Wind power aggregation power probability prediction method and device, equipment and storage medium
Through the Gaussian Copula model and the scale-up algorithm, the dependence relationship and probability distribution between wind farms are generated, which solves the uncertainty problem of wind power prediction and realizes high-precision probability prediction of wind power polymerization power.
Patent Information
- Application Number
- CN202411787202.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-06
- Publication Date
- 2025-05-02
AI Technical Summary
The randomness, volatility and intermittentity of wind power have led to challenges in supply and demand balance and stability of power grids, and traditional point prediction methods cannot effectively reflect the uncertainty of wind power.
Wind speed prediction data and historical observation data are used to generate the power probability prediction data of each wind farm. The dependence between wind farms is described through the Gaussian Copula model, and the reference scenario is extended to the target wind farm position using the rising scale algorithm, and finally the target power probability is added to obtain the power probability prediction results of wind farm aggregation.
The probability prediction of wind power polymerization power is realized, and the probability distribution is inferred using historical data of similar wind farms, which enhances the flexibility and accuracy of prediction, breaks the fixed number limit, and is suitable for the aggregation of wind farms of different scales.
Smart Images
Figure CN119921295A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of power grid power prediction, and in particular to a method, device, equipment and storage medium for probabilistic prediction of wind power aggregation power. Background Art
[0002] In recent years, with the rapid growth of population and economic development, new energy has become an important energy strategy for carbon emission reduction. Wind power, as an important new energy power generation method, has received policy support and rapid development in many countries due to its advantages of being renewable, green and environmentally friendly, and highly energy efficient. However, wind power has significant randomness, volatility, intermittency, and poor power regulation capabilities. With the incorporation of a large number of wind power systems into the power system, the real-time dynamic balance problem of wind power output has brought new challenges to the safe and economic operation of the power grid.
[0003] The rapid development of wind power has increased the difficulty of dispatching and managing the supply and demand balance of the power system. The power fluctuation of wind power may affect the load characteristics and power flow distribution of the power grid, and in some cases may even cause instability in the power grid. In order to cope with the safe and stable operation problems caused by the change of wind power output, it is particularly important to accurately predict wind power. This not only helps to provide effective data support for microgrid energy management, power grid pricing and load management, but also helps to formulate reasonable dispatching plans, realize the effective regulation of wind power systems, promote the consumption of wind power, and improve economic benefits.
[0004] For power systems with large-scale wind power access, it is more important to predict the aggregate power of the total output of multiple wind power systems in a certain area. However, wind power-related power forecasting faces a series of challenges, including complex spatial correlation, data sparsity, insufficient data support for newly developed areas, mutual influence between multiple wind farms, and quantification of uncertain factors such as wind speed changes. Traditional power forecasting methods mainly predict a single value at each future time point, so they are called point forecasts or deterministic forecasts. However, this method cannot fully reflect the uncertainty of wind power. Summary of the invention
[0005] The purpose of this application is to overcome the deficiencies in the above-mentioned prior art and to provide a method, device, equipment and storage medium for predicting wind power aggregation probability.
[0006] The present application provides a method for predicting wind power aggregation probability, comprising:
[0007] Generate power probability forecast data for each wind farm at each time point using wind speed forecast data and historical observation data;
[0008] Derived a Gaussian Copula according to the power probability prediction data and the acquired observation data, and converted the Gaussian Copula into a multivariate normal distribution of a covariance matrix and a mean vector;
[0009] According to the multivariate normal distribution, extracting samples from the Gaussian Copula as reference scenarios, wherein the reference scenarios include dependencies between wind farms;
[0010] Using an upscaling algorithm, the reference scenario is extended to all target wind farm locations, and a target power probability of the target wind farm is determined according to a reference power probability in the reference scenario;
[0011] All the target power probabilities are summed up to obtain the power probability prediction result of the wind farm aggregation.
[0012] Optionally, the wind speed prediction data and historical observation data are used to generate power probability prediction data for each wind farm at each time point, including:
[0013] The wind speed prediction data and historical observation data are trained using a machine learning model to generate the power probability prediction data.
