Wind turbine generator operation control optimization method and system based on deep learning

Abstract

The invention relates to the field of wind turbine generator control, in particular to a wind turbine generator operation control optimization method and system based on deep learning, and the method comprises the steps: obtaining the vibration, bearing temperature and generator winding temperature of a wind turbine generator, and the wind speed, wind direction and air density of the wind turbine generator; inputting all acquired data into a control parameter acquisition model, and outputting an ideal control parameter set of the current wind turbine generator; calculating an accurate rated wind speed; adjusting the control frequency based on the wind speed and the wind direction; each control parameter is adjusted; the wind turbine generator operation control optimization system comprises a data monitoring module, a control parameter acquisition module, a control frequency adjustment module and a pitch angle adjustment module. The method has the advantages that limitation of a traditional single-machine control strategy is broken through, especially when the wind speed is lower than the rated wind speed, the ideal increment of the pitch angle is actively adjusted, and therefore the total generating capacity of the wind power plant is improved.

Description

Technical Field

[0001] The present invention relates to the field of wind turbine control, and in particular to a method and system for optimizing wind turbine operation control based on deep learning. Background Art

[0002] With the transformation of the global energy structure, wind energy, as a clean and renewable energy, has increased its share in the power system year by year. Large wind farms achieve large-scale power generation by clustering wind turbines, but in actual operation, the wake effect of upstream wind turbines will have a significant shielding effect on downstream units. Studies have shown that the wake effect can cause the output power of downstream units to decrease by 10% to 30%, seriously affecting the overall power generation efficiency of the wind farm. Traditional wind turbine control strategies usually adopt a fixed parameter adjustment mode based on a single-machine model, such as maintaining constant power output by adjusting the pitch angle above the rated wind speed, or pursuing maximum wind energy capture below the rated wind speed. However, this type of method ignores the dynamic coupling relationship of the wake effect and the real-time influence of environmental parameters, resulting in low coordination efficiency between units.

[0003] In recent years, the initial application of deep learning technology in wind farm optimization has shown potential, such as predicting short-term wind speed or optimizing the pitch angle of a single machine through neural networks; however, existing research still has two major limitations: first, the control model focuses on the state parameters of a single machine (such as speed and power), and lacks explicit modeling of the wake coupling effect between units; second, the input feature dimension is single, and it fails to effectively integrate the multimodal data of mechanical health indicators such as vibration and temperature with environmental parameters; especially under conditions below the rated wind speed, traditional methods usually aim to maximize the power of a single machine, while ignoring the global optimization of the power loss of downstream units. This "local optimal" strategy will cause hidden losses in the total power generation of the wind farm.

[0004] In the field of wake effect modeling, existing technologies mostly rely on simplified analytical models or computational fluid dynamics (CFD) simulations. Although the analytical model has a fast calculation speed, it assumes that the wake diffuses linearly, which makes it difficult to reflect the real flow characteristics under complex terrain and dynamic wind conditions; although the CFD method has high accuracy, it takes too long to calculate and cannot meet real-time control requirements; therefore, in order to solve the above problems, the present invention proposes a wind turbine operation control optimization method and system based on deep learning. Summary of the invention

[0005] The present invention integrates the real-time operating status parameters of the wind turbine such as vibration, temperature, wind speed, etc., combined with environmental characteristics such as air density and wind direction vector, to dynamically perceive the coupling relationship between the health of the unit and the environmental conditions, and generate a more adaptable set of ideal control parameters such as pitch angle and yaw angle. The core advantage lies in the construction of a deep collaborative optimization framework for multi-source heterogeneous data, breaking through the limitations of traditional single-machine control strategies. Especially when the wind speed is lower than the rated wind speed, a screening mechanism for the unit with the greatest impact of the wake effect is introduced, and the pitch angle is actively adjusted to an ideal value-added, thereby increasing the total power generation of the wind farm.

[0006] A wind turbine operation control optimization method based on deep learning, comprising:

[0007] For any wind turbine in the wind farm, the vibration, bearing temperature, generator winding temperature and wind speed of the wind turbine are obtained at the current monitoring time point, and the current wind direction and air density are obtained at the same time. All the data obtained at the current monitoring time point are input into the control parameter acquisition model, and the ideal control parameter set of the current wind turbine is output through the control parameter acquisition model, including pitch angle, yaw angle, generator speed and generator torque;

[0008] Calculates accurate rated wind speed based on current air density;

[0009] For any wind turbine generator set, based on the wind speed and wind direction of the wind turbine generator set, the control frequency of the wind turbine generator set is adjusted;

[0010] For any wind turbine, at the current control time point, if the wind speed of the wind turbine collected most recently is greater than or equal to the precise rated wind speed, the ideal control parameter set is directly applied to adjust the control parameters of the current wind turbine;

[0011] Otherwise, select the downstream unit with the greatest impact on the current wind turbine, and at the current control time point, sort the wind speed obtained by the current wind turbine and the wind speed obtained by the downstream unit with the greatest impact in the recent period of time into upstream unit wind speed time series data and downstream unit wind speed time series data in chronological order; obtain the ideal value-added pitch angle through the pitch angle adjustment model; modify the ideal control parameter set, and use the modified ideal control parameter set to adjust various control parameters of the current wind turbine.

