Method, device, equipment and medium for short-term wind power prediction considering wake and obstruction

Abstract

The invention relates to the technical field of wind power generation, and in particular to a method, device, equipment and medium for short-term wind power prediction considering wake and blockage. The method comprises: collecting external environment and operation data of each wind turbine in an offshore wind farm; obtaining wind farm inflow information data and performing preprocessing; obtaining wind farm wind turbine layout and unit parameters, establishing an engineering analytical model in which a Jensen wake model and a cylindrical vortex blockage model are coupled, and obtaining inflow wind speed and wind direction values ​​of the wind turbine based on the wind farm inflow information data; constructing an offshore wind power prediction model based on a conditional generative adversarial network; using a one-dimensional random noise vector obeying a standard normal distribution, the external environment of the wind turbine on the prediction day, the inflow wind speed and the sine and cosine values ​​of the wind direction as model inputs, and the actual active power of the offshore wind farm on the prediction day as output, to train the power prediction model; and using the trained power prediction model to predict the offshore wind farm.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power generation, and in particular to a method, device, equipment and medium for predicting short-term wind power taking into account wake and blockage. Background Art

[0002] my country has abundant offshore wind energy resources, and the offshore wind power industry has become an important strategic support for promoting the green transformation of my country's power system and ensuring the development of a low-carbon economy. Compared with onshore wind energy resources, offshore wind power has unique advantages such as higher power generation, less land resource occupation and proximity to the eastern load center. However, marine environmental conditions change frequently, and offshore wind farms are susceptible to multi-scale spatial and temporal factors such as the atmospheric boundary layer, wake and blockage effects, the operating status of wind turbines, and ocean currents. Wind conditions are highly random, seasonal, and highly volatile, which brings huge challenges to the development, utilization, and management of offshore wind power.

[0003] At the same time, the high proportion and large-scale wind power grid connection makes it very difficult to balance the supply and demand of electricity, and also affects the safe and stable operation of the power system. The short-term prediction time scale is generally one to two days in the future, and is mainly used for the day-ahead arrangement and dispatch of offshore wind turbines. The wake effect and the blocking effect are common aerodynamic phenomena inside wind farms, and are also important factors affecting the production of electricity by wind farms. Ignoring either of them will lead to an overestimation of the power prediction results of the entire wind farm. Therefore, it is very necessary to incorporate the wake effect and the blocking effect as physical factors into the short-term power prediction method of wind power.

[0004] The information disclosed in this background technology section is only intended to deepen the understanding of the overall background technology of the present invention, and should not be regarded as acknowledging or suggesting in any form that the information constitutes the prior art already known to those skilled in the art. Summary of the invention

[0005] The present invention provides a method, device, equipment and medium for predicting short-term wind power taking into account wake and blockage, thereby effectively solving the problems in the background technology.

[0006] In order to achieve the above object, the technical solution adopted by the present invention is: a method for short-term wind power prediction considering wake and blockage, comprising the following steps:

[0007] Collect the external environment and operation data of each wind turbine in the offshore wind farm; obtain the wind farm flow information data and perform pre-processing;

[0008] Obtain the layout and unit parameters of the wind farm wind turbines, establish an engineering analytical model that couples the Jensen wake model and the cylindrical vortex blocking model, and obtain the inflow wind speed and wind direction values ​​of the wind turbine units based on the preprocessed wind farm inflow information data;

[0009] Construct an offshore wind power prediction model based on conditional generative adversarial networks;

[0010] The power prediction model is trained by taking a one-dimensional random noise vector that obeys a standard normal distribution, the external environment of the wind turbine generator set on the forecast day, the sine and cosine values ​​of the inflow wind speed and wind direction as model inputs, and the actual active power of the offshore wind farm on the forecast day as output;

[0011] The trained power prediction model is used to predict the offshore wind farm, and the power prediction results of the wind farm at each time on the prediction day are obtained.

[0012] Furthermore, the external environment includes: external wind speed, wind direction, and turbulence intensity of each wind turbine;

[0013] The operation data include: active output of each wind turbine;

[0014] The wind farm inflow information data includes: hourly wind speed data in longitude and latitude directions at heights of 2m, 10m and 50m.

[0015] Furthermore, the wind farm incoming flow information data is obtained and preprocessed, and the preprocessing process includes:

[0016] Eliminate the two types of shutdown data and power limit points, where the wind speed is less than the cut-in wind speed and greater than the cut-out wind speed;

[0017] Treat negative data values ​​uniformly as zero;

[0018] Adjust the time resolution accuracy of each operating data to the preset value and fill in the missing data;

[0019] The longitude and latitude wind speed data in the wind farm inflow information data are converted into wind speed and wind direction angle, and converted to the height of 90m at the wind turbine hub using the following formula:

[0020] V 90m =1.484V 10m ;

[0021] Where: V 90m , V 10m The wind speed values ​​at the heights of 90m and 10m respectively;

[0022] The input wind speed and turbulence intensity data are standardized, the wind direction is represented by its sine and cosine values, and the active output data is proportionally scaled to between [0,1].

[0023] Furthermore, after collecting the external environment and operating data of each wind turbine in the offshore wind farm, representative wind turbines are selected according to the correlation between the output power of each wind turbine and the total output of the wind farm. In subsequent model training, the external environment, inflow wind speed and wind direction sine and cosine values ​​of the representative wind turbines are used as model inputs.

