A wind speed prediction and correction method and system for wind farms based on wind speed-error feature fusion

Abstract

This invention discloses a wind speed prediction correction method and system for wind farms based on wind speed-error feature fusion. The method includes: extracting two types of error distribution features based on historical windows: the quantile features of prediction errors under the wind speed conditions at the prediction time and the quantile statistical features of the error sequence in the lookback time window, explicitly characterizing the dynamic change law of the error; then, constructing a wind speed prediction correction model, where two Bi-LSTM encoders extract the temporal features of wind speed and error information respectively, and concatenate them with the error distribution features at the time step, achieving feature fusion through linear mapping; then, an LSTM decoder takes the fused features and the future predicted wind speed as input, and outputs the corrected wind speed prediction result, realizing the correction of the day-ahead wind speed prediction deviation. By introducing wind speed condition error modeling and the wind speed-error feature fusion mechanism, the accuracy and robustness of wind speed prediction at wind farms are improved, providing higher quality wind speed data support for wind power prediction and system scheduling.

Description

Technical Field

[0001] This invention belongs to the field of wind speed prediction technology for wind farms, and particularly relates to a wind speed prediction and correction method and system for wind farms based on wind speed-error feature fusion. Background Technology

[0002] The accuracy of wind speed forecasting directly affects the precision of wind power forecasting, and is a crucial link in ensuring the safety of power grid dispatch and the capacity for renewable energy absorption. During wind speed forecasting, due to the high complexity of atmospheric motion and the influence of various factors such as topography, underlying surface characteristics, and observation errors, numerical weather prediction (NWP) products often exhibit systematic biases. This leads to significant differences between predicted and actual wind speed values, and the forecast errors show obvious time-series characteristics and cumulative effects.

[0003] Traditional wind speed forecasting correction methods typically rely on historical observation data and employ statistical regression, machine learning, or deep learning models to correct forecast biases. These methods primarily use wind speed forecast values ​​or forecast errors for modeling, neglecting the conditional distribution characteristics of errors as wind speed conditions change. This results in insufficient adaptability of the models under complex weather conditions and limited correction effectiveness. In particular, in multi-step forecasting tasks, the problem of error accumulation becomes more prominent, making it difficult to consistently guarantee forecast accuracy.

[0004] Furthermore, wind speed prediction errors depend not only on the current wind speed conditions but also on the systematic bias characteristics of the prediction model itself and the trend of weather evolution. Modeling methods that rely solely on wind speed or historical errors cannot fully explore the intrinsic relationship between the evolution of prediction errors and future wind speed trends.

[0005] Therefore, there is an urgent need for a correction method that can comprehensively consider the changes in wind speed prediction information and error characteristics, especially in multi-step prediction scenarios. This method should be able to dynamically correct prediction errors through feature fusion modeling, thereby improving the overall accuracy and robustness of wind speed prediction. Summary of the Invention

[0006] This invention provides a wind farm wind speed prediction and correction method and system based on wind speed-error feature fusion, which solves the technical problems of insufficient modeling of the conditional dependence between wind speed state and prediction error and limited correction effect in existing wind speed prediction and correction methods.

[0007] In a first aspect, the present invention provides a wind farm wind speed prediction and correction method based on wind speed-error feature fusion, comprising:

[0008] Collect measured wind speeds at the hub of the target wind farm at 5 minutes for D historical sample days. Predicted wind speed at the wheel hub And calculate the single-step prediction error at each time step. , To form an error sequence ,in, This represents the total number of time steps in the sample.

[0009] For the measured wind speed Predicted wind speed and error sequence The measured wind speed was normalized to the [0,1] interval using the max-min normalization method to obtain the normalized wind speed. Normalized predicted wind speed and the normalized error sequence ;

[0010] Extract the time points before the start of the measurement respectively time step Error conditional distribution characteristics and lookback window The error quantile statistical characteristics are obtained, and the error conditional distribution characteristics and the error quantile statistical characteristics are concatenated to obtain the prediction error feature sequence. ;

[0011] The normalized measured wind speed sequence Error sequence Predicted wind speed sequence and prediction error feature sequence The data is input into a preset wind speed prediction and correction model, which outputs the wind speed prediction and correction results for the wind farm. ,in, For the length of the historical retrospective window, This is for predicting the time window length.

[0012] Secondly, the present invention provides a wind farm wind speed prediction and correction system based on wind speed-error feature fusion, comprising:

[0013] The data acquisition module is configured to collect the measured wind speed at the hub level for 5 minutes over D historical sample days at the target wind farm. Predicted wind speed at the wheel hub And calculate the single-step prediction error at each time step. , To form an error sequence ,in, This represents the total number of time steps in the sample.