[0012] Preferably, applying the trained pitch angle adjustment model to obtain the ideal pitch angle increment includes:

[0013] Select the geographical center of the wind farm as the coordinate origin, establish the wind farm coordinate system with due east as the X-axis direction and due north as the Y-axis direction, and the unit length of the wind farm coordinate system is the rotor diameter of the wind turbine; subtract the coordinates of the current wind turbine from the coordinates of the downstream unit with the largest impact to obtain the relative position vector; and at the same time obtain the wind direction vector with a modulus length of 1 in the wind farm coordinate system of the current wind direction;

[0014] The upstream unit wind speed time series data, the downstream unit wind speed time series data, the relative position vector, the air density and the wind direction vector are input into the pitch angle adjustment model, and the ideal increment of the current wind turbine pitch angle is output.

[0015] Preferably, the accurate rated wind speed is calculated based on the current air density, and the calculation formula is:

[0016] ;

[0017] In the formula, For accurate rated wind speed, is the current air density, is the default standard air density, It is the original design rated wind speed.

[0018] Preferably, for any wind turbine set, based on the wind speed and wind direction of the wind turbine set, the control frequency of the wind turbine set is adjusted, including:

[0019] Set the length to The sliding time window is For any wind turbine, , in the latest sliding time window Get the wind speed time series data of the current wind turbine and wind direction time series data , =1, 2, …, ; Indicates the number of monitoring time points in a sliding time window; Indicates the current time;

[0020] Calculate wind speed time series data Standard Deviation , using the formula Calculate the turbulence compensation coefficient ; express Wind speed time series data The average value of

[0021] Using the formula Calculate wind speed fluctuation intensity ; Wind speed fluctuation intensity Normalize to obtain the corrected wind speed fluctuation intensity ;

[0022] Calculate wind direction change , using the formula Calculate the intensity of wind direction fluctuations, is the effective cumulative damping factor of wind direction; Normalize to obtain the corrected wind direction fluctuation intensity ;

[0023] Using the formula Calculate the intensity of fusion fluctuations , ; Based on the preset minimum monitoring frequency and the maximum monitoring frequency , using the formula Calculate the adjusted control frequency of the current wind turbine ; is the control coefficient.

[0024] Preferably, the downstream unit with the greatest impact on the current wind turbine generator set is selected, including:

[0025] Filter out wind turbines whose distance to the current wind turbine is less than the effective distance threshold Other candidate wind turbines include: is the rotor diameter; for any candidate wind turbine, based on the current wind turbine position and the current candidate wind turbine locations , calculate the distance between the two ;

[0026] Based on relative position vector and wind direction vector , for the relative position vector Unitize , calculate the alignment ;

[0027] Using the formula Calculate the degree to which the current candidate wind turbine is affected by the wake effect of the current wind turbine, where: is the distance attenuation coefficient;

[0028] The candidate wind turbine that is most affected by the wake effect of the current wind turbine is taken as the downstream unit with the greatest impact on the current wind turbine.

[0029] Preferably, the training process of the control parameter acquisition model includes:

[0030] Initialize the parameters in the control parameter acquisition model;

[0031] Obtaining a number of first training samples with labeled ideal control parameter sets, each of which includes a wind turbine vibration, a bearing temperature, a generator winding temperature, a wind speed and direction, and an air density of the wind turbine;

[0032] Dividing all acquired first training samples into a first training set and a first validation set;

[0033] The control parameter acquisition model is trained by using a first training set, and then the control parameter acquisition model is verified by using a first verification set to obtain a first verification result;

[0034] It is determined whether the obtained first verification result meets the preset first training condition. If so, the trained control parameter acquisition model is output; if not, the control parameter acquisition model is continuously trained through the first training set.

[0035] Preferably, the pitch angle adjustment model is established based on a convolutional neural network, including a second input layer, a feature extraction layer, a feature splicing layer, a second fully connected layer and a second output layer.