[0014] Optionally deriving a Gaussian Copula and converting the Gaussian Copula into a multivariate normal distribution of a covariance matrix and a mean vector includes:
[0015] The Gaussian Copula is given by a covariance matrix, and the covariance matrix is adjusted by a forgetting factor λ, where λ∈[0,1].
[0016] Optionally, the forgetting factor λ is set to 0.3.
[0017] Optionally, the reference scenario is extended to all target wind farm locations using an upscaling algorithm, including:
[0018] The current output power probability distribution of all wind farms in a specific area is determined using the measurement data, installed capacity information and location information of the reference wind farm and the reference scenario.
[0019] Optionally, after deriving a Gaussian Copula based on the power probability prediction data and the acquired observation data, converting the Gaussian Copula into a multivariate normal distribution of a covariance matrix and a mean vector, the method further includes:
[0020] The predicted probability distribution is adjusted according to the historical observations to ensure that the predicted probability distribution is consistent with the historical observations.
[0021] Optionally, it also includes:
[0022] A leave-one-out cross-validation method is used to train a calibration function, and the calibration function is applied to adjust the power probability prediction data of the target wind farm, thereby correcting deviations caused by system errors.
[0023] The present application also provides a wind power aggregation power probability prediction device, comprising:
[0024] The model module uses wind speed forecast data and historical observation data to generate power probability forecast data for each wind farm at each time point;
[0025] A conversion module, which derives a Gaussian Copula according to the power probability prediction data and the acquired observation data, and converts the Gaussian Copula into a multivariate normal distribution of a covariance matrix and a mean vector;
[0026] A sampling module extracts samples from the Gaussian Copula as reference scenarios according to the multivariate normal distribution, wherein the reference scenarios include dependencies between wind farms;
[0027] An expansion module, using an upscaling algorithm, expands the reference scenario to all target wind farm locations, and determines a target power probability of the target wind farm according to a reference power probability in the reference scenario;
[0028] The aggregation module adds up all the target power probabilities to obtain a power probability prediction result of the wind farm aggregation.
[0029] The present application also provides a wind power aggregation power probability prediction device, comprising:
[0030] Memory;
[0031] A processor is used to retrieve from the memory a computer executable program of the above-mentioned wind power aggregation power probability prediction method to execute: using wind speed prediction data and historical observation data to generate power probability prediction data for each wind farm at each time point; deriving a Gaussian Copula based on the power probability prediction data and the acquired observation data, and converting the Gaussian Copula into a multivariate normal distribution of a covariance matrix and a mean vector; according to the multivariate normal distribution, extracting samples from the Gaussian Copula as reference scenarios, wherein the reference scenarios contain dependencies between wind farms; using an upscaling algorithm, extending the reference scenarios to all target wind farm locations, and determining the target power probability of the target wind farm based on the reference power probability in the reference scenarios; and summing up all the target power probabilities to obtain a power probability prediction result for wind farm aggregation.
[0032] The present application also provides a storage medium, comprising: a computer executable program stored therein, wherein the computer executable program is used to be called by a processor to execute the steps of the above-mentioned wind power aggregation power probability prediction method.
[0033] The beneficial effects of this application are:
[0034] The present application provides a method for predicting the power probability of wind power aggregation, including: using wind speed prediction data and historical observation data to generate power probability prediction data for each wind farm at each time point; deriving a Gaussian Copula based on the power probability prediction data and the acquired observation data, and converting the Gaussian Copula into a multivariate normal distribution of a covariance matrix and a mean vector; extracting samples from the Gaussian Copula as a reference scenario based on the multivariate normal distribution, and the reference scenario contains dependencies between wind farms; using a scaling algorithm, extending the reference scenario to all target wind farm locations, and determining the target power probability of the target wind farm based on the reference power probability in the reference scenario; summing up all the target power probabilities to obtain a power probability prediction result for wind farm aggregation. The present application uses historical data of similar wind farms to infer their probability distribution to achieve effective prediction; it realizes flexible aggregation of wind farms of different sizes, breaks the fixed number limit, and enhances the universality of the application. At the same time, the constructed spatial correlation model deeply depicts the interaction between wind farms and significantly improves the prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0035] Figure 1 It is a schematic diagram of the wind power aggregation power probability prediction process in this application;
[0036] Figure 2 This is a schematic diagram of the time series of quantile 1 in the experiment in this application;
[0037] Figure 3 This is a schematic diagram of the experimental 2-quantile time series in this application;
[0038] Figure 4 It is a schematic diagram of the reliability in this application;
[0039] Figure 5 This is a schematic diagram of the experimental 3-quantile time series in this application;
[0040] Figure 6 This is a schematic diagram of the results of Experiment 2 in this application. DETAILED DESCRIPTION
[0041] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that various forms of implementing the present disclosure should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.
