A method for predicting wind power under short-term scales

Abstract

The present invention discloses a method for predicting wind power under short-term scales, which relates to the field of smart grids and includes: collecting the initial predicted wind speed, measured wind speed, and actual power at a set sampling interval, preprocessing the qualified input data, calculating the predicted wind speed deviation according to the preprocessed initial predicted wind speed and measured wind speed using the first system deviation formula, and then correcting the initial predicted wind speed to obtain the corrected predicted wind speed; calculating the initial predicted power using the improved Richards model according to the corrected predicted wind speed; calculating the power prediction deviation according to the initial predicted power and actual power using the second system deviation formula, and then correcting the initial predicted power to obtain the corrected predicted power; detecting whether the corrected predicted power is qualified, and if it is qualified, outputting the corrected predicted power at a set interval. The present invention suppresses the expansion of prediction deviation through two-stage system deviation correction and can improve the accuracy of short-term wind power prediction.

Description

[0001] This application is a divisional application. The application number of the original application is 202410475895.8, the application date is April 19, 2024, and the invention title is "A Two-Stage Short-Term Wind Power Prediction Method and Device". Technical Field

[0002] The present invention relates to the technical field of smart grids, and more specifically, to a method for predicting wind power that meets the short-term scale. Background Art

[0003] Wind power prediction refers to establishing a prediction model for the output power of a wind farm using data such as the historical power, historical wind speed, terrain, numerical weather prediction, and operating status of wind turbines. Using wind speed, power, or numerical weather prediction data as the input of the model, combined with the equipment status and operating conditions of the wind farm units, the future output power of the wind farm is obtained. Among them, short-term scale (0 - 72 hours) wind power prediction highly depends on the predicted wind speed provided by weather forecasting.

[0004] Classic wind power curve fitting techniques include polynomial regression models, piecewise linear models, Logistic function models, and non-parametric regression models. However, polynomial regression models, piecewise linear models, Logistic function models, etc. cannot consider the uncertainty of the operating status of wind turbines; non-parametric regression models such as kernel density estimation methods have large estimation deviations in the low wind speed section and near-rated wind speed section, and are also sensitive to local outliers; while methods such as machine learning are sensitive to data quality and parameter selection and have high requirements for computing resources. In terms of the uncertainty of weather forecasting, signal decomposition methods and prediction error probability distribution functions are the most commonly used tools. However, signal decomposition methods and probability distribution functions also require a large amount of actual observation data and are easily affected by outliers.

[0005] Therefore, how to improve the accuracy of short-term wind power prediction is an urgent problem for those skilled in the art. Summary of the Invention

[0006] In view of this, the present invention provides a method for predicting wind power that meets the short-term scale. First, an uncertain term is introduced into the Richards curve, which characterizes the uncertainty of the unit output while characterizing the basic S-shaped characteristics of wind power. Secondly, by estimating and correcting the system prediction deviation in two links of wind speed and power prediction, the optimization of the final prediction result is realized, thereby improving the accuracy of short-term wind power prediction.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] The present invention discloses a two-stage short-term wind power prediction method, asFigure 1 As shown in the figure, the specific steps are as follows:

[0009] Data input: According to the set sampling interval, collect the initial predicted wind speed, measured wind speed, and actual power to obtain the input data;

[0010] Data verification: Detect whether the input data is qualified. If it is qualified, perform preprocessing; otherwise, issue a warning.

[0011] Predicted wind speed correction: According to the preprocessed initial predicted wind speed and measured wind speed, use the first system deviation formula to calculate the predicted wind speed deviation; correct the initial predicted wind speed according to the predicted wind speed deviation to obtain the corrected predicted wind speed.

[0012] Power prediction and correction: According to the corrected predicted wind speed, use the improved Richards model to calculate the initial predicted power; according to the initial predicted power and the actual power, use the second system deviation formula to calculate the power prediction deviation; correct the initial predicted power according to the power prediction deviation to obtain the corrected predicted power.

[0013] Data output: Detect whether the corrected predicted power is qualified. If it is qualified, output the corrected predicted power at the set interval.

[0014] Furthermore, the initial predicted wind speed and the measured wind speed are input by a weather forecasting workstation in Safety Zone III / IV through a reverse isolation device, and the actual power is input by a power prediction workstation in Safety Zone II through a forward isolation device.

