as the inter-cluster distance; the calculation formula for the maximum value of the distance between the power generation of any two wind farms in the cluster is:

[0087] D i,j =max{d(S A ,S B ),S A ∈C i ,S B ∈C j}

[0088] Where D i,j Represents cluster C i and C j The distance between A ,S B ) represents cluster C i and C j Any wind farm S A and S B The distance between the generating power.

[0089] Since the geographical location of the wind farm cannot truly describe the similarity of the power generation of the wind farm, the present invention measures the similarity of the power generation of different wind farms by the Pearson correlation coefficient; the calculation formula of the Pearson correlation coefficient is:

[0090]

[0091] Where Y A and Y B Represents wind farm S A and S B The generated power sequence is E(·), E(·) represents the expectation, μ and σ represent the mean and standard deviation of the power sequence respectively.

[0092] Since the meanings of distance and similarity are exactly opposite, that is, the similarity is low at long distances and high at short distances, the distance between the output powers of two wind farms is determined as:

[0093] d(S A ,S B )=1-|ρ(Y A ,Y B )|

[0094] Among them, d(S A ,S B ) is smaller, the smaller the distance between the wind farm power generation is, and the higher the similarity between the power is.

[0095] The process of using hierarchical clustering to divide sub-regions includes:

[0096] Step 1: Calculate the distance d (S) between any two wind farms in the region. A ,S B );

[0097] Step 2: Initialize all wind farms into clusters C1, C2, ..., C N ;

[0098] Step 3: Calculate the distance D between any two clusters;

[0099] Step 4: Merge the nodes with the smallest Di,j Two clusters C i and C j For a cluster C ij ;

[0100] Step5: Repeat Step3 and Step4 until only one cluster remains.

[0101] All wind farms are clustered into M clusters through hierarchical clustering, corresponding to M sub-regions. The wind farms included in the mth sub-region are Where N m Indicates the number of wind farms in the mth sub-region.

[0102] 2) Selection of reference stations

[0103] Selecting a reference station includes: how many reference stations to select and how to determine the reference stations

[0104] Select the number of reference stations, that is, determine the number of reference stations, including:

[0105] The number of sub-regions is gradually increased, and the regional aggregate power probability prediction results are evaluated by continuous ranked probability score (CRPS).

[0106] Perform CRPS evaluation on the random sub-region M to obtain the aggregate power probability prediction result of the evaluated region;

[0107] And gradually increase the number of sub-areas. When the number of sub-areas increases for 5 consecutive times and the aggregated power probability prediction result of the area evaluated by CRPS no longer decreases, the number of wind farms in the sub-area corresponding to the time when the aggregated power probability prediction result no longer decreases is used as the benchmark station number.

[0108] Specifically, when the number of sub-regions M=1, the corresponding regional aggregation power probability prediction result when the number of sub-regions evaluated by continuous graded probability scoring is 1 is b1; when the number of sub-regions M=2, the corresponding regional aggregation power probability prediction result when the number of sub-regions evaluated by continuous graded probability scoring is 2 is b2; when the number of sub-regions M=3, the corresponding regional aggregation power probability prediction result when the number of sub-regions evaluated by continuous graded probability scoring is 3 is b3; when the number of sub-regions is 4, the corresponding regional aggregation power probability prediction result when the number of sub-regions evaluated by continuous graded probability scoring is M=4 is b4; when the number of sub-regions is M=5, the corresponding regional aggregation power probability prediction result when the number of sub-regions evaluated by continuous graded probability scoring is 5 is b5; when the number of sub-regions is M=6, the corresponding regional aggregation power probability prediction result when the number of sub-regions evaluated by continuous graded probability scoring is 6 is b6, and b6≥b1;

[0109] The number of sub-regions M=6 corresponding to the regional aggregate power probability prediction result b6 is taken as the number of reference sites.

[0110] This process not only ensures the accuracy of regional power prediction but also reduces the number of reference stations as much as possible and simplifies the model scale.

