Wind farm multi-objective yaw wake optimization method, electronic device and storage medium

Abstract

The application relates to the technical field of wind power generation, and particularly provides a wind farm multi-target yaw wake optimization method, an electronic device and a storage medium, and aims to solve the technical problem that the service life of a wind turbine is significantly shortened due to the fact that the existing yaw control strategy considers the total power generation but does not consider the fatigue damage of the unit. To this end, the wind farm multi-target yaw wake optimization method comprises the following steps: obtaining unit information, inflow condition information and a fatigue damage proxy model of a target wind farm; establishing an initial population based on the unit information; starting from the initial population, a genetic algorithm is used to perform target iteration to maximize a first optimization target and minimize a second optimization target; when the first optimization target and the second optimization target are converged, a Pareto frontier of multi-target optimization is obtained; and the optimal yaw angle combination of the target wind farm is obtained based on the Pareto frontier. By taking the unit fatigue damage into the control optimization target, the total power generation is improved while the unit fatigue damage is effectively inhibited.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wind power generation, and particularly provides a wind farm multi-objective yaw wake optimization method, an electronic device and a storage medium. BACKGROUND

[0002] Wind energy is an important driving force to address climate change and achieve clean energy transformation, while the wake effect in the wind farm is a key obstacle to maximize the overall benefits of the wind farm, which makes active wake control of the wind farm more valued.

[0003] The existing yaw control strategy usually only maximizes the overall power generation of the wind farm, and such a single-objective strategy ignores the fatigue damage accumulation problem of the unit in the complex wake environment, which may cause the service life of the wind turbine to be significantly shortened, and affect the life cycle economy of the wind farm. SUMMARY

[0004] In order to overcome the above defects, the present application is proposed to solve or at least partially solve the technical problem that the existing yaw control strategy considers the overall power generation of the wind farm without considering the fatigue damage of the unit, which may cause the service life of the wind turbine to be significantly shortened. The present application provides a wind farm multi-objective yaw wake optimization method, an electronic device and a storage medium.

[0005] In a first aspect, the present application provides a wind farm multi-objective yaw wake optimization method, the method comprising:

[0006] obtaining unit information, inflow condition information and a fatigue damage proxy model of a target wind farm;

[0007] establishing an initial population based on the unit information, wherein the initial population comprises a plurality of individuals, and each individual is a yaw angle combination of all units in the target wind farm;

[0008] starting from the initial population, using a genetic algorithm to perform target iteration to maximize a first optimization target and minimize a second optimization target, wherein the first optimization target is the total output power of the wind farm corresponding to each individual, and the second optimization target is the fatigue damage of the wind farm corresponding to each individual, and the total output power and the fatigue damage of the wind farm are determined based on the inflow condition information and the fatigue damage proxy model of the target wind farm, respectively;

[0009] when the first optimization target and the second optimization target are converged, obtaining a Pareto front of multi-objective optimization;

[0010] obtaining an optimal yaw angle combination of the target wind farm based on the Pareto front.

[0011] In one embodiment of the wind farm multi-objective yaw wake optimization method, the fatigue damage proxy model is constructed by the following steps:

[0012] create a three-dimensional turbulent wind speed matrix under different random seeds to obtain a turbulent wind pre-computation library;

[0013] generate multiple working conditions using a low-discrepancy sequence;

[0014] simulate loads of the units of the target wind farm according to the multiple working conditions and the turbulent wind pre-computation library by using a multi-body dynamics simulation software to form an initial data set;

[0015] obtain a simulation data set based on the initial data set;

[0016] build a fully connected neural network model;

[0017] train the fully connected neural network model based on the simulation data set to obtain the fatigue damage proxy model.

[0018] In an embodiment of the wind farm multi-target yaw wake optimization method, simulating loads of the units of the target wind farm according to the multiple working conditions and the turbulent wind pre-computation library includes: in the process of simulating loads by using the multi-body dynamics simulation software, performing multiple load simulations for each working condition, and using different random seeds of turbulent wind randomly extracted from the turbulent wind pre-computation library as inlet turbulent wind for each simulation; and / or

[0019] The fully connected neural network model includes one input layer, at least one hidden layer, and one output layer. The input layer includes four nodes, respectively representing incoming flow wind speed, incoming flow turbulent intensity, unit yaw angle, and thrust coefficient. The output layer includes two nodes, respectively representing blade root combined equivalent fatigue damage and tower bottom combined equivalent fatigue damage.

[0020] In an embodiment of the wind farm multi-target yaw wake optimization method, the use of a genetic algorithm to iteratively maximize a first optimization target and minimize a second optimization target includes:

[0021] setting a constraint condition, the constraint condition being a change range of the yaw angle of each wind turbine unit;

[0022] updating an initial population according to a Gaussian random number under the constraint condition;

[0023] forming a new optimization population from the updated initial population;

[0024] determining the first optimization target and the second optimization target according to the new optimization population, the incoming flow condition information, and the fatigue damage proxy model;

[0025] Maximizing the target value of the first optimization target and minimizing the target value of the second optimization target are taken as optimization targets for target iterative optimization.

