A method for predicting wind turbine output loss in small sample conditions under cold waves

Abstract

The present invention proposes a method for predicting wind turbine output loss in small sample conditions of cold waves. In order to make up for the deficiencies in the current research on wind turbine output loss prediction under extreme weather conditions such as cold waves, on the one hand, the amount of data on extreme weather conditions such as cold waves is expanded to provide a broad and solid data foundation for wind turbine output loss prediction; on the other hand, a prediction model for wind turbine output loss is established to predict the wind turbine output loss when a cold wave occurs. This method uses a meta-learning-adversarial generative network to expand and enhance the required data during the cold wave period, and studies the influence mechanism of various meteorological factors on wind turbine output during the cold wave period and normal period respectively. Two models are established to predict the wind turbine output values ​​under extreme weather conditions and non-extreme weather conditions respectively, and the two are subtracted to obtain the wind turbine output loss value. Through this method, sufficient estimation and analysis of wind turbine output can be provided when responding to cold wave events, and appropriate and timely response strategies can be given.

Description

Technical Field

[0001] The present invention belongs to the field of power system operation and control, and in particular relates to a method for predicting wind turbine output loss in a small sample situation during a cold wave. Background Art

[0002] At present, due to the increasingly prominent global issues such as energy crisis and global warming, the energy structure has been transformed from traditional energy such as coal to new energy such as wind power and photovoltaics. Wind power stands out from many new energy sources with its broad industrial foundation, solid technical foundation, and abundant resources, and has become the pillar energy of the power system. The size of wind power output mainly depends on the wind and meteorological resources of the local environment. Abundant wind resources can promote the development of the local wind power industry. When wind resources are abundant, wind power output is high; when wind resources are scarce, wind power output is low. This results in the output of wind turbines cannot be artificially controlled like traditional energy such as thermal power, but has certain volatility, randomness and uncontrollability, which brings threats and challenges to the safe and stable operation of the power system.

[0003] my country has a unique geographical location and topographic conditions. In winter, under the influence of Siberian high pressure, cold air flows smoothly from north to south into the inland, bringing abundant wind energy resources; the eastern coastal areas are also very rich in wind energy resources due to the temperature difference between sea and land and the additive effect of monsoon climate. Wind energy resources are particularly abundant in the "Three Norths" (Northwest, Northeast, North China), Inner Mongolia, Southwest, Xinjiang and other places. From this point of view, my country's wind power has great development potential and a favorable natural environment for development. However, in the above areas, especially in winter, it is often accompanied by strong winds and cooling. In the arid Northwest and North China, strong winds often bring dust or sandstorms. When cold air is active, the temperature drops sharply, and the low temperature generally lasts for one to several days. The temperature drops more in the Northwest and North China, and sub-zero temperatures often cause frost, accompanied by rainfall, causing ice on the blades, affecting the aerodynamic and mechanical properties of wind turbines. Therefore, despite the abundant wind energy resources, extreme weather conditions such as cold waves occasionally occur due to the influence of terrain and meteorological environment. Once they occur, not only will the wind turbine output be greatly uncertain, but it may even damage the unit's electrical equipment and secondary measuring components. In addition, forced power outages during strong winds will increase the risk of large-scale power outages, further affecting the stability and safety of the power system and causing immeasurable losses.

[0004] However, there are few studies on wind turbine output under extreme weather conditions such as cold waves. Not only is there a lack of sufficient research on the impact of cold waves on wind turbines, but there is also little estimated output of wind turbines when a cold wave is about to occur. This leads to a lack of sufficient estimated analysis of wind turbine output when responding to cold wave events, and it is difficult to provide appropriate and timely response strategies. Therefore, there is an urgent need for a method that can predict wind turbine output loss under cold wave weather conditions. Summary of the invention

[0005] In order to make up for the current deficiencies in the research on prediction of wind turbine output loss under extreme weather conditions such as cold waves, the purpose of the present invention is: on the one hand, to expand the amount of data on extreme weather conditions such as cold waves, and provide a broad and solid data foundation for the prediction of wind turbine output loss; on the other hand, to establish a prediction model for wind turbine output loss to predict the wind turbine output loss when a cold wave occurs. This method uses a meta-learning-adversarial generative network to expand and enhance the required data during the cold wave period, and studies the influence mechanism of various meteorological factors on wind turbine output during the cold wave period and normal period respectively. Two models are established to respectively predict the wind turbine output values ​​under extreme weather conditions and non-extreme weather conditions, and the two are subtracted to obtain the wind turbine output loss value.

