Global wind power plant power prediction pre-training method based on data generation technology

Abstract

The invention provides a global wind power plant power prediction pre-training method based on a data generation technology, and the method comprises the steps: generating massive simulation wind power plant output data in a global range based on historical meteorological data and physical models of various wind speeds, wake flows, air densities, temperatures and the like, and fitting simulation data based on an improved Transform model. By adopting the global wind power plant power prediction pre-training method based on the data generation technology, the method can be suitable for any wind power plant without training of historical data of a real wind power plant, and a good prediction effect is obtained; and the problem of low universality caused by high customization of the current wind power short-term prediction model can be effectively solved.

Description

Technical Field

[0001] The present invention relates to the technical field of short-term power prediction of wind farms, and in particular to a global wind farm power prediction pre-training method based on data generation technology. Background Art

[0002] Short-term wind power forecasting is crucial for the reliable and economic operation of low-carbon power systems. Short-term wind power forecasting refers to predicting the 96-point power output value from 00:00 to 23:45 the next day at a fixed time point on the day (usually 8-12 am). At present, developers of short-term wind farm power forecasting models develop customized forecasting models based on factors such as different geographical characteristics of wind farms, different seasons, and different wind turbines. The model usually takes the numerical weather forecast (NWP) near the wind farm as input and outputs the power output of the next day. At present, developers of short-term wind farm power forecasting models develop customized forecasting models based on factors such as different geographical characteristics of wind farms, different seasons, and different wind turbines. The model is only applicable to specific wind farms. If it is forcibly relocated, the prediction accuracy will drop sharply. However, there is no universal short-term wind power forecasting model that can be applied to any wind farm and achieve good prediction results. Summary of the invention

[0003] The purpose of the present invention is to provide a global wind farm power prediction pre-training method based on data generation technology, which can be applied to any wind farm without the need for historical data of real wind farms for training and achieve good prediction results. It can effectively solve the problem of low universality caused by the high customization of current wind power short-term prediction models.

[0004] To achieve the above object, the present invention provides a global wind farm power prediction pre-training method based on data generation technology, comprising the following steps:

[0005] Generate pre-training data, wherein the pre-training data includes wind farm meteorological data, meteorological data, and power data;

[0006] The generation of meteorological data for wind farms includes selecting n points as wind farms within the world's longitude and latitude range, generating corresponding installed capacity, obtaining meteorological grid data within the global range, and interpolating the meteorological grid data to wind farms through linear interpolation, bilinear interpolation and nearest neighbor interpolation methods;

[0007] Generate power data from meteorological data, including determining the model of wind turbines in the wind farm, defining wind speed, wake, air density, temperature, and power model, and judging whether to smooth the wind power curve based on the power model, and finally obtaining the historical power output data of n wind farms;

[0008] Based on the historical output data of wind farms, and by fitting these historical data through the improved Transformer model, a global wind farm power prediction pre-training model is obtained.

[0009] Preferably, the wind speed model is divided into four types: logarithmic model, Hellman model, linear interpolation extrapolation model and logarithmic interpolation extrapolation model;

[0010] The calculation formula of the Logarithmic model is as follows:

[0011]

[0012] In the formula, w data Represents the wind speed of the data, w hub The wind speed h at the hub height hub Indicates the hub height, h data Indicates wind speed height, z 0 represents the roughness length;

[0013] The calculation formula of the Hellman model is as follows:

[0014]

[0015] In the formula, α is the Herman index, which is divided into onshore and offshore;

[0016] The calculation formula of the Linear interpolation extrapolation method is as follows:

[0017]

[0018] The calculation formula of the Logarithmic interpolation extrapolation method is as follows:

[0019]

[0020] In the formula, w 100,data Indicates the wind speed at a height of 100 meters, w 10,data Indicates the wind speed at a height of 10 meters, h 100,data Indicates 100 meters altitude, h 10,data Indicates a height of 10 meters.