[0042] Please refer to Figure 1 and Figure 2 As shown, the present application provides a method for predicting wind power aggregation probability, comprising:
[0043] S101 . Generate power probability prediction data for each wind farm at each time point using wind speed prediction data and historical observation data.
[0044] Using existing wind speed forecast data and historical observation data, a probabilistic forecast for each reference wind farm is generated at each time point. These forecasts will serve as the basis for modeling and aggregation using Copula theory in subsequent steps.
[0045] Wind speed forecast data: comes from meteorological models or professional wind speed forecast services, providing wind speed forecast values for a period of time in the future (such as the next few hours, days or weeks).
[0046] Historical observation data: Actual wind speed and power output data of each wind farm over a period of time. This data is used to train the model so that it can learn the complex relationship between wind speed and power output.
[0047] According to the characteristics of the data and the requirements of the problem, choose an appropriate prediction model. These models can be based on statistics (such as time series models, regression models, etc.) or machine learning (such as neural networks, support vector machines, etc.).
[0048] The model is trained using wind speed forecast data and historical observation data as input and actual power output as output. During the training process, the model learns the mapping relationship between wind speed and power output and is able to generate power forecasts based on new wind speed forecast data.
[0049] Unlike traditional point forecasting, which predicts a single value at each future point in time, probabilistic forecasting takes into account the uncertainty of the forecast. Therefore, in this step, the model generates multiple power forecast values for each point in time and assigns a probability to each value. This is usually achieved by predicting a continuous probability distribution or quantile.
[0050] After model training and prediction, the output will be the probability prediction results of each reference wind farm at each time point. These results can be presented in the form of tables, graphs or databases, including the predicted power value and its corresponding probability for each wind farm and each time point.
[0051] S102, deriving a Gaussian Copula according to the power probability prediction data and the acquired observation data, and converting the Gaussian Copula into a multivariate normal distribution of a covariance matrix and a mean vector.
[0052] The uncertainty of the forecast can be described using different quantities, such as probability distributions, quantiles or scenarios. Quantiles describe the probability of exceeding a certain power level. Scenarios or trajectories describe possible future developments and also contain information about spatial dependencies. In this application, scenarios are used to describe the uncertainty and interdependencies of many wind farm forecasts, considering that they are suitable for the expansion, upgrading and aggregation of wind farms.
[0053] The correlation between individual sites is crucial for upscaling and aggregation. Copula can be used to describe the correlation structure between individual sites of a wind farm. More precisely, this application uses Gaussian Copula to estimate the covariance matrix of forecast uncertainty between wind farms. Once the Gaussian Copula and the joint distribution of all forecasts are known, samples can be drawn from the joint distribution, which can be upscaled to the target wind farm. From the upscaled samples, the distribution can be reconstructed.
[0054] The specific Gaussian Copula is determined by the following steps:
[0055] S201. Determine a cumulative distribution function.
[0056] Calculate the cumulative distribution function F of wind farm power based on quantitative prediction t (i) Estimated value:
[0057]
[0058] Among them, the function F t (i) It is obtained by calculating m quantized values q for each time step t and wind field i. α,t (i) It is estimated by linear interpolation.
[0059] S202. Convert to a uniform random variable.
[0060] In order to estimate the covariance, we must use the cumulative distribution function F t (i) The random power variable Y t (i)Convert to a uniformly distributed random variable U t (i) , the expression is as follows:
[0061]
[0062] Applying this formula to the observed power value Y t (i) And get a new time series U with uniform distribution t (i) .