[0024] Furthermore, when selecting the representative unit, the correlation between the output power of each wind turbine and the total output of the wind farm is determined by Pearson correlation analysis, including:

[0025]

[0026] Where: P k is the power sequence of the kth wind turbine, P farm is the power sequence of the wind farm, n is the number of elements in the sequence, t is the time; the correlation coefficient r k ∈[-1, 1], r k The larger it is, the stronger the correlation between the wind farm power and the wind power of the representative unit is.

[0027] Furthermore, the construction of an offshore wind power prediction model based on a conditional generative adversarial network includes:

[0028] The generator model is composed of two layers of parallel gated recurrent unit GRU networks and one layer of fully connected network, and the internal structure of the power prediction model is constructed based on a one-dimensional convolutional neural network and a channel attention layer.

[0029] Furthermore, the wind farm wind turbine layout includes:

[0030] The longitude and latitude information of each wind turbine;

[0031] The unit parameters include: the unit's rated power and rated wind speed, cut-in and cut-out wind speeds, hub height, rotor diameter, and theoretical wind power and thrust coefficient curves.

[0032] Furthermore, the engineering analytical model for coupling the wake model and the cylindrical vortex blocking model comprises:

[0033] In the Jensen wake model, the wake velocity at the axial distance x of the wind rotor is:

[0034]

[0035] Where: U ∞ is the free flow wind speed at the fan hub height, C T is the thrust coefficient of the fan, R is the radius of the wind wheel, k w is the wake expansion coefficient;

[0036] The cylindrical vortex blocking model assumes that the wake geometry is a cross-sectional radius equal to the rotor radius R and has a constant tangential vorticity γ t The semi-infinite cylinder, where the tangential vorticity is calculated as:

[0037]

[0038] Where: is the average axial inductance in the rotor plane;

[0039] The induced velocity at the axial distance x of the wind wheel is:

[0040]

[0041] Where: r is the radial position, K and Π are the second and third complete elliptic integrals respectively, k is the ellipse parameter,

[0042] Based on the above formula, when the radial position is 0, the simplified form of the velocity near the wake area is:

[0043]

[0044] The simplified form of the far wake velocity is:

[0045]

[0046] Using Taylor's formula to expand, we can get the simplified form:

[0047]

[0048] In the process of coupling the wake and blocking models, it is assumed that the induced velocity of the vortex column model is zero in the semi-infinite cylindrical volume and on the wind rotor plane. The velocity field of the cylindrical vortex model is used to describe the blocking effect upstream of the unit. At the same time, the Jensen wake model is used to represent the wake field downstream of the unit. The two velocity fields in other areas are directly linearly superimposed. The preprocessed wind farm inflow information data is input into the model to obtain the inflow wind speed and wind direction values ​​at the representative unit location.

[0049] Furthermore, when training the power prediction model, a one-dimensional random noise vector obeying a standard normal distribution, the hourly wind speed and turbulence intensity, the inflow wind speed and the sine and cosine values ​​of the wind direction of the representative unit on the prediction day are used as inputs of the model, and the actual active power of the offshore wind farm with a time interval of one hour on the prediction day is used as output, and the one-dimensional random noise vector contains 24 elements.

[0050] Furthermore, in the construction of the offshore wind power prediction model based on the conditional generative adversarial network, the loss functions of the generator and the discriminator of the conditional generative adversarial network are respectively expressed as:

[0051]

[0052]

[0053] The specific function expression of the adversarial loss based on the conditional generation adversarial network architecture is:

[0054]

[0055] Where: L(D, G) is the objective function of the conditional generative adversarial network, and Represent the standard normal distribution Pz and the real data distribution P r The expected value of the distribution, x is the real wind power data, y is the conditional variable, z is a noise sequence that obeys the standard normal distribution, D(x|y) is the similarity between the actual wind power sample and the real wind power population, D(G(z|y)|y) is the similarity between the predicted wind power sample and the real wind power population. In the training of the conditional generative adversarial network, the generator G is committed to minimizing the above objective function, but the discriminator D is committed to maximizing the above formula, thus forming a binary minimax game;

[0056] Add the loss function to its objective function. The Huber loss function includes:

[0057]

[0058] Where: δ is a parameter, p and are one-dimensional vectors consisting of the actual value and predicted value of offshore wind power, pi is the actual value of offshore wind power at the i-th moment, is the predicted value of offshore wind power at the i-th moment output by the CGAN generated network;

[0059] The final loss function of the generator of the conditional generative adversarial network is:

[0060] L G =λ1L′ G +λ2L Huber ((G(z|y),x);

[0061] Wherein: λ1 and λ2 are weight coefficients, λ1+λ2=1.

[0062] The present invention also includes a wind power short-term power prediction device taking into account wake and blockage, using the above method, including:

[0063] The collection unit is used to collect the external environment and operation data of each wind turbine in the offshore wind farm; obtain the wind farm flow information data and perform pre-processing;

[0064] An engineering analytical model modeling unit is used to obtain the layout and unit parameters of the wind farm wind turbines, establish an engineering analytical model that couples the Jensen wake model and the cylindrical vortex blocking model, and obtain the inflow wind speed and wind direction values ​​of the wind turbine units based on the preprocessed wind farm inflow information data;

[0065] A power prediction model modeling unit, used to build an offshore wind power prediction model based on a conditional generative adversarial network;

[0066] A training unit is used to train the power prediction model by taking a one-dimensional random noise vector obeying a standard normal distribution, the external environment of the wind turbine generator set on the forecast day, the sine and cosine values ​​of the inflow wind speed and wind direction as model inputs, and the actual active power of the offshore wind farm on the forecast day as output;

[0067] The prediction unit is used to use the trained power prediction model to predict the offshore wind farm and obtain the power prediction results of the wind farm at each time on the prediction day.