[0021] Preferably, the temperature model uses a linear gradient to calculate the temperature at the hub height, and the calculation formula is as follows:

[0022] T hub =T air -0.0065·(h hub -h data ) (5);

[0023] Where, T air The temperature at 2m, T hub is the air temperature at hub height.

[0024] Preferably, the air density model uses the barometric pressure altitude equation or the ideal gas equation to calculate the air density at the hub height, which is divided into the atmospheric pressure model Barometric and the ideal gas model Ideal gas;

[0025] The calculation formula of the Barometric model is as follows:

[0026]

[0027] In the formula, ρ hub is the air density at the hub height; p is the surface pressure, ρ 0 、T 0 and p 0 are respectively the air density, temperature and air pressure of the surrounding environment;

[0028] The calculation formula of the Ideal gas model is as follows:

[0029]

[0030] Preferably, the power model calculates the active output of the wind farm through a power curve or a power coefficient curve, which is divided into a power curve model and a power coefficient curve model;

[0031] The calculation formula of the power factor curve model is as follows:

[0032]

[0033] Where P is the active output of the wind turbine; w hub,i , i=1,2,3 represents the wind speed model; ρ hub,m , m=1,2, represents the air density model; d rotor is the rotor diameter of the wind turbine; cp is the power coefficient. The power coefficient curve model requires the wind speed power curve of the wind turbine, directly converts the wind speed at the hub height into active output, and corrects the wind speed at the hub height based on the air density model at the hub height. The specific calculation formula is as follows:

[0034]

[0035] In the formula, v stdis the standard wind speed, v site To correct the wind speed for density, p(w) is calculated as follows:

[0036]

[0037] The power output is zero for wind speeds above the maximum wind speed and below the minimum wind speed given in the power curve. The output of the wind farm is obtained by multiplying the active output of the wind turbines by the number of wind turbines:

[0038] P farm =P·k (11);

[0039] The number of wind turbines in a wind farm is obtained by dividing the installed capacity of the wind farm by the rated capacity of the wind turbines:

[0040] k=C / P rated (12);

[0041] In the formula, k represents the number of wind turbines in the wind farm, C represents the rated capacity of the wind farm, and P rated Indicates the rated capacity of the wind turbine.

[0042] Preferably, the wind power curve is smoothed, and the specific calculation formula is:

[0043]

[0044] Where P smoothed is the smoothed power curve, Δw i w i and w i+1 The length of the interval between, μ is the mean, σ is the standard deviation, where the standard deviation is calculated as follows:

[0045] σ=0.6·0.2·w (14). Preferably, the historical data are fitted by improving the Transformer model, including

[0046] First, the wind farm weather forecast data and historical power output data are segmented into non-overlapping patches through patch partitioning;

[0047] Then, it is projected into a feature vector of any dimension through linear embedding as the input of the Transformer module;

[0048] Finally, after passing through layers 2 and 3 consisting of patch aggregation and Transformer modules, the next day's power output is finally output.

[0049] A global wind farm power forecast pre-training system based on data generation technology, including

[0050] The data generation module is used to generate pre-training data, wherein the pre-training data includes wind farm meteorological data, meteorological data, and power data;

[0051] The data generation module includes a wind farm meteorological data generation unit and a meteorological data generation unit;

[0052] The fitting module is used to fit the historical data based on the historical output data of wind farms by improving the Transformer model, thereby obtaining a pre-trained model for global wind farm power prediction.

[0053] An electronic device comprises a memory and a processor, wherein a computer program is stored in the memory, and when the processor calls the computer program in the memory, the steps of a global wind farm power prediction pre-training method based on data generation technology are implemented.

[0054] A storage medium stores computer executable instructions, which, when loaded and executed by a processor, implement the steps of a global wind farm power prediction pre-training method based on data generation technology.