[0014] The normalization module is configured to normalize the measured wind speed. Predicted wind speed and error sequence The measured wind speed was normalized to the [0,1] interval using the max-min normalization method to obtain the normalized wind speed. Normalized predicted wind speed and the normalized error sequence ;

[0015] The stitching module is configured to extract data before the start time point. time step Error conditional distribution characteristics and lookback window The error quantile statistical characteristics are obtained, and the error conditional distribution characteristics and the error quantile statistical characteristics are concatenated to obtain the prediction error feature sequence. ;

[0016] The output module is configured to output the normalized measured wind speed sequence. Error sequence Predicted wind speed sequence and prediction error feature sequence The data is input into a preset wind speed prediction and correction model, which outputs the wind speed prediction and correction results for the wind farm. ,in, For the length of the historical retrospective window, This is for predicting the time window length.

[0017] Thirdly, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the wind farm wind speed prediction and correction method based on wind speed-error feature fusion according to any embodiment of the present invention.

[0018] Fourthly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor performs the steps of the wind farm wind speed prediction and correction method based on wind speed-error feature fusion according to any embodiment of the present invention.

[0019] The wind speed prediction and correction method and system for wind farms based on wind speed-error feature fusion proposed in this application have the following beneficial effects:

[0020] 1. By jointly modeling the error and wind speed, a conditional probability density function of the prediction error under wind speed conditions is constructed, and multi-quantile statistical features are extracted to explicitly reflect the distribution trend of the prediction error with wind speed, thereby enhancing the model's ability to understand the dynamic relationship between wind speed and error.

[0021] 2. By using a bidirectional LSTM structure to encode the historical measured wind speed and prediction error sequences respectively, and combining the error distribution quantile features, deep fusion of multi-source features is achieved through feature concatenation and linear mapping. The wind speed prediction error is dynamically corrected in the decoding stage, and the context information is effectively used to improve the correction effect.

[0022] 3. The error distribution characteristics under the current wind speed conditions are dynamically introduced at each prediction time step, enabling the model to update the correction strategy over time, and showing stronger robustness and consistency in multi-step prediction scenarios.

[0023] 4. By introducing error probability distribution modeling under wind speed conditions and combining it with a multi-source time series feature fusion mechanism, this invention fully considers the dynamic coupling relationship between error and wind speed in the modeling stage, enabling the model to have stronger error identification and correction capabilities. Attached Figure Description

[0024] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 A flowchart of a wind farm wind speed prediction and correction method based on wind speed-error feature fusion provided in an embodiment of the present invention;

[0026] Figure 2 A flowchart for extracting error conditional distribution features and error quantile statistical features according to a specific embodiment of the present invention is provided.

[0027] Figure 3 A flowchart of a wind speed prediction and correction model is provided for one embodiment of the present invention;

[0028] Figure 4 A flowchart of a wind speed prediction and correction method according to a specific embodiment of the present invention is provided.

[0029] Figure 5 A bar chart of evaluation indexes for each model in a specific embodiment of the present invention is provided;

[0030] Figure 6 A structural block diagram of a wind farm wind speed prediction and correction system based on wind speed-error feature fusion is provided in an embodiment of the present invention.

[0031] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] Please see Figure 1 The diagram shows a flowchart of a wind farm wind speed prediction and correction method based on wind speed-error feature fusion according to this application.

[0034] like Figure 1 As shown, the wind speed prediction and correction method for wind farms based on wind speed-error feature fusion specifically includes the following steps:

[0035] Step S101: Collect the measured wind speed at the hub of the target wind farm at 5 minutes for D historical sample days. Predicted wind speed at the wheel hub And calculate the single-step prediction error at each time step. , To form an error sequence ,in, This represents the total number of time steps for the sample.

[0036] Step S102, measure the wind speed Predicted wind speed and error sequence The measured wind speed was normalized to the [0,1] interval using the max-min normalization method to obtain the normalized wind speed. Normalized predicted wind speed and the normalized error sequence .

[0037] Step S103: Extract the values ​​before the start time point. time step Error conditional distribution characteristics and lookback window The error quantile statistical characteristics are obtained, and the error conditional distribution characteristics and the error quantile statistical characteristics are concatenated to obtain the prediction error feature sequence. .