[0006] In order to achieve the above purpose, the technical solution adopted by the present invention is:

[0007] S1: Construct a cold wave meteorological characterization vector to identify the occurrence of cold wave events, define typical meteorological change parameters and discrimination thresholds;

[0008] S2: Construct a data set containing the NWP weather forecast value and wind turbine output value of the previous day;

[0009] S3: According to the influence of NWP meteorological factors on wind turbine output, the meteorological factors with the greatest influence on wind turbine output are selected as typical features to construct a typical feature set;

[0010] S4: Expand the dataset under cold wave weather conditions; perform random undersampling on samples under normal weather conditions;

[0011] S5: establishing network mappings according to the expanded sample set under cold wave weather conditions obtained in S4 and the sample set under normal weather conditions obtained by random undersampling, respectively; and training to obtain cold wave and normal weather neural network models;

[0012] S6: Input the set of typical meteorological factors of cold waves in NWP into the cold wave model in S5, and input the typical meteorological factor set data of normal weather into the normal weather model in S5, and obtain two wind turbine output prediction values ​​respectively; subtract the wind turbine output prediction values ​​under normal weather conditions and cold wave weather conditions to obtain the wind turbine output loss prediction value under cold wave weather conditions.

[0013] Based on the above scheme, there are 24 NWP meteorological factors, including air temperature, momentum flux, wind speed at 170m, wind speed at 100m, wind speed at 30m, wind speed at 10m, surface wind speed, wind direction at 170m, wind direction at 100m, wind direction at 30m, wind direction at 10m, surface wind direction, sea level pressure, cloud cover, sensible heat flux, latent heat flux, shortwave radiation, longwave radiation, surface pressure, PRT, large-scale precipitation, convective precipitation, 2m air temperature, and humidity.

[0014] On the basis of the above scheme, in S1: based on the NWP data of the previous day, the cold wave meteorological characterization vector H of the day is defined as shown in formula (1):

[0015]

[0016] Where, t n is the negative temperature distance, Δt is the steep drop in temperature within a day, R day is the daily cumulative rainfall, is the daily average wind speed;

[0017] Temperature negative distance:

[0018]

[0019] This formula represents the difference between the daily average temperature and the multi-year average temperature of the decade in which the day is located, where: represents the daily average temperature, Indicates the multi-year average temperature of the decade in which the day is located;

[0020] Temperature drop within one day:

[0021] Δt=t max -t min (3)

[0022] Where, t max and t min They are the maximum and minimum NWP temperature values, respectively;

[0023] The discrimination threshold is shown in formula (4):

[0024]

[0025] In the formula, χ t is the temperature negative distance judgment threshold, χ Δt is the threshold value for judging the degree of temperature drop, χ R is the total rainfall threshold for one day, χ S is the wind speed judgment threshold;

[0026] If the identification threshold is met as shown in formula (4), it is considered that a cold wave occurred during this period, otherwise it is a normal weather condition.

[0027] Based on the above solution, in S2: Count the NWP meteorological data of a certain wind farm area M = {M (i) ,i=1,2,...,24},M (i) represents the value of different meteorological factors; the measured output data {P} of the wind turbines in the wind farm are counted; the data set is shown in formulas (5) and (6); where S n is the data set under normal weather conditions, S c This is a dataset under cold wave weather conditions;

[0028]

[0029]

[0030] Where M c and P c They are NWP meteorological data and wind turbine output data under cold wave weather conditions;

[0031] M n and P n They are NWP meteorological data and wind turbine output data under normal weather conditions respectively.

[0032] Based on the above scheme, the time resolution is 15min, and the NWP meteorological data and wind turbine output data correspond in the time series;

[0033] Among them, the dataset S c There are l data in the dataset S n There are k pieces of data.