[0026] In an embodiment of the wind farm multi-objective yaw wake optimization method, the determining of the first optimization target and the second optimization target based on the new optimization population, the inflow condition information, and the fatigue damage proxy model of the target wind farm comprises:

[0027] Based on the yaw angles of each wind turbine in the new optimization population and the inflow condition information, the local inflow wind speed, the inflow turbulence intensity, and the operating condition parameters of each wind turbine in the wind farm are calculated using a wake model;

[0028] Based on the local inflow wind speed and the operating condition parameters, the output power of each unit is determined;

[0029] Based on the output power of each unit, the total output power of the wind farm is determined;

[0030] Based on the local inflow wind speed, the inflow turbulence intensity, and the operating condition parameters, the fatigue damage of the wind turbine is calculated using the fatigue damage proxy model;

[0031] Based on the fatigue damage of the wind turbine, the fatigue damage of the wind farm is determined.

[0032] In an embodiment of the wind farm multi-objective yaw wake optimization method, the inflow condition information includes the overall inflow wind speed, the inflow turbulence intensity, and the inflow wind direction angle of the wind farm, and the operating condition parameters include the thrust coefficient and the power coefficient;

[0033] The calculation of the local inflow wind speed, the inflow turbulence intensity, and the operating condition parameters of each wind turbine in the wind farm comprises:

[0034] Using a wake model, the dimensionless speed deficit and the additional turbulence standard deviation at a specified spatial position downstream of the wind turbine are calculated based on the overall inflow wind speed, the inflow turbulence intensity, and the inflow wind direction angle of the wind farm;

[0035] Based on the dimensionless speed deficit and the additional turbulence standard deviation, the local inflow wind speed and the inflow turbulence intensity are determined according to a wake superposition model;

[0036] Based on the yaw angles of each wind turbine in the new optimization population, the thrust coefficient and the power coefficient of each unit are determined.

[0037] In an embodiment of the wind farm multi-objective yaw wake optimization method, the determination of the total output power of the wind farm based on the output power of each unit comprises: taking the sum of the output power of each unit in the target wind farm as the total output power of the wind farm.

[0038] In an embodiment of the wind farm multi-objective yawing wake optimization method, the fatigue damage of the wind turbine includes a blade root combined equivalent fatigue damage and a tower bottom combined equivalent fatigue damage; the fatigue damage of the wind turbine is calculated based on the local inflow wind speed, the inflow turbulence intensity and the operating condition parameters by using the fatigue damage surrogate model, including: inputting the local inflow wind speed, the inflow turbulence intensity and the operating condition parameters into the fatigue damage surrogate model, and outputting the blade root combined equivalent fatigue damage and the tower bottom combined equivalent fatigue damage; and / or

[0039] The fatigue damage of the wind farm is determined based on the fatigue damage of the wind turbine, including: taking the maximum fatigue damage of the wind turbine in the target wind farm as the fatigue damage of the wind farm.

[0040] In a second aspect, an electronic device is provided, including:

[0041] at least one processor;

[0042] and a memory in communication with the at least one processor;

[0043] wherein the memory stores a computer program, and the computer program is executed by the at least one processor to implement the wind farm multi-objective yawing wake optimization method.

[0044] In a third aspect, a computer readable storage medium is provided, which stores a plurality of program codes, and the program codes are adapted to be loaded and run by a processor to implement the wind farm multi-objective yawing wake optimization method.

[0045] The above one or more technical solutions of the present application have at least one or more of the following advantages

[0046] Advantages:

[0047] The wind farm multi-objective yawing wake optimization method provided by the present application includes: obtaining unit information, inflow condition information and fatigue damage surrogate model of a target wind farm; establishing an initial population based on the unit information; starting from the initial population, using a genetic algorithm to iterate the target to maximize the first optimization target and minimize the second optimization target; when the first optimization target and the second optimization target converge, obtaining a Pareto frontier of multi-objective optimization; and obtaining an optimal yaw angle combination of the target wind farm based on the Pareto frontier. By using the fatigue damage surrogate model based on deep learning, the fatigue damage of the unit is included in the control optimization target, which effectively suppresses the fatigue damage of the unit while improving the total power generation. BRIEF DESCRIPTION OF DRAWINGS

[0048] The disclosure of the present application will become more apparent from the following description with reference to the attached drawings. It is readily understood by those skilled in the art that the drawings are only for the purpose of illustration and are not intended to limit the scope of protection of the present application. In addition, similar numbers in the figures are used to represent similar components, wherein:

[0049] Figure 1 is the main flowchart of the wind farm multi-objective yawing wake optimization method in an embodiment of the present application;

[0050] Figure 2 is the schematic diagram of the Mann turbulent box spanwise and vertical coverage in an embodiment of the present application;

[0051] Figure 3 is the Halton sequence sampling diagram of the working condition parameters in an embodiment of the present application;

[0052] Figure 4 is the fatigue damage proxy model based on deep learning in an embodiment of the present application;

[0053] Figure 5(a) is a schematic diagram of wind farm arrangement in an embodiment of the present application;

[0054] Figure 5(b) is a schematic diagram of wind turbine thrust coefficient and power coefficient curve in an embodiment of the present application;

[0055] Figure 6 is the Pareto frontier solution distribution obtained by yaw control optimization considering fatigue damage in an embodiment of the present application;

[0056] Figure 7 is the optimal yaw control strategy in an embodiment of the present application;

[0057] Figure 8(a) is a schematic diagram of the output power of each row of units in the field before and after yaw optimization in an embodiment of the present application;

[0058] Figure 8(b) and Figure 8(c) are schematic diagrams of equivalent fatigue damage change in an embodiment of the present application;

[0059] Figure 9 is a schematic diagram of the structure of an electronic device in an embodiment of the present application. DETAILED DESCRIPTION

[0060] Some embodiments of the present application will be described below with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are only used to explain the technical principles of the present application and are not intended to limit the scope of protection of the present application.