[0036] Preferably, the training process of the pitch angle adjustment model includes:

[0037] A plurality of second training samples with ideal pitch angle increments marked are obtained, each of which contains the upstream unit wind speed time series data, the downstream unit wind speed time series data, the wind direction vector, the air density and the relative position vector;

[0038] Dividing all acquired second training samples into a second training set and a second validation set;

[0039] The pitch angle adjustment model is trained by using the second training set, and the pitch angle adjustment model is verified by using the second verification set to obtain a second verification result;

[0040] It is determined whether the obtained second verification result meets the preset second training condition. If so, the trained pitch angle adjustment model is output; if not, the pitch angle adjustment model is continuously trained through the second training set.

[0041] A wind turbine operation control optimization system based on deep learning, comprising:

[0042] The data monitoring module is used to obtain the vibration of any wind turbine in the wind farm, the temperature of the bearing, the temperature of the generator winding, and the wind speed to which the wind turbine is subjected at the current monitoring time point, as well as the current wind direction and air density;

[0043] A control parameter acquisition module is used to input all data acquired at the current monitoring time point into a control parameter acquisition model, and output an ideal control parameter set of the current wind turbine through the control parameter acquisition model, including pitch angle, yaw angle, generator speed and generator torque;

[0044] A control frequency adjustment module is used to adjust the control frequency of any wind turbine generator set based on the wind speed and wind direction of the wind turbine generator set;

[0045] A pitch angle adjustment module, including a rated wind speed adjustment unit and a pitch angle adjustment unit;

[0046] Rated wind speed adjustment unit, used to calculate and obtain accurate rated wind speed based on current air density;

[0047] The pitch angle adjustment unit is used for any wind turbine set. At the current control time point, if the wind speed of the wind turbine set collected most recently at the current control time point is greater than or equal to the precise rated wind speed, then the ideal control parameter set is directly applied to adjust the various control parameters of the current wind turbine set; otherwise, the downstream unit with the largest impact on the current wind turbine set is selected, and at the current control time point, the wind speed obtained by the current wind turbine set and the wind speed obtained by the downstream unit with the largest impact in the recent period are sorted in time to form the upstream unit wind speed time series data and the downstream unit wind speed time series data; the ideal value-added of the pitch angle is obtained through the pitch angle adjustment model, the ideal control parameter set is modified according to the ideal value-added of the pitch angle, and the modified ideal control parameter set is applied to adjust the various control parameters of the current wind turbine set.

[0048] The present invention has the following advantages:

[0049] 1. The present invention integrates the real-time operating status parameters of the wind turbine such as vibration, temperature, wind speed, etc., combined with environmental characteristics such as air density and wind direction vector, to dynamically perceive the coupling relationship between the health of the unit and the environmental conditions, and generate a more adaptable ideal control parameter set such as pitch angle and yaw angle. The core advantage lies in the construction of a deep collaborative optimization framework for multi-source heterogeneous data, breaking through the limitations of traditional single-machine control strategies. Especially when the wind speed is lower than the rated wind speed, a screening mechanism for the unit with the greatest influence of the wake effect is introduced, and the pitch angle is actively adjusted to an ideal value-added, thereby increasing the total power generation of the wind farm.

[0050] 2. The present invention analyzes the two-dimensional fluctuation characteristics of wind speed and direction through a real-time sliding window, and combines the turbulence compensation coefficient to adaptively correct the wind speed intensity and the wind direction damping factor to suppress invalid angle jumps, thereby realizing a fully data-driven wind turbine control frequency adjustment, and adjusting the control frequency on demand based on the intensity of environmental fluctuations, which can balance the control accuracy and system energy consumption; it can quickly improve the response speed to sudden strong turbulence or continuous wind direction deviation, and automatically reduce the computing load to low-frequency small disturbances, thereby reducing frequency vibration losses and extending the life of the unit while ensuring the accuracy of wake coordinated control. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is a schematic diagram of the structure of a wind turbine operation control optimization system based on deep learning adopted in an embodiment of the present invention. DETAILED DESCRIPTION

[0052] In order to enable persons skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0053] Embodiment 1, a wind turbine operation control optimization method based on deep learning, comprising:

