A Wind Power Prediction Method and System for Abnormal Weather Conditions

Abstract

The present invention relates to a wind power prediction method and system for abnormal weather conditions. The method includes determining the unit shutdown risk; determining the wind speed characterization matrix of the wind farm; constructing an abnormal weather feature evaluation system; based on the abnormal weather feature evaluation system, performing abnormal weather identification and classification based on rolling AP clustering; estimating the degree of output limitation according to the unit shutdown risk and the output data set under abnormal weather conditions; performing data augmentation by using a deep convolutional generative adversarial network according to the output data set and random variables under abnormal weather conditions; reconstructing the data set according to the augmented data, the output data set under normal weather conditions, and the estimated degree of output limitation; training a deep time series network model according to the reconstructed data set; and determining the prediction result of the wind power according to the trained deep time series network model and the abnormal weather identification and classification result. The present invention can improve the reliability of wind power prediction under abnormal weather conditions.

Description

Technical Field

[0001] The present invention relates to the field of power prediction, and particularly to a wind power prediction method and system for abnormal weather conditions. Background Art

[0002] With the continuous expansion of the scale of the wind power industry, the problems brought about by the high proportion of wind power grid connection have become increasingly prominent. Accurately predicting wind power and incorporating it into the dispatching plan are important prerequisites for improving the wind power acceptance capacity of the receiving-end power grid and enhancing the operational safety and economy of the power system. However, under abnormal weather conditions such as sandstorm-rainstorm-icing, the current wind power prediction method still has deficiencies in estimating the degree of restriction of the unit output, making it difficult to ensure the prediction accuracy and reliability under such weather conditions, and it is difficult to provide continuous and reliable inertial response and frequency support for the power grid.

[0003] The current wind power prediction methods are generally divided into physical model methods and statistical model methods. Numerical weather prediction is a typical representative of the physical model prediction method. This method mainly converts wind energy into kinetic energy through information such as the topography, temperature, and air pressure near the wind farm, and then converts it into power through the power curve of the wind turbine. The statistical model method is a method that discovers the potential relationship between power generation and meteorological information through neural networks and other means to achieve prediction. The main application models include: machine learning models such as support vector machines and XGBoost models, as well as traditional deep learning models such as artificial neural networks and long short-term neural networks (LSTM). Through the above models, feature extraction and dimensionality reduction are performed on the original data, and dynamic modeling of the power output of the wind farm is carried out to achieve the prediction of wind power.

[0004] However, the above technologies mainly rely on historical operation data with a long time scale to train a model with strong generalization ability. The average training error is the smallest under long-term historical operation data, mainly for the prediction scenario of the output of wind turbines under normal wind speed conditions. However, for the prediction scenario of abnormal weather conditions such as sandstorm-thunderstorm-icing, there are still the following problems:

[0005] 1. Under abnormal weather conditions, the accuracy of the wind farm's wind measurement device is easily affected by changes in the types and concentrations of atmospheric aerosol particles, and even fails suddenly due to factors such as icing and sandstorms, increasing the uncertainty of power prediction.

[0006] 2. Under abnormal weather conditions, the meteorological parameters such as wind speed and temperature fluctuate greatly and have poor regularity, resulting in weak recognition ability and low classification accuracy of the current prediction technology for abnormal weather conditions, and it is difficult to conduct research and modeling on abnormal weather condition data.

[0007] 3. In the historical operation data of the wind farm, the proportion of abnormal weather condition data is relatively low. It is difficult for the prediction model to fully analyze the distribution law of a small number of samples and establish an accurate mapping relationship, increasing the prediction error.

[0008] 4. Under abnormal weather conditions, the output of the unit is affected by factors such as mechanical damage and changes in the blade cross-sectional shape. The actual output characteristics will change, resulting in phenomena such as a significant reduction in output or even shutdown. The current technology has not considered the impact of such factors on power prediction.

[0009] To solve the above problems, there is an urgent need for a wind power prediction method or system for abnormal weather conditions. Summary of the Invention

[0010] The object of the present invention is to provide a wind power prediction method and system for abnormal weather conditions, which can improve the reliability of wind power prediction under abnormal weather conditions.