[0038] In this step, such as Figure 2 As shown, the start time point is defined. The length before time is The backtracking window, for the backtracking window Each historical point in time Extract the predicted wind speed and the corresponding prediction error to obtain the error sample pair set. ,in, For the time points within the backtracking window Predicted wind speed, For the time points within the backtracking window The prediction error;

[0039] In the error sample pair set The difference between the predicted wind speed value at the current time and the value at the last moment shall not exceed a threshold. Historical predicted wind speeds were used to obtain a subset of error sample pairs. ;

[0040] Based on the one-dimensional kernel density estimation, an error sample pair subset is established at the current time. Predicted wind speed Conditional probability density function under given conditions The expression is:

[0041] ,

[0042] In the formula, Let be the conditional probability density function, representing the probability density function given the current time. Predicted wind speed Under these conditions, prediction error The probability density, For prediction error, For the number of samples, The bandwidth parameter for kernel density estimation. For kernel functions;

[0043] conditional probability density function Integrating yields the cumulative distribution function. , Let be the conditional cumulative distribution function, representing the conditional cumulative distribution function at a given current time. Measured wind speed Under these conditions, prediction error The cumulative probability distribution, Let be the conditional probability density function. For the current moment The measured wind speed, As the integral variable, the 25th, 50th, and 75th quantiles are then calculated to represent the error distribution trend under the current predicted wind speed, thus forming the error conditional distribution characteristic vector. The expression is:

[0044] ,

[0045] ,

[0046] In the formula, Let m be the 25th percentile level of the prediction error corresponding to similar predicted wind speeds within the backtracking window. Given the inverse cumulative distribution function, calculate the error value corresponding to the 0.25 quantile level. The prediction error at time m corresponds to the 50th percentile level of the prediction error for similar predicted wind speeds within the backtracking window. Given the inverse cumulative distribution function, calculate the error value corresponding to the 0.5 quantile level. The prediction error at time m corresponds to the 75th percentile of similar predicted wind speeds within the backtracking window. Given the inverse cumulative distribution function, calculate the error value corresponding to the 0.75 quantile level;

[0047] For the start time Extraction length is Prediction error sequence within the lookback window This constitutes a subset of errors. and the error subset Extracting the error quantile statistical feature vector ;

[0048] The normalized error conditional distribution eigenvector Sum of error quantile statistical eigenvectors By splicing the images together, the predicted time can be obtained. Prediction error feature sequence The expression is:

[0049] ,

[0050] In the formula, This represents the 25th percentile level of the normalized lookback window prediction error. This represents the 50th percentile level of the normalized lookback window prediction error. This represents the 75th percentile level of the normalized lookback window prediction error.

[0051] It should be noted that the aforementioned subset of errors Extracting the error quantile statistical feature vector include:

[0052] Subset of errors Sort in ascending order to obtain the subset of errors within the lookback window. The sequence sorted in ascending order The expression is:

[0053] ,

[0054] In the formula, This indicates that the subset of errors within the looking-back window will be considered. Sort in ascending order;

[0055] The 25th percentile, 50th percentile, and 75th percentile are extracted from the sorted sequence as error quantile statistical features according to their proportional positions. The expression is as follows:

[0056] ,

[0057] In the formula, Indicates rounding down. For the subset of errors within the retrospective window The ranking error is 25% in the ascending order sequence. For the subset of errors within the retrospective window The rank error is 50% in the sequence after ascending order. For the subset of errors within the retrospective window 75% error in the sequence after ascending order;

[0058] The error quantile statistical characteristic vector at prediction time m is composed of the 25th, 50th, and 75th quantiles. .

[0059] Step S104: Normalize the measured wind speed sequence Error sequence Predicted wind speed sequence and prediction error feature sequence The data is input into a preset wind speed prediction and correction model, which outputs the wind speed prediction and correction results for the wind farm. .

[0060] In this step, please refer to Figure 3 The wind speed prediction and correction model includes an encoder module, a feature fusion module, and a decoder module.

[0061] in, For the length of the historical retrospective window, To predict the time window length, the normalized measured wind speed sequence is used. Error sequence Predicted wind speed sequence and prediction error feature sequence The data is input into a preset wind speed prediction and correction model, which outputs the wind speed prediction and correction results for the wind farm. include:

[0062] The measured wind speed sequence was analyzed separately. and error sequence Feature extraction is performed to obtain the full sequence features of the error. Wind speed full-series characteristics The expression is:

[0063] ,

[0064] ,

[0065] In the formula, This is a normalized error sequence for the bidirectional LSTM-encoded lookback window. This is a bidirectional LSTM-encoded lookback window normalized measured wind speed sequence. Let be a one-dimensional real number space, representing a space of length . , This represents the number of hidden units in the forward and backward directions of the Bi-LSTM.