[0063] S203. Transformation of normally distributed variables.
[0064] Next, the uniformly distributed random variable U t (i) Transformed into a normally distributed variable V t (i) , the expression is as follows:
[0065]
[0066] Among them, Φ -1 (x)=√2erf -1 (2x-1), V (i) ≈N(0,1).
[0067] S204. Estimate Copula.
[0068] When all variables are transformed into normally distributed variables, the linkages coupling the predictions of all reference wind farms can be determined.
[0069] Gaussian Copula is given by the covariance matrix, and the calculation formula is:
[0070] V t =(V t 1 ,...,V t n ),
[0071]
[0072] Among them, the forgetting factor λ∈[0,1] sets the influence of previous estimates. Usually, when λ is 0.3, good results can be obtained. If the data analyzed is highly volatile, the smaller the λ value, the better.
[0073] Furthermore, the covariance is normalized to compensate for the slight deviation in the update step. The normalization formula is as follows:
[0074]
[0075] in, is element-wise division, σ t is the standard deviation vector, which is the square root of the diagonal elements of t.
[0076] The copula of all wind farms can now be written as the covariance matrix (denoted as ∑t in the following content) and the mean vector μ t =0 multivariate normal distribution.
[0077] S103 . According to the multivariate normal distribution, extract samples from the Gaussian Copula as reference scenarios, where the reference scenarios include dependencies between wind farms.
[0078] Extracting scenes from the Copula involves drawing samples from the multivariate normal distribution defined by the estimated covariance matrix and transforming these samples back into the distribution of the original power values. The following is a detailed description of this step:
[0079] Specific steps to extract scenes from Copula:
[0080] Draw m samples from the multivariate normal distribution N(0,∑t) estimated in the above covariance matrix formula and normalization formula. Here,∑t is the covariance matrix given by the Gaussian Copula, which describes the correlation structure of the forecast uncertainty between wind farms.
[0081] These extracted samples are called "scenarios" and they contain information about the dependencies between wind farms. Due to the covariance matrix, these samples are correlated, i.e. they reflect the interrelationships between the power variations of wind farms.
[0082] The samples drawn are normally distributed variables. To transform them back into the distribution of the original power values, the same steps as when determining the copula need to be applied, but in the reverse order.
[0083] Specifically, first, through the inverse process of the above normal distribution variable formula, the normal distribution variable V t (i) Convert back to a uniformly distributed random variable U t (i) This process involves the inverse error function (erf -1 ) application.
[0084] Then, these uniformly distributed random variables U t (i) It can further be used to reconstruct the distribution of original power values.
[0085] The above process will continue in all time steps. Once new observation data is available, the covariance matrix ∑t will be updated according to the above normalization formula to reflect the latest correlation information between wind farms.
[0086] S104: Use an upscaling algorithm to expand the reference scenario to all target wind farm locations, and determine a target power probability of the target wind farm according to the reference power probability in the reference scenario.
[0087] Upscaling is an extrapolation of an estimate of the overall result from the partial results. It is used when not all information about the overall result is available.
[0088] The upscaling algorithm uses the measured data, installed capacity information and location information of some reference wind farms and determines the current output power of all wind farms in a specific area. The goal of this application is to use this method for probability upscaling, that is, to find the total output power probability distribution of all wind farms or combinations.
[0089] The power of the target wind farm at time t Power of the reference wind farm weighted by different parameters Sure:
[0090]
[0091] in, is the measured wind energy of the jth wind farm among n reference wind farms, normalized to its installed capacity. j，t is the state of the measurement (0 = fault, 1 = normal), automatically taking into account faults or erroneous measurements; the reference wind farm is indexed with j and the target wind farm is indexed with i. Matrix a i，j corresponds to the distance-dependent weighting factor, which is defined as:
[0092]
[0093] Among them, d i，j is the distance between the reference wind farm j and the target wind farm i; r is the attenuation factor, which is set to r = 50 km here and can have different values for other settings. The correction coefficient k in formula (6) is i,t is chosen so that the weights always sum to 1 (normalization).