[0111] How to determine the reference station specifically includes:

[0112] Sub-region C m Wind farms in Sort the aggregate power probability prediction results based on the continuous hierarchical probability score evaluation to obtain the ranking of the aggregate power probability prediction results based on the continuous hierarchical probability score evaluation for:

[0113]

[0114] Sub-region C m Wind farms in According to the correlation between power generation and regional power measured by Pearson correlation coefficient, the correlation ranking between power generation and regional power measured by Pearson correlation coefficient is obtained. for:

[0115]

[0116] In the formula, r1 CRPS and r1 Pearson Represent sub-regions C m The ranking of the aggregated power probability prediction results based on the continuous hierarchical probability score evaluation and the ranking of the Pearson correlation coefficient of the first wind farm in the graph; the smaller the ranking, the higher the prediction accuracy or the higher the power correlation.

[0117] The ranking of the aggregated power probability prediction results based on the continuous hierarchical probability score evaluation and the ranking of the Pearson correlation coefficient of the same wind farm are added together to obtain the comprehensive ranking R of the wind farms in each sub-region. m :

[0118]

[0119] Select comprehensive ranking R m The smallest wind farm in m As sub-area C m The benchmark station of wind farm S m The overall ranking is r m =min{r∈R m}.

[0120] Determine the power generation probability distribution parameters of the reference station, including:

[0121] The power generation of the benchmark station is predicted by the probability prediction method of the linear spline quantile function, and the prediction result of the benchmark station is expressed as P m (y m ;Ω m ), m=1,2,...,M, where Ω m =(γ m ,β m ,ξ m ) represents the parameter of the quantile function.

[0122] For the spline quantile function with K knots, the corresponding parameters Uniquely represents the reference station S m Probability prediction results;

[0123] The corresponding reference station S m The probability distribution parameters of the power generation include the cumulative distribution function and quantile function of the power generation, which are F m (y m ) and Q m (Ω m ).

[0124] like Figure 2 As shown in FIG. 1 , the probability upscaling model considering spatial correlation constructed by the present invention converts the input matrix Λ into the probability distribution parameter Ω of the regional aggregation power through the nonlinear mapping composed of MLP and projection layer, where:

[0125] The input matrix Λ includes the parameters of the probability distribution of the power generation of the reference station [Ω1,...,Ω M] and the spatial correlation characteristics of the benchmark stations extracted by the Gaussian Copula function Θ G , set the number of parameters of the probability distribution of the power generation of the benchmark station to 2K+1.

[0126] Since the input matrix Λ is a two-dimensional matrix, it is necessary to flatten the input matrix Λ into a one-dimensional vector of the lower triangular matrix L, where the lower triangular matrix L only takes the elements of the main diagonal and below it. The lower triangular matrix L is input into the MLP, and the quantile function parameters Ω=(γ, β, ξ) of the output regional aggregate power are estimated by minimizing the continuous hierarchical probability scoring loss function.

[0127] The nonlinear mapping involves learning the input features through the MLP and representing the learning process of the MLP as an intermediate variable h:

[0128] h=ρ MLP (Λ; θ MLP )

[0129] In the formula, ρ MLP and θ MLP They represent the nonlinear mapping and its parameters learned by MLP respectively.

[0130] The nonlinear mapping also includes a projection layer composed of an affine transformation and an activation function, which transforms the intermediate variable h into the quantile function parameter Ω of the regional aggregation power:

[0131] Ω=ρ proj (h; θ proj )

[0132] In the formula, ρ proj and θ proj They represent the nonlinear mapping of the projection layer and its parameters respectively.

[0133] In the projection layer, the activation function type depends on the constraints of the quantile function parameters. For example, the quantile function parameters require that each element in the vector is positive and the sum is 1, so it is generated by the softmax activation function. Other parameters are similar. Positive numbers are generated by the softplus activation function, and non-negative numbers are generated by the ReLU activation function.

[0134] The parameter θ is adjusted by the Adam algorithm MLP and θ proj Optimize, through the optimized parameters θ MLP and θ proj , and optimize the constructed probability upscaling model taking into account spatial correlation.

[0135] According to the optimized probability upscaling model considering spatial correlation, the power forecast result of the sub-region is obtained as the probability distribution parameter of the regional aggregate power. The 50% quantile is taken from the quantile function to obtain the final deterministic forecast result.

[0136] The regional distributed power generation power upscaling prediction method proposed in the present invention taking into account spatial correlation can achieve accurate prediction of regional renewable energy output.