[0068] The present invention also includes a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method described above is implemented.

[0069] The present invention also includes a storage medium on which a computer program is stored. When the computer program is executed by a processor, the method described above is implemented.

[0070] The beneficial effects of the present invention are as follows: the present invention obtains the layout of wind turbines and unit parameters of the wind farm, establishes an engineering analytical model coupled with the Jensen wake model and the cylindrical vortex blocking model, obtains the inflow wind speed and wind direction values ​​of the wind turbine unit based on the preprocessed wind farm inflow information data, and constructs an offshore wind power prediction model based on a conditional generative adversarial network. The conditional generative adversarial network is combined with its advantages in extracting high-dimensional features and being good at generating data. At the same time, the calculation results of the engineering analytical model are used as part of the input of the prediction model, thereby realizing an effective combination of the analytical model and the deep learning method. This can not only effectively improve the prediction accuracy of the day-ahead active output of the offshore wind farm, but also enhance the interpretability and generalization of the prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0072] Figure 1 is a flow chart of the method in Example 1;

[0073] Figure 2 It is a structural schematic diagram of the device in Example 1;

[0074] Figure 3 Schematic diagram of the process in Example 2;

[0075] Figure 4 This is a distribution diagram of 15 units in the offshore wind farm in Example 2;

[0076] Figure 5 The Pearson correlation coefficient heat map of the output power of each wind turbine and the total output of the wind farm in Example 2;

[0077] Figure 6 The structure of the CGAN prediction model in Example 2;

[0078] Figure 7 is the day-ahead forecast result of the entire wind power in Example 2;

[0079] Figure 8 A schematic diagram of the structure of a computer device. DETAILED DESCRIPTION

[0080] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0081] Embodiment 1:

[0082] like Figure 1 As shown: A method for short-term wind power prediction considering wake and obstruction includes the following steps:

[0083] Collect the external environment and operation data of each wind turbine in the offshore wind farm; obtain the wind farm flow information data and perform pre-processing;

[0084] Obtain the layout and unit parameters of the wind farm wind turbines, establish an engineering analytical model that couples the Jensen wake model and the cylindrical vortex blocking model, and obtain the inflow wind speed and wind direction values ​​of the wind turbines based on the preprocessed wind farm inflow information data;

[0085] Construct an offshore wind power prediction model based on conditional generative adversarial networks;

[0086] The power prediction model is trained by taking a one-dimensional random noise vector that follows a standard normal distribution, the external environment of the wind turbine on the forecast day, the sine and cosine values ​​of the inflow wind speed and wind direction as the model input, and the actual active power of the offshore wind farm on the forecast day as the output;

[0087] The trained power prediction model is used to predict the offshore wind farm and obtain the power prediction results of the wind farm at each time on the prediction day.

[0088] By obtaining the layout and unit parameters of wind farm wind turbines, an engineering analytical model coupled with the Jensen wake model and the cylindrical vortex blockage model is established, and the inflow wind speed and wind direction values ​​of the wind turbine are obtained based on the preprocessed wind farm inflow information data. An offshore wind power prediction model based on a conditional generative adversarial network is constructed. The advantage of the conditional generative adversarial network in extracting high-dimensional features is combined, and the calculation results of the engineering analytical model are used as part of the prediction model input, realizing an effective combination of analytical models and deep learning methods. This can not only effectively improve the prediction accuracy of the day-ahead active output of offshore wind farms, but also enhance the interpretability and generalization of the prediction model.

[0089] In this embodiment, the external environment includes: external wind speed, wind direction, and turbulence intensity of each wind turbine;

[0090] Operational data include: active output of each wind turbine;

[0091] The wind farm inflow information data includes hourly wind speed data in the longitude and latitude directions at heights of 2m, 10m and 50m.

[0092] Among them, the wind farm flow information data is obtained and preprocessed. The preprocessing process includes:

[0093] Eliminate the two types of shutdown data and power limit points, where the wind speed is less than the cut-in wind speed and greater than the cut-out wind speed;

[0094] Treat negative data values ​​uniformly as zero;

[0095] Adjust the time resolution accuracy of each operating data to the preset value and fill in the missing data;

[0096] The longitude and latitude wind speed data in the wind farm inflow information data are converted into wind speed and wind direction angle, and converted to the height of 90m at the wind turbine hub using the following formula:

[0097] V 90m =1.484V 10m ;

[0098] Where: V90m , V 10m These are the wind speed values ​​at heights of 90m and 10m respectively;

[0099] The input wind speed and turbulence intensity data are standardized, the wind direction is represented by its sine and cosine values, and the active output data is proportionally scaled to between [0,1].

[0100] After collecting the external environment and operation data of each wind turbine in the offshore wind farm, representative wind turbines are selected according to the correlation between the output power of each wind turbine and the total output of the wind farm. In the subsequent model training, the external environment, inflow wind speed and wind direction sine and cosine values ​​of the representative wind turbines are used as model inputs.