[0094]
[0095] This upscaling method is applied to each of the previously generated scenarios. By upscaling these scenarios to the target wind farm, we extend each scenario (including only the reference wind farm before upscaling) to a scenario that includes all wind farms (reference wind farm and target wind farm).
[0096] The extended scenarios are calibrated to ensure that the predicted probability distribution is consistent with historical observations.
[0097] If the observed measurement frequencies do not match the predicted probabilities, calibration can be used to adjust the predictions based on historical measurements, further improving the reliability of the probabilistic forecasts.
[0098] For each time step t and each target wind farm i, the scenario members Sort by their values and get Make
[0099] For each time step t and each wind farm i, the sorted scenario members Interpreted as m' quantiles, each with probability k / (m'+1).
[0100] Predicted value does not necessarily correspond to the nominal probability α caused by the systematic error in the upscaling step, i.e. To correct for these deviations, the calibration function can be used To adjust the dimension
[0101] For the estimation of the calibration function, this application uses quantum regression, the expression:
[0102]
[0103] ρ α (x) = x·(α-1 x＜0 ).
[0104] Among them, ρ α (x) is the test function, the indicator function 1 x<0 If x < 0, it is 1, otherwise it is 0. Make optimizations.
[0105] Predictors:
[0106]
[0107] The prediction factors are composed of different upscaled quantile predictions of the reference wind farm, using the leave-one-out method, as shown below, the effect distance vector
[0108] Effect distance:
[0109]
[0110] Predictors and d effhave the same length. Through quantile regression, θ α The parameters are optimized. For this function, the Polynomial method is selected.
[0111]
[0112] The effective distance can be understood as: the farther the reference wind farm is, the greater their distance d k,l The larger is the value, the smaller is their relative contribution to the calibration. Therefore, the calibration depends on the distance d from the target wind farm k to all reference wind farms l. k,l .
[0113] This application uses the leave-one-out cross-validation method to train the calibration function given by the quantum regression formula:
[0114] First, the upscaling of the reference wind field is additionally determined.
[0115] The specific method is to select a wind field l from the reference wind field set {1,...,n} in turn, and estimate its predicted value by amplifying the predictions of the remaining wind fields {1,...,n}\{l}. Therefore, in addition to the observed value y t In addition, the upscaled prediction values of the reference wind field is also known. On this data, the quantum regression formula is trained on all measured and upscaled values simultaneously. The result is a generalized calibration model that is ultimately used to calibrate the quantitative predictions for the target wind farm.
[0116] S105: sum up all the target power probabilities to obtain a power probability prediction result of the wind farm aggregation.
[0117] Aggregate the forecasts of all wind farms. Sum up the scenarios of all wind farms to get the aggregated power forecast of the entire wind farm.
[0118] The probabilistic forecast of aggregated wind power is determined from the forecasts of the individual wind farms. To do this, the individual probabilistic forecasts must be aggregated taking into account their relevance.
[0119] When summarizing the prediction results of the reference wind field, y with time index t, wind field index i and member index k is determined according to the above steps. t (i,k) According to each member index and time, only the wind farm y t (agg,k) Just sum the scenes.
[0120] Aggregate reference wind farm and target wind farm:
[0121]
[0122] The first sum is the sum of the scenarios for the reference wind farm I, and the second sum is the sum of the calibration scenarios for the target wind farm J. The scene of the target wind farm is estimated by enlarging and calibrating the scene of the reference wind farm.
[0123] test:
[0124] The experiments in this application use wind farms based on 20 available measurements from a country. Depending on the experiment, a part of the wind farms are used as reference wind farms, whose power measurements are explicitly used, while the remaining wind farms are target wind farms, whose power measurements are only used for validation. In the experiments, a mix of reference wind farms and wind farms without measurements is used. The goal is to predict the aggregate power of all wind farms based on the predictions of the reference wind farms.
[0125] Experiment 1: Upscaling probability prediction. Probabilistic predictions are made for all reference wind fields. The predictions for the target wind field are made based on these reference wind fields.