[0055] Therefore, the present invention adopts the above-mentioned global wind farm power prediction pre-training method based on data generation technology, which can be applied to any wind farm without the need for historical data of real wind farms for training, and can achieve good prediction results. It can effectively solve the problem of low universality caused by the high customization of current wind power short-term prediction models. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 Generate a flow chart for pre-training data;

[0057] Figure 2 It is an improved Transformer model. DETAILED DESCRIPTION

[0058] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.

[0059] Unless otherwise defined, technical or scientific terms used in the present invention shall have the common meanings understood by one having ordinary skills in the field to which the present invention belongs.

[0060] Embodiment 1

[0061] The present invention is a global wind farm power prediction pre-training method based on data generation technology, comprising the following steps:

[0062] Pre-training data generation: The overall process of pre-training data generation technology is as follows Figure 1As shown in Figure 1, it is divided into two parts: generating wind farm meteorological data and generating power data from meteorological data. In the part of generating wind farm meteorological data, first, n points are randomly selected as wind farms within the range of longitude ([0,360]) and latitude ([90N,90S]) in the world, and the wind farms are generated in [C min ,C max ] randomly generates the installed capacity of the wind farm within the range, where C min Indicates the lower limit of installed capacity, C min Indicates the upper limit of installed capacity. Secondly, obtain the ERA5 open source historical meteorological grid data of ECMWF (European Centre for Medium-Range Weather Forecasts). The spatial range of the data is global, the spatial resolution is 0.25×0.25°, the temporal resolution is 1 hour, and the variables include the u component of 100-meter wind, the v component of 100-meter wind, the u component of 10-meter wind, the v component of 10-meter wind, the 2-meter temperature and surface pressure. The u component and v component of 100-meter wind are converted into the 100-meter wind speed w through formula (1), and the u component and v component of 10-meter wind are converted into the 10-meter wind speed w. Finally, the meteorological grid data is interpolated to the wind farm through linear interpolation, bilinear interpolation and nearest neighbor interpolation methods (the selection probability of each method is 1 / 3).

[0063]

[0064] Meteorological data generates power data, including:

[0065] The first step is to determine the model of the wind turbine in the wind farm and obtain information such as the hub height, rotor radius, rated power, wind speed power curve, etc. of the wind turbine of this model. The number of wind turbines in the wind farm is obtained by dividing the installed capacity of the wind farm by the rated capacity of the wind turbine:

[0066] k=C / P rated (2);

[0067] In the formula, k represents the number of wind turbines in the wind farm, C represents the rated capacity of the wind farm, and P rated Indicates the rated capacity of the wind turbine.

[0068] The second step is to define the wind speed model and convert the 100-meter and 10-meter wind speed data to the hub height of the wind turbine. The wind speed model is divided into four types: Logarithmic, Hellman, Linear interpolationextrapolation and Logarithmic interpolation extrapolation. The calculation formula of the Logarithmic model is as follows:

[0069]

[0070] In the formula, w data Represents the wind speed of the data, w hub Indicates the wind speed at hub height in m / s; h hub Indicates the hub height, h data Indicates wind speed height, in m; z 0 Represents the roughness length. The roughness length is divided into 9 levels, 0.0002, 0.0024, 0.03, 0.055, 0.1, 0.2, 0.4, 0.6, 1.6. The probability of selecting the first 6 levels is 0.95, and the probability of the last three levels is 0.05. The calculation formula of the Hellman model is as follows:

[0071]

[0072] Where α is the Herman index, which is divided into onshore (subject to a normal distribution with a mean of 1 / 7 and a variance of 0.01) and offshore (subject to a normal distribution with a mean of 1 / 9 and a variance of 0.01), and the probability of selecting both is 1 / 2. The historical meteorological data has wind speeds at 100 meters and 10 meters. The interpolation extrapolation method can combine the wind speed data of 100 meters and 10 meters to obtain the wind speed at the approximate hub height. The calculation formula of the linear interpolation extrapolation method is as follows:

[0073]

[0074] The calculation formula of the Logarithmic interpolation extrapolation method is as follows:

[0075]

[0076] In the formula, w 100,data Indicates the wind speed at a height of 100 meters, w 10,data Indicates the wind speed at a height of 10 meters, h 100,data Indicates 100 meters altitude, h 10,dataIndicates a height of 10 meters.