[0011] To achieve the above object, the present invention provides the following solutions:

[0012] A wind power prediction method for abnormal weather conditions, comprising:

[0013] Detect outliers in the unit wind speed data to determine abnormal data; determine the data anomaly ratio according to the abnormal data; and determine the unit shutdown risk according to the data anomaly ratio; the unit wind speed data includes: unit wind speed, anemometer tower wind speed, and incoming flow direction;

[0014] Correct the abnormal data; determine the average wind speed at different unit positions according to the incoming flow direction; and determine the wind farm wind speed characterization matrix according to the average wind speed at different unit positions and the corrected unit wind speed data; the unit positions include: front-section units, middle-section units, and rear-section units;

[0015] Construct an abnormal weather feature evaluation system according to the wind farm wind speed characterization matrix, unit rotation speed, power, pitch angle, temperature, and humidity; the abnormal weather feature evaluation system includes: theoretical power deviation amplitude, tip speed, blade temperature, and pitch angle change amplitude;

[0016] According to the abnormal weather feature evaluation system, identify and classify abnormal weather based on rolling AP clustering to determine the output data set under normal weather conditions and the output data set under abnormal weather conditions;

[0017] Estimate the degree of output limitation according to the unit shutdown risk and the output data set under abnormal weather conditions;

[0018] According to the output data set under abnormal weather conditions and random variables, use a deep convolutional generative adversarial network for data augmentation;

[0019] Reconstruct the data set according to the augmented data, the output data set under normal weather conditions, and the estimated degree of output limitation;

[0020] Train a deep time series network model based on the reconstructed dataset; the trained deep time series network model is used to output the wind farm power probability;

[0021] Determine the prediction result of wind power according to the trained deep time series network model and the abnormal weather recognition and classification results.

[0022] Optionally, the outlier detection of the unit wind speed data to determine abnormal data; determine the data anomaly ratio according to the abnormal data; and determine the unit shutdown risk according to the data anomaly ratio, specifically including:

[0023] Use the formula Determine the data anomaly ratio;

[0024] Use the formula Determine the unit shutdown risk;

[0025] where d e is the data anomaly ratio, l is the number of measurement time nodes within the time range, m is the total number of units in the wind farm, is whether the unit i is missing at time t, if missing, it is 1, otherwise, it is 0, WT t i is the unit shutdown risk of unit i at time t, and D is the shutdown threshold.

[0026] Optionally, the construction of an abnormal weather feature evaluation system according to the wind farm wind speed characterization matrix, unit speed, power, pitch angle, temperature, and humidity, specifically including:

[0027] Use the formula Determine the theoretical power deviation amplitude;

[0028] Use the formula Determine the tip speed;

[0029] Use the formula Determine the blade temperature;

[0030] Use the formula Δθ t =θ t -θ t-1 Determine the pitch angle change amplitude;

[0031] where Δp is the theoretical power deviation amplitude, v blade is the tip speed, T blade is the blade temperature, Δθ t is the pitch angle change amplitude, p actual is the actual output power of the wind farm, C P is the wind energy utilization coefficient, θ is the average pitch angle of the units in the field, R is the blade length of the unit, v is the incoming flow wind speed, ω is the wind turbine angular velocity, T avis the measured temperature of the wind farm, P av is the measured pressure of the wind farm, θ t and θ t-1 are the pitch angles at the current time t and the previous time t - 1 respectively, and ρ is the air density.

[0032] Optionally, according to the abnormal weather feature evaluation system, the output data sets under normal weather conditions and abnormal weather conditions are determined based on the abnormal weather recognition and classification of rolling AP clustering, specifically including:

[0033] According to the abnormal weather feature evaluation system, calculate the weather feature similarity at different times with the Pearson coefficient as the index;

[0034] Divide the abnormal weather feature evaluation system into h parts, and use the median of the weather similarity in each part as the self-similarity;

[0035] Perform AP clustering on each part of the data to obtain the clustering centers;

[0036] According to the clustering centers, determine the new user set; if the new user set is greater than the set value, return to the step of dividing the abnormal weather feature evaluation system into h parts and using the median of the weather similarity in each part as the self-similarity; otherwise, initialize the self-similarity of the new user set.

[0037] Perform AP clustering on the new user set to obtain the secondary clustering centers and the number of clusters;

[0038] Calculate the clustering quality index according to the secondary clustering centers and the number of clusters;

[0039] Update the self-similarity;

[0040] Judge whether the number of clusters is 2. If it is 2, compare the clustering quality indexes corresponding to different numbers of clusters and determine the best number of clusters to complete the abnormal weather recognition and classification; otherwise, return to the step of performing AP clustering on the new user set to obtain the secondary clustering centers and the number of clusters.

[0041] Optionally, the estimation of the output limitation degree according to the unit shutdown risk and the output data set under abnormal weather conditions specifically includes:

[0042] Use the formula to determine the estimated output limitation degree of the unit;

[0043] where F is the estimated output limitation degree of the unit, m is the total number of units in the wind farm, WT t i is the unit shutdown risk of unit i at time t, K is the weather feature index, if the weather feature is abnormal weather, the value is 0, otherwise, it is 1.