[0061] In the description of the present application, "module", "processor" can include hardware, software or a combination of both. A module can include hardware circuit, various suitable sensors, communication port, memory, and can also include software part such as program code, and can also be a combination of software and hardware. The processor can be a central processor, microprocessor, image processor, digital signal processor or any other suitable processor. The processor has data and / or signal processing functions. The processor can be implemented in software, hardware or a combination of both. The non-transitory computer readable storage medium includes any suitable medium that can store program code, such as magnetic disk, hard disk, optical disk, flash memory, read-only memory, random access memory, etc. The term "A and / or B" means all possible combinations of A and B, such as only A, only B or A and B. The term "at least one A or B" or "at least one of A and B" has similar meaning as "A and / or B", which can include only A, only B or A and B. The singular form of the term "one", "this" can also include plural forms.

[0062] Currently, the traditional yaw control strategy usually only maximizes the total power generation. This single-objective strategy ignores the fatigue damage accumulation of the unit in the complex wake environment, which may lead to a significant reduction in the structural life of the wind turbine and affect the life cycle economy of the wind farm. In addition, due to the increase of wake turbulence in the field and the complex influence of unit control action on wind turbine fatigue damage, the conventional wind turbine fatigue damage empirical model cannot accurately quantify the unit load, while the higher-precision load simulation method requires too much computing resources and cannot meet the real-time requirements of wind farm control. At present, there is a lack of a fatigue damage rapid evaluation method that can effectively evaluate the fatigue damage of the unit and is suitable for the control optimization process.

[0063] To solve the above problems, the technical field provides a wind farm multi-objective yaw wake control optimization method based on a fatigue damage surrogate model. The fatigue damage of the unit is included in the control optimization target by using a fatigue damage surrogate model based on deep learning, which effectively suppresses the fatigue damage of the unit while improving the total power generation.

[0064] Referring to the accompanying Figure 1 , Figure 1 Figure 1 is a schematic diagram of the main process of a wind farm multi-objective yaw wake optimization method according to an embodiment of the present application.

[0065] As shown in Figure 1 , the wind farm multi-objective yaw wake optimization method in the embodiment of the present application mainly includes the following steps S10-S50.

[0066] Step S10: Obtain the unit information, inflow condition information and fatigue damage surrogate model of the target wind farm.

[0067] Step S20: establishing an initial population based on the unit information, wherein the initial population contains a plurality of individuals, each individual being a yaw angle combination of all units in the target wind farm.

[0068] Step S30: starting from the initial population, performing target iteration using a genetic algorithm to maximize a first optimization target and minimize a second optimization target, wherein the first optimization target is a total output power of the wind farm corresponding to each individual, and the second optimization target is a fatigue damage of the wind farm corresponding to each individual, the total output power and the fatigue damage being determined based on the inflow condition information and a fatigue damage proxy model of the target wind farm.

[0069] Step S40: obtaining a Pareto front of multi-objective optimization when the first optimization target and the second optimization target are both converged.

[0070] Step S50: obtaining an optimal yaw angle combination of the target wind farm based on the Pareto front.

[0071] Based on the above steps S10-S50, first, the unit information, the inflow condition information and the fatigue damage proxy model of the target wind farm are obtained; an initial population is established based on the unit information; starting from the initial population, a genetic algorithm is used to perform target iteration to maximize a first optimization target and minimize a second optimization target; when the first optimization target and the second optimization target are both converged, a Pareto front of multi-objective optimization is obtained; and an optimal yaw angle combination of the target wind farm is obtained based on the Pareto front. By using a fatigue damage proxy model based on deep learning, the fatigue damage of the unit is included in the control optimization target, which effectively suppresses the fatigue damage of the unit while improving the total power generation.

[0072] The above steps S10-S50 will be further described below.

[0073] Specifically, in the above step S10, the unit information of the target wind farm can include the number of units in the target wind farm, the unit coordinates and the unit model of each wind turbine unit.

[0074] The inflow condition information can include the overall incoming wind speed, the incoming turbulence intensity and the incoming wind direction angle of the wind farm. In one embodiment, the inflow condition information to be optimized of the target wind farm can be that the incoming wind direction is 270 degrees (west wind), the incoming wind speed is 8 m / s, and the incoming turbulence intensity is 0.06.

[0075] The fatigue damage proxy model is a deep learning model for predicting the fatigue damage of the unit, which is established in advance. Exemplarily, a fully connected neural network, a convolutional neural network, a recurrent neural network, etc. can be used as an example of the fatigue damage proxy model.

[0076] In addition, the fatigue damage proxy model can be obtained through the following steps S101 to S106.

[0077] Step S101: Create a three-dimensional turbulent wind speed matrix under different random seeds to obtain a turbulent wind pre-computation library.

[0078] Specifically, a tool such as a Mann turbulent generator can be used to generate a set of three-dimensional turbulent wind speed matrices with spatial distribution characteristics based on multiple different random seeds, i.e. wind speed fields in the flow direction, spanwise direction, and vertical direction. All these turbulent wind data sets form a turbulent wind pre-computation library for subsequent simulation, wherein the spanwise and vertical ranges of each turbulent box cover all the load sampling points of the wind turbine model blades and tower. Exemplarily, Figure 2 which can be a schematic diagram of the spanwise and vertical coverage of the Mann turbulent box.