[0077] Logarithmic and Hellman have two methods each (from 100-meter wind speed to hub-height wind speed and from 10-meter wind speed to hub-height wind speed), Linear interpolation extrapolation and Logarithmic interpolation extrapolation have one method each, for a total of 6 methods, and the probability of selecting each method is 1 / 6.

[0078] The third step is to define the temperature model, which uses a linear gradient to calculate the temperature at the hub height. The calculation formula is as follows:

[0079] T hub =T air -0.0065·(h hub -h data ) (7);

[0080] Where, T air The temperature at 2m, T hub is the air temperature at the hub height, both in K.

[0081] The fourth step is to define the air density model. This model uses the pressure altitude equation or the ideal gas equation to calculate the air density at the hub height. It is divided into Barometric (atmospheric pressure) and Ideal gas (ideal gas). The probability of selecting both models is 1 / 2. The calculation formula of the Barometric model is as follows:

[0082]

[0083] In the formula, ρ hub is the air density at the hub height, in kg / m 3 ; p is the surface pressure, unit is Pa; ρ 0 、T 0 and p 0 They are the air density (default value is 1.225), temperature (default value is 288.15) and air pressure (default value is 101330) of the surrounding environment. The calculation formula of the Ideal gas model is as follows:

[0084]

[0085] The fifth step is to define the wake model, which reduces the wind speed through the wind efficiency curve. The wind efficiency curve is divided into Knorr and Dena. The two wind efficiency curves are calculated by averaging the wind efficiency curves of 12 wind farms in Germany (Dena) and more than 2,000 wind farms in Germany (Knorr). The probability of selecting each model is 1 / 2.

[0086] The sixth step is to define the power model, which calculates the active output of the wind farm through the power curve or power coefficient curve, which is divided into Power curve and Power coefficient curve. The calculation formula of Power coefficient curve is as follows:

[0087]

[0088] Where P is the active output of the wind turbine; w hub,i , i=1,2,3 represents the wind speed model; ρ hub,m , m=1,2, represents the air density model; d rotor is the rotor diameter of the wind turbine; cp is the power coefficient. The power coefficient curve model requires the wind speed power curve of the wind turbine, directly converts the wind speed at the hub height into active output, and corrects the wind speed at the hub height based on the air density model at the hub height. The specific calculation formula is as follows:

[0089]

[0090] In the formula, v std is the standard wind speed, v site To correct the wind speed for density, p(w) is calculated as follows:

[0091]

[0092] After obtaining the power data, a post-processing operation will be performed. The power output of wind speeds above the maximum wind speed and below the minimum wind speed given in the power curve is zero. The selection probability of the two models, Power curve and Power coefficient curve, is 1 / 2. The output of the wind farm is obtained by multiplying the active output of the wind turbine by the number of wind turbines:

[0093] P farm =P·k (13);

[0094] Step 7: Smoothing the wind power curve. The specific calculation formula is:

[0095]

[0096] Where P smoothed is the smoothed power curve, Δw i w i and w i+1 The length of the interval between, μ is the mean, and σ is the standard deviation. The calculation formula of the standard deviation is as follows:

[0097] σ = 0.6 · 0.2 · w (15);

[0098] The smoothing process is to take the spatial distribution of wind speed into account, and the probability of whether to choose to perform smoothing is 1 / 2. Finally, the historical active output data of the wind farm is output.