[0015] Furthermore, in the data verification, the qualification rate calculation formula is:

[0016]

[0017] where N 总 is the total number of sampling points within the detection time period, N 缺测点 is the total number of missing measurement points within the detection time period, and N 离群点 is the total number of outliers within the detection time period; when the qualification rate is greater than or equal to 90%, data preprocessing is performed; otherwise, a warning is issued.

[0018] The outliers within the detection time period are the mutation points in the sliding correlation coefficient r VP of the measured wind speed and the actual power, and the calculation formula for the sliding correlation coefficient r VP is:

[0019]

[0020] In the formula, V i and P iThey are the values of the time series of the measured wind speed and the actual power in a sliding window with a width of k, and is the average value of the points within the corresponding sliding window; during the process of the detection window sliding forward, the points where the sliding correlation coefficient is less than 0.9 are recorded as mutation points.

[0021] Furthermore, the preprocessing is as follows: the replacement value of the unqualified data points is generated by the adjacent weighted average method, and the unqualified data points are replaced;

[0022] The calculation formula for the replacement value is:

[0023]

[0024] where y is the replacement value, y -2 , y -1 , y 1 and y 2 are the values of the 4 points adjacent to the unqualified point y 0 , w -2 , w -1 , w 1 and w 2 are the weight values corresponding to the adjacent points.

[0025] Furthermore, the first system deviation formula is:

[0026]

[0027] where E is the system deviation between the initial predicted wind speed and the measured wind speed, n is the length of the wind speed time series, V i , V' i are the measured wind speed and the initial predicted wind speed at the i-th moment respectively;

[0028] The correction of the initial predicted wind speed includes: using the ARIMA(p,d,q) model to estimate the time series of the wind speed prediction deviation E w , and calculating the corrected predicted wind speed V' w ' according to the time series of E i and the initial predicted wind speed. The specific calculation formula is:

[0029]

[0030] where φ i is the autoregressive term coefficient, indicating the influence of the previous observed values of the time series; p is the number of autoregressive terms; q is the number of moving average terms; d is the number of differences required for the series to be stationary; L is the lag operator L(y t ) = y t-1 , L k is the k-th power of the lag operator, indicating y tLag by k time steps, L k (y t ) = y t-k ; θ i is the coefficient of the moving average term, representing the influence of the previous q random shocks on the current value; ε t is the random error term.

[0031] Furthermore, the calculation formula for the initial predicted power is:

[0032]

[0033] In the formula, P' i represents the initial predicted power at time i, P max is the rated power of the wind turbine, a is the shape parameter, e is the natural constant, V″ i is the corrected predicted wind speed, v 0 is the cut-in wind speed of the wind turbine, K is the scale function, used to characterize the uncertainty of the unit output at different wind speeds, and its expression is:

[0034]

[0035] In the formula, σ is the scale parameter, and x is the fluctuation range of the wind speed;

[0036] The second system deviation formula is:

[0037]

[0038] In the formula, E' is the system deviation between the initial predicted power and the measured power, n' is the length of the power time series, P i , P' i are the measured power and the initial predicted power at time i, respectively;

[0039] The corrected initial predicted power includes: using the ARIMA(p, d, q) model to estimate the time series of the power prediction deviation E p , and calculating the corrected predicted power P″ p according to the time series of E i , and the specific calculation formula is:

[0040]

[0041] In the formula, φ i is the autoregressive term coefficient,, representing the influence of the previous observations of the time series; p is the number of autoregressive terms; q is the number of moving average terms; d is the number of differences required for series stationarity; L is the lag operator L(y t ) = y t-1 , L k is the k-th power of the lag operator, representing yt Lagged by k time steps, L k (y t ) = y t-k ; θ i is the coefficient of the moving average term, representing the influence of the previous q random shocks on the current value; ε t is the random error term.