[0137] like Figure 3 As shown in the figure, the solution of the upscaling prediction method is to select representative new energy benchmark stations from the region through regional division, predict the power of the benchmark stations and expand the prediction results to the regional level. In essence, the process of selecting benchmark stations in the upscaling prediction method is to extract regional features and reduce dimensions, so this method can be applied to the regional power prediction problem of large-scale new energy power generation clusters.

[0138] The present invention also proposes a regional distributed power generation power upscaling prediction system, comprising:

[0139] A regional division module is used to obtain the actual wind farm data of regional distributed power sources and divide the regions to determine the number of sub-regions;

[0140] The reference station number determination module is used to perform continuous hierarchical probability scoring evaluation on random sub-areas, obtain the aggregate power probability prediction result of the evaluated area, and gradually increase the number of sub-areas until the prediction result no longer decreases, and the number of sub-areas corresponding to the prediction result no longer decreases is used as the reference station number;

[0141] The reference station selection module is used to sort the number of sub-regions according to the aggregate power probability prediction results and the correlation between wind farm power and regional power according to the number of reference stations, and determine the corresponding ranking of wind farms in each sub-region; wherein, the smaller the ranking, the higher the power correlation; according to the ranking of wind farms in each sub-region, by determining the comprehensive ranking of wind farms in each sub-region, the wind farm with the smallest comprehensive ranking is used as the reference station of the sub-region;

[0142] The regional power prediction module is used to determine the probability distribution parameters of the power generation of the benchmark station according to the benchmark station in the sub-region; by extracting the spatial correlation characteristics of the benchmark station, the probability distribution parameters of the power generation of the benchmark station and the spatial correlation characteristics are input into the constructed probability upscaling model considering spatial correlation for prediction, so as to obtain the power prediction results of the sub-region.

[0143] Since renewable energy sources such as wind power and solar energy are greatly affected by the weather, large-scale grid connection of renewable energy power generation also shows uncertainty. The regional distributed power source power upscaling prediction method considering spatial correlation proposed in the present invention can overcome the problem of unpredictable power caused by the lack of collected data of distributed power sources in the region. By performing upscaling prediction of regional energy power through correlation analysis, the accuracy of distributed power source prediction can be improved. High-precision power prediction can help the power grid to reasonably arrange the access and scheduling of renewable energy, help the power grid to carry out effective backup capacity planning, ensure the reliability of power supply when the output of renewable energy fluctuates, and provide strong impetus for the digitalization level of the power grid.

[0144] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

[0145] In addition, unless otherwise specified, all technical and scientific terms used in the present invention have the same meanings as commonly understood by those skilled in the art to which the present invention belongs. All documents mentioned in this specification are incorporated by reference to disclose and describe the methods related to the documents. In the event of any conflict with any incorporated document, the content of this specification shall prevail.

Claims

1. A method for predicting regional distributed power generation power upscaling, characterized in that: The following steps are involved: Obtain the actual wind farm data of regional distributed power sources and divide the regions to determine the number of sub-regions; Conduct continuous hierarchical probability scoring assessment on random sub-areas, obtain the aggregate power probability prediction results of the assessed areas, and gradually increase the number of sub-areas until the prediction results no longer decrease. The number of sub-areas corresponding to the time when the prediction results no longer decrease is used as the number of benchmark sites. According to the number of benchmark stations, the number of sub-regions is sorted according to the aggregated power probability prediction results and the correlation between wind farm power and regional power to determine the corresponding ranking of wind farms in each sub-region; the smaller the ranking, the higher the power correlation; according to the corresponding ranking of wind farms in each sub-region, the comprehensive ranking of wind farms in each sub-region is determined, and the wind farm with the smallest comprehensive ranking is used as the benchmark station of the sub-region; Determine the power generation probability distribution parameters of the reference station according to the reference station in the sub-region; By extracting the spatial correlation characteristics of the benchmark station, the probability distribution parameters and spatial correlation characteristics of the power generation of the benchmark station are input into the constructed probability upscaling model considering spatial correlation for prediction, and the power prediction results of the sub-region are obtained.