[0099] Fitting the wind farm pre-trained model: This part fits n sets of pre-trained data based on the improved Transformer model. The improved Transformer model is as follows: Figure 2 As shown in the figure. First, the hourly weather forecast data of the wind farm for the next 24 hours (M×N×24, M is the number of meteorological elements, N is the number of grid points in the meteorological grid data) and the historical 24-hour 15-minute power output data (1×96) are divided into non-overlapping patches through patch partitioning, and then projected to feature vectors of any dimension through linear embedding as the input of the Transformer module, and then pass through layer 2 and layer 3 composed of patch aggregation and Transformer modules, and finally output the power output of 96 points for the next day. The Transformer module consists of a multi-head self-attention mechanism (MSA) module and a nonlinear multi-layer perceptron (MLP) module. Layer regularization (LN) is applied before each MSA module and each MLP, and residual connection is applied after each module.

[0100] A global wind farm power forecast pre-training system based on data generation technology, including

[0101] The data generation module is used to generate pre-training data, wherein the pre-training data includes wind farm meteorological data, meteorological data, and power data;

[0102] The data generation module includes a wind farm meteorological data generation unit and a meteorological data generation unit;

[0103] The fitting module is used to fit the historical data based on the historical output data of wind farms by improving the Transformer model, thereby obtaining a pre-trained model for global wind farm power prediction.

[0104] Since the embodiments of the apparatus part correspond to the embodiments of the method part, please refer to the description of the embodiments of the method part for the embodiments of the apparatus part, which will not be repeated here.

[0105] The present application also provides a storage medium on which a computer program is stored, and when the computer program is executed, the steps provided in the above embodiment can be implemented. The storage medium may include: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.

[0106] The present application also provides an electronic device, which may include a memory and a processor, wherein a computer program is stored in the memory, and when the processor calls the computer program in the memory, the steps provided in the above embodiment may be implemented. Of course, the electronic device may also include various network interfaces, power supplies and other components.

[0107] Therefore, the present invention adopts the above-mentioned global wind farm power prediction pre-training method based on data generation technology, which can be applied to any wind farm without the need for historical data of real wind farms for training, and can achieve good prediction results. It can effectively solve the problem of low universality caused by the high customization of current wind power short-term prediction models.

[0108] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A global wind farm power forecasting pre-training method based on data generation technology, characterized in that: The following steps are involved: Generate pre-training data, wherein the pre-training data includes wind farm meteorological data, meteorological data, and power data; The generation of meteorological data for wind farms includes selecting n points as wind farms within the world's longitude and latitude range, generating corresponding installed capacity, obtaining meteorological grid data within the global range, and interpolating the meteorological grid data to wind farms through linear interpolation, bilinear interpolation and nearest neighbor interpolation methods; Generate power data from meteorological data, including determining the model of wind turbines in the wind farm, defining wind speed, wake, air density, temperature, and power model, and judging whether to smooth the wind power curve based on the power model, and finally obtaining the historical power output data of n wind farms; Based on the historical output data of wind farms, and by fitting these historical data through the improved Transformer model, a global wind farm power prediction pre-training model is obtained.

2. The global wind farm power prediction pre-training method based on data generation technology according to claim 1 is characterized in that: Wind speed models are divided into four types: Logarithmic model, Hellman model, Linear interpolation extrapolation model and Logarithmic interpolation extrapolation model; The calculation formula of the Logarithmic model is as follows: In the formula, w data Represents the wind speed of the data, w hub The wind speed h at the hub height hub Indicates the hub height, h data represents the wind speed height, z0 represents the roughness length; The calculation formula of the Hellman model is as follows: In the formula, α is the Herman index, which is divided into onshore and offshore; The calculation formula of the Linear interpolation extrapolation method is as follows: The calculation formula of the Logarithmic interpolation extrapolation method is as follows: In the formula, w 100,data Indicates the wind speed at a height of 100 meters, w 10,data Indicates the wind speed at a height of 10 meters, h 100,data Indicates 100 meters altitude, h 10,data Indicates a height of 10 meters.