[0079] In one embodiment, the spanwise length of the generated Mann turbulent box can be 127.6m, the vertical length can be 145.2m, and the flow direction length can be 10240m, to cover the load sampling points of the NREL-5MW unit blades and tower, and suitable for the inflow wind development length of the preset time length simulation under different wind speeds, such as the inflow wind development length of 10 minutes simulation.

[0080] Step S102: Generate multiple working conditions using a low-discrepancy sequence.

[0081] The low-discrepancy sequence can be a Halton sequence with random offset. The Halton sequence with random offset is an improved low-discrepancy sequence used to generate more uniformly distributed sample points in multi-dimensional space. The Halton sequence itself is a series of points constructed based on different prime number bases, which have lower discrepancy than completely random sampling points in a unit hypercube, thus providing more uniform space filling. Exemplarily, Figure 3 which can be a sampling diagram of the Halton sequence.

[0082] The working condition refers to the combination of various inflow conditions and control states that the wind turbine may encounter during operation, such as the inflow wind speed U0(z wind ), the inflow turbulence intensity Iu0(z wind ), the wind shear exponent α, and the unit yaw angle γ, i.e. the Halton sequence with random offset generates multiple uniformly distributed working condition sampling points in four-dimensional space.

[0083] Step S103: Use multi-body dynamics simulation software to simulate the load of the target wind farm unit according to the multiple working conditions and the turbulent wind pre-computation library to form an initial data set.

[0084] In one specific embodiment, the load simulation of the unit of the target wind farm according to the plurality of working conditions and the turbulent wind pre-calculation library comprises: in the process of the load simulation by using the multi-body dynamics simulation software, performing multiple load simulations for each working condition, and each simulation using different random seeds of the turbulent wind randomly extracted from the turbulent wind pre-calculation library as the inlet turbulent flow.

[0085] Specifically, according to the working conditions generated in the step S102, the load simulation of a preset time length can be performed by using the OpenFAST software, wherein the preset time length can be a preset value, for example, 10 minutes, 12 minutes, 15 minutes, etc. can be taken as an example of the preset time length. Exemplarily, in each working condition, N wind different random seeds of the turbulent wind are randomly sampled in the turbulent wind pre-calculation library, and after the wind profile is superimposed according to the incoming flow speed, the incoming turbulent flow intensity and the wind shear exponent, and the turbulent standard deviation is scaled, 10 minutes of load simulation is performed respectively as the inlet turbulent flow to obtain an initial data set.

[0086] In one embodiment, the high-fidelity multi-body dynamics software can be the OpenFAST software, and the value range of the four working condition parameters required for the load simulation can be the incoming flow speed U0(z wind )∈[3m / s,25m / s], the incoming turbulent flow intensity Iu0(z wind )∈[0.02,0,26], the wind shear exponent α∈[0.1,0.26], and the unit yaw angle γ∈[-10°,30°].

[0087] Step S104: obtaining a simulation data set based on the initial data set.

[0088] Specifically, the initial data set is obtained by performing N wind 10-minute load simulations for each working condition in the step S103, and finally the input parameters and the output parameters of each working condition in the initial data set are taken as the average values of the N wind simulation results, thereby obtaining a simulation data set. That is, the data set input parameters in the simulation data set are the incoming flow speed U0(the average value of the wind wheel disc multi-point), the incoming turbulent flow intensity Iu0(the average value of the wind wheel disc multi-point), the unit yaw angle γ, and the thrust coefficient C T (the derived quantity of the OpenFAST load calculation); and the data set output parameters are the blade root combined equivalent fatigue damage and the tower bottom combined equivalent fatigue damage.

[0089] Step S105: building a full connection neural network model.

[0090] In one specific embodiment, the fully connected neural network model comprises an input layer, at least one hidden layer and an output layer, the input layer contains four nodes, respectively representing the incoming flow wind speed, the incoming flow turbulence intensity, the unit yaw angle and the thrust coefficient; the output layer contains two nodes, respectively representing the blade root synthetic equivalent fatigue damage and the tower bottom synthetic equivalent fatigue damage. Exemplarily, Figure 4 The fully connected neural network model can be used as an example.

[0091] In this embodiment, the fully connected neural network model is used to input the wind turbine operating environment and state parameters (wind speed, turbulence, yaw angle, thrust coefficient), and to predict the fatigue damage degree of two key structural parts (blade root and tower bottom) according to the input information. The input layer is used to receive the original data input; the hidden layer includes one or more intermediate layers, which are used to extract features and perform nonlinear transformation; and the output layer outputs the final prediction result.

[0092] Step S106: training the fully connected neural network model based on the simulation data set to obtain a fatigue damage proxy model.

[0093] Specifically, the neural network is trained using the simulation data set, the network weights are optimized through the back propagation algorithm, and the loss function can be selected as the mean square error (MSE) to measure the difference between the predicted value and the true value; overfitting is prevented on the validation set, and the final performance is evaluated on the test set; and finally a fast prediction model is obtained, which can replace the time-consuming multi-body dynamics simulation and be used for real-time or batch prediction of turbine fatigue damage.

[0094] The above is a further description of step S10, and the following continues to further describe step S20.

[0095] For the above step S20, according to the number N of target wind farm units T , an initial optimization variable, i.e. a yaw angle variable of all zeros, is generated, and an initial population of genetic algorithm is established, wherein the initial population contains N individuals, each individual is a combination of yaw angles of all units in the wind farm representing a yaw control strategy of the wind farm under the current incoming flow condition.