[0101] When selecting representative units, the Pearson correlation analysis is used to determine the correlation between the output power of each wind turbine and the total output of the wind farm, including:

[0102]

[0103] Where: P k is the power sequence of the kth wind turbine, P farm is the power sequence of the wind farm, n is the number of elements in the sequence, t is the time; the correlation coefficient r k ∈[-1, 1], r k The larger it is, the stronger the correlation between the wind farm power and the wind power of the representative unit is.

[0104] In this embodiment, an offshore wind power prediction model based on a conditional generative adversarial network is constructed, including:

[0105] The generator model is composed of two layers of parallel gated recurrent unit (GRU) networks and one layer of fully connected network, and the internal structure of the power prediction model is constructed based on a one-dimensional convolutional neural network and a channel attention layer.

[0106] The wind farm wind turbine layout includes:

[0107] The longitude and latitude information of each wind turbine;

[0108] The unit parameters include: the unit's rated power and rated wind speed, cut-in and cut-out wind speeds, hub height, rotor diameter, and theoretical wind power and thrust coefficient curves.

[0109] An engineering analytical model that couples the wake model and the cylindrical vortex blocking model is established, including:

[0110] In the Jensen wake model, the wake velocity at the axial distance x of the wind rotor is:

[0111]

[0112] Where: U ∞ is the free flow wind speed at the fan hub height, C T is the thrust coefficient of the fan, R is the radius of the wind wheel, k w is the wake expansion coefficient;

[0113] The cylindrical vortex blocking model assumes that the wake geometry is a semi-infinite cylinder with a cross-sectional radius equal to the rotor radius R and a constant tangential vorticity γt, where the tangential vorticity calculation formula is:

[0114]

[0115] Where: is the average axial inductance in the rotor plane;

[0116] The induced velocity at the axial distance x of the wind wheel is:

[0117]

[0118] Where: r is the radial position, K and Π are the second and third complete elliptic integrals respectively, k is the ellipse parameter,

[0119]

[0120] Based on the above formula, when the radial position is 0, the simplified form of the velocity near the wake area is:

[0121]

[0122] The simplified form of the far wake velocity is:

[0123]

[0124] Using Taylor's formula to expand, we can get the simplified form:

[0125]

[0126] In the process of coupling the wake and blocking models, it is assumed that the induced velocity of the vortex column model is zero in the semi-infinite cylindrical volume and on the wind rotor plane. The velocity field of the cylindrical vortex model is used to describe the blocking effect upstream of the unit. At the same time, the Jensen wake model is used to represent the wake field downstream of the unit. The two velocity fields in other areas are directly linearly superimposed. The preprocessed wind farm inflow information data is input into the model to obtain the inflow wind speed and wind direction values ​​at the representative unit location.

[0127] When training the power prediction model, a one-dimensional random noise vector that obeys the standard normal distribution, the hourly wind speed and turbulence intensity of the representative unit on the prediction day, the sine and cosine values ​​of the inflow wind speed and wind direction are used as the input of the model, and the actual active power of the offshore wind farm with a time interval of one hour on the prediction day is used as the output. The one-dimensional random noise vector contains 24 elements.

[0128] In this embodiment, the mean absolute error L is used to train the power prediction model. MAE , mean absolute percentage error L MAPE , Root mean square error L RMSE and the normalized root mean square error L NRMSE As an evaluation criterion, determine whether the prediction accuracy of the model has been trained to meet expectations:

[0129]

[0130]

[0131]

[0132]

[0133] Where: y i is the target value of offshore wind power at the i-th moment; is the predicted value of offshore wind power at the i-th moment; n p For the number of prediction time steps, take 24.

[0134] In constructing an offshore wind power prediction model based on a conditional generative adversarial network, the loss functions of the generator and discriminator of the conditional generative adversarial network are expressed as:

[0135]

[0136]

[0137] The specific function expression of the adversarial loss based on the conditional generation adversarial network architecture is:

[0138]

[0139] Where: L(D,G) is the objective function of the conditional generative adversarial network, and Represent the standard normal distribution P z With the real data distribution P rThe expected value of the distribution, x is the real wind power data, y is the conditional variable, z is a noise sequence that obeys the standard normal distribution, D(x|y) is the similarity between the actual wind power active power sample and the real wind power population, D(G(z|y)|y) is the similarity between the predicted wind power active power sample and the real wind power population. In the training of the conditional generative adversarial network, the generator G is committed to minimizing the above objective function, but the discriminator D is committed to maximizing the above formula, thus forming a binary minimax game.

[0140] Among them, the loss function is added to its objective function, and the Huber loss function includes:

[0141]

[0142] Where: δ is a parameter, p and are one-dimensional vectors consisting of the actual value and predicted value of offshore wind power, respectively. i is the actual value of offshore wind power at the i-th moment, is the predicted value of offshore wind power at the i-th moment output by the CGAN generated network.

[0143] As a preference for the above embodiment, the final loss function of the generator of the conditional generative adversarial network is:

[0144] L G =λ1L′ G +λ2L Huber ((G(z|y),x);

[0145] Wherein: λ1 and λ2 are weight coefficients, λ1+λ2=1.