[0034] On the basis of the above scheme, in S3, a typical feature set is selected from the data sets of the NWP weather forecast value and the wind turbine output value obtained in S2, respectively, to represent the significant characteristics of the two types of weather conditions;

[0035] First, the following two regression models are constructed. Formula (7) is a constrained regression, and formula (8) is an unconstrained regression.

[0036]

[0037]

[0038] In the formula, and are the values ​​of the wind turbine output constrained regression and unconstrained regression models at time t, a and b represent the regression coefficients of the wind turbine output variable and the NWP meteorological variable, respectively. D is the maximum lag time of the wind turbine output, S is the maximum lag time of the NWP factor, and δ t is white noise;

[0039] The residual squares of the two regression models are calculated according to formula (9), and the F statistic is calculated according to formula (10);

[0040]

[0041]

[0042] According to the F statistic, meteorological factors with a confidence level of more than 99% are selected as the typical feature sets of the two weather conditions, among which the typical feature set of normal weather is T n ={T (i) ,i=1,2,...,I n};

[0043] The typical characteristic set of cold wave weather is T c ={T (i) ,i=1,2,...,I c}, I n <24, I c <24;

[0044]

[0045]

[0046] Formula (11) and Formula (12) represent the typical meteorological feature sets under normal weather conditions and cold wave weather conditions.

[0047] Based on the above scheme, in S4: for the data set under normal weather conditions, a random undersampling method is used to obtain the data from S n The NWP meteorological data and wind turbine output data under some time periods are randomly selected to obtain a new sample set. As shown in formula (13);

[0048]

[0049] In the formula, represents the sample set after undersampling. There are K data in the sample set. It can be seen that K <k;

[0050] For the dataset S under cold wave weather conditions c , using adversarial generative neural network toc The data set is expanded to obtain the expanded sample set The adversarial neural network is used to learn the data distribution characteristics under cold wave weather conditions, and simulates and generates more similar samples based on the data distribution characteristics to form a new sample set with the original sample set, as shown in formula (14);

[0051]

[0052] In the formula, It represents the expanded sample set. There are L data in the sample set, so L>l.

[0053] On the basis of the above scheme, in S5: using the extreme learning machine algorithm to respectively learn the mapping relationship between the typical meteorological feature set under cold wave weather conditions and normal weather conditions and the measured output value of the wind turbine;

[0054] Formula (15) represents the mapping between typical meteorological conditions during cold waves and wind turbine output.

[0055] Formula (16) represents the mapping between typical meteorological conditions and wind turbine output under normal weather conditions, and f1 and f2 represent the extreme learning machine algorithm;

[0056]

[0057]

[0058] Based on the above scheme, in S6: by giving the NWP meteorological data of the cold wave weather The typical feature sets under the two weather conditions are respectively input into the two models trained in S5, as shown in formula (17):

[0059]

[0060] In the formula, and are the typical characteristic set data of NWP in cold wave weather and normal weather during the forecast period, is the predicted value of wind turbine output under normal weather conditions, is the predicted value of wind turbine output under cold wave weather conditions, P loss is the predicted value of wind turbine output loss.

[0061] On the basis of the above scheme, the amount of data after the cold wave weather samples are expanded is equal to the amount of data after undersampling in normal weather, that is, L=K.

[0062] The prior art basis and data sources used in the present invention are:

[0063] 1. Use the 24-dimensional NWP data of the day before, which includes 24 meteorological factors, including temperature, momentum flux, wind speed at 170m, wind speed at 100m, wind speed at 30m, wind speed at 10m, surface wind speed, wind direction at 170m, wind direction at 100m, wind direction at 30m, wind direction at 10m, surface wind direction, sea level pressure, cloud cover, sensible heat flux, latent heat flux, shortwave radiation, longwave radiation, surface pressure, prt, large-scale precipitation, convective precipitation, 2m temperature, and humidity.

[0064] 2. Use actual measured output data of wind turbines affected by cold waves.

[0065] 3. Use the Granger causality test method using mathematical statistics tools.

[0066] 4. Use deep learning neural network algorithms: adversarial neural networks, extreme learning machines.