[0054] For any wind turbine in the wind farm, the wind turbine vibration, bearing temperature, generator winding temperature and wind speed of the wind turbine are obtained at the current monitoring time point, and the current wind direction and air density are also obtained. These data are important indicators of the operating status of the wind turbine, and can directly or indirectly reflect the health status and operating efficiency of the wind turbine. For example, the vibration of the wind turbine can reveal whether there is abnormal wear or imbalance of mechanical parts. The bearing temperature and generator winding temperature can indicate whether the key components are overheated, thereby avoiding potential failure risks. Wind speed, wind direction and air density are key environmental factors that affect the power generation efficiency of wind turbines. Accurately obtaining these data will help with subsequent precise control. The data obtained at the current monitoring time point will be used to monitor the wind turbine's operating status. The unit vibration, bearing temperature, generator winding temperature, wind speed, wind direction and air density are input into the control parameter acquisition model, and the ideal control parameter set of the current wind turbine is output through the control parameter acquisition model, including pitch angle, yaw angle, generator speed and generator torque; these parameters are key variables for wind turbine operation control, and by adjusting them, the operation state of the wind turbine can be optimized so that it can achieve the best balance between power generation efficiency and equipment life under the current wind conditions and environmental conditions; for example, reasonable pitch angle adjustment can enable the wind rotor to capture the maximum wind energy at different wind speeds; the optimization of the yaw angle ensures that the wind rotor is always facing the incoming wind direction, thereby improving the utilization rate of wind energy; the control of generator speed and torque directly affects the output quality of electric energy and the operating stability of the equipment;

[0055] The precise rated wind speed is calculated based on the original design rated wind speed of the wind turbine and the current air density. The original design rated wind speed is the wind speed corresponding to the rated power output of the wind turbine under standard air density conditions. It is an important parameter in the design stage of the wind turbine and reflects the performance characteristics of the wind turbine under ideal conditions. However, in actual operation, the air density will change due to factors such as altitude, temperature, and humidity, and the change in air density directly affects the density of wind energy and the power output characteristics of the wind turbine. Therefore, in order to make the control strategy of the wind turbine more in line with the actual operating conditions, the original design rated wind speed needs to be corrected according to the current air density to obtain the precise rated wind speed. speed; Specifically, a decrease in air density will lead to a decrease in wind energy density at the same wind speed, and the output power of the wind turbine will also decrease accordingly; conversely, an increase in air density will increase wind energy density and output power; by combining the original design rated wind speed with the current air density for calculation, an accurate rated wind speed that more accurately reflects the performance of the wind turbine under the current environmental conditions can be obtained; this accurate rated wind speed will serve as an important basis for subsequent control strategy adjustments, such as directly applying the ideal control parameter set when the wind speed is high, and adjusting the pitch angle by considering factors such as the wake effect when the wind speed is low, so as to ensure that the wind turbine can achieve the optimal operating state under different environmental conditions, and improve power generation efficiency and equipment reliability;

[0056] For any wind turbine generator set, based on the wind speed and wind direction of the wind turbine generator set, the control frequency of the wind turbine generator set is adjusted;

[0057] For any wind turbine, at the current control time point, if the wind speed of the wind turbine collected most recently at the current control time point is greater than or equal to the precise rated wind speed, the ideal control parameter set is directly applied to adjust the control parameters of the current wind turbine; because the wind turbine is already in full power state at this time, it needs to adjust the pitch angle to maintain the rated power. In order to avoid the risk of overload due to the health status of the unit such as vibration and temperature, it is difficult to continue to adjust the pitch angle on this basis, and the benefits obtained from the adjustment at this time may not be obvious;

[0058] Otherwise, the downstream unit with the greatest impact on the current wind turbine is selected. Specifically, based on the current wind direction, the wind turbine that is most affected by the wake effect of the current wind turbine is selected as the downstream unit with the greatest impact. Then, at the current control time point, the wind speed obtained by the current wind turbine is sorted in time to form the upstream unit wind speed time series data, and the wind speed obtained by the downstream unit with the greatest impact in the recent period is sorted in time to form the downstream unit wind speed time series data. At the same time, the relative position vector of the current wind turbine and its downstream unit with the greatest impact is obtained, and the wind direction is converted into a wind direction vector. The upstream unit wind speed time series data and the downstream unit wind speed time series data are combined. The time series data, relative position vector, air density and wind direction vector are input into the pitch angle adjustment model, and the ideal value-added of the pitch angle of the current wind turbine is output. The ideal control parameter set is modified according to the ideal value-added of the pitch angle, and the modified ideal control parameter set is applied to adjust the various control parameters of the current wind turbine. Its essence is to use the time series convolution model to dynamically capture the spatiotemporal correlation of wake transmission, and correct the weight of the wind direction vector through air density. Finally, the pitch angle adjustment value optimized by disturbance simulation can significantly reduce the wake shielding effect while slightly reducing the single-machine power (fine-tuning the pitch angle) to achieve global power gain.

[0059] Obtain the relative position vector of the current wind turbine and its downstream turbine with the greatest impact, and convert the wind direction into a wind direction vector, including:

[0060] The geographical center of the wind farm is selected as the coordinate origin, and the wind farm coordinate system is established with the east as the X-axis direction and the north as the Y-axis direction. The unit length of the wind farm coordinate system is the rotor diameter of the wind turbine. The coordinate of the downstream unit with the largest impact is subtracted from the coordinate of the current wind turbine to obtain the relative position vector.