[0066] Specifically, Bidirectional Long Short-Term Memory (Bi-LSTM) is a recurrent neural network architecture used to model time-series data. It processes sequences simultaneously in both forward and reverse time order, thus capturing contextual dependencies more comprehensively. Bi-LSTM consists of two LSTM sub-networks that independently extract features in the forward and reverse time directions, respectively, given the current input. The hidden state of the previous moment The hidden state at the next moment Its forward LSTM generates a sequence of hidden states: Inverse LSTM generates the hidden state sequence: The final output is the result of concatenating the two: This structure can capture information from both the past and the future, making it more suitable for modeling complex historical sequence features. Therefore, Bi-LSTM is used to encode the temporal features of historical wind speed sequences and historical error sequences, respectively.

[0067] The error full sequence features extracted by the encoder module Wind speed full-series characteristics Repeatedly replicated along the time dimension Next, extending to the prediction window length, the expression is:

[0068] ,

[0069] In the formula, To extract the encoder module Repeated copying along the time dimension Second-rate, To extract the encoder module Repeated copying along the time dimension Second-rate, To predict the time window length, The back-lookout window prediction error features extracted from the encoder. The lookback window error features extracted from the encoder;

[0070] Will , With prediction error feature sequence The input features are spliced ​​together according to time steps to form a fused input feature. Then, feature fusion is performed based on linear transformation to obtain fused features. The expression is:

[0071] ,

[0072] ,

[0073] ,

[0074] In the formula, The weight matrix is ​​the linear transformation matrix. This is the bias term for the linear transformation;

[0075] Fusion features With predicted wind speed sequence By performing time-step concatenation, a fused feature sequence is obtained. And as input to the decoder module, the expression is:

[0076] ,

[0077] The fused feature sequence is processed by a single-layer LSTM unit in the decoder module. The hidden state sequence is obtained, and then mapped to the wind speed prediction and correction result of the wind farm through a fully connected mapping layer. The expression is:

[0078] ,

[0079] In the formula, Output the hidden state sequence for the LSTM. Processing fused feature sequences for LSTM , This is the weight matrix of the fully connected mapping layer. This is the bias term for the fully connected mapping layer.

[0080] Specifically, Long Short-Term Memory (LSTM) networks are a variant of Recurrent Neural Networks (RNNs) designed to address the problem of long-term dependencies in sequential data. Compared to traditional RNNs, LSTMs effectively alleviate the vanishing gradient problem and are suitable for modeling complex time series such as wind power variation patterns. An LSTM network consists of multiple cascaded LSTM units, each updating the input sequence at each time step t. Given the current input... The hidden state of the previous moment Memory state Output the current hidden state. and memory state .

[0081] In summary, please refer to the method and process outlined in this application. Figure 4 First, historical measured and predicted wind speed data from wind farms are collected, and the prediction error is calculated. Second, based on the current predicted wind speed status and historical windows, two types of error distribution features are extracted: the quantile features of the prediction error under the wind speed conditions at the prediction time and the quantile statistical features of the error sequence in the lookback window, explicitly characterizing the dynamic changes in the error. Then, a wind speed prediction correction model is constructed. Two Bi-LSTM encoders extract the temporal features of wind speed and error information, respectively, and concatenate them with the error distribution features at the time step, achieving feature fusion through linear mapping. Finally, the LSTM decoder takes the fused features and the future predicted wind speed as input and outputs the corrected wind speed prediction result, thus correcting the day-ahead wind speed prediction deviation. By introducing wind speed condition error modeling and a wind speed-error feature fusion mechanism, the accuracy and robustness of wind speed prediction at wind farms are improved, providing higher-quality wind speed data support for wind power prediction and system scheduling.

[0082] In one specific embodiment, to verify the application effect of the prediction model of the present invention in actual engineering, the following comparison method was selected for experimental verification:

[0083] BiLSTM-LSTM-Full: This paper proposes a method that uses a Bi-LSTM encoder and an LSTM decoder to integrate historical measured wind speed, historical prediction error, predicted wind speed, as well as conditional error quantile features and error statistical features, and uses them as a complete correction model.

[0084] BiLSTM-LSTM-NoErr: Only input historical measured wind speed and predicted wind speed, remove all error information (including historical error sequences and error distribution characteristics), and is used to evaluate the impact of error input on model performance.