[0067] Beneficial effects of the present invention:

[0068] Due to the small number of extreme weather events such as cold waves and insufficient data, there is not only a lack of sufficient research on the impact of cold waves on wind turbines, but also less estimated output of wind turbines when a cold wave is about to occur. The prediction method of the present invention can provide a sufficient estimated analysis of wind turbine output when a cold wave event occurs, and provide an appropriate and timely response strategy. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] The present invention has the following accompanying drawings:

[0070] Figure 1 A process model for predicting wind turbine output loss under small sample conditions of cold waves. DETAILED DESCRIPTION

[0071] The following combination Figure 1 The present invention is described in further detail.

[0072] A method for predicting wind turbine output loss in a small sample of cold waves comprises the following steps:

[0073] S1. Based on the definition and standards of cold wave events, starting from the meteorological characteristics of cold waves, a meteorological characterization vector for distinguishing the occurrence of cold wave events is constructed, and typical meteorological change parameters and discrimination thresholds are defined to provide a discrimination method for identifying the occurrence of cold wave events.

[0074] S2. Collect wind farm operation information and the weather forecast information of the station, build a data set containing the NWP weather forecast value and wind turbine output value of the previous day, and divide it into two categories according to the method in S1. One is the weather forecast and wind turbine output data set under normal weather conditions, and the other is the weather forecast and wind turbine output data set under extreme weather conditions such as cold waves.

[0075] S3. Analyze the meteorological factors that affect wind turbine output under two weather conditions, use mathematical statistics methods to quantify the degree of influence of 24 NWP meteorological factors on wind turbine output under two weather conditions, select the meteorological factors with the greatest influence on wind turbine output as typical features, and construct a typical feature set.

[0076] S4. Expand the dataset under cold wave weather conditions. Use the generative adversarial network to simulate the distribution of similar meteorological data and wind turbine output data under cold wave weather conditions to generate more cold wave meteorological data and corresponding cold wave wind turbine output data. Perform random undersampling on samples under normal weather conditions. This way, the number of samples in the two categories is consistent.

[0077] S5. Based on the expanded sample set under cold wave weather conditions obtained in S4 and the sample set under normal weather conditions obtained by undersampling, network mapping is established respectively. In the case of cold wave, the input is the set of typical meteorological factors of NWP in cold wave, and the output is the wind turbine output value in cold wave; in normal weather conditions, the input is the set of typical meteorological factors of NWP in normal weather, and the output is the wind turbine output value in normal period. Two neural network models are obtained by training.

[0078] S6. According to the method provided in S1, the occurrence of cold wave events is predicted, and the set of typical meteorological factors of cold wave in NWP is input into the cold wave model in S5, and the set of typical meteorological factors of normal weather is input into the normal weather model in S5, and two wind turbine output prediction values ​​are obtained respectively. The wind turbine output prediction values ​​under normal weather conditions and cold wave weather conditions are subtracted to obtain the wind turbine output loss prediction value under cold wave weather conditions.

[0079] On the basis of the above technical scheme, in S1, the standard for analyzing the definition of cold wave events is studied, and the intensity and severity of cold air activities are studied. The characteristics of NWP meteorological data during the cold wave period are explored, and the characterization vector that can clearly characterize the occurrence of cold wave weather conditions is extracted and constructed, and the characterization parameters are defined. The typical meteorological characteristics of cold waves are a large-scale temperature drop, accompanied by strong winds, even blizzards, low temperatures and heavy rainfall. Based on the NWP data of the day before, the cold wave meteorological characterization vector H for one day is defined as shown in formula (1-3).

[0080]

[0081] in,

[0082] ①Temperature negative distance:

[0083]

[0084] This formula represents the difference between the daily average temperature and the multi-year average temperature of the decade in which the day is located, where: represents the daily average temperature, Indicates the multi-year average temperature of the decade in which the day falls.

[0085] ②The degree of temperature drop within one day:

[0086] Δt=t max -t min (3)

[0087] Where, t max and t min They are the maximum and minimum NWP temperatures, respectively.

[0088] ③R day is the daily accumulated rainfall;

[0089] ④ is the daily average wind speed.