[0042] Furthermore, in the data output, the calculation formula for the output qualification rate is:

[0043]

[0044] where N' 总 is the total number of output time points within the prediction time period, N' 缺测点 is the total number of missing points within the prediction time period, N' 离群点 is the total number of outliers within the prediction time period; when the qualification rate is less than 95%, a warning is issued, and when it is greater than or equal to 90%, the power prediction result is output to the power prediction workstation through the reverse isolation device and sent to the dispatching center by the power prediction workstation;

[0045] The outliers within the prediction time period are the mutation points in the sliding correlation coefficient r between the corrected predicted wind speed and the corrected predicted power. The calculation formula for the sliding correlation coefficient r is:

[0046]

[0047] In the formula, V″ i and P″ i are the values of the time series of the corrected predicted wind speed and the corrected predicted power within a sliding window with a width of k, and are the average values of the points within the corresponding sliding window; during the forward sliding of the detection window, the points where the sliding correlation coefficient is less than 0.9 are recorded as mutation points.

[0048] The present invention also discloses a two-stage short-term wind power prediction device, including:

[0049] Data input module: Collect the initial predicted wind speed, measured wind speed, and actual power at a set sampling interval to obtain input data;

[0050] Data verification module: Detect whether the input data is qualified. If it is qualified, perform data preprocessing; otherwise, issue a warning;

[0051] Predicted wind speed correction module: Calculate the predicted wind speed deviation according to the initial predicted wind speed and the measured wind speed after preprocessing using the first system deviation formula; correct the initial predicted wind speed according to the predicted wind speed deviation to obtain the corrected predicted wind speed;

[0052] Power prediction and correction module: Predict the wind speed according to the correction, and calculate the initial predicted power using the improved Richards model; Calculate the power prediction deviation according to the initial predicted power and the actual power using the second system deviation formula; Correct the initial predicted power according to the power prediction deviation to obtain the corrected predicted power;

[0053] Data output module: Detect whether the corrected predicted power is qualified, and if it is qualified, output the corrected predicted power at a set interval.

[0054] As can be seen from the above technical solutions, compared with the prior art, the present invention discloses a two-stage short-term wind power prediction method and device. Based on the improved Richards model, while ensuring the basic S-shaped characteristics of the wind power curve, it takes into account the uncertainty of the unit operating state. At the same time, through two-stage system deviation correction, it suppresses the deviation propagation in the wind power prediction process, enhances the stability and on-site applicability of the power prediction result while ensuring the prediction accuracy; Moreover, the device proposed by the present invention is an independent module, which is connected to the power prediction system through a forward and reverse isolation device, ensuring the security of key data and being compatible with the existing prediction system. The present invention is applicable to both single-unit and whole-station short-term wind power prediction correction, can improve the prediction accuracy and stability, increase the electricity sales revenue of the wind farm, and reduce the grid operation cost and operation risk. Description of the Drawings

[0055] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained according to the provided drawings.

[0056] Figure 1 It is a schematic diagram of the overall process of the prediction method in the embodiment of the present invention.

[0057] Figure 2 It is a schematic diagram of the overall structure of the prediction device in the embodiment of the present invention.

[0058] Figure 3 It is a schematic diagram of the implementation effect in the embodiment of the present invention. Detailed Embodiments

[0059] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0060] An embodiment of the present invention discloses a two-stage short-term wind power prediction method, as Figure 1 shown, the specific steps are as follows:

[0061] Data input: According to the set sampling interval, collect the initial predicted wind speed, measured wind speed, and actual power to obtain the input data;

[0062] Data verification: Detect whether the input data is qualified. If it is qualified, perform preprocessing; otherwise, issue a warning.

[0063] Predicted wind speed correction: According to the preprocessed initial predicted wind speed and measured wind speed, use the first system deviation formula to calculate the predicted wind speed deviation; correct the initial predicted wind speed according to the predicted wind speed deviation to obtain the corrected predicted wind speed;

[0064] Power prediction and correction: According to the corrected predicted wind speed, use the improved Richards model to calculate the initial predicted power; according to the initial predicted power and actual power, use the second system deviation formula to calculate the power prediction deviation; correct the initial predicted power according to the power prediction deviation to obtain the corrected predicted power;

[0065] Data output: Detect whether the corrected predicted power is qualified. If it is qualified, output the corrected predicted power at the set interval.

[0066] In a specific embodiment, as Figure 2 shown, the initial predicted wind speed and measured wind speed are input by a weather forecasting workstation located in security zone III / IV through a reverse isolation device, and the actual power is input by a power prediction workstation located in security zone II through a forward isolation device, so as to ensure the security of key data and be compatible with the existing prediction system.