[0085] BiLSTM-LSTM-NoErrCDF: While inputting historical measured wind speed and predicted wind speed, it retains the historical error sequence, but does not input the conditional quantile features and quantile statistical features of the error. It only models based on the original error value to verify the role of distribution information.

[0086] LSTM-LSTM-Full: Replaces the Bi-LSTM in the encoder with a unidirectional LSTM, while keeping the rest of the structure and input unchanged. It is used to analyze the impact of the bidirectional LSTM structure on the modeling effect.

[0087] GRU-LSTM-Full: The encoder is replaced with a GRU, while the decoder remains an LSTM. All inputs are kept unchanged to verify the effectiveness of the model used in this paper.

[0088] TCN-LSTM-Full: The encoder is replaced with TCN, while the decoder remains LSTM. All inputs are kept unchanged to verify the effectiveness of the model used in this paper.

[0089] The comparison between BiLSTM-LSTM-Ful and BiLSTM-LSTM-NoErrCDF in the above comparative models aims to verify the contribution of error-related characteristics to the wind speed prediction correction performance. (See Table 1 and...) Figure 5 As can be seen, the BiLSTM-LSTM-Full model significantly outperforms the model with reduced error features across all metrics, improving MAPE by 2.79 and 2.34 percentage points, RMSE by 40.00% and 20.35%, and MAE by 43.42% and 22.81%. This result demonstrates that introducing the original historical sequence of prediction errors, as well as the conditional quantile and quantile statistical features of the errors, is crucial for improving prediction correction performance. It helps the model characterize the error structure from multiple perspectives, capture the evolution of errors, and thus adjust the predicted wind speed more accurately. This verifies the effectiveness of introducing conditional quantile and statistical features of errors in improving prediction correction accuracy.

[0090] The comparison of BiLSTM-LSTM-Full with LSTM-LSTM-Full, GRU-LSTM-Full, and TCN-LSTM-Full models above aims to verify the impact of encoder structure selection on model performance. (See Table 1 and...) Figure 5As can be seen, the BiLSTM-LSTM-Full model slightly outperforms other variants in terms of MAPE, RMSE, and MAE. Specifically, the MAPE performance is improved by approximately 6.61, 7.05, and 6.57 percentage points compared to the LSTM, GRU, and TCN models, respectively; the RMSE performance is improved by approximately 1.72%, 3.93%, and 1.17%; and the MAE performance is improved by approximately 1.53%, 7.86%, and 1.53%. This result demonstrates that the Bi-LSTM encoder possesses a stronger ability to capture temporal dependencies in historical information modeling, and can simultaneously utilize past and future contextual information. This helps generate more stable and high-quality wind speed feature representations, further enhancing the decoder's ability to correct future errors, thus validating the effectiveness of the model selected in this paper.

[0091] The comparison between BiLSTM-LSTM-Full and BiLSTM-LSTM-NoErr in the above comparative models aims to verify the role of the error sequence in wind speed prediction correction and to validate the effectiveness of the joint error-wind speed modeling proposed in this paper. (From Table 1 and...) Figure 5 As can be seen, after removing the error input, the improvement in MAPE decreased to 2.19 percentage points, RMSE decreased by only 17.54%, and MAE decreased by 20.18%, showing a significant performance degradation compared to the complete model that incorporates error information. This indicates that wind speed prediction error itself is a variable with significant pattern and temporal correlation. Explicitly modeling it helps the model better capture the generation mechanism and changing trend of the deviation, thereby making more accurate corrections for future wind speeds. This verifies the effectiveness of the joint modeling of wind speed and power in this paper, effectively improving the prediction correction effect.

[0092] Table 1

[0093] ,

[0094] Among them, MAPE (Mean Absolute Percentage Error) represents the average percentage of the error between the predicted and actual values, and is used to measure the accuracy of the prediction.

[0095] RMSE (Root Mean Square Error): The root mean square error. It represents the square root of the average of the squares of the differences between the predicted and actual values, and is used to measure the accuracy of the prediction.

[0096] MAE (Mean Absolute Error): The average absolute error. It represents the average absolute value of the error between the predicted and actual values, and is used to measure the average magnitude of the prediction error.

[0097] Please see Figure 6The diagram shows a structural block diagram of a wind farm wind speed prediction and correction system based on wind speed-error feature fusion according to this application.

[0098] like Figure 6 As shown, the wind speed prediction and correction system 200 for wind farms includes a data acquisition module 210, a normalization module 220, a splicing module 230, and an output module 240.