[0061] The direction of the wind direction vector indicates the actual blowing direction of the wind, and its modulus is constant at 1.

[0062] Calculate the accurate rated wind speed based on the original design rated wind speed of the wind turbine and the current air density, including:

[0063] Get the current air density ;

[0064] The rated power formula of a wind turbine is: ;

[0065] In the formula, is the default standard air density, , is the wind energy utilization coefficient, is the swept area of ​​the wind wheel, is the original design rated wind speed;

[0066] When the air density changes to In order to maintain the rated power unchanged, the new rated power formula is ;

[0067] Combining the two formulas, we can eliminate , we can get ,Right now: ;

[0068] Finally, the formula is derived , and the accurate rated wind speed is calculated from this .

[0069] For any wind turbine generator set, based on the wind speed and wind direction of the wind turbine generator set, the control frequency of the wind turbine generator set is adjusted, including:

[0070] Set the length to The sliding time window is For any wind turbine, , in the latest sliding time window Get the wind speed time series data of the current wind turbine and wind direction time series data , =1, 2, …, ; Indicates the number of monitoring time points in a sliding time window; Indicates the current time;

[0071] Calculate wind speed time series data Standard Deviation , using the formula Calculate the turbulence compensation coefficient , this coefficient quantifies the relative intensity of turbulence through the ratio of the standard deviation to the average wind speed, eliminating the interference of the magnitude of the wind speed itself on the fluctuation assessment, and enhancing the comparability of fluctuations under different wind speed environments; when the ratio of the standard deviation to the average wind speed is less than 0.05, it is forced to take 0.05, which not only avoids the numerical instability of the denominator approaching zero under low wind speeds, but also retains the basic response capability for stable meteorology, ensuring that the model can reasonably trigger frequency adjustment in both strong turbulence (such as offshore gusts) and light wind environments; express Wind speed time series data The average value of

[0072] Using the formula Calculate wind speed fluctuation intensity ; Wind speed fluctuation intensity Normalize to obtain the corrected wind speed fluctuation intensity ;

[0073] Calculate wind direction change , using the formula Calculate the intensity of wind direction fluctuations, is the effective cumulative damping factor of wind direction, which suppresses the interference of single large-angle wind direction jump with an exponential decay function. When the single wind direction changes more than (for example, set to 30°), its contribution decays rapidly with the increase of the angle, thereby filtering out occasional abnormal disturbances; at the same time, it retains the cumulative effect of small-angle continuous fluctuations (such as slow shift of wind direction), which is closer to the sensitivity of the wake to continuous wind direction changes; and the intensity of wind direction fluctuations is Normalize to obtain the corrected wind direction fluctuation intensity ;

[0074] Using the formula Calculate the intensity of fusion fluctuations , ; Based on the preset minimum monitoring frequency and the maximum monitoring frequency , using the formula Calculate the adjusted control frequency of the current wind turbine ; is the control coefficient, which adjusts the slope of the mapping curve from the fusion fluctuation intensity to the control frequency. The larger the value, the steeper the change of the frequency in the vicinity. For example, When set to 5, a value slightly higher than 1 will quickly approach the maximum monitoring frequency , adapt to the sudden increase of fluctuations (such as typhoon passing); otherwise, low The value makes the frequency transition smooth, which is conducive to smoothing frequent small fluctuations (such as daily gusts). The essence is to balance the response speed and stability through the curvature of the Sigmoid function.

[0075] The wind turbine that is most affected by the wake effect of the current wind turbine is taken as the downstream unit with the greatest impact, including:

[0076] Filter out wind turbines whose distance to the current wind turbine is less than the effective distance threshold Other candidate wind turbines include: is the wind wheel diameter;

[0077] For any candidate wind turbine, based on the current wind turbine position and the current candidate wind turbine locations , calculate the distance between the two ;

[0078] Based on relative position vector and wind direction vector , for the relative position vector Unitize , calculate the alignment ;

[0079] Using the formula Calculate the degree to which the current candidate wind turbine is affected by the wake effect of the current wind turbine, where: is the distance attenuation coefficient;

[0080] The candidate wind turbine that is most affected by the wake effect of the current wind turbine is taken as the downstream unit with the greatest impact on the current wind turbine.