[0090] Cold wave weather conditions will produce a certain temperature negative distance, and the temperature will drop rapidly within a day. In some parts of my country, strong winds and heavy rainfall are often accompanied. The discrimination thresholds of the above characterization parameters are given to identify cold wave weather events. The threshold conditions are shown in formula (4):

[0091]

[0092] In the formula, χ t is the temperature negative distance judgment threshold, χ Δt is the threshold value for judging the degree of temperature drop, χ R is the total rainfall threshold for one day, χ S is the wind speed judgment threshold. If the above judgment conditions are met, it can be considered that a cold wave has occurred during this period, otherwise it is a normal weather condition.

[0093] Based on the above technical solution, in S2, the NWP meteorological data of a wind farm area is counted M = {M (i) ,i=1,2,...,24},M (i) represents the numerical value of different meteorological factors; and the measured output data {P} of the wind turbines in the wind farm are counted. The data set is shown in formulas (5) and (6). n is the data set under normal weather conditions, S c This is a dataset under cold wave weather conditions.

[0094]

[0095]

[0096] Where M c and P c They are NWP meteorological data and wind turbine output data under cold wave weather conditions. n and P n They are NWP meteorological data and wind turbine output data under normal weather conditions, with a time resolution of 15 minutes. There are 96 data in total in one day. NWP meteorological data and wind turbine output data correspond to each other in the time series. c There are l data in the dataset S n There are k data in . Generally speaking, the value of l is much smaller than the value of k.

[0097] On the basis of the above technical solution, in S3, typical feature sets are selected from the meteorological factor sets in the two sample sets obtained in S2 to represent the significant characteristics of the two types of weather conditions. The Granger causality verification method is used to verify the causal relationship between the meteorological factors and the wind turbine output under the two weather conditions, and the correlation between the 24 meteorological factors and the wind turbine output is calculated. The calculation method is as follows: first, the following two regression models are constructed, formula (7) is a constrained regression, and formula (8) is an unconstrained regression.

[0098]

[0099]

[0100] In the formula, and are the values ​​of the wind turbine output constrained regression and unconstrained regression models at time t, a and b represent the regression coefficients of the wind turbine output variable and the NWP meteorological variable, respectively. D is the maximum lag time of the wind turbine output, S is the maximum lag time of the NWP factor, and δ t is white noise.

[0101] The residual squares of the two regression models are calculated according to formula (9), and the F statistic is calculated according to formula (10).

[0102]

[0103]

[0104] According to the F statistic, meteorological factors with a confidence level of more than 99% are selected as the typical feature sets of the two weather conditions, among which the typical feature set of normal weather is T n ={T (i) ,i=1,2,...,I n}; The typical characteristic set of cold wave weather is T c ={T (i) ,i=1,2,...,I c}, I n <24, I c <24. The typical feature sets under the two weather conditions basically represent all the information contained in the 24 meteorological factors, and at the same time have the greatest reference contribution to the wind turbine output value.

[0105]

[0106]

[0107] Formula (11) and Formula (12) represent the typical meteorological feature sets under normal weather conditions and cold wave weather conditions. The meteorological factors in these two typical feature sets are not exactly the same.

[0108] On the basis of the above technical solutions, in S4, in order to solve the data distribution problem of unbalanced data volume in the two types of data sets, the following measures are taken for the two types of data sets. For the data set under normal weather conditions, since the data volume is large, random undersampling is adopted. n The NWP meteorological data and wind turbine output data under some time periods are randomly selected to obtain a new sample set. As shown in formula (13).

[0109]

[0110] In the formula, represents the sample set after undersampling. There are K data in the sample set. It can be seen that K <k。

[0111] For the dataset S under cold wave weather conditions c , due to the small amount of data, it is not only difficult to meet the needs of later analysis and modeling, but also the data volume is quite different from that under normal weather conditions. Therefore, adversarial generative neural network is used to train S c The data set is expanded to obtain the expanded sample set The adversarial neural network is used to learn the data distribution characteristics under cold wave weather conditions, and simulates the generation of more similar samples based on the data distribution characteristics and combines them with the original sample sets to form a new sample set, as shown in formula (14).

[0112]

[0113] In the formula, It represents the expanded sample set. There are L data in the sample set, so L>l.