[0099] Among them, the data acquisition module 210 is configured to collect the measured wind speed at the hub level for 5 minutes of D historical sample days of the target wind farm. Predicted wind speed at the wheel hub And calculate the single-step prediction error at each time step. , To form an error sequence ,in, This represents the total number of time steps in the sample.

[0100] Normalization module 220, configured to normalize measured wind speed Predicted wind speed and error sequence The measured wind speed was normalized to the [0,1] interval using the max-min normalization method to obtain the normalized wind speed. Normalized predicted wind speed and the normalized error sequence ;

[0101] The splicing module 230 is configured to extract data before the start time point. time step Error conditional distribution characteristics and lookback window The error quantile statistical characteristics are obtained, and the error conditional distribution characteristics and the error quantile statistical characteristics are concatenated to obtain the prediction error feature sequence. ;

[0102] Output module 240 is configured to output the normalized measured wind speed sequence. Error sequence Predicted wind speed sequence and prediction error feature sequence The data is input into a preset wind speed prediction and correction model, which outputs the wind speed prediction and correction results for the wind farm. .

[0103] It should be understood that Figure 6 The modules and references described in the document Figure 1 The steps described in the text correspond to those in the method described above. Therefore, the operations, features, and corresponding technical effects described above also apply to the method described in the text. Figure 6 The various modules in the document will not be described in detail here.

[0104] In other embodiments, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor performs the wind speed prediction and correction method for wind farms based on wind speed-error feature fusion in any of the above method embodiments.

[0105] In one embodiment, the computer-readable storage medium of the present invention stores computer-executable instructions, which are configured as follows:

[0106] Collect measured wind speeds at the hub of the target wind farm at 5 minutes for D historical sample days. Predicted wind speed at the wheel hub And calculate the single-step prediction error at each time step. , To form an error sequence ,in, This represents the total number of time steps in the sample.

[0107] For the measured wind speed Predicted wind speed and error sequence The measured wind speed was normalized to the [0,1] interval using the max-min normalization method to obtain the normalized wind speed. Normalized predicted wind speed and the normalized error sequence ;

[0108] Extract the time points before the start of the measurement respectively time step Error conditional distribution characteristics and lookback window The error quantile statistical characteristics are obtained, and the error conditional distribution characteristics and the error quantile statistical characteristics are concatenated to obtain the prediction error feature sequence. ;

[0109] The normalized measured wind speed sequence Error sequence Predicted wind speed sequence and prediction error feature sequence The data is input into a preset wind speed prediction and correction model, which outputs the wind speed prediction and correction results for the wind farm. .

[0110] Computer-readable storage media may include a stored program area and a stored data area, wherein the stored program area may store an operating system and an application program required for at least one function; the stored data area may store data created based on the use of the wind farm wind speed prediction and correction system based on wind speed-error feature fusion. Furthermore, the computer-readable storage medium may include high-speed random access memory, and may also include memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, the computer-readable storage medium may optionally include memory remotely disposed relative to a processor, which can be connected to the wind farm wind speed prediction and correction system based on wind speed-error feature fusion via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0111] Figure 7 This is a schematic diagram of the structure of the electronic device provided in the embodiment of the present invention, such as... Figure 7 As shown, the device includes a processor 310 and a memory 320. The electronic device may also include an input device 330 and an output device 340. The processor 310, memory 320, input device 330, and output device 340 can be connected via a bus or other means. Figure 7 Taking a bus connection as an example, the memory 320 is the computer-readable storage medium described above. The processor 310 executes various server functions and data processing by running non-volatile software programs, instructions, and modules stored in the memory 320, thereby implementing the wind speed prediction and correction method for wind farms based on wind speed-error feature fusion as described in the above embodiment. The input device 330 can receive input digital or character information and generate key signal inputs related to user settings and function control of the wind farm wind speed prediction and correction system based on wind speed-error feature fusion. The output device 340 may include a display screen or other display device.

[0112] The aforementioned electronic device can execute the method provided in the embodiments of the present invention, and has the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this embodiment can be found in the method provided in the embodiments of the present invention.

[0113] In one implementation, the above-described electronic device is applied to a wind farm wind speed prediction and correction system based on wind speed-error feature fusion, and is used as a client. It includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to:

[0114] Collect measured wind speeds at the hub of the target wind farm at 5 minutes for D historical sample days. Predicted wind speed at the wheel hub And calculate the single-step prediction error at each time step. , To form an error sequence ,in, This represents the total number of time steps in the sample.