[0096] Exemplarily, real number coding can be used to generate yaw angle variables of all zeros to establish the initialization population. The initialization population contains multiple individuals, each individual is a row vector containing N T values (i.e. ), which respectively correspond to a yaw control strategy. Wherein, the real number coding means that each gene value of the individual is represented by a floating point number.

[0097] The above is a further description of step S20, and the following continues to further describe step S30.

[0098] The step S30 can be implemented by the following steps S301-S305.

[0099] Step S301: Set a constraint condition, which is the variation range of the yaw angle of each wind turbine.

[0100] Specifically, set a constraint condition for the optimization variable, which can be the variation range γ of the yaw angle of the wind turbine i ∈ [0°, 30°], i = 1, 2, …, N T , where N T is the number of wind turbines.

[0101] Step S302: Update the initial population according to the Gaussian random number under the constraint condition.

[0102] Specifically, change the yaw angle of each unit in the initial population randomly under the constraint condition to generate new optimization variables For example, for a unit i, the updated yaw angle γ i ′ is obtained by adding a Gaussian random offset to the original yaw angle γ i : γ′ i = γ i + Gaussian(σ γ ), where Gaussian(σ γ ) is a Gaussian random number with 0 as the center and σ γ as the standard deviation.

[0103] Step S303: Form a new optimization population from the updated initial population.

[0104] Specifically, according to the population size N, form a new optimization population from the updated N optimization variables (yaw angle combinations).

[0105] Step S304: Determine the first optimization target and the second optimization target according to the new optimization population, the inflow condition information, and the fatigue damage proxy model, the first optimization target being the total output power of the wind farm corresponding to each individual, and the second optimization target being the fatigue damage of the wind farm corresponding to each individual.

[0106] The step S304 can be implemented by the following steps S3041-S3045.

[0107] Step S3041: Based on the yaw angles of each wind turbine in the new optimization population and the inflow condition information, use the wake model to calculate the local inflow wind speed, inflow turbulence intensity, and operating condition parameters of each wind turbine in the wind farm. The inflow condition information includes the overall inflow wind speed, inflow turbulence intensity, and inflow wind direction angle of the wind farm, and the operating condition parameters include the thrust coefficient and power coefficient.

[0108] The step S3041 can be implemented by the following steps S30411 to S30413.

[0109] Step S30411: Using the wake model, for each wind turbine in the wind farm, calculate the non-dimensional speed deficit and additional turbulence standard deviation at a specified spatial location downstream of the wind turbine according to the overall inflow wind speed, inflow turbulence intensity and inflow wind direction angle of the wind farm.

[0110] The specified spatial location downstream of the wind turbine can be a certain point or a certain position downstream of the wind turbine.

[0111] Specifically, first obtain the overall inflow wind speed U ∞ (z wind ), the inflow turbulence intensity Iu ∞ (z wind ) and the inflow wind direction angle θ wind of the wind farm. For any wind turbine in the wind farm, the non-dimensional speed deficit and the additional turbulence standard deviation Δσ u at a certain point (x, y, z) downstream of the wind turbine are calculated using the wake model (with the tower bottom of the wind turbine as the origin):

[0112]

[0113] In the formula, ΔU / U0 is the non-dimensional speed deficit, and Δσ u / U0 is the additional turbulence standard deviation; U0 is the inflow wind speed of the wind turbine, i.e. the average wind speed on the disc plane of the wind turbine, for a single wind turbine not affected by the wake, U0 = mean(U ∞ | disk ); dx, dy and dz are the streamwise, spanwise and vertical distances, respectively, dx = x-x T , dy = y-y c , dz = z-z T , where x T and z T are the streamwise and vertical positions of the hub center of the wind turbine, and y c is the spanwise offset of the wake center.

[0114] The non-dimensional speed deficit can be further expressed as:

[0115]

[0116] where C T is the thrust coefficient of the wind turbine; D is the diameter of the wind turbine; γ is the yaw angle of the wind turbine; y c is the spanwise position of the wake center of the wind turbine; z h is the hub height of the wind turbine; σ y and σ zrespectively are the spanwise and vertical Gaussian standard deviations of the wake velocity.

[0117] Considering the linear expansion of the wake boundary in the far wake region and adding a gradual correction in the near wake region, the following quasi-linear relationship is formed:

[0118]

[0119] where k w is the wake expansion rate; x NW is the length of the near wake region; σ NW is the standard deviation of the wake at the starting point of the wake expansion (x = x NW ).

[0120] The length of the near wake region and the standard deviation of the starting point of the wake expansion are determined by the following formula:

[0121]

[0122] where Iu0 is the inflow turbulence intensity of the wind turbine rotor, i.e., the average turbulence intensity of the rotor disc; α = 0.9; β = 0.077.

[0123] The yaw wake center offset y c (x) is calculated by spanwise velocity integration:

[0124]

[0125] where V c (x) is the spanwise velocity of the wake center corresponding to each flow direction position.

[0126] The additional turbulence standard deviation calculation formula is:

[0127]

[0128] where ΔIu max is the flow direction peak value of the additional turbulence intensity in the wake region; is the spanwise distribution function; δ h (r) is the vertical correction function; is the distance of a point (x, y, z) in the wake region from the wake center.

[0129] where the flow direction peak value is calculated by the following formula:

[0130]

[0131] where Iu0 is the inflow turbulence intensity of the wind turbine rotor, i.e., the average turbulence intensity of the rotor disc, which is taken as

[0132] The spanwise distribution function is:

[0133]

[0134] where r 1 / 2 is the wake half-width; σ T is the Gaussian standard deviation of the additional turbulence intensity peak outward.