[0114] The number of samples in the two sample sets obtained in this way is consistent, that is, K=L, which ensures the balance of sample data under cold wave weather conditions and normal weather conditions, and also meets the data volume requirements for modeling in subsequent steps, ensuring its rationality and effectiveness.

[0115] On the basis of the above technical solution, in S5, the extreme learning machine algorithm is used to learn the mapping relationship between the typical meteorological feature set and the measured output value of the wind turbine under the cold wave weather conditions and the normal weather conditions. Formula (15) represents the mapping between the typical meteorological conditions during the cold wave period and the wind turbine output. Formula (16) represents the mapping between the typical meteorological conditions under normal weather conditions and the wind turbine output. f1 and f2 represent the extreme learning machine algorithm.

[0116]

[0117]

[0118] Based on the above technical solution, in S6, the NWP meteorological data of cold wave weather is given The typical feature sets under the two weather conditions are respectively input into the two models trained in S5, as shown in formula (17):

[0119]

[0120] In the formula, and are the typical characteristic set data of NWP in cold wave weather and normal weather during the forecast period, is the predicted value of wind turbine output under normal weather conditions, is the predicted value of wind turbine output under cold wave weather conditions, P loss is the predicted value of wind turbine output loss.

[0121] It can be considered It should be the output of the fan under normal weather conditions. Due to the influence of cold weather, the fan output is reduced to Therefore, the loss prediction value of the present invention is calculated based on the typical feature set of different meteorological conditions to obtain the difference.

[0122] The above implementation modes are only used to illustrate the present invention, but not to limit the present invention. Ordinary technicians in the relevant technical field can make various changes and modifications without departing from the essence and scope of the present invention. Therefore, all equivalent technical solutions also belong to the scope of the present invention. The scope of patent protection of the present invention should be limited by the claims.

[0123] The contents not described in detail in this specification belong to the prior art known to professional and technical personnel in this field.

Claims

1. A method for predicting wind turbine output loss in a small sample of cold waves, characterized in that: The following steps are involved: S1: Construct a cold wave meteorological characterization vector to identify the occurrence of cold wave events, define typical meteorological change parameters and discrimination thresholds; S2: Construct a data set containing the weather forecast value and wind turbine output value of the day before the NWP day; S3: According to the influence of NWP meteorological factors on wind turbine output, the meteorological factors with the greatest influence on wind turbine output are selected as typical features to construct a typical feature set; S4: Expand the dataset under cold wave weather conditions; perform random undersampling on samples under normal weather conditions; S5: establishing network mappings according to the expanded sample set under cold wave weather conditions obtained in S4 and the sample set under normal weather conditions obtained by random undersampling, respectively; and training to obtain cold wave and normal weather neural network models; S6: Input the set of typical meteorological factors of cold waves in NWP into the cold wave model in S5, and input the typical meteorological factor set data of normal weather into the normal weather model in S5, and obtain two wind turbine output prediction values ​​respectively; subtract the wind turbine output prediction values ​​under normal weather conditions and cold wave weather conditions to obtain the wind turbine output loss prediction value under cold wave weather conditions.

2. A method for predicting wind turbine output loss in a small sample of cold waves as claimed in claim 1, characterized in that: There are 24 NWP meteorological factors.

3. A method for predicting wind turbine output loss in a small sample of cold waves as claimed in claim 2, characterized in that: In S1: Based on the data of the day before NWP, the cold wave meteorological characterization vector of the day is defined As shown in formula (1): ; In the formula, is the negative temperature distance, The temperature drops sharply in one day. is the daily cumulative rainfall, is the daily average wind speed; Temperature negative distance: ; This formula represents the difference between the daily average temperature and the multi-year average temperature of the decade in which the day is located, where: represents the daily average temperature, Indicates the multi-year average temperature of the decade in which the day is located; Temperature drop within one day: ; In the formula, and They are the maximum and minimum NWP temperature values, respectively; The discrimination threshold is shown in formula (4): ; In the formula, is the temperature negative distance judgment threshold, is the threshold for judging the degree of temperature drop, is the total rainfall threshold for one day, is the wind speed judgment threshold; If the identification threshold is met as shown in formula (4), it is considered that a cold wave occurred during this period, otherwise it is a normal weather condition.