[0146] like Figure 2 As shown, this embodiment also includes a wind power short-term power prediction device considering wake and obstruction, using the above method, including:

[0147] The collection unit is used to collect the external environment and operation data of each wind turbine in the offshore wind farm; obtain the wind farm flow information data and perform pre-processing;

[0148] The engineering analytical model modeling unit is used to obtain the layout and unit parameters of the wind farm wind turbines, establish an engineering analytical model that couples the Jensen wake model and the cylindrical vortex blocking model, and obtain the inflow wind speed and wind direction values ​​of the wind turbines based on the preprocessed wind farm inflow information data;

[0149] A power prediction model modeling unit, used to build an offshore wind power prediction model based on a conditional generative adversarial network;

[0150] A training unit is used to train a power prediction model by taking a one-dimensional random noise vector that obeys a standard normal distribution, the external environment of the wind turbine generator set on the forecast day, the sine and cosine values ​​of the inflow wind speed and wind direction as model inputs, and the actual active power of the offshore wind farm on the forecast day as output;

[0151] The prediction unit is used to use the trained power prediction model to predict the offshore wind farm and obtain the power prediction results of the wind farm at each time on the prediction day.

[0152] Embodiment 2:

[0153] like Figure 3 As shown, this embodiment includes a method for short-term offshore wind power prediction considering wake and blocking effects, including the following steps:

[0154] Step 1: Collect the external environment and operation data of the 15 wind turbines contained in the offshore wind farm from the SCADA (Supervisory Control And Data Acquisition) system, and obtain the wind farm inflow information data from the MERRA-2 atmospheric reanalysis dataset produced by NASA's Global Modeling and Assimilation Office. Perform outlier cleaning, missing value processing, and time resolution adjustment on the SCADA data of each wind turbine, and convert and height convert the wind speed data in the MERRA-2 dataset.

[0155] The dataset records the measured data of 15 Sinovel doubly-fed wind turbines SL1-SL15 with a rated capacity of 3MW at Longyuan Offshore Wind Farm in Rudong County, Nantong City, Jiangsu Province, China, from January 1, 2014 to October 31, 2015, with a sampling time step of 10 minutes for a total of 669 days, including three types of data: external wind speed (Wspd), turbulence intensity (Ti) and active power output (Patv). See the unit distribution diagram for details. Figure 4 In terms of data processing, the SCADA data of 15 units were processed separately, and two types of shutdown data, including wind speed less than the cut-in wind speed of 3m / s and greater than the cut-out wind speed of 25m / s, and various power limit points were eliminated, and negative data values ​​were uniformly treated as zero values; in order to address the problem of missing some SCADA data, the time resolution of each operating characteristic data was reduced from 10 minutes to 1 hour.

[0156] The longitude and latitude wind speed data in the MERRA-2 dataset are converted into wind speed and wind direction angles, and converted to the height of 90m at the wind turbine hub using the following formula:

[0157] V 90m =1.484V 10m (1)

[0158] Where: V90m , V 10m They are the wind speed values ​​at 90m and 10m in the MERRA-2 dataset.

[0159] The input wind speed and turbulence intensity data are standardized, the wind direction is represented by its sine and cosine values, and the active output data is proportionally scaled to [0,1].

[0160] Step 2: Analyze the correlation between the output power of each wind turbine and the total output of the wind farm based on the Pearson correlation method, and select the representative wind farm unit with the largest correlation.

[0161] The Pearson correlation coefficient calculation formula is as follows:

[0162]

[0163] Where: P k is the power sequence of the kth wind turbine, P farm is the power sequence of the wind farm, n is the number of elements in the sequence, and t is the time. k ∈[-1, 1], r k The larger it is, the stronger the correlation between the wind farm power and the reference unit wind power is, and the smaller the error is when upscaling is used to predict the active power of the wind farm.

[0164] According to the above formula, the Pearson correlation coefficient of the output power time series of each wind turbine and the total output of the wind farm is calculated as follows: Figure 5 .according to Figure 5 The calculation results of the correlation coefficient show that SL8 with a correlation coefficient of 0.96 is selected as the representative unit of the wind farm.

[0165] Step 3: Obtain the wind farm wind turbine layout and unit parameters, establish an engineering analytical model that couples the Jensen wake model and the cylindrical vortex blocking model, and obtain the inflow wind speed and wind direction values ​​at the representative unit location based on the preprocessed MERRA-2 inflow information data.

[0166] Among them, the wind farm wind turbine layout includes: the longitude and latitude information of each wind turbine; the unit parameters include: the unit's rated power and rated wind speed, cut-in and cut-out wind speeds, hub height and thrust coefficient curve.

[0167] The Jensen wake model assumes that the wake velocity loss downstream of the wind turbine satisfies the "top hat" distribution. The wake velocity at the axial distance x of the wind rotor is

[0168]

[0169] Where: U ∞ is the incoming wind speed at the fan hub height, CT is the thrust coefficient of the fan, R is the radius of the wind wheel, k w is the wake expansion coefficient, which is generally taken as 0.05 for offshore wind farms.

[0170] The cylindrical vortex blocking model assumes that the wake geometry is a cross-sectional radius equal to the rotor radius R and has a constant tangential vorticity γ t The semi-infinite cylinder, where the tangential vorticity is calculated as:

[0171]

[0172] Where: is the average axial inductance in the rotor plane.