[0044] Optionally, determining the prediction result of wind power according to the trained deep time series network model and the abnormal weather recognition and classification result specifically includes:

[0045] Obtain the real-time data of the wind farm and the set of abnormal weather feature evaluation indicators;

[0046] According to the real-time data of the wind farm and the set of abnormal weather feature evaluation indicators, use the trained deep time series network model to determine the wind farm power probability;

[0047] Use the abnormal weather recognition and classification result to correct the wind farm power probability and determine the prediction result of the wind power.

[0048] A wind power prediction system for abnormal weather conditions, which is applied to the above-mentioned wind power prediction method for abnormal weather conditions. The system includes:

[0049] A unit wind speed data detection module, which is used to detect abnormal values of the unit wind speed data, determine abnormal data; determine the data abnormality ratio according to the abnormal data; and determine the unit shutdown risk according to the data abnormality ratio; the unit wind speed data includes: unit wind speed, anemometer tower wind speed, and incoming flow wind direction;

[0050] A wind farm wind speed characterization matrix determination module, which is used to correct the abnormal data; determine the average wind speed at different unit positions according to the incoming flow wind direction; and determine the wind farm wind speed characterization matrix according to the average wind speed at different unit positions and the corrected unit wind speed data; the unit positions include: front-section units, middle-section units, and rear-section units;

[0051] An abnormal weather feature evaluation system construction module, which is used to construct an abnormal weather feature evaluation system according to the wind farm wind speed characterization matrix, unit rotation speed, power, pitch angle, temperature, and humidity; the abnormal weather feature evaluation system includes: theoretical power deviation amplitude, tip speed, blade temperature, and pitch angle change amplitude;

[0052] An abnormal weather recognition and classification result determination module, which is used to determine the output data set under normal weather conditions and the output data set under abnormal weather conditions based on the abnormal weather recognition and classification based on rolling AP clustering according to the abnormal weather feature evaluation system;

[0053] An output limitation degree estimation module, which is used to estimate the output limitation degree according to the unit shutdown risk and the output data set under abnormal weather conditions;

[0054] A data augmentation module, which is used to perform data augmentation according to the output data set under abnormal weather conditions and random variables by using a deep convolutional generative adversarial network;

[0055] A dataset reconstruction module, configured to reconstruct a dataset according to the augmented data, the output dataset under normal weather conditions, and the estimation of the output limitation degree;

[0056] A trained deep time-series network model determination module, configured to train a deep time-series network model according to the reconstructed dataset; the trained deep time-series network model is used to output the wind farm power probability;

[0057] A prediction result determination module, configured to determine the prediction result of the wind power according to the trained deep time-series network model and the abnormal weather recognition and classification result.

[0058] Optionally, the unit wind speed data detection module specifically includes:

[0059] A data anomaly ratio determination unit, configured to use the formula to determine the data anomaly ratio;

[0060] A unit shutdown risk determination unit, configured to use the formula to determine the unit shutdown risk;

[0061] where d e is the data anomaly ratio, l is the number of measurement time nodes within the time range, m is the total number of units in the wind farm, is whether the unit i is missing at time t, if missing, it is 1, otherwise it is 0, WT t i is the unit shutdown risk of the unit i at time t, and D is the shutdown threshold.

[0062] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0063] A wind power prediction method and system for abnormal weather conditions provided by the present invention solve the problem of a relatively high probability of inaccurate and invalid wind measurement data of wind turbines in abnormal weather conditions through the detection and correction of outliers. By combining meteorological factors with state parameters such as the rotational speed and power of the unit through an abnormal weather feature evaluation system, the impact of weather conditions on the unit can be further identified, thus providing richer effective indicators for the degree of output limitation and the identification of abnormal weather conditions. The weather recognition and classification method based on rolling AP clustering makes up for the disadvantage of low accuracy of the clustering results of AP clustering when the sample size is large, and realizes the accurate discrimination of abnormal weather conditions. The abnormal weather operation data expansion method using a deep convolutional generative adversarial network (DCGAN) improves the fitting quality of the data distribution law through a convolutional network and realizes the training of the data generator through the generative adversarial network structure. The quantile correction link is carried out using the recognition and classification results of abnormal weather. The lower limit of the power prediction with a 95% confidence level is used as the output value of the predicted power during abnormal weather periods, further ensuring the reliability of the predicted power in abnormal weather conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] 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 use in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0065] Figure 1 Schematic flow chart of a wind power prediction method for abnormal weather conditions provided by the present invention;

[0066] Figure 2 Schematic overall flow chart of a wind power prediction method for abnormal weather conditions provided by the present invention;

[0067] Figure 3 Schematic diagram of the structure of a deep convolutional generative adversarial network;

[0068] Figure 4 Schematic diagram of the structure of a wind power prediction system for abnormal weather conditions provided by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0069] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to 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 shall fall within the protection scope of the present invention.