[0081] The control parameter acquisition model is established based on a fully connected neural network, including the first input layer, the first hidden layer, the second hidden layer and the output layer; the first input layer is used to receive the vibration of the unit, the bearing temperature, the generator winding temperature, the wind speed, the wind direction and the air density; the first hidden layer is used to transform the input data by applying a nonlinear activation function to capture the complex relationship between the input data; the second hidden layer is used to further process the features output by the first hidden layer and learn a more advanced feature representation; the first output layer is used to output the ideal control parameter set of the wind turbine;

[0082] The training process of the control parameter acquisition model includes:

[0083] Initialize the parameters in the control parameter acquisition model;

[0084] Obtaining a number of first training samples with labeled ideal control parameter sets, each of which includes a wind turbine vibration, a bearing temperature, a generator winding temperature, a wind speed and direction, and an air density of the wind turbine;

[0085] Dividing all acquired first training samples into a first training set and a first validation set;

[0086] The control parameter acquisition model is trained by using a first training set, and then the control parameter acquisition model is verified by using a first verification set to obtain a first verification result;

[0087] It is determined whether the obtained first verification result meets the preset first training condition. If so, the trained control parameter acquisition model is output; if not, the control parameter acquisition model is continuously trained through the first training set.

[0088] The pitch angle adjustment model is established based on the CNN model, including the second input layer, feature extraction layer, feature splicing layer, second fully connected layer and second output layer; the input layer is used to receive the upstream unit wind speed time series data, the downstream unit wind speed time series data, relative position vector, air density and wind direction vector;

[0089] The feature extraction layer includes a parallel time series branch and a static feature branch, the time series branch includes a convolution layer and a pooling layer, and the static feature branch includes a first fully connected layer;

[0090] The convolution layer is used to extract the time series characteristics of the upstream unit wind speed time series data and the downstream unit wind speed time series data, and output a multi-dimensional time series feature map; the pooling layer is used to compress the time series dimension and enhance the robustness of the key features, and output the time series feature vector by performing global average pooling on the multi-dimensional time series feature map; the first fully connected layer is used to extract the static feature vectors of the relative position vector, air density and wind direction vector; the feature fusion layer is used to splice the time series feature vector with the static feature vector to obtain the fused feature vector; the second fully connected layer is used to further extract the high-dimensional feature representation of the fused feature vector; the second output layer is used to output the ideal value of the pitch angle.

[0091] The training process of the pitch angle adjustment model includes:

[0092] A number of second training samples with ideal pitch angle increments marked are obtained, each of which contains the upstream unit wind speed time series data, the downstream unit wind speed time series data, the wind direction vector, the air density and the relative position vector, specifically including:

[0093] Use open source wind turbine simulation tools to perform multi-unit coupling simulation and randomly generate training scenarios. Each training scenario is a second training sample, including randomly generated upstream unit wind speed time series data, downstream unit wind speed time series data, wind direction vector and air density, and the relative position vector of the upstream and downstream units; the upstream unit wind speed and the downstream unit wind speed are always lower than the precise rated wind speed of the current training scenario.

[0094] Fix the upstream unit pitch angle to the optimal design value , simulate and obtain the original active power of the upstream unit and the original downstream unit active power , and calculate the original total power ;

[0095] Apply pitch angle disturbance to the upstream unit , for each , simulate and calculate the active power of the new upstream unit and the active power of downstream units , and calculate the new total power ;

[0096] For any training scenario, if the current training scenario has a new total power Greater than the original total power , then the new total power The largest As the sample label of the current training scene; if the current training scene does not have a new total power Greater than the original total power , then As the sample label of the current training scene;

[0097] Dividing all acquired second training samples into a second training set and a second validation set;

[0098] The pitch angle adjustment model is trained by using the second training set, and the pitch angle adjustment model is verified by using the second verification set to obtain a second verification result;

[0099] It is determined whether the obtained second verification result meets the preset second training condition. If so, the trained pitch angle adjustment model is output; if not, the pitch angle adjustment model is continuously trained through the second training set.

[0100] Embodiment 2, a wind turbine operation control optimization system based on deep learning, such as Figure 1 As shown, including:

[0101] The data monitoring module is used to obtain the unit vibration, bearing temperature, generator winding temperature and wind speed of any wind turbine in the wind farm at the current monitoring time point, and obtain the current wind direction and air density;

[0102] The control parameter acquisition module is used to input the unit vibration, bearing temperature, generator winding temperature, wind speed, wind direction and air density obtained at the current monitoring time point into the control parameter acquisition model, and output the ideal control parameter set of the current wind turbine through the control parameter acquisition model, including pitch angle, yaw angle, generator speed and generator torque;

[0103] A control frequency adjustment module is used to adjust the control frequency of any wind turbine generator set based on the wind speed and wind direction of the wind turbine generator set;

[0104] A pitch angle adjustment module, including a rated wind speed adjustment unit and a pitch angle adjustment unit;

[0105] A rated wind speed adjustment unit is used to calculate the accurate rated wind speed using the original design rated wind speed of the wind turbine and the current air density;