[0115] For the measured wind speed Predicted wind speed and error sequence The measured wind speed was normalized to the [0,1] interval using the max-min normalization method to obtain the normalized wind speed. Normalized predicted wind speed and the normalized error sequence ;

[0116] Extract the time points before the start of the measurement respectively time step Error conditional distribution characteristics and lookback window The error quantile statistical characteristics are obtained, and the error conditional distribution characteristics and the error quantile statistical characteristics are concatenated to obtain the prediction error feature sequence. ;

[0117] The normalized measured wind speed sequence Error sequence Predicted wind speed sequence and prediction error feature sequence The data is input into a preset wind speed prediction and correction model, which outputs the wind speed prediction and correction results for the wind farm. .

[0118] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0119] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A wind farm wind speed prediction rectification method based on wind speed-error characteristic fusion, characterized in that, Comprising: Collecting 5-minute level measured wind speed at hub of D historical sample days of target wind farm station and predicted wind speed at hub and calculating single-step prediction error of each time , forming error sequence wherein, is total time step of sample observed wind speed , predicted wind speed and error sequence The maximum-minimum normalization method is used to normalize the observed wind speed , predicted wind speed and error sequence to the interval [0, 1]. Extract the time points before the start of the measurement respectively time step Error conditional distribution characteristics and lookback window The error quantile statistical characteristics are obtained, and the error conditional distribution characteristics and the error quantile statistical characteristics are concatenated to obtain the prediction error feature sequence. ; a normalized measured wind speed sequence , an error sequence , a predicted wind speed sequence , and a predicted error feature sequence is input into a preset wind speed prediction rectification model, and the wind speed prediction rectification model outputs a wind farm station wind speed prediction rectification result , wherein, is a historical lookback window length, is a prediction time window length, and the wind speed prediction rectification model comprises an encoder module, a feature fusion module, and a decoder module; The normalized measured wind speed sequence , an error sequence , a predicted wind speed sequence , and a predicted error feature sequence is input into a preset wind speed prediction correction model, and the wind speed prediction correction model outputs a wind farm station wind speed prediction correction result comprises: The measured wind speed sequence and the error sequence are respectively subjected to feature extraction, to obtain error full-sequence features and wind speed full-sequence features ; error full sequence features extracted by the encoder module wind speed full sequence features repeatedly copied along the time dimension times, extending to the prediction time window length; Will , and the prediction error feature sequence , spliced by time steps, forming a fusion input feature , and then performing feature fusion according to linear transformation to obtain a fusion feature ; concatenate the fusion features with the predicted wind speed sequence to obtain a fusion feature sequence as input of the decoder module; The fusion feature sequence is processed by a single-layer LSTM unit in a decoder module to obtain a hidden state sequence , and the hidden state sequence is mapped to a wind farm station wind speed prediction correction result via a fully connected mapping layer .

2. The wind farm wind speed prediction deviation correction method based on wind speed-error characteristic fusion according to claim 1, characterized in that, The error condition distribution feature and the error quantile statistical feature of the time step before the respective extraction starting time point The error condition distribution feature and the error quantile statistical feature of the time step before the respective extraction starting time point The error condition distribution feature and the error quantile statistical feature of the time step before the respective extraction starting time point The error condition distribution feature and the error quantile statistical feature of the time step before the respective extraction starting time point The error condition distribution feature and the error quantile statistical feature of the time step before the respective extraction starting time point Define a time point for starting measurement A backtracking window with a length of Before the time point, for each historical time point In the backtracking window , extract the predicted wind speed and the predicted error corresponding to the predicted wind speed, to obtain an error sample pair set , wherein, is the predicted wind speed at the time point in the backtracking window, is the predicted error at the time point in the backtracking window; In the error sample pair set , the historical predicted wind speed whose difference with the predicted wind speed value of the current time is not more than a threshold value is reserved, to obtain an error sample pair sub-set . ; According to a one-dimensional kernel density estimation to establish a conditional probability density function of the error sample pair subset at the current time under the predicted wind speed of the current time , expressed as: ， wherein is the conditional probability density function, representing the probability density of the prediction error given the predicted wind speed at the current time instant is the prediction error, is the prediction error, is the number of samples, is the bandwidth parameter of the kernel density estimation, is the kernel function; conditional probability density function Integrating yields the cumulative distribution function. , Let be the conditional cumulative distribution function, representing the conditional cumulative distribution function at a given current time. Measured wind speed Under these conditions, prediction error The cumulative probability distribution, Let be the conditional probability density function. For the current moment The measured wind speed, As the integral variable, the 25th, 50th, and 75th quantiles are then calculated to represent the error distribution trend under the current predicted wind speed, thus forming the error conditional distribution characteristic vector. The expression is: ， ， wherein, is the 25% quantile level of the prediction error corresponding to the similar prediction wind speed of the forecasted wind speed at time m in the lookback window, is the inverse cumulative distribution function, calculating the error value corresponding to the 0.25 quantile level, is the 50% quantile level of the prediction error corresponding to the similar prediction wind speed of the forecasted wind speed at time m in the lookback window, is the inverse cumulative distribution function, calculating the error value corresponding to the 0.5 quantile level, is the 75% quantile level of the prediction error corresponding to the similar prediction wind speed of the forecasted wind speed at time m in the lookback window, is the inverse cumulative distribution function, calculating the error value corresponding to the 0.75 quantile level. For the start time Extraction length is Prediction error sequence within the lookback window This constitutes a subset of errors. and the error subset Extracting the error quantile statistical feature vector ; the normalized error condition feature vector and the error quantile statistic feature vector perform splicing to obtain a prediction error feature sequence at the prediction time , and the expression is: ， wherein Q25is the 25th percentile level of the normalized lookback window prediction error, Q50is the 50th percentile level of the normalized lookback window prediction error, Q75is the 75th percentile level of the normalized lookback window prediction error.