[0070] The object of the present invention is to provide a wind power prediction method and system for abnormal weather conditions, which can improve the reliability of wind power prediction under abnormal weather conditions.

[0071] To make the above objects, features and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0072] Figure 1 It is a schematic flowchart of a wind power prediction method for abnormal weather conditions provided by the present invention. Figure 2 It is a schematic overall flowchart of a wind power prediction method for abnormal weather conditions provided by the present invention. As Figure 1 and Figure 2 shown, the wind power prediction method for abnormal weather conditions provided by the present invention includes:

[0073] S101. Detect outliers in the unit wind speed data to determine abnormal data; determine the data abnormality ratio according to the abnormal data; and determine the unit shutdown risk according to the data abnormality ratio; the unit wind speed data includes: unit wind speed, anemometer tower wind speed, and incoming flow direction.

[0074] S101 solves the problem that the probability of inaccurate and invalid anemometry data of wind turbines is relatively large under abnormal weather conditions. And by horizontally comparing the real-time data of each unit among the units and vertically comparing with the historical measured data, the detection of outliers is realized. If there are abnormal unit measurement values at many moments, an abnormal risk alarm information of the unit operation state is reported.

[0075] S101 specifically includes:

[0076] Judge whether there are outliers and missing values in the wind turbine wind speed, anemometer tower wind speed, and incoming flow direction data. Calculate the data missing length, that is, the time range covered by the missing data, and at the same time calculate the number of units with abnormal data at the same moment in the wind farm.

[0077] Use the formula to determine the data abnormality ratio.

[0078] Use the formula to determine the unit shutdown risk.

[0079] where de is the data anomaly ratio, l is the number of measurement time nodes within the time range, and m is the total number of units in the wind farm. indicates whether unit i is missing at time t. If it is missing, it is 1; otherwise, it is 0, WT t i is the unit shutdown risk of unit i at time t, and D is the shutdown threshold.

[0080] S102. Correct the abnormal data; determine the average wind speed at different unit positions according to the incoming wind direction; and determine the wind farm wind speed characterization matrix based on the average wind speed at different unit positions and the corrected unit wind speed data. The unit positions include: front-segment units, middle-segment units, and rear-segment units.

[0081] S102 considers the wake effect and the incoming wind direction to perform real-time characterization of the wind farm wind speed state, reduces the wind speed information at a large number of units to three categories, further eliminates potential measurement errors, and reduces the input data dimension.

[0082] Use the formula to determine the wind farm wind speed characterization matrix

[0083] where are the number of front-segment units, the number of middle-segment units, and the number of rear-segment units under the wind direction dir, respectively.

[0084] S103. Construct an abnormal weather feature evaluation system based on the wind farm wind speed characterization matrix, unit speed, power, pitch angle, temperature, and humidity. The abnormal weather feature evaluation system includes: theoretical power deviation amplitude, tip speed, blade temperature, and pitch angle change amplitude.

[0085] S103 specifically includes:

[0086] Use the formula to determine the theoretical power deviation amplitude. The theoretical power deviation amplitude is used to characterize the degree of output limitation.

[0087] Use the formula to determine the tip speed. The tip speed is used to characterize air resistance.

[0088] Use the formula to determine the blade temperature. The blade temperature is used to evaluate the probability of blade icing in rain and snow weather.

[0089] Use the formula Δθ t =θ t -θ t-1 to determine the pitch angle change amplitude. The pitch angle change amplitude is used to characterize the change amplitude of the optimal pitch angle.

[0090] where Δp is the theoretical power deviation amplitude, v blade is the tip speed, T blade is the blade temperature, Δθ t is the pitch angle change amplitude, p actual is the actual output power of the wind farm, C P is the wind energy utilization coefficient, θ is the average pitch angle of the units in the farm, R is the blade length of the unit, v is the incoming flow wind speed, ω is the wind turbine angular velocity, T av is the measured temperature of the wind farm, P av is the measured pressure of the wind farm, θ t and θ t-1 are the pitch angles at the current time t and the previous time t - 1 respectively, and ρ is the air density.