3. The global wind farm power prediction pre-training method based on data generation technology according to claim 1 is characterized in that: The temperature model uses a linear gradient to calculate the temperature at the hub height using the following formula: T hub =T air -0.0065·(h hub -h data ) (5); Where, T air The temperature at 2m, T hub is the air temperature at hub height.

4. The global wind farm power prediction pre-training method based on data generation technology according to claim 1 is characterized in that: The air density model uses the pressure altitude equation or the ideal gas equation to calculate the air density at the hub height. It is divided into the atmospheric pressure model Barometric and the ideal gas model Ideal gas. The calculation formula of the Barometric model is as follows: In the formula, ρ hub is the air density at the hub height; p is the surface pressure, ρ0, T0 and p0 are the air density, temperature and air pressure of the surrounding environment respectively; The calculation formula of the Idealgas model is as follows:

5. The global wind farm power prediction pre-training method based on data generation technology according to claim 1 is characterized in that: The power model calculates the active output of the wind farm through the power curve or power coefficient curve, which is divided into the power curve model and the power coefficient curve model; The calculation formula of the power factor curve model is as follows: Where P is the active output of the wind turbine; w hub,i , i=1,2,3 represents the wind speed model; ρ hub,m , m=1,2, represents the air density model; d rotor is the rotor diameter of the wind turbine; cp is the power coefficient. The power coefficient curve model requires the wind speed power curve of the wind turbine, directly converts the wind speed at the hub height into active output, and corrects the wind speed at the hub height based on the air density model at the hub height. The specific calculation formula is as follows: In the formula, v std is the standard wind speed, v site To correct the wind speed for density, p(w) is calculated as follows: The power output is zero for wind speeds above the maximum wind speed and below the minimum wind speed given in the power curve. The output of the wind farm is obtained by multiplying the active output of the wind turbines by the number of wind turbines: P farm =P·k (11); The number of wind turbines in a wind farm is obtained by dividing the installed capacity of the wind farm by the rated capacity of the wind turbines: k=C / P rated (12); In the formula, k represents the number of wind turbines in the wind farm, C represents the rated capacity of the wind farm, and P rated Indicates the rated capacity of the wind turbine.

6. The global wind farm power prediction pre-training method based on data generation technology according to claim 1 is characterized in that: The wind power curve is smoothed, and the specific calculation formula is: Where P smoothed is the smoothed power curve, Δw i w i and w i+1 The length of the interval between, μ is the mean, σ is the standard deviation, where the standard deviation is calculated as follows: σ=0.6·0.2·w(14).

7. The global wind farm power prediction pre-training method based on data generation technology according to claim 1 is characterized in that: By improving the Transformer model to fit these historical data, including First, the wind farm weather forecast data and historical power output data are segmented into non-overlapping patches through patch partitioning; Then, it is projected into a feature vector of any dimension through linear embedding as the input of the Transformer module; Finally, after passing through layers 2 and 3 consisting of patch aggregation and Transformer modules, the next day's power output is finally output.

8. A global wind farm power prediction pre-training system based on data generation technology, characterized in that: include The data generation module is used to generate pre-training data, wherein the pre-training data includes wind farm meteorological data, meteorological data, and power data; The data generation module includes a wind farm meteorological data generation unit and a meteorological data generation unit; The fitting module is used to fit the historical data based on the historical output data of wind farms by improving the Transformer model, thereby obtaining a pre-trained model for global wind farm power prediction.

9. An electronic device, characterized in that: The method comprises a memory and a processor, wherein a computer program is stored in the memory, and when the processor calls the computer program in the memory, the steps of the global wind farm power prediction pre-training method based on the data generation technology as claimed in any one of claims 1 to 7 are implemented.

10. A storage medium, characterized in that: The storage medium stores computer executable instructions, which, when loaded and executed by a processor, implement the steps of the global wind farm power forecasting pre-training method based on data generation technology as claimed in any one of claims 1 to 7.

Images