[0135] Considering that the velocity deficit profile of the ZPA wake model presents an approximate ellipse, the additional turbulence intensity is modified based on the standard deviation of the wind speed profile obtained from the ZPA model as follows:

[0136]

[0137] Further, the vertical correction function is added as:

[0138]

[0139] where a is the azimuth angle, i.e., the angle between the radial line and the positive direction of the y-axis; and k1 is a proportional coefficient.

[0140] The proportional coefficient k1 is:

[0141]

[0142] Step S30412: According to the wake superposition model, the local inflow wind speed and the inflow turbulence intensity are determined based on the dimensionless velocity deficit and the additional turbulence standard deviation.

[0143] Specifically, for any wind turbine i, the local inflow wind speed and the turbulence intensity are the average wind speed and the average turbulence intensity at multiple equivalent inflow points on the wind wheel disc, where the wind speed and the turbulence intensity at each equivalent inflow point (x, y, z) are calculated using the wake superposition model:

[0144]

[0145] where is the local inflow wind speed at each equivalent inflow point (x, y, z); is the inflow turbulence intensity at each equivalent inflow point (x, y, z); is the combined wake velocity deficit of the upstream multiple units at point (x, y, z); is the combined wake additional turbulence standard deviation of the upstream multiple units at point (x, y, z).

[0146] where the combined wake velocity deficit The following process is adopted for iterative solution:

[0147] 1) Obtain the velocity deficit ΔU i (x, y, z) and the convection velocity of the upstream units at point (x, y, z):

[0148]

[0149] 2) Initialize the inflow velocity

[0150] 3) Update the synthetic velocity deficit:

[0151]

[0152] 4) Update the inflow velocity:

[0153]

[0154] 5) If is true, then use to calculate the wake zone wind velocity U w (x,y,z), if not, then set and go back to step 2).

[0155] The synthetic wake additional turbulence intensity is calculated by:

[0156]

[0157] where is the additional turbulence intensity caused by the upstream i-th turbine at point (x,y,z).

[0158] Step S30413: Based on the yaw angle of each wind turbine in the new optimized population, determine the thrust coefficient and power coefficient of each turbine, and obtain the turbine operating condition parameters.

[0159] Specifically, the turbine operating condition parameters can further include the yaw angle of the turbine. For a certain turbine i, the yaw angle γ i is the control variable of the yaw wake control, which is a known quantity. Further, the thrust coefficient and the power coefficient under the yaw condition can be calculated.

[0160]

[0161] wherein, and are the thrust coefficient and the power coefficient of the turbine i without yaw, respectively, which are obtained by interpolation on the wind speed-thrust coefficient curve and the wind speed-power coefficient curve through the inflow wind speed U0.

[0162] Step S3042: Based on the local inflow wind speed and the operating condition parameters, determine the output power of each turbine.

[0163] Specifically, according to the power coefficient of the unit and the type information of the unit, the output power of each wind turbine (taking unit i as an example) is calculated:

[0164]

[0165] In the formula, P i is the output power of the unit i; p is the air density; is the power coefficient of the unit i; A i is the swept area of the unit i; is the local inflow wind speed of the unit i.

[0166] Step S3043: determining the total output power of the wind farm based on the output power of each unit.

[0167] In one specific embodiment of the present application, the determination of the total output power of the wind farm based on the output power of each unit comprises: summing the output power of each unit in the target wind farm as the total output power of the wind farm.

[0168] Specifically, the total output power of the wind farm is the sum of the output power of each wind turbine in the wind farm:

[0169]

[0170] In the formula, P total is the total output power of the wind farm; N T is the total number of units in the wind farm.

[0171] Step S3044: calculating the fatigue damage of the wind turbine by using a fatigue damage proxy model based on the local inflow wind speed, the inflow turbulence intensity and the operating condition parameters.

[0172] In one specific embodiment of the present application, the fatigue damage of the wind turbine includes the blade root synthetic equivalent fatigue damage and the tower bottom synthetic equivalent fatigue damage; the calculation of the fatigue damage of the wind turbine by using the fatigue damage proxy model based on the local inflow wind speed, the inflow turbulence intensity and the operating condition parameters comprises: inputting the local inflow wind speed, the inflow turbulence intensity and the operating condition parameters into the fatigue damage proxy model to output the blade root synthetic equivalent fatigue damage and the tower bottom synthetic equivalent fatigue damage.

[0173] Specifically, the local inflow wind speed and the operating condition parameters (such as the local inflow wind speed the inflow turbulence intensity the thrust coefficient of the unit and the yaw angle γ i ) can be input into the fatigue damage proxy model (neural network) to perform a feedforward operation without updating the model parameters to obtain the synthetic equivalent fatigue damage DEL Bldand the tower bottom synthetic equivalent fatigue damage DEL Twr .

[0174] Step S3045: determining the fatigue damage of the wind farm based on the fatigue damage of the wind turbine.

[0175] In one embodiment of the present application, the fatigue damage of the wind farm is determined based on the fatigue damage of the wind turbine, including: taking the maximum fatigue damage of the wind turbine in the target wind farm as the fatigue damage of the wind farm.

[0176] Specifically, the fatigue damage of the wind farm can be the maximum fatigue damage among the wind turbines:

[0177]

[0178] F total is the fatigue damage of the wind farm; and are the blade root synthetic equivalent fatigue damage and the tower bottom synthetic equivalent fatigue damage of the wind turbine i, respectively.