2. The method for predicting regional distributed power generation power upscaling according to claim 1, characterized in that: Determining the number of sub-areas comprises the following steps: According to the actual wind farm data of regional distributed power sources obtained, including the geographical distribution data of the stations, the region is divided by the inter-cluster distance of hierarchical clustering to obtain the wind farm data of each cluster; According to the wind farm data of each cluster, the maximum value of the distance between the power generation of any two wind farms in the cluster is determined and used as the inter-cluster distance; The calculation formula for the maximum distance between the power generation of any two wind farms in the cluster is: D i,j =max{d(S A ,S B ),S A ∈C i ,S B ∈C j } Where D i,j Represents cluster C i and C j The distance between A ,S B ) represents cluster C i and C j Any wind farm S A and S B The distance between the generating power; The similarity of power generation of different wind farms is measured by Pearson correlation coefficient, and the distance between the output power of two wind farms is determined; The calculation formula of the Pearson correlation coefficient is: Where Y A and Y B Represents wind farm S A and S B The power generation sequence of , E(·) represents the expectation, μ and σ represent the mean and standard deviation of the power sequence respectively; The distance between the output powers of the two wind farms is: d(S A ,S B )=1-|ρ(Y A ,Y B )| Distance d(S A ,S B ) is smaller, indicating that the distance between the power generation of wind farms is smaller and the similarity between the powers is higher; Among them, hierarchical clustering includes the following steps: Step 1: Calculate the distance d (S) between any two wind farms in the region. A ,S B ); Step 2: Initialize all wind farms into clusters C1, C2, ..., C N ; Step 3: Calculate the distance D between any two clusters; Step 4: Merge the nodes with the smallest D i,j Two clusters C i and C j For a cluster C ij ; Step 5: Repeat Step 3 and Step 4 until only one cluster remains; All wind farms are clustered into M clusters through hierarchical clustering, corresponding to M sub-regions. The wind farms included in the mth sub-region are Where N m Indicates the number of wind farms in the mth sub-region.

3. The method for predicting regional distributed power generation power upscaling according to claim 2, characterized in that: The method of taking the number of the sub-area corresponding to when the prediction result no longer decreases as the number of reference stations specifically includes: Perform continuous hierarchical probability scoring evaluation on the random sub-region M to obtain the aggregate power probability prediction result of the evaluated region; And gradually increase the number of sub-regions. When the number of sub-regions increases five times in a row and the aggregate power probability prediction results of the regions evaluated by continuous graded probability scores no longer decrease, that is: When the number of sub-regions M=1, the corresponding regional aggregate power probability prediction result when the number of sub-regions evaluated by continuous graded probability scoring is 1 is b1; when the number of sub-regions M=2, the corresponding regional aggregate power probability prediction result when the number of sub-regions evaluated by continuous graded probability scoring is 2 is b2; when the number of sub-regions M=3, the corresponding regional aggregate power probability prediction result when the number of sub-regions evaluated by continuous graded probability scoring is 3 is b3; when the number of sub-regions is 4, the corresponding regional aggregate power probability prediction result when the number of sub-regions evaluated by continuous graded probability scoring is M=4 is b4; when the number of sub-regions M=5, the corresponding regional aggregate power probability prediction result when the number of sub-regions evaluated by continuous graded probability scoring is 5 is b5; when the number of sub-regions M=6, the corresponding regional aggregate power probability prediction result when the number of sub-regions evaluated by continuous graded probability scoring is 6 is b6, and b6≥b1; The number of sub-regions M=6 corresponding to the regional aggregate power probability prediction result b6 is taken as the number of reference sites.

4. The method for predicting regional distributed power generation power upscaling according to claim 3, characterized in that: Determining the ranking of wind farms in each sub-region specifically includes: Sub-region C m Wind farms in Sort the aggregate power probability prediction results based on the continuous hierarchical probability score evaluation to obtain the ranking of the aggregate power probability prediction results based on the continuous hierarchical probability score evaluation for: Sub-region C m Wind farms in According to the correlation between power generation and regional power measured by Pearson correlation coefficient, the correlation ranking between power generation and regional power measured by Pearson correlation coefficient is obtained. for: In the formula, and Represent sub-regions C m The ranking of the aggregated power probability prediction results of the first wind farm based on the continuous hierarchical probability score evaluation and the correlation ranking of the Pearson correlation coefficient; the smaller the ranking, the higher the prediction accuracy or the higher the power correlation.

5. The method for predicting regional distributed power generation power upscaling according to claim 4, characterized in that: The smallest wind farm in the comprehensive ranking is used as the benchmark site in the sub-region, including: The ranking of the aggregate power probability prediction results based on the continuous hierarchical probability score evaluation and the correlation ranking of the Pearson correlation coefficient of the same wind farm are added together to obtain the comprehensive ranking R of the wind farms in the sub-region. m : Select comprehensive ranking R m The smallest wind farm in m As sub-area C m The reference station of the wind farm S m The overall ranking of m =min{r∈R m }.