[0067] In a specific embodiment, in data verification, the qualification rate calculation formula is:

[0068]

[0069] where N 总 is the total number of sampling points within the detection time period, N 缺测点 is the total number of missing measurement points within the detection time period, N 离群点 is the total number of outliers within the detection time period; when the qualification rate is greater than or equal to 90%, data preprocessing is performed; otherwise, a warning is issued.

[0070] The outliers within the detection time period are the mutation points of the sliding correlation coefficient r between the measured wind speed and the actual power. The sliding correlation coefficient r VP is as defined in VP The calculation formula is:

[0071]

[0072] where V i and P i are the values of the time series of the measured wind speed and the actual power within a sliding window of width k respectively. and are the average values of the points within the corresponding sliding window. During the process of the detection window sliding forward, the points where the sliding correlation coefficient is less than 0.9 are recorded as mutation points.

[0073] In a specific embodiment, the preprocessing is as follows: the replacement values of the unqualified data points are generated by using the adjacent weighted average method, and the unqualified data points are replaced.

[0074] The calculation formula of the replacement value is:

[0075]

[0076] where y is the replacement value, y -2 , y -1 , y 1 and y 2 are the values of the 4 points adjacent to the unqualified point y 0 , w -2 , w -1 , w 1 and w 2 are the weight values of the corresponding adjacent points. After data verification, the time series of the initial predicted wind speed V', the measured wind speed V, and the actual power P are obtained.

[0077] In a specific embodiment, the first system deviation formula is:

[0078]

[0079] where E is the system deviation between the initial predicted wind speed and the measured wind speed, n is the length of the wind speed time series, V i , V' i are the measured wind speed and the initial predicted wind speed at the i-th moment respectively;

[0080] Correcting the initial predicted wind speed includes: using the ARIMA(p,d,q) model to estimate the time series of the wind speed prediction deviation E w , and calculating the corrected predicted wind speed V″ according to the time series of E w and the initial predicted wind speedi , the specific calculation formula is:

[0081]

[0082] In the formula, φ i is the autoregressive term coefficient, representing the influence of previous observations of the time series; p is the number of autoregressive terms; q is the number of moving average terms; d is the number of differences required for series stationarity; L is the lag operator L(y t ) = y t-1 , L k is the k - th power of the lag operator, representing that y t lags k time steps, L k (y t ) = y t-k ; θ i is the moving average term coefficient, representing the influence of the previous q random shocks on the current value; ε t is the random error term. According to the historical wind speed of the input data, predict L k (E w ), and output L(E w ), and adopt the automatic fitting method to determine the values of the model parameters p, d, q, and establish the ARIMA(p, d, q) model. Substitute L k-1 (y t ) back into the model, and the predicted value can be obtained, which is the value of the time series of E w at the i - th moment.

[0083] Specifically, first, in the stage of evaluating data stationarity, perform at least one unit root test on the time series data of the wind speed prediction deviation E w . If the series is not stationary, perform successive difference operations on the original data until a stationary series is obtained; the number of difference operations performed is determined as the difference - order parameter d. Then, set the candidate ranges of the autoregressive term p and the moving average term q, generally positive integers in [1, 5]; search for the parameter combination, and find the p, q combination when the residual is minimized during the process of fitting historical data. The residual calculation formula is:

[0084]

[0085] In the formula, N is the number of points of the fitting historical data.

[0086] In a specific embodiment, the calculation formula for the initial predicted power is:

[0087]

[0088] In the formula, P' i represents the initial predicted power according to the i - th moment, P maxis the rated power of the fan, a is the shape parameter, e is the natural constant, V' i ' is the corrected predicted wind speed, v 0 is the cut-in wind speed of the fan, K is the scale function, which is used to characterize the uncertainty of the unit output at different wind speeds. Its expression is:

[0089]

[0090] In the formula, σ is the scale parameter, and x is the fluctuation range of the wind speed;

[0091] Among them, the determination methods of the undetermined parameters σ and a are as follows:

[0092] Set the candidate ranges of σ and a. Generally, the shape parameter a is between [0.5, 2], and σ is between [0, 3]. Search for the parameter combination, the combination of σ and a with the minimum energy loss in the process of fitting historical data. The energy loss calculation formula is:

[0093]

[0094] It means that in the initial prediction process, the prediction accuracy of the low wind speed section is abandoned, and more attention is paid to the prediction accuracy of the medium and high wind speed sections and the periods with more generated electric energy. In particular, for a wind farm with non-uniform turbine models, the turbines are classified first according to the cut-in wind speed, and then the initial power prediction is carried out separately according to the categories.