[0091] S104. According to the abnormal weather feature evaluation system, determine the output data set under normal weather conditions and the output data set under abnormal weather conditions based on the abnormal weather recognition and classification of rolling AP clustering. Rolling AP clustering can make up for the shortcoming of the low accuracy of the clustering result of AP clustering when the sample scale is large, and achieve the accurate discrimination of the abnormal weather state. Rolling AP clustering mainly consists of two layers of AP clustering and the evaluation of clustering quality indicators. The specific operation steps are as follows:

[0092] Step 4.1: Calculate the weather feature similarity s ρ (A i , A j ) at different times with the Pearson coefficient as the index. The calculation method is as follows:

[0093]

[0094] In the formula, A i and A j are the weather features at two times to be calculated, and are respectively the c-th element in A i and A j , and are the weather feature means, and Γ is the number of weather features.

[0095] Step 4.2: Divide the data set into h parts, and use the median of the weather similarity in each part as the self-similarity.

[0096]

[0097] In the formula, is the j-th user in the H γ -th part.

[0098] Step 4.3: Perform AP clustering on each part of the data to obtain the clustering centers

[0099] Step (a): Calculate the similarity of each part and the responsibility degree

[0100]

[0101] In the formula, is the data to be calculated, is the data to be used as the data center

[0102] Step (b): Adjust the data center, and iteratively calculate the responsibility degree and credibility of the clustering result. The structure of the N γ th iteration is as follows

[0103]

[0104] In the formula, λ is the damping factor, which is used to accelerate the convergence speed of AP clustering

[0105]

[0106] Step (c): Judge whether the iteration of the responsibility degree and credibility of the data set is stable. If it is stable, calculate the clustering center set by the following formula If then the number of stable times Otherwise

[0107]

[0108] Step (d): If N γ reaches the maximum set number of times, go to step (e). Otherwise, N γ = N γ + 1, and return to step (b).

[0109] Step (e): Obtain the clustering center set E new = [E1, E1,..., E B .

[0110] Step 4.4: Establish a new user set from the clustering center set. If the number of sets is greater than the set value, go to step 4.2. Otherwise, go to step 4.5

[0111] Step 4.5: Initialize the user self-similarity, and let where is the jth data sequence in E new and is the jth data sequence in E newThe initialized median value.

[0112] Step 4.6: Perform AP clustering on E new to obtain stable secondary clustering centers and the number of clusters

[0113] Step 4.7: Calculate the clustering quality index from the clustering results:

[0114]

[0115] where d out (A i ) is the average distance between classes, and d in (A i ) is the average distance between data within a class.

[0116] Step 4.8: Update the user self-similarity using Equation (8).

[0117]

[0118] where μ m is the similarity median and Q is the number of clusters.

[0119] Step 4.9: Determine whether the number of clusters is 2. If so, go to Step 4.10; otherwise, return to Step 4.6.

[0120] Step 4.10: Compare the clustering quality indices corresponding to different numbers of clusters and determine the optimal number of clusters c * .

[0121] c * = argmax Q Z Q,av (10)

[0122] Finally, divide the dataset into the output dataset under abnormal weather conditions and the output dataset under normal weather conditions based on the clustering results.

[0123] S105. Estimate the degree of output limitation according to the unit shutdown risk and the output dataset under abnormal weather conditions.

[0124] S105 specifically includes:

[0125] Use the formula to determine the estimated degree of output limitation of the unit.

[0126] where F is the estimated degree of output limitation of the unit, m is the total number of units in the wind farm, and WT t iThe unit shutdown risk of unit i at time t, K is the weather characteristic index. If the weather characteristic is abnormal weather, the value is 0; otherwise, it is 1.

[0127] S106. According to the output data set and random variables under abnormal weather conditions, use a deep convolutional generative adversarial network for data augmentation.

[0128] As Figure 3 shown, the deep convolutional generative adversarial network is an improvement of GAN. It introduces a convolutional network into the structure of GAN to improve the quality of the data generated by GAN based on the strong feature extraction ability of the convolutional network.

[0129] The deep convolutional generative adversarial network takes the output data set under abnormal weather conditions as real data and inputs it into the DCGAN. Secondly, the generator samples from the random noise distribution and maps through the deep convolutional network, and its output result is

[0130] Set the discrimination condition of the discriminator as: when the input is real data, the discrimination result is 1; when the input data is a sample of the generator, the discrimination result is 0, that is and

[0131]

[0132] In the formula, is the activation function, x and z are real data and random noise samples, W D , b D , W G , b G are the weight and bias parameters of the generative adversarial network and the discriminator network.

[0133] The optimization goal of the generator is to deceive the discriminator as much as possible so that the output result corresponding to the pseudo sample is 1. The optimization goal of the discriminator is to correctly distinguish whether the data comes from real samples or pseudo samples from the generator. Define the optimization objective function as:

[0134] In the formula, P data represents the real data distribution, P z represents the sampling noise distribution. When the two tend to be consistent, the DCGAN reaches the optimal solution.