[0106] The pitch angle adjustment unit is used for any wind turbine set. At the current control time point, if the wind speed of the wind turbine set is greater than or equal to the precise rated wind speed, the ideal parameter control set is directly applied to adjust the various control parameters of the current wind turbine set; if the wind speed of the wind turbine set is less than the precise rated wind speed, the downstream unit with the greatest impact on the current wind turbine set is selected. Specifically, based on the current wind direction, the wind turbine set that is most affected by the wake effect of the current wind turbine set is selected as the downstream unit with the greatest impact, and then the wind speeds obtained by the current wind turbine set in the recent period are sorted by time to form the wind speed time series of the upstream unit. The wind speeds of the downstream unit with the largest influence in the recent period are sorted in time to form the downstream unit wind speed time series data, and the relative position vector of the current wind turbine unit and its downstream unit with the largest influence is obtained, and the wind direction is converted into a wind direction vector; the upstream unit wind speed time series data, the downstream unit wind speed time series data, the relative position vector, the air density and the wind direction vector are input into the pitch angle adjustment model, and the ideal value-added pitch angle of the current wind turbine unit is output. The ideal control parameter set is modified according to the ideal value-added pitch angle, and the modified ideal control parameter set is applied to adjust the various control parameters of the current wind turbine unit.

[0107] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all such improvements and changes should fall within the scope of protection of the appended claims of the present invention. Parts not described in detail in this specification belong to the prior art known to those skilled in the art.

Claims

1. A wind turbine operation control optimization method based on deep learning, characterized in that: include: For any wind turbine in the wind farm, the vibration, bearing temperature, generator winding temperature and wind speed of the wind turbine are obtained at the current monitoring time point, and the current wind direction and air density are obtained at the same time. All the data obtained at the current monitoring time point are input into the control parameter acquisition model, and the ideal control parameter set of the current wind turbine is output through the control parameter acquisition model, including pitch angle, yaw angle, generator speed and generator torque; Calculates accurate rated wind speed based on current air density; For any wind turbine generator set, based on the wind speed and wind direction of the wind turbine generator set, the control frequency of the wind turbine generator set is adjusted; For any wind turbine, at the current control time point, if the wind speed of the wind turbine collected most recently is greater than or equal to the precise rated wind speed, the ideal control parameter set is directly applied to adjust the control parameters of the current wind turbine; Otherwise, select the downstream unit with the greatest impact on the current wind turbine, and at the current control time point, sort the wind speed obtained by the current wind turbine and the wind speed obtained by the downstream unit with the greatest impact in the recent period of time into upstream unit wind speed time series data and downstream unit wind speed time series data in chronological order; obtain the ideal value-added pitch angle through the pitch angle adjustment model; modify the ideal control parameter set, and use the modified ideal control parameter set to adjust various control parameters of the current wind turbine.

2. The wind turbine operation control optimization method based on deep learning according to claim 1 is characterized in that: Apply the trained pitch angle adjustment model to obtain the ideal value of the pitch angle, including: Select the geographical center of the wind farm as the coordinate origin, establish the wind farm coordinate system with due east as the X-axis direction and due north as the Y-axis direction, and the unit length of the wind farm coordinate system is the rotor diameter of the wind turbine; subtract the coordinates of the current wind turbine from the coordinates of the downstream unit with the largest impact to obtain the relative position vector; and at the same time obtain the wind direction vector with a modulus length of 1 in the wind farm coordinate system of the current wind direction; The upstream unit wind speed time series data, the downstream unit wind speed time series data, the relative position vector, the air density and the wind direction vector are input into the pitch angle adjustment model, and the ideal increment of the current wind turbine pitch angle is output.

3. The wind turbine operation control optimization method based on deep learning according to claim 2 is characterized in that: Calculate the precise rated wind speed based on the current air density using the following formula: ； In the formula, For accurate rated wind speed, is the current air density, is the default standard air density, It is the original design rated wind speed.

4. The wind turbine operation control optimization method based on deep learning according to claim 3 is characterized in that: For any wind turbine generator set, based on the wind speed and wind direction of the wind turbine generator set, the control frequency of the wind turbine generator set is adjusted, including: Set the length to The sliding time window is For any wind turbine, , in the latest sliding time window Get the wind speed time series data of the current wind turbine and wind direction time series data , =1, 2, …, ; Indicates the number of monitoring time points in a sliding time window; Indicates the current time; Calculate wind speed time series data Standard Deviation , using the formula Calculate the turbulence compensation coefficient ; express Wind speed time series data The average value of Using the formula Calculate wind speed fluctuation intensity ; Wind speed fluctuation intensity Normalize to obtain the corrected wind speed fluctuation intensity ; Calculate wind direction change , using the formula Calculate the intensity of wind direction fluctuations, is the effective cumulative damping factor of wind direction; Normalize to obtain the corrected wind direction fluctuation intensity ; Using the formula Calculate the intensity of fusion fluctuations , ; Based on the preset minimum monitoring frequency and the maximum monitoring frequency , using the formula Calculate the adjusted control frequency of the current wind turbine ; is the control coefficient.