[0173] The induced velocity at the axial distance x of the wind rotor is

[0174]

[0175] Where: r is the radial position, K and Π are the second and third complete elliptic integrals respectively, k is the ellipse parameter,

[0176] Based on the above formula, when the radial position is 0, the simplified form of the velocity near the wake area is

[0177]

[0178] The simplified form of the velocity in the far wake region is

[0179]

[0180] The simplified form of Taylor's formula is:

[0181]

[0182] In the process of coupling the wake and blocking models, it is assumed that the induced velocity of the vortex column model is zero in the semi-infinite cylindrical volume and on the wind rotor plane. The velocity field of the cylindrical vortex model is used to describe the blocking effect upstream of the unit. At the same time, the Jensen wake model is used to represent the wake field downstream of the unit. The two velocity fields in other areas are directly superimposed linearly. The preprocessed MERRA-2 inflow information data is input into the model to obtain the inflow wind speed and wind direction values ​​at the representative unit location.

[0183] Step 4: Construct an offshore wind power short-term prediction model based on conditional generative adversarial network. This model combines a one-dimensional convolutional neural network and a GRU (Gate Recurrent Unit) network, and uses a two-layer parallel GRU network as the main part of the generator network. The internal structure of the discriminant model is constructed based on a one-dimensional convolutional neural network, and a channel attention layer is added to help the CNN network better focus on important feature channels and improve the classification performance of the discriminant model. The activation function of the output layer is set to the Sigmoid function, which maps the variable to between 0 and 1. The closer to 1, the closer the input wind power prediction value is to the true value. The overall structure of the model is shown in Figure 6 .

[0184] In order to solve the problem that the traditional GAN ​​(Generative Adversarial Networks) model cannot capture the temporal correlation of the original sequence well, the CGAN (Conditional Adversarial Networks) model uses the GRU network, which has excellent results in time series prediction, as the main part of the generator, and then transforms it into a one-dimensional time series of predicted wind power through a fully connected layer.

[0185] In general, the loss functions of the generator and the discriminator can be expressed as:

[0186]

[0187]

[0188] The specific function expression of the adversarial loss based on the conditional generation adversarial network architecture is:

[0189]

[0190] Where: L(D,G) is the objective function of CGAN, and Respectively represent the standard normal distribution Pz and the real data distribution P rThe expected value of the distribution, x is the actual wind power data, y is the conditional variable (wind speed and turbulence intensity), z is a noise sequence that obeys the standard normal distribution, D(x|y) is the similarity between the actual wind power active power sample and the actual wind power population, and D(G(z|y)|y) is the similarity between the predicted wind power active power sample and the actual wind power population. In the training of CGAN, the generator G is committed to minimizing the above objective function, but the discriminator D is committed to maximizing the above formula, thus forming a binary minimax game. After calculating the loss value, feedback is given in the form of a gradient function to update and optimize the network parameters of the generator and discriminator. When CGAN completes training, the two output probabilities of the discriminator fluctuate around 0.5, indicating that the model has reached the "Nash equilibrium" state at this time.

[0191] Relying solely on the discriminant feedback from the discriminator may not be enough for the generator to capture the real data distribution, so the Huber loss function is additionally used as the supervised loss term in the generator loss function. Huber loss combines the good properties of MSE (mean square error) and MAE, has better robustness to wind power time series, and can avoid losing the minimum value when training the model using the gradient descent method. The Huber loss function is defined as follows:

[0192]

[0193] Where: δ is a parameter, p and are one-dimensional vectors consisting of the actual value and predicted value of offshore wind power, pi is the actual value of offshore wind power at the i-th moment, is the predicted value of offshore wind power at the i-th moment output by the CGAN generated network.

[0194] Combined with formula (9), the final loss function of the generator is:

[0195] L G =λ1L′ G +λ2L Huber (G(z|y),x) (13)

[0196] Wherein: λ1 and λ2 are weight coefficients, λ1+λ2=1.

[0197] Step 5: Set the CGAN model hyperparameters, such as the number of input and output channels of the convolution layer in the discriminator, the convolution kernel size, the feature dimension of the GRU hidden layer in the generator, etc., and initialize the weights of the generator loss function. Construct a combined data set containing the SCADA wind speed, turbulence intensity data and incoming flow characteristics of representative units, and divide the data set into training set, validation set and test set. A one-dimensional random noise vector obeying a standard normal distribution, the hourly external environment of the representative unit on the forecast day, the sine and cosine values ​​of the inflow wind speed and wind direction are used as model inputs, and the actual active power of the offshore wind farm on the forecast day is used as output to complete the training of the power prediction model.

[0198] The training set is the data of 2014, the validation set is the data of January 2015, and the test set is the data of February to October 2015. The input data of the generator consists of two parts: a random noise vector that follows the standard normal distribution and a two-dimensional matrix consisting of conditional variables containing two types of meteorological data, and a two-dimensional matrix containing representative unit inflow data, both of which have a dimension of 3×24; the input of the discriminator includes the real wind power data spliced ​​with the conditional variables and the power prediction value generated by the generator, both of which have a dimension of 3×24. The input wind speed and turbulence intensity data are standardized, the wind direction is represented by its sine and cosine values, and the active output data is proportionally scaled to [0,1].

[0199] The output of the prediction model is a 1×24 vector, which corresponds to the wind farm power prediction sequence for each hour of the day to be predicted.