[0126] Experiment 2: Probabilistic aggregation. All wind farms are reference wind farms for which probabilistic forecasts can be made directly. The goal is to determine the probabilistic forecast of the total input power.
[0127] Experiment 3: Upscaling Probabilistic Forecasts and Probabilistic Aggregation. Not all wind farms are reference wind farms, and the probabilistic forecasts for non-reference wind farms are estimated by upscaling the forecasts from known reference wind farms to these target wind farms and then performing probabilistic aggregation.
[0128] Regarding the results of Experiment 1, Figure 3 The upscaled forecasts show systematic deviations from the observations. This indicates that a single upscaling is not very reliable. The calibrated forecasts significantly widen the spread, removing the bias and improving the reliability. The reason these biases are more pronounced is that relatively few reference wind farms were used. In practice, more reference wind farms would be selected, resulting in more accurate results than shown here.
[0129] Regarding the results of Experiment 2, Figure 4 It can be seen that the method of probability aggregation works well and generally captures the composite observations well. However, at some points, the outer quantiles do not seem to be wide enough to capture the observed power spectrum. This fact is also reflected in Figure 5 In the confidence diagram given on the left, the sum of the power of all wind farms is based only on the individual probabilistic forecasts of the single wind farm involved (i.e., each wind farm is considered as a reference wind farm), and the observed errors are most likely due to errors in the basic reference forecast.
[0130] Regarding the results of Experiment 3, Figure 6In , we observe that if the calibration step is skipped, the observed forecast performance is very similar to that obtained in Experiment 2. This manipulation works well, and the observed forecast errors are mainly due to errors in the base reference forecast. The systematic deviations from the synthetic observations and the fact that the range of the forecasts at some time points is not wide enough to capture the observed power are reflected in Figure 4 In the reliability diagram on the right.
[0131] Compared with the results of Experiments 2 and 3 without calibration, the calibration upgrade scenario achieves the best performance. This shows the importance of the calibration step, which improves the reliability of the prediction and may fix errors in correlations that were not fully modeled before. The results show that the proposed method has good performance and has great potential in the field of wind power forecasting.
[0132] The present application also provides a wind power aggregation power probability prediction device, comprising:
[0133] The model module uses wind speed forecast data and historical observation data to generate power probability forecast data for each wind farm at each time point;
[0134] A conversion module, which derives a Gaussian Copula according to the power probability prediction data and the acquired observation data, and converts the Gaussian Copula into a multivariate normal distribution of a covariance matrix and a mean vector;
[0135] A sampling module extracts samples from the Gaussian Copula as reference scenarios according to the multivariate normal distribution, wherein the reference scenarios include dependencies between wind farms;
[0136] An expansion module, using an upscaling algorithm, expands the reference scenario to all target wind farm locations, and determines a target power probability of the target wind farm according to a reference power probability in the reference scenario;
[0137] The aggregation module adds up all the target power probabilities to obtain a power probability prediction result of the wind farm aggregation.
[0138] The present application also provides a wind power aggregation power probability prediction device, comprising:
[0139] Memory;
[0140] A processor is used to retrieve from the memory a computer executable program of the above-mentioned wind power aggregation power probability prediction method to execute: using wind speed prediction data and historical observation data to generate power probability prediction data for each wind farm at each time point; deriving a Gaussian Copula based on the power probability prediction data and the acquired observation data, and converting the Gaussian Copula into a multivariate normal distribution of a covariance matrix and a mean vector; according to the multivariate normal distribution, extracting samples from the Gaussian Copula as reference scenarios, wherein the reference scenarios contain dependencies between wind farms; using an upscaling algorithm, extending the reference scenarios to all target wind farm locations, and determining the target power probability of the target wind farm based on the reference power probability in the reference scenarios; and summing up all the target power probabilities to obtain a power probability prediction result for wind farm aggregation.
[0141] The present application also provides a storage medium, comprising: a computer executable program stored therein, wherein the computer executable program is used to be called by a processor to execute the steps of the above-mentioned wind power aggregation power probability prediction method.