5. The wind turbine operation control optimization method based on deep learning according to claim 4 is characterized in that: Select the downstream units with the greatest impact on the current wind turbines, including: Filter out wind turbines whose distance to the current wind turbine is less than the effective distance threshold Other candidate wind turbines include: is the rotor diameter; for any candidate wind turbine, based on the current wind turbine position and the current candidate wind turbine locations , calculate the distance between the two ; Based on relative position vector and wind direction vector , for the relative position vector Unitize , calculate the alignment ; Using the formula Calculate the degree to which the current candidate wind turbine is affected by the wake effect of the current wind turbine, where: is the distance attenuation coefficient; The candidate wind turbine that is most affected by the wake effect of the current wind turbine is taken as the downstream unit with the greatest impact on the current wind turbine.

6. A wind turbine operation control optimization method based on deep learning according to claim 5, characterized in that: The training process of the control parameter acquisition model includes: Initialize the parameters in the control parameter acquisition model; Obtaining a number of first training samples with labeled ideal control parameter sets, each of which includes a wind turbine vibration, a bearing temperature, a generator winding temperature, a wind speed and direction, and an air density of the wind turbine; Dividing all acquired first training samples into a first training set and a first validation set; The control parameter acquisition model is trained by using a first training set, and then the control parameter acquisition model is verified by using a first verification set to obtain a first verification result; It is determined whether the obtained first verification result meets the preset first training condition. If so, the trained control parameter acquisition model is output; if not, the control parameter acquisition model is continuously trained through the first training set.

7. The wind turbine operation control optimization method based on deep learning according to claim 6 is characterized in that: The pitch angle adjustment model is established based on a convolutional neural network, including a second input layer, a feature extraction layer, a feature splicing layer, a second fully connected layer and a second output layer.

8. The wind turbine operation control optimization method based on deep learning according to claim 7 is characterized in that: The training process of the pitch angle adjustment model includes: A plurality of second training samples with ideal pitch angle increments marked are obtained, each of which contains the upstream unit wind speed time series data, the downstream unit wind speed time series data, the wind direction vector, the air density and the relative position vector; Dividing all acquired second training samples into a second training set and a second validation set; The pitch angle adjustment model is trained by using the second training set, and the pitch angle adjustment model is verified by using the second verification set to obtain a second verification result; It is determined whether the obtained second verification result meets the preset second training condition. If so, the trained pitch angle adjustment model is output; if not, the pitch angle adjustment model is continuously trained through the second training set.

9. A wind turbine operation control optimization system based on deep learning, characterized in that: The system is used to implement a wind turbine operation control optimization method based on deep learning as described in any one of claims 1 to 8, comprising: The data monitoring module is used to obtain the vibration, bearing temperature, generator winding temperature and wind speed of any wind turbine in the wind farm at the current monitoring time point, and obtain the current wind direction and air density; A control parameter acquisition module is used to input all data acquired at the current monitoring time point into a control parameter acquisition model, and output an ideal control parameter set of the current wind turbine through the control parameter acquisition model, including pitch angle, yaw angle, generator speed and generator torque; a control frequency adjustment module is used to adjust the control frequency of any wind turbine based on the wind speed and wind direction of the wind turbine; A pitch angle adjustment module, including a rated wind speed adjustment unit and a pitch angle adjustment unit; Rated wind speed adjustment unit, used to calculate and obtain accurate rated wind speed based on current air density; The pitch angle adjustment unit is used for any wind turbine set. At the current control time point, if the wind speed collected by the wind turbine set at the most recent time is greater than or equal to the precise rated wind speed, the ideal control parameter set is directly applied to adjust the various control parameters of the current wind turbine set; otherwise, the downstream unit with the largest impact on the current wind turbine set is selected, and at the current control time point, the wind speed obtained by the current wind turbine set and the wind speed obtained by the downstream unit with the largest impact in the most recent period are sorted in time to form the upstream unit wind speed time series data and the downstream unit wind speed time series data; the ideal value-added of the pitch angle is obtained through the pitch angle adjustment model, the ideal control parameter set is modified according to the ideal value-added of the pitch angle, and the modified ideal control parameter set is applied to adjust the various control parameters of the current wind turbine set.

Images