[0135] Based on the DCGAN network, input random variables to generate operation data under abnormal weather conditions.

[0136] S107. According to the augmented data, the output data set under normal weather conditions, and the output limitation prediction, perform data set reconstruction.

[0137] S108. Train the deep time series network model LSTM based on the reconstructed data set. The trained deep time series network model is used to output the wind farm power probability.

[0138] By introducing structures such as forget gates, LSTM selectively stores information and extends the time series information learning range of the recurrent neural network.

[0139] One layer of the LSTM neural network consists of a set of recurrent structures, and each recurrent structure includes: an input gate, an output gate, a forget gate, and a cell state module.

[0140] The input gate determines how much of the data currently input into the neural network can be saved to the cell state. By inputting all the input parameters x at the current time t and the output parameter h of the hidden state at the previous time t-1 , the calculation of the input gate is realized:

[0141] i t =σ(W i [h t-1 ,x t )+b i );

[0142] The forget gate controls the amount of data from the previous time saved to the data at this time. Using the tanh activation function, the calculation formula is as follows:

[0143] f t =σ(W f [h t-1 ,x t )+b f );

[0144] C t '=tanh(W c [h t-1 ,x t )+b c );

[0145] The output gate controls how much of the current cell state is output to the current data. The calculation formula of the output gate O t is as follows:

[0146] o t =σ(W o [h t-1 ,x t )+b o );

[0147] The LSTM neural network includes the transfer of long-term memory and short-term memory, and realizes the transfer of long-term memory by calculating the input gate, output gate, and memory information at the previous moment. The short-term memory realizes the transfer by calculating the output gate and activating the information in the long-term memory. The transfer formulas for realizing long-term memory and short-term memory are as follows:

[0148] C t = f t C t-1 + i t C t ';

[0149] h t = o t tanh(C t );

[0150] Among them, W i , W f , W C , W o are the weight matrices in the input gate, forget gate, and output gate respectively; x t is the input parameter; h t is the output parameter in the hidden state; b is the bias. Finally, the power probability distribution of the wind farm is output.

[0151] After completing the optimization for the set number of times, the decentralized wind power prediction model is trained and can perform real-time prediction.

[0152] S109. Determine the wind power prediction result according to the trained deep time series network model and the abnormal weather recognition and classification results.

[0153] S109 specifically includes:

[0154] Obtain the real-time data of the wind farm and the set of abnormal weather feature evaluation indicators.

[0155] According to the real-time data of the wind farm and the set of abnormal weather feature evaluation indicators, use the trained deep time series network model to determine the wind farm power probability.

[0156] Use the abnormal weather recognition and classification results to correct the wind farm power probability and determine the wind power prediction result.

[0157] Use the formula for correction.

[0158] In the formula, P PR_max is the power value with the maximum probability in the power probability distribution; P PR_95％ - is the power lower limit with a 95% confidence level, F ab is the abnormal weather discrimination result. If it is abnormal weather, its value is 1, otherwise it is 0.

[0159] Figure 4 The following is a schematic structural diagram of a wind power prediction system for abnormal weather conditions provided by the present invention. As Figure 4 shown, a wind power prediction system for abnormal weather conditions provided by the present invention is applied to the above-mentioned wind power prediction method for abnormal weather conditions. The system includes:

[0160] A unit wind speed data detection module 401, which is used to detect outliers in the unit wind speed data to determine abnormal data; determine the data abnormality ratio according to the abnormal data; and determine the unit shutdown risk according to the data abnormality ratio; the unit wind speed data includes: unit wind speed, anemometer tower wind speed, and incoming flow wind direction.

[0161] A wind farm wind speed characterization matrix determination module 402, which is used to correct the abnormal data; determine the average wind speed at different unit positions according to the incoming flow wind direction; and determine the wind farm wind speed characterization matrix according to the average wind speed at different unit positions and the corrected unit wind speed data; the unit positions include: front-section units, middle-section units, and rear-section units.

[0162] An abnormal weather feature evaluation system construction module 403, which is used to construct an abnormal weather feature evaluation system according to the wind farm wind speed characterization matrix, unit rotation speed, power, pitch angle, temperature, and humidity; the abnormal weather feature evaluation system includes: theoretical power deviation range, tip speed, blade temperature, and pitch angle change range.

[0163] An abnormal weather recognition and classification result determination module 404, which is used to determine the output data set under normal weather conditions and the output data set under abnormal weather conditions based on the abnormal weather recognition and classification based on rolling AP clustering according to the abnormal weather feature evaluation system.