[0179] Step S305: performing target iterative optimization with the maximum target value of the first optimization target and the minimum target value of the second optimization target as the optimization target.

[0180] Specifically, the target iterative optimization is performed with the maximum target value of the first optimization target and the minimum target value of the second optimization target as the optimization target, and the yaw angle of each wind turbine is adjusted through multiple iterations until the total power of the wind farm calculated under the current yaw angle of the wind turbine is maximum and the synthetic fatigue damage of all wind turbines is minimum.

[0181] Exemplarily, the genetic algorithm can be NSGA-II algorithm; the new wind farm yaw angle combination can be obtained based on the constraint range of the yaw angle through crossover and mutation; the wind farm yaw angle combination adopts real number coding, and the yaw angle is generated in the range of 0 to 30 degrees, i.e. i γ ∈[0°,30°], the yaw angle is obtained by adding a Gaussian random offset to the yaw angle in the last generation population.

[0182] The above is the description of step S30, and then step S40 is further described.

[0183] For the above step S40, when the first optimization target and the second optimization target converge, the optimization process is ended, and the Pareto front of multi-objective optimization is obtained, each individual on the Pareto front is a certain wind farm yaw control strategy, including the yaw angle of each wind turbine.

[0184] For the above step S50, since the Pareto frontier includes multiple individuals, finally the optimal yaw control strategy of the wind farm can be selected according to actual requirements, that is, the optimal yaw angle combination of the target wind farm.

[0185] In order to solve the technical problems of "the existing yaw control strategy usually only maximizes the total power generation of the wind farm, and cannot take into account the fatigue damage of the wind turbine under the influence of the increased turbulence of the wake in the field and the control action of the unit", and "there is currently a lack of a fatigue damage rapid evaluation method which can effectively evaluate the fatigue damage and is suitable for the control optimization process", the application first establishes a fatigue damage proxy model based on deep learning, and applies a genetic algorithm to maximize the target value of the first optimization objective and minimize the target value of the second optimization objective as the optimization objective, and multi-objective optimization is performed on the yaw control strategy of the wind farm. Finally, the yaw control strategy of the wind farm can be determined on the obtained multi-objective optimization Pareto frontier solution according to actual requirements. Through the above method, the total output power and fatigue damage of the wind farm are considered comprehensively, the fatigue damage of the wind farm is reduced while ensuring the total output power of the wind farm, the operating life of the wind farm is prolonged, and the comprehensive benefits of the wind farm in the whole life cycle are improved.

[0186] In one embodiment, the units in the wind farm are arranged in the form of 3x8, 24 NREL-5MW type wind turbines are installed, the wind farm unit point is shown in FIG. 5(a), the basic parameters of the wind farm are shown in Table 1, the power and thrust curves of the wind turbine are shown in FIG. 5(b), and the Pareto frontier solution distribution obtained by multi-objective yaw control optimization is shown in FIG. 5(c). Figure 6

[0187] Table 1 Wind farm parameters

[0188]

[0189]

[0190] In the target wind farm, the yaw angle change range γ of the wind turbine is set to γ i ∈ [0°, 30°], the incoming wind speed U ∞ (z wind ) = 8 m / s, the incoming turbulence intensity Iu ∞ (z wind ) = 0.06, and multi-objective yaw control optimization is performed. The Pareto frontier solution distribution obtained by optimization is shown in FIG. 5(c), wherein the control strategy with the largest power generation improvement amplitude under the premise of no fatigue damage increase, and the yaw angle distribution of each unit under the control strategy is shown in FIG. 5(d). Figure 6 Figure 7 Figure 8(a) , 8(b) ​​​As shown in FIG. 8(c), the total output power corresponding to the optimal yaw control strategy of the embodiment is 25705.62kW, which is increased by 0.52% compared with the non-yaw state, and the fatigue damage is reduced by 1.30%. The optimal yaw control strategy of the embodiment can relieve the fatigue damage of the wind turbine while increasing the output power of the wind farm, and is expected to further improve the total benefit of the wind farm in the whole life cycle.

[0191] It should be noted that, although the above embodiments describe the steps in a specific order, those skilled in the art can understand that, in order to achieve the effects of the present application, the steps do not have to be executed in this order, and they can be executed simultaneously (in parallel) or in other orders, and these changes are within the protection scope of the present application.

[0192] Further, the present application also provides an electronic device, which can include at least one processor; and a memory connected with the at least one processor in communication; wherein the memory stores a computer program, and the computer program is executed by the at least one processor to implement the wind farm multi-objective yaw wake optimization method of any of the above embodiments. Referring to Figure 9 As shown in FIG. 10, Figure 9 The structure of the electronic device is shown in the embodiment of FIG. 10, which includes a processor 100 and a memory 200.

[0193] Further, the present application also provides a computer readable storage medium. In a computer readable storage medium embodiment according to the present application, the computer readable storage medium can be configured to store a program for executing the wind farm multi-objective yaw wake optimization method of the above method embodiments, which can be loaded and run by a processor to implement the wind farm multi-objective yaw wake optimization method described above. For ease of illustration, only the parts related to the embodiments of the present application are shown, and the specific technical details are not disclosed, please refer to the method part of the embodiments of the present application. The computer readable storage medium can be a memory device formed by various electronic devices, and optionally, the computer readable storage medium in the embodiments of the present application is a non-transitory computer readable storage medium.