[0142] The above are only specific embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A method for predicting wind power aggregation probability, characterized in that: include: Generate power probability forecast data for each wind farm at each time point using wind speed forecast data and historical observation data; Derived a Gaussian Copula according to the power probability prediction data and the acquired observation data, and converted the Gaussian Copula into a multivariate normal distribution of a covariance matrix and a mean vector; According to the multivariate normal distribution, extracting samples from the Gaussian Copula as reference scenarios, wherein the reference scenarios include dependencies between wind farms; Using an upscaling algorithm, the reference scenario is extended to all target wind farm locations, and a target power probability of the target wind farm is determined according to a reference power probability in the reference scenario; All the target power probabilities are summed up to obtain the power probability prediction result of the wind farm aggregation.
2. According to claim 1, a wind power aggregation power probability prediction method is characterized in that: Use wind speed forecast data and historical observation data to generate power probability forecast data for each wind farm at each time point, including: The wind speed prediction data and historical observation data are trained using a machine learning model to generate the power probability prediction data.
3. According to claim 1, a wind power aggregation power probability prediction method is characterized in that A Gaussian Copula is derived and converted into a multivariate normal distribution of a covariance matrix and a mean vector, including: The Gaussian Copula is given by a covariance matrix, and the covariance matrix is adjusted by a forgetting factor λ, where λ∈[0,1].
4. According to claim 3, a wind power aggregation power probability prediction method is characterized in that: The forgetting factor λ is set to 0.
3.
5. According to claim 1, a wind power aggregation power probability prediction method is characterized in that: Using an upscaling algorithm, the reference scenario is extended to all target wind farm locations, including: The current output power probability distribution of all wind farms in a specific area is determined using the measurement data, installed capacity information and location information of the reference wind farm and the reference scenario.
6. A wind power aggregate power probability prediction method according to claim 1, characterized in that: The method further comprises: deriving a Gaussian Copula based on the power probability prediction data and the acquired observation data, and converting the Gaussian Copula into a multivariate normal distribution of a covariance matrix and a mean vector, The predicted probability distribution is adjusted according to the historical observations to ensure that the predicted probability distribution is consistent with the historical observations.
7. A wind power aggregation power probability prediction method according to claim 1, characterized in that: Also includes: A leave-one-out cross-validation method is used to train a calibration function, and the calibration function is applied to adjust the power probability prediction data of the target wind farm, thereby correcting deviations caused by system errors.
8. A wind power aggregation power probability prediction device, characterized in that: include: The model module uses wind speed forecast data and historical observation data to generate power probability forecast data for each wind farm at each time point; A conversion module, which derives a Gaussian Copula according to the power probability prediction data and the acquired observation data, and converts the Gaussian Copula into a multivariate normal distribution of a covariance matrix and a mean vector; A sampling module extracts samples from the Gaussian Copula as reference scenarios according to the multivariate normal distribution, wherein the reference scenarios include dependencies between wind farms; An expansion module, using an upscaling algorithm, expands the reference scenario to all target wind farm locations, and determines a target power probability of the target wind farm according to a reference power probability in the reference scenario; The aggregation module adds up all the target power probabilities to obtain a power probability prediction result of the wind farm aggregation.
9. A wind power aggregation power probability prediction device, characterized in that: include: Memory; A processor, for retrieving from the memory a computer executable program of a method for predicting the power probability of wind power aggregation as described in any one of claims 1 to 7, to execute: using wind speed prediction data and historical observation data to generate power probability prediction data for each wind farm at each time point; deriving a Gaussian Copula based on the power probability prediction data and the acquired observation data, and converting the Gaussian Copula into a multivariate normal distribution of a covariance matrix and a mean vector; extracting samples from the Gaussian Copula as a reference scenario based on the multivariate normal distribution, wherein the reference scenario contains dependencies between wind farms; using an upscaling algorithm, extending the reference scenario to all target wind farm locations, and determining a target power probability of a target wind farm based on a reference power probability in the reference scenario; and summing up all the target power probabilities to obtain a power probability prediction result for wind farm aggregation.
10. A storage medium, characterized in that: include: A computer executable program is stored, and the computer executable program is used to be called by a processor to execute the steps of a wind power aggregation power probability prediction method as described in any one of claims 1 to 7.