[0164] An output limitation degree estimation module 405, which is used to estimate the output limitation degree according to the unit shutdown risk and the output data set under abnormal weather conditions.

[0165] A data augmentation module 406, which is used to perform data augmentation according to the output data set under abnormal weather conditions and random variables by using a deep convolutional generative adversarial network.

[0166] A data set reconstruction module 407, which is used to reconstruct the data set according to the augmented data, the output data set under normal weather conditions, and the output limitation degree estimation.

[0167] A trained deep time series network model determination module 408, which is used to train a deep time series network model according to the reconstructed data set; the trained deep time series network model is used to output the wind farm power probability.

[0168] A prediction result determination module 409, configured to determine a prediction result of wind power according to the trained deep time series network model and the abnormal weather recognition and classification result.

[0169] The unit wind speed data detection module 401 specifically includes:

[0170] A data anomaly ratio determination unit, configured to use the formula to determine the data anomaly ratio.

[0171] A unit shutdown risk determination unit, configured to use the formula to determine the unit shutdown risk.

[0172] where d e is the data anomaly ratio, l is the number of measurement time nodes within the time range, m is the total number of units in the wind farm, indicates whether the unit i is missing at time t. If it is missing, it is 1; otherwise, it is 0. WT t i is the unit shutdown risk of unit i at time t, and D is the shutdown threshold.

[0173] In the present specification, each embodiment is described in a progressive manner. The key point of each embodiment is the difference from other embodiments. The same or similar parts among the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0174] In this article, specific examples are used to elaborate on the principle and implementation manner of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A wind power prediction method for abnormal weather conditions, characterized in that, Including: Performing outlier detection on the wind speed data of the unit to determine abnormal data; Determining the data anomaly ratio according to the abnormal data; And determining the unit shutdown risk according to the data anomaly ratio; The wind speed data of the unit includes: the wind speed of the unit, the wind speed of the anemometer tower, and the incoming flow wind direction; Correcting the abnormal data; determining the average wind speed at different unit positions according to the incoming flow wind direction; and determining the wind farm wind speed characterization matrix according to the average wind speed at different unit positions and the corrected wind speed data of the unit; the unit positions include: the front-section unit, the middle-section unit, and the rear-section unit; Constructing an abnormal weather feature evaluation system according to the wind farm wind speed characterization matrix, the unit rotational speed, power, pitch angle, temperature, and humidity; the abnormal weather feature evaluation system includes: the theoretical power deviation amplitude, the tip speed, the blade temperature, and the pitch angle change amplitude; According to the abnormal weather feature evaluation system, determining the output data set under normal weather conditions and the output data set under abnormal weather conditions based on rolling AP clustering for abnormal weather identification and classification; Estimating the output limitation degree according to the unit shutdown risk and the output data set under abnormal weather conditions; Performing data augmentation according to the output data set under abnormal weather conditions and random variables by using a deep convolutional generative adversarial network; Reconstructing the data set according to the augmented data, the output data set under normal weather conditions, and the output limitation degree estimation; Training a deep time series network model according to the reconstructed data set; the trained deep time series network model is used to output the wind farm power probability; Determining the prediction result of the wind power according to the trained deep time series network model and the abnormal weather identification and classification result; The constructing an abnormal weather feature evaluation system according to the wind farm wind speed characterization matrix, the unit rotational speed, power, pitch angle, temperature, and humidity specifically includes: Using the formula to determine the theoretical power deviation magnitude; Use the formula to determine the tip speed; Using the formula to determine the blade temperature; Use the formula Δθ t = θ t - θ t-1 to determine the pitch angle change amplitude; Among them, Δp is the theoretical power deviation amplitude, v blade is the tip speed, T blade is the blade temperature, Δθ t is the pitch angle change amplitude, p actual is the actual output power of the wind farm, C P is the wind energy utilization coefficient, θ is the average pitch angle of the units in the field, R is the blade length of the unit, v is the incoming flow wind speed, ω is the wind turbine angular velocity, T av is the measured temperature of the wind farm, P av is the measured pressure of the wind farm, θ t and θ t-1 are the pitch angles at the current moment t and the previous moment t - 1 respectively, and ρ is the air density.

2. The wind power prediction method for abnormal weather conditions according to claim 1, wherein, The performing outlier detection on the wind speed data of the unit to determine abnormal data; determining the data anomaly ratio according to the abnormal data; and determining the unit shutdown risk according to the data anomaly ratio specifically includes: Use the formula to determine the data anomaly ratio; Use the formula to determine the unit shutdown risk; Among them, de is the data anomaly ratio, l is the number of measurement time nodes within the time range, and m is the total number of units in the wind farm. Indicates whether unit i is missing at time t. If it is missing, it is 1; otherwise, it is 0. Is the unit shutdown risk of unit i at time t, and D is the shutdown threshold.