[0095] The second system deviation formula is:

[0096]

[0097] In the formula, E' is the system deviation between the initial predicted power and the measured power, n' is the length of the power time series, P i , P' i are the measured power and the initial predicted power at the i-th moment respectively;

[0098] Correct the initial predicted power, including: using the ARIMA(p, d, q) model to estimate the time series of the power prediction deviation E p According to the time series of E p and the initial predicted power, calculate the corrected predicted power P″ i , and the specific calculation formula is:

[0099]

[0100] In the formula, φ i is the autoregressive term coefficient, which represents the influence of the previous observed values of the time series; p is the number of autoregressive terms; q is the number of moving average terms; d is the number of differences required for the series to be stationary; L is the lag operator L(y t ) = y t-1 , Lk is the k-th power of the lag operator, representing y t lags k time steps, L k (y t ) = y t-k ; θ i is the coefficient of the moving average term, representing the impact of the previous q random shocks on the current value; ε t is the random error term. Predict L based on the historical power of the input data k (E p ), and output L(E p ). Using the automatic fitting method, determine the values of the model parameters p, d, q, and establish the ARIMA(p, d, q) model. Substitute L k-1 (y t ) back into the model to obtain the predicted value is the value of the time series of E p at the i-th moment.

[0101] Specifically, the values of the model parameters p, d, q are determined by the automatic fitting method. First, in the stage of evaluating data stationarity, perform at least one unit root test on the time series data of the power prediction deviation E p . If the series is not stationary, perform successive differencing operations on the original data until a stationary series is obtained; the number of differencing operations performed is determined as the differencing order parameter d. Then, set the candidate ranges of the autoregressive term p and the moving average term q, generally positive integers in [1, 5]; search for the parameter combination, and find the p, q combination when the residual is minimized during the process of fitting historical data. The residual calculation formula is:

[0102]

[0103] where N' is the number of points of the fitting historical data.

[0104] In a specific embodiment, in the data output, the calculation formula for the output qualification rate is:

[0105]

[0106] where N' 总 is the total number of output time points within the prediction time period, N' 缺测点 is the total number of missing points within the prediction time period, N' 离群点 is the total number of outliers within the prediction time period; when the qualification rate is less than 95%, a warning is issued, and when it is greater than or equal to 90%, the power prediction result is output to the power prediction workstation through the reverse isolation device and sent to the dispatching center by the power prediction workstation;

[0107] Outliers within the prediction time period are the mutation points in the sliding correlation coefficient r of the corrected predicted wind speed and the corrected predicted power. The calculation formula for the sliding correlation coefficient r is as follows:

[0108]

[0109] In the formula, V' i ' and P' i ' are the values of the time series of the corrected predicted wind speed and the corrected predicted power in a sliding window with a width of k, and are the average values of the points within the corresponding sliding window; during the process of the detection window sliding forward, the points where the sliding correlation coefficient is less than 0.9 are recorded as mutation points.

[0110] In a specific embodiment, the implementation effect of a two-stage short-term wind power prediction method is as Figure 3 shown. After the prediction is optimized by the method of the present invention, the RMSE error drops by 6.3%; the systematic deviation before optimization is 3.10 MW, and the systematic deviation after optimization is -0.13 MW.