[0200] Step 6: After the training is completed, the optimal weight of the model loss function is generated according to the validation error and the model hyperparameters are adjusted. The CGAN model is used to predict the wind power of the entire offshore wind farm on the day before. After denormalization, the active power prediction results of the entire wind farm at each time on the prediction day are obtained. Figure 7 .

[0201] The mean absolute error (MAE), mean absolute percentage error (MAPE), root mean square error (RMSE) and normalized root mean square error (NRMSE) are used as the standards for measuring the accuracy of the model. Since each measurement standard has different characteristics, combining the four comprehensive evaluation indicators can obtain better accuracy measurement results. The smaller the error value, the higher the prediction accuracy. The function expressions of each indicator are as follows:

[0202]

[0203]

[0204]

[0205]

[0206] Where: y i is the target value of offshore wind power at the i-th moment, is the predicted value of offshore wind power at the i-th moment; n p For the number of prediction time steps, take 24.

[0207] To prove the effectiveness of the present invention, the wind farm measured total active power and predicted total power data from the 635th to 641st day (September 27th to October 3rd, 2015) in the test set were selected as verification samples, and the prediction accuracy of the present invention method was compared with that of the traditional time series prediction models LSTM (Long Short-Term Memory Neural Network), GRU (Gated Recurrent Unit Network) and Conditional Generative Adversarial Network (FCCGAN) built with a fully connected layer. Each model was trained based on the same external environment and power data, and the prediction results are shown in Table 1. Compared with the model that does not consider the inflow information of the representative unit, the calculated wind speed and wind direction angle sine and cosine values ​​determined by the engineering model are used as additional inputs of the prediction model, and the output power prediction accuracy of each model is improved, and it can be seen that the prediction effect of the prediction model used in the present invention is the best.

[0208] Table 1 Statistics of average power prediction errors of different models from day 635 to day 641

[0209]

[0210] See also Figure 8 A computer device 400 provided in an embodiment of the present application includes: a processor 410 and a memory 420, wherein the memory 420 stores a computer program executable by the processor 410, and when the computer program is executed by the processor 410, the above method is executed.

[0211] The embodiment of the present application further provides a storage medium 430 on which a computer program is stored. When the computer program is run by the processor 410, the above method is executed.

[0212] Among them, the storage medium 430 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable red-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, disk or optical disk.

[0213] In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. "Multiple" means two or more, unless otherwise clearly and specifically defined.

[0214] In the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0215] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, unless they are contradictory.

[0216] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code that includes one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred embodiments of the present invention includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present invention belong.

[0217] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways if necessary, and then stored in a computer memory.

[0218] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0219] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0220] The storage medium mentioned above may be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the present invention. A person of ordinary skill in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A method for short-term wind power prediction considering wake and blockage, characterized in that: The steps include: Collect the external environment and operation data of each wind turbine in the offshore wind farm; obtain the wind farm flow information data and perform pre-processing; Obtain the layout and unit parameters of the wind farm wind turbines, establish an engineering analytical model that couples the Jensen wake model and the cylindrical vortex blocking model, and obtain the inflow wind speed and wind direction values ​​of the wind turbine units based on the preprocessed wind farm inflow information data; Construct an offshore wind power prediction model based on conditional generative adversarial networks; The power prediction model is trained by taking a one-dimensional random noise vector that obeys a standard normal distribution, the external environment of the wind turbine generator set on the forecast day, the sine and cosine values ​​of the inflow wind speed and wind direction as model inputs, and the actual active power of the offshore wind farm on the forecast day as output; Use the trained power prediction model to predict the offshore wind farm to obtain power prediction results of the wind farm at each time on the prediction day; After collecting the external environment and operation data of each wind turbine in the offshore wind farm, a representative wind turbine is selected according to the correlation between the output power of each wind turbine and the total output of the wind farm. In subsequent model training, the external environment, inflow wind speed and wind direction sine and cosine values ​​of the representative wind turbine are used as model inputs; The engineering analytical model for establishing the coupling between the Jensen wake model and the cylindrical vortex blocking model includes: In the Jensen wake model, the wake velocity at the axial distance x of the wind rotor is: Where: U ∞ is the free flow wind speed at the fan hub height, C T is the thrust coefficient of the fan, R is the radius of the wind wheel, k w is the wake expansion coefficient; The cylindrical vortex blocking model assumes that the wake geometry is a cross-sectional radius equal to the rotor radius R and has a constant tangential vorticity γ t The semi-infinite cylinder, where the tangential vorticity is calculated as: Where: is the average axial inductance in the rotor plane; The induced velocity at the axial distance x of the wind wheel is: Where: r is the radial position, K and Π are the second and third complete elliptic integrals respectively, k is the ellipse parameter, Based on the above formula, when the radial position is 0, the simplified form of the velocity near the wake area is: The simplified form of the far wake velocity is: The simplified form obtained by expanding the Taylor formula is: In the process of coupling the wake and blocking models, it is assumed that the induced velocity of the vortex column model is zero in the semi-infinite cylindrical volume and on the wind rotor plane. The velocity field of the cylindrical vortex model is used to describe the blocking effect upstream of the unit. At the same time, the Jensen wake model is used to represent the wake field downstream of the unit. The two velocity fields in other areas are directly superimposed linearly. The preprocessed wind farm inflow information data is input into the model to obtain the inflow wind speed and wind direction values ​​at the representative unit location.