3. A wind power prediction method for abnormal weather conditions according to claim 1, characterized in that, The determining the output data set under normal weather conditions and the output data set under abnormal weather conditions based on rolling AP clustering for abnormal weather identification and classification according to the abnormal weather feature evaluation system specifically includes: Calculating the weather feature similarity at different moments with the Pearson coefficient as an index according to the abnormal weather feature evaluation system; Dividing the abnormal weather feature evaluation system into h parts, and taking the median of the weather similarity in each part as the self-similarity; Performing AP clustering on each part of the data to obtain the clustering center; According to the clustering center, determining a new user set; if the new user set is greater than the set value, then return to the step of dividing the abnormal weather feature evaluation system into h parts and taking the median of the weather similarity in each part as the self-similarity; otherwise, initializing the self-similarity of the new user set; Performing AP clustering on the new user set to obtain the secondary clustering center and the number of clusters; Calculating the clustering quality index according to the secondary clustering center and the number of clusters; Updating the self-similarity; Determine whether the number of clusters is 2. If it is 2, compare the clustering quality indicators corresponding to different numbers of clusters, and determine the optimal number of clusters to complete the identification and classification of abnormal weather; otherwise, return to the step of performing AP clustering on the new user set to obtain the secondary clustering center and the number of clusters.

4. A wind power prediction method for abnormal weather conditions according to claim 1, characterized in that The estimation of the output limitation degree according to the unit shutdown risk and the output data set under abnormal weather conditions specifically includes: Using the formula to determine the estimated degree of output limitation of the unit; Among them, F is the estimated output limitation degree of the unit, m is the total number of units in the wind farm, WT t i is the unit shutdown risk of unit i at time t, K is the weather characteristic index. If the weather characteristic is abnormal weather, the value is 0; otherwise, it is 1.

5. A wind power prediction method for abnormal weather conditions according to claim 1, characterized in that The determination of the prediction result of wind power according to the trained deep time series network model and the abnormal weather identification and classification results specifically includes: Obtain the real-time data of the wind farm and the set of abnormal weather feature evaluation indicators; According to the real-time data of the wind farm and the set of abnormal weather feature evaluation indicators, use the trained deep time series network model to determine the power probability of the wind farm; Use the abnormal weather identification and classification results to correct the power probability of the wind farm and determine the prediction result of wind power.

6. A wind power prediction system for abnormal weather conditions, which is applied to a wind power prediction method for abnormal weather conditions described in any one of claims 1-5, and is characterized in that, The system includes: A unit wind speed data detection module, which is used to detect abnormal values of the unit wind speed data, determine the abnormal data; determine the data abnormality ratio according to the abnormal data; and determine the unit shutdown risk according to the data abnormality ratio; the unit wind speed data includes: unit wind speed, anemometer tower wind speed, and incoming flow direction; A wind farm wind speed characterization matrix determination module, which is used to correct the abnormal data; determine the average wind speed at different unit positions according to the incoming flow direction; and determine the wind farm wind speed characterization matrix according to the average wind speed at different unit positions and the corrected unit wind speed data; the unit positions include: front-section units, middle-section units, and rear-section units; An abnormal weather feature evaluation system construction module, which is used to construct an abnormal weather feature evaluation system according to the wind farm wind speed characterization matrix, unit rotation speed, power, pitch angle, temperature, and humidity; the abnormal weather feature evaluation system includes: theoretical power deviation amplitude, tip speed, blade temperature, and pitch angle change amplitude; An abnormal weather identification and classification result determination module, which is used to determine the output data set under normal weather conditions and the output data set under abnormal weather conditions based on the abnormal weather identification and classification based on rolling AP clustering according to the abnormal weather feature evaluation system; An output limitation degree estimation module, which is used to estimate the output limitation degree according to the unit shutdown risk and the output data set under abnormal weather conditions; A data augmentation module, which is used to perform data augmentation according to the output data set under abnormal weather conditions and random variables using a deep convolutional generative adversarial network; A data set reconstruction module, which is used to reconstruct the data set according to the augmented data, the output data set under normal weather conditions, and the output limitation degree estimation; A trained deep time series network model determination module, which is used to train a deep time series network model according to the reconstructed data set; the trained deep time series network model is used to output the power probability of the wind farm; A prediction result determination module, which is used to determine the prediction result of wind power according to the trained deep time series network model and the abnormal weather identification and classification results.

Images