[0111] The present invention also discloses a two-stage short-term wind power prediction device, including:

[0112] Data input module: Collect the initial predicted wind speed, measured wind speed, and actual power at a set sampling interval to obtain input data;

[0113] Data verification module: Detect whether the input data is qualified. If it is qualified, perform data preprocessing; otherwise, issue a warning;

[0114] Predicted wind speed correction module: According to the preprocessed initial predicted wind speed and measured wind speed, use the first system deviation formula to calculate the predicted wind speed deviation; correct the initial predicted wind speed according to the predicted wind speed deviation to obtain the corrected predicted wind speed;

[0115] Power prediction and correction module: According to the corrected predicted wind speed, use the improved Richards model to calculate the initial predicted power; according to the initial predicted power and the actual power, use the second system deviation formula to calculate the power prediction deviation; correct the initial predicted power according to the power prediction deviation to obtain the corrected predicted power;

[0116] Data output module: Detect whether the corrected predicted power is qualified. If it is qualified, output the corrected predicted power at a set interval.

[0117] In this specification, the various embodiments are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same or similar parts among the various embodiments, reference can be made to each other. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple. For the relevant parts, reference can be made to the description in the method section.

[0118] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the broadest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for predicting wind power under short - term scales, characterized in that, the specific steps are as follows: Data input: According to the set sampling interval, collect the initial predicted wind speed, measured wind speed, and actual power to obtain input data; Data verification: Detect whether the input data is qualified. If it is qualified, perform pre - processing; otherwise, issue a warning; Predicted wind speed correction: According to the pre - processed initial predicted wind speed and measured wind speed, use the first system deviation formula to calculate the predicted wind speed deviation; correct the initial predicted wind speed according to the predicted wind speed deviation to obtain the corrected predicted wind speed; Power prediction and correction: According to the corrected predicted wind speed, use the improved Richards model to calculate the initial predicted power; according to the initial predicted power and actual power, use the second system deviation formula to calculate the power prediction deviation; correct the initial predicted power according to the power prediction deviation to obtain the corrected predicted power; Data output: Detect whether the corrected predicted power is qualified. If it is qualified, output the corrected predicted power at the set interval; wherein, the first system deviation formula is: Where E is the systematic deviation between the initial predicted wind speed and the measured wind speed, n is the length of the wind speed time series, V i 、V' i are the measured wind speed and the initial predicted wind speed at time i respectively; Calibrate the initial predicted wind speed, including: using the ARIMA(p,d,q) model to estimate the wind speed prediction deviation E w of the time series, and perform at least one unit root test on the time series data of the wind speed prediction deviation E w If the time series is not stationary, perform successive differencing operations on the original data until a stationary series is obtained. The number of differencing operations performed is determined as the number of differences d required for series stationarization; set the number of autoregressive terms p and the number of moving average terms q as positive integers in [1,5]; search for parameter combinations, and estimate the p and q combinations when the residual is minimized during the fitting of historical data. The residual calculation formula is: In the formula, N is the number of points for fitting historical data; According to E w calculate the corrected predicted wind speed V″ based on the time series and the initial predicted wind speed i , and the specific calculation formula is as follows: where φ i is the autoregressive term coefficient, representing the influence of previous observations of the time series; p is the number of autoregressive terms; q is the number of moving average terms; d is the number of differences required for series stationarity; L is the lag operator L(y t ) = y t-1 , L k is the k-th power of the lag operator, representing y t lagging k time steps, L k (y t ) = y t-k ; θ i is the moving average term coefficient, representing the influence of the previous q random shocks on the current value; ε t is the random error term.

2. A method for predicting wind power under short - term scales according to claim 1, characterized in that, the initial predicted wind speed and measured wind speed are input by a weather forecasting workstation in safety zone III / IV through a reverse isolation device, and the actual power is input by a power prediction workstation in safety zone II through a forward isolation device.

3. A method for predicting wind power under short - term scales according to claim 1, characterized in that, in data verification, the qualification rate calculation formula is: Among them, N 总 is the total number of sampling points within the detection time period, N 缺测点 is the total number of missing measurement points within the detection time period, N 离群点 is the total number of outliers within the detection time period; when the pass rate is greater than or equal to 90%, data preprocessing is performed, otherwise a warning is issued; The outliers within the detection time period are the mutation points in the sliding correlation coefficient r between the measured wind speed and the actual power, and the calculation formula for the sliding correlation coefficient r is as follows: VP VP ​​ where V i and P i are the values of the time series of the measured wind speed and the actual power in a sliding window with a width of k, and are the average values of the points within the corresponding sliding window; during the process of the detection window sliding forward, the points where the sliding correlation coefficient is less than 0.9 are recorded as mutation points.