2. The method for short-term wind power prediction considering wake and blockage according to claim 1, characterized in that: The external environment includes: external wind speed, wind direction, and turbulence intensity of each wind turbine; The operation data include: active output of each wind turbine; The wind farm inflow information data includes: hourly wind speed data in longitude and latitude directions at heights of 2m, 10m and 50m.

3. The method for short-term wind power prediction considering wake and blockage according to claim 2, characterized in that: The wind farm incoming flow information data is obtained and preprocessed, and the preprocessing process includes: Eliminate the two types of shutdown data and power limit points, where the wind speed is less than the cut-in wind speed and greater than the cut-out wind speed; Treat negative data values ​​uniformly as zero; Adjust the time resolution accuracy of each operating data to the preset value and fill in the missing data; The longitude and latitude wind speed data in the wind farm inflow information data are converted into wind speed and wind direction angle, and converted to the height of 90m at the wind turbine hub using the following formula: V 90m =1.484V 10m ; Where: V 90m , V 10m These are the wind speed values ​​at heights of 90m and 10m respectively; The input wind speed and turbulence intensity data are standardized, the wind direction is represented by its sine and cosine values, and the active output data is proportionally scaled to between [0,1].

4. The method for short-term wind power prediction considering wake and blockage according to claim 1, characterized in that: When selecting the representative units, the correlation between the output power of each wind turbine and the total output of the wind farm is determined through Pearson correlation analysis, including: Where: P k is the power sequence of the kth wind turbine, P farm is the power sequence of the wind farm, n is the number of elements in the sequence, t is the time; the correlation coefficient r k ∈[-1, 1], r k The larger it is, the stronger the correlation between the wind farm power and the wind power of the representative unit is.

5. The method for short-term wind power prediction considering wake and blockage according to claim 1, characterized in that: The construction of an offshore wind power prediction model based on a conditional generative adversarial network includes: The generator model is composed of two layers of parallel gated recurrent unit GRU networks and one layer of fully connected network, and the internal structure of the power prediction model is constructed based on a one-dimensional convolutional neural network and a channel attention layer.

6. The method for short-term wind power prediction considering wake and blockage according to claim 1, characterized in that: The wind farm wind turbine layout includes: The longitude and latitude information of each wind turbine; The unit parameters include: rated power and rated wind speed of the unit, cut-in and cut-out wind speeds, hub height, rotor diameter, theoretical wind power and thrust coefficient curve.

7. The method for short-term wind power prediction considering wake and blockage according to claim 1, characterized in that: When training the power prediction model, a one-dimensional random noise vector obeying a standard normal distribution, the hourly wind speed and turbulence intensity, inflow wind speed and wind direction sine and cosine values ​​of the representative unit on the prediction day are used as inputs of the model, and the actual active power of the offshore wind farm with a time interval of one hour on the prediction day is used as output, and the one-dimensional random noise vector contains 24 elements.

8. The method for short-term wind power prediction considering wake and blockage according to claim 1, characterized in that: In the offshore wind power prediction model based on the conditional generative adversarial network, the loss functions of the generator and the discriminator of the conditional generative adversarial network are respectively expressed as: The specific function expression of the adversarial loss based on the conditional generation adversarial network architecture is: Where: L(D, G) is the objective function of the conditional generative adversarial network, and Represent the standard normal distribution P z With the real data distribution P r The expected value of the distribution, x is the real wind power data, y is the conditional variable, z is a noise sequence that obeys the standard normal distribution, D(x|y) is the similarity between the actual wind power sample and the real wind power population, D(G(z|y)y) is the similarity between the predicted wind power sample and the real wind power population. In the training of the conditional generative adversarial network, the generator G is committed to minimizing the above objective function, but the discriminator D is committed to maximizing the above formula, thus forming a binary maximum and minimum game; Add the loss function to its objective function. The Huber loss function includes: Where: δ is a parameter, p and are one-dimensional vectors consisting of the actual and predicted values ​​of offshore wind farm power, respectively, i is the actual value of offshore wind power at the i-th moment, is the predicted value of offshore wind power at the i-th moment output by the CGAN generated network; The final loss function of the generator of the conditional generative adversarial network is: L G =λ1L′ G +λ2L Huber (G(z|y,),x); Wherein: λ1 and λ2 are weight coefficients, λ1+λ2=1.

9. A wind power short-term power prediction device considering wake and blockage, characterized in that: Use of the method according to any one of claims 1 to 8, comprising: The collection unit is used to collect the external environment and operation data of each wind turbine in the offshore wind farm; obtain the wind farm flow information data and perform pre-processing; An engineering analytical model modeling unit is used to obtain the layout and unit parameters of the wind farm wind turbines, establish an engineering analytical model that couples the Jensen wake model and the cylindrical vortex blocking model, and obtain the inflow wind speed and wind direction values ​​of the wind turbine units based on the preprocessed wind farm inflow information data; A power prediction model modeling unit, used to build an offshore wind power prediction model based on a conditional generative adversarial network; A training unit is used to train the power prediction model by taking a one-dimensional random noise vector obeying a standard normal distribution, the external environment of the wind turbine generator set on the forecast day, the sine and cosine values ​​of the inflow wind speed and wind direction as model inputs, and the actual active power of the offshore wind farm on the forecast day as output; The prediction unit is used to use the trained power prediction model to predict the offshore wind farm and obtain the power prediction results of the wind farm at each time on the prediction day.

10. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 8 is implemented.

11. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images