4. A method for predicting wind turbine output loss in a small sample of cold waves as claimed in claim 2, characterized in that: In S2: Statistics of NWP meteorological data in a certain wind farm area Indicates the values ​​of different meteorological factors; statistics on the measured output data of wind turbines in the wind farm ; The data set is shown in formulas (5) and (6); is a dataset under normal weather conditions. This is a dataset under cold wave weather conditions; ; ; In the formula, and They are NWP meteorological data and wind turbine output data under cold wave weather conditions; and They are NWP meteorological data and wind turbine output data under normal weather conditions respectively.

5. A method for predicting wind turbine output loss in a small sample of cold waves as claimed in claim 4, characterized in that: The time resolution is 15 minutes, and the NWP meteorological data and wind turbine output data correspond to each other in the time series; Among them, the data set There are Data set There are k Piece of data.

6. A method for predicting wind turbine output loss in a small sample of cold waves as claimed in claim 4, characterized in that: In S3, typical feature sets are selected from the data sets of the NWP weather forecast values ​​and wind turbine output values ​​obtained in S2 to represent the significant characteristics of the two types of weather conditions respectively; First, the following two regression models are constructed. Formula (7) is a constrained regression, and formula (8) is an unconstrained regression. ; ; In the formula, and They are respectively the constrained regression model and the unconstrained regression model of the wind turbine output value. t The value of the moment, and They represent the regression coefficients of wind turbine output variables and NWP meteorological variables, D is the maximum lag time of the fan output, S is the maximum lag time of the NWP factor, and is white noise; The residual squares of the two regression models are calculated according to formula (9), and the F statistic is calculated according to formula (10); ; ; According to the F statistic, meteorological factors with a confidence level of more than 99% are selected as the typical feature sets of the two weather conditions. The typical feature set of normal weather is ; The typical characteristics of cold wave weather are , ; ; ; Equations (11) and (12) represent the typical meteorological feature sets under normal weather conditions and cold wave weather conditions.

7. A method for predicting wind turbine output loss in a small sample of cold waves as claimed in claim 4, characterized in that: In S4: For the data set under normal weather conditions, a random undersampling method is used to obtain The NWP meteorological data and wind turbine output data under some time periods are randomly selected to obtain a new sample set. , as shown in formula (13); ; In the formula, Represents the sample set after undersampling, and the sample set contains K Data, we can know ; For the dataset under cold wave weather conditions , using adversarial generative neural network to The data set is expanded to obtain the expanded sample set ; The adversarial neural network is used to learn the data distribution characteristics under cold wave weather conditions, and simulates and generates more similar samples based on the data distribution characteristics to form a new sample set with the original sample set, as shown in formula (14); ; In the formula, Represents the expanded sample set, which contains L Data, we can know L>l .

8. A method for predicting wind turbine output loss in a small sample of cold waves as claimed in claim 6, characterized in that: In S5: using an extreme learning machine algorithm to respectively learn the mapping relationship between a typical meteorological feature set under cold wave weather conditions and normal weather conditions and the measured output value of the wind turbine; Formula (15) represents the mapping between typical meteorological conditions during cold waves and wind turbine output. Formula (16) represents the mapping between typical meteorological conditions and wind turbine output under normal weather conditions. and represents the extreme learning machine algorithm; ; 。 9. A method for predicting wind turbine output loss in a small sample of cold waves as claimed in claim 1, characterized in that: In S6: the NWP meteorological data of the cold wave weather is given , respectively inputting its typical feature sets under the two weather conditions into the two models trained in the mapping model established in S5, as shown in formula (17): ; In the formula, and are the typical characteristic set data of NWP in cold wave weather and normal weather during the forecast period, is the predicted value of wind turbine output under normal weather conditions, is the predicted value of wind turbine output under cold wave weather conditions, is the predicted value of wind turbine output loss.

10. A method for predicting wind turbine output loss in a small sample of cold waves as claimed in claim 7, characterized in that: The amount of data after the cold wave weather sample expansion is equal to the amount of data after the normal weather undersampling, that is, L=K .

Images