4. A method for predicting wind power under short - term scales according to claim 1, characterized in that, the pre - processing is: adopting the adjacent weighted average method to generate replacement values for unqualified data points and replace the unqualified data points; the calculation formula for the replacement value is: where y is the replacement value, y -2 , y -1 , y 1 and y 2 are the values of the 4 points adjacent to the nonconforming point y 0 , w -2 , w -1 , w 1 and w 2 are the weight values corresponding to the adjacent points.

5. A method for predicting wind power under short - term scales according to claim 1, characterized in that, the calculation formula for the initial predicted power is: Wherein, P' i represents the initial predicted power at the i-th moment, P max is the rated power of the wind turbine, a is the shape parameter, e is the natural constant, v″ u is the corrected predicted wind speed, v 0 is the cut-in wind speed of the wind turbine, K is the scale function used to characterize the uncertainty of the unit output under different wind speeds, and its expression is: In the formula, σ is the scale parameter, and x is the fluctuation range of the wind speed; the second system deviation formula is: where E' is the system deviation between the initial predicted power and the measured power, n' is the length of the power time series, and P i , P' i are the measured power and the initial predicted power at the i-th moment, respectively; Calibrate the initial predicted power, including: using the ARIMA(p,d,q) model to estimate the power prediction deviation E p of the time series, and calculating the calibrated predicted power P″ according to the time series of E p and the initial predicted power i , and the specific calculation formula is: where φ i is the autoregressive term coefficient, representing the influence of previous observations of the time series; p is the number of autoregressive terms; q is the number of moving average terms; d is the number of differences required for series stationarity; L is the lag operator L(y t ) = y t-1 , L k is the k-th power of the lag operator, representing y t lagging k time steps, L k (y t ) = y t-k ; θ i is the moving average term coefficient, representing the influence of the previous q random shocks on the current value; ε t is the random error term.

6. A method for predicting wind power under short - term scales according to claim 5, characterized in that, the determination method for the scale parameter σ and the shape parameter a is: set the candidate ranges of σ and a, the shape parameter a is between [0.5, 2], and the scale parameter σ is between [0, 3]; search for the parameter combination, the combination of σ and a with the minimum energy loss during the process of fitting historical data, and the energy loss calculation formula is: Among them, for a wind farm with non - uniform turbine models, first classify the turbines according to the cut - in wind speed, and then perform initial power prediction for each category separately.

7. A method for predicting wind power under short - term scales according to claim 5, characterized in that, the values of the autoregressive term number p, the moving average term number q, and the number of differences d required for sequence stationarity are determined by an automatic fitting method, specifically: Perform at least one unit root test on the time series data of the power prediction deviation E p If the sequence is not stationary, perform successive differencing operations on the original data until a stationary sequence is obtained; the number of differencing operations performed is determined as the number of differences d required for sequence stationarization; Set the candidate ranges of the number of autoregressive terms \(p\) and the number of moving average terms \(q\) as positive integers in \([1, 5]\); search for the parameter combination, that is, the \(p, q\) combination when the residual is minimized during the process of fitting historical data. The formula for the residual is: In the formula, \(N'\) is the number of points of the fitted historical data.

8. A method for predicting wind power to meet the short-term scale according to claim 1, characterized in that, In the data output, the formula for calculating the qualification rate is: Among them, N' 总 is the total number of output time points within the prediction time period, and N' 缺测点 is the total number of missing points within the prediction time period, and N' 离群点 is the total number of outliers within the prediction time period; when the qualification rate is less than 95%, a warning is issued, and when it is greater than or equal to 90%, the power prediction result is output to the power prediction workstation through the reverse isolation device and sent to the dispatching center by the power prediction workstation; The outlier points within the prediction time period are the mutation points in the sliding correlation coefficient \(r\) of the corrected predicted wind speed and the corrected predicted power. The formula for the sliding correlation coefficient \(r\) is: where V' i ' and P' i ' are the values of the time series of the corrected predicted wind speed and the corrected predicted power in a sliding window of width k, and is the average value of the points within the corresponding sliding window; during the process of the detection window sliding forward, the points where the sliding correlation coefficient is less than 0.9 are recorded as mutation points.

Images