6. The method for predicting regional distributed power generation power upscaling according to claim 5, characterized in that: Determining the probability distribution parameters of the power generation of the reference station specifically includes: The power generation of the benchmark station is predicted by the probability prediction method of the linear spline quantile function, and the prediction result of the benchmark station is expressed as P m (y m ;Ω m ), m=1,2,...,M, where Ω m =(γ m ,β m ,ξ m ) represents the parameter of the quantile function; For the spline quantile function with K knots, the corresponding parameters Uniquely represents the reference station S m Probability prediction results; The corresponding reference station S m The probability distribution parameters of the power generation include the cumulative distribution function and quantile function of the power generation, which are F m (y m ) and Q m (Ω m ).

7. The method for predicting regional distributed power generation power upscaling according to claim 6, characterized in that: The constructed probability upscaling model considering spatial correlation specifically includes: The probability upscaling model transforms the input matrix Λ into the probability distribution parameter Ω of regional aggregate power through the nonlinear mapping composed of MLP and projection layer, where: The input matrix Λ includes the parameters of the probability distribution of the power generation of the reference station [Ω1, ..., Ω M ] and the spatial correlation characteristics of the benchmark stations extracted by the Gaussian Copula function Θ G ; Flatten the input matrix Λ into a one-dimensional vector of the lower triangular matrix L, input the lower triangular matrix L into the MLP, and output the quantile function parameters Ω=(γ, β, ξ) of the regional aggregate power, where the quantile function parameters Ω=(γ, β, ξ) are estimated by minimizing the continuous hierarchical probability scoring loss function; The nonlinear mapping includes learning the input features through MLP, and representing the learning process of MLP as an intermediate variable h: h=ρ MLP (L;θ MLP ) In the formula, ρ MLP and θ MLP They represent the nonlinear mapping and its parameters learned by MLP respectively; The nonlinear mapping also includes a projection layer composed of an affine transformation and an activation function, which transforms the intermediate variable h into a quantile function parameter Ω of the regional aggregation power: Oh = p proj (h;θ proj ) In the formula, ρ proj and θ proj They represent the nonlinear mapping of the projection layer and its parameters respectively; In the projection layer, the activation function type depends on the constraints on the quantile function parameters; The parameter θ is adjusted by the Adam algorithm MLP and θ proj Optimize, through the optimized parameters θ MLP and θ proj , and optimize the constructed probability upscaling model taking into account spatial correlation.

8. The method for predicting regional distributed power generation power upscaling according to claim 7, characterized in that: The obtaining of the power prediction result of the sub-region specifically includes: According to the optimized probability upscaling model considering spatial correlation, the power prediction result of the sub-region is obtained as the probability distribution parameter of the regional aggregate power.

9. A regional distributed power generation power upscaling prediction system, characterized in that: include: A regional division module is used to obtain the actual wind farm data of regional distributed power sources and divide the regions to determine the number of sub-regions; The reference station number determination module is used to perform continuous hierarchical probability scoring evaluation on random sub-areas, obtain the aggregate power probability prediction result of the evaluated area, and gradually increase the number of sub-areas until the prediction result no longer decreases, and the number of sub-areas corresponding to the prediction result no longer decreases is used as the reference station number; The reference station selection module is used to sort the number of sub-regions according to the aggregate power probability prediction results and the correlation between wind farm power and regional power according to the number of reference stations, and determine the corresponding ranking of wind farms in each sub-region; wherein, the smaller the ranking, the higher the power correlation; according to the ranking of wind farms in each sub-region, by determining the comprehensive ranking of wind farms in each sub-region, the wind farm with the smallest comprehensive ranking is used as the reference station of the sub-region; The regional power prediction module is used to determine the probability distribution parameters of the power generation of the benchmark station based on the benchmark station in the sub-region; by extracting the spatial correlation characteristics of the benchmark station, the probability distribution parameters of the power generation of the benchmark station and the spatial correlation characteristics are input into the constructed probability upscaling model considering spatial correlation for prediction, so as to obtain the power prediction results of the sub-region.

Images