3. The wind farm wind speed prediction deviation correction method based on wind speed-error characteristic fusion according to claim 2, characterized in that, The pair error subset The error sub-set is extracted Comprises: the subset of errors arranged in ascending order, obtaining a subset of errors within the look-back time window the sequence arranged in ascending order , the expression being ， In the formula, denotes the subset of error values within the lookback window arranged in ascending order; The 25% quantile, 50% quantile and 75% quantile are extracted from the proportional position of the sorting sequence as the error quantile statistical features, and the expression is: ， wherein denotes rounding down, error subset within the lookback window the 25% ranked errors in the sorted sequence in ascending order, error subset within the lookback window the 50% ranked errors in the sorted sequence in ascending order, error subset within the lookback window the 75% ranked errors in the sorted sequence in ascending order. The error quantile statistical feature vector of the prediction time m is composed of the 25% quantile, the 50% quantile, and the 75% quantile .

4. A wind farm wind speed prediction correction system based on wind speed-error characteristic fusion, characterized in that, Comprising: The collection module is configured to collect 5-minute-level measured wind speed at the hub of D historical sample days of a target wind power station and the predicted wind speed at the hub , and calculate the single-step prediction error of each time , , form an error sequence , is the total number of sample time steps a normalization module configured to normalize the measured wind speed , the predicted wind speed and the error sequence to the interval [0, 1] to obtain normalized measured wind speed , normalized predicted wind speed and normalized error sequence ; The stitching module is configured to extract data before the start time point. time step Error conditional distribution characteristics and lookback window The error quantile statistical characteristics are obtained, and the error conditional distribution characteristics and the error quantile statistical characteristics are concatenated to obtain the prediction error feature sequence. ; an output module configured to input the normalized measured wind speed sequence , the error sequence , the predicted wind speed sequence , and the predicted error feature sequence into a preset wind speed prediction rectification model, wherein the wind speed prediction rectification model outputs a wind farm station wind speed prediction rectification result , wherein is a historical lookback window length, is a prediction time window length, and the wind speed prediction rectification model comprises an encoder module, a feature fusion module, and a decoder module. The normalized measured wind speed sequence , an error sequence , a predicted wind speed sequence , and a predicted error feature sequence is input into a preset wind speed prediction correction model, and the wind speed prediction correction model outputs a wind farm station wind speed prediction correction result comprises: The measured wind speed sequence and the error sequence are respectively subjected to feature extraction, to obtain error full-sequence features and wind speed full-sequence features ; error full sequence features extracted by the encoder module wind speed full sequence features repeatedly copied along the time dimension extended to the prediction time window length; Will , and the prediction error feature sequence , spliced by time steps, forming a fusion input feature , and then performing feature fusion according to linear transformation to obtain a fusion feature ; concatenate the fusion features with the predicted wind speed sequence to obtain a fusion feature sequence as input of the decoder module; The fusion feature sequence is processed by a single-layer LSTM unit in a decoder module to obtain a hidden state sequence , and the hidden state sequence is mapped to a wind farm station wind speed prediction correction result via a fully connected mapping layer .

5. An electronic device, comprising: Comprising: At least one processor, and a memory connected with the at least one processor in communication, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1 to 3.

6. A computer-readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to implement the method of any one of claims 1 to 3.

Images