[0194] Further, it should be understood that, since the setting of each module is only to illustrate the functional units of the device of the present application, the corresponding physical device of the module can be the processor itself, or a part of software, a part of hardware, or a part of combination of software and hardware in the processor. Therefore, the number of each module in the figure is only illustrative.

[0195] So far, the technical solutions of the present application have been described in combination with the specific embodiments shown in the drawings, but it is easy for those skilled in the art to understand that the protection scope of the present application is obviously not limited to these specific embodiments. Those skilled in the art can make equivalent changes or replacements to the related technical features without departing from the principles of the present application, and the technical solutions after the changes or replacements will all fall within the protection scope of the present application.

Claims

1. A wind farm multi-objective yaw wake optimization method, characterized in that, The method comprises: obtaining unit information, inflow condition information and fatigue damage proxy model of a target wind farm; establishing an initial population based on the unit information, wherein the initial population comprises a plurality of individuals, and each individual is a yaw angle combination of all units in the target wind farm; starting from the initial population, using a genetic algorithm to perform target iteration to maximize a first optimization target and minimize a second optimization target, wherein the first optimization target is the total output power of the wind farm corresponding to each individual, and the second optimization target is the fatigue damage of the wind farm corresponding to each individual, and the total output power and the fatigue damage of the wind farm are determined based on the inflow condition information and the fatigue damage proxy model of the target wind farm respectively; constructing the fatigue damage proxy model by the following steps: obtaining a unit load simulation data set of a target wind farm; building a fully connected neural network model; training the fully connected neural network model based on the unit load simulation data set to obtain the fatigue damage proxy model; determining the second optimization target by the following steps, comprising: determining a new optimization population based on the initial population; based on the yaw angle of each wind turbine in the new optimization population and the inflow condition information, calculating the local inflow wind speed, inflow turbulence intensity and operating condition parameters of each wind turbine in the wind farm; based on the local inflow wind speed, inflow turbulence intensity and operating condition parameters, calculating the fatigue damage of the wind turbine using the fatigue damage proxy model; determining the fatigue damage of the wind farm based on the fatigue damage of the wind turbine; when the first optimization target and the second optimization target converge, obtaining a multi-objective optimization Pareto front; obtaining the optimal yaw angle combination of the target wind farm based on the Pareto front.

2. The wind farm multi-objective yawing wake optimization method of claim 1, wherein, The unit load simulation data set of the target wind farm is obtained by the following steps: creating a three-dimensional turbulence wind speed matrix under different random seeds to obtain a turbulence wind precalculation library; generating multiple working conditions using low difference sequence; using multi-body dynamics simulation software to perform load simulation on the units of the target wind farm according to the multiple working conditions and the turbulence wind precalculation library to form an initial data set; obtaining a simulation data set based on the initial data set.

3. The wind farm multi-objective yawing wake optimization method of claim 2, wherein, The load simulation of the units of the target wind farm according to the multiple working conditions and the turbulence wind precalculation library comprises: in the process of load simulation using multi-body dynamics simulation software, performing multiple load simulations for each working condition, and using different random seeds of turbulence wind randomly selected from the turbulence wind precalculation library as the inlet turbulence for each simulation; and / or The fully connected neural network model comprises an input layer, at least one hidden layer and an output layer, the input layer comprises four nodes representing incoming flow speed, incoming turbulence intensity, unit yaw angle and thrust coefficient respectively; the output layer comprises two nodes representing blade root combined equivalent fatigue damage and tower bottom combined equivalent fatigue damage.

4. The wind farm multi-objective yawing wake optimization method of claim 1, wherein, The use of a genetic algorithm to perform target iteration to maximize a first optimization target and minimize a second optimization target comprises: setting a constraint condition, the constraint condition being the change range of the yaw angle of each wind turbine; Under the constraint condition, the initial population is updated according to a Gaussian random number; The updated initial population is grouped into a new optimization population; A first optimization target and a second optimization target are determined according to the new optimization population, the inflow condition information and the fatigue damage proxy model; Target iterative optimization is performed with the optimization target being maximizing the target value of the first optimization target and minimizing the target value of the second optimization target.

5. The wind farm multi-objective yawing wake optimization method of claim 4, wherein, The first optimization target is determined according to the new optimization population and based on the inflow condition information, and includes: Based on the yaw angle of each wind turbine in the new optimization population and the inflow condition information, a wake model is used to calculate the local inflow wind speed, inflow turbulence intensity and operating condition parameters of each wind turbine in the wind farm; Based on the local inflow wind speed and operating condition parameters, the output power of each wind turbine is determined; Based on the output power of each wind turbine, the total output power of the wind farm is determined.

6. The wind farm multi-objective yawing wake optimization method of claim 5, wherein, The inflow condition information includes the overall inflow wind speed, inflow turbulence intensity and inflow wind direction angle of the wind farm, and the operating condition parameters include the thrust coefficient and power coefficient; The calculation of the local inflow wind speed, inflow turbulence intensity and operating condition parameters of each wind turbine in the wind farm includes: Using a wake model, the dimensionless speed deficit and additional turbulence standard deviation at a specified spatial position downstream of the wind turbine are calculated according to the overall inflow wind speed, inflow turbulence intensity and inflow wind direction angle; Based on the dimensionless speed deficit and additional turbulence standard deviation, the local inflow wind speed and inflow turbulence intensity are determined according to a wake superposition model; Based on the yaw angle of each wind turbine in the new optimization population, the thrust coefficient and power coefficient of each wind turbine are determined.

7. The wind farm multi-objective yawing wake optimization method of claim 5, wherein, The determination of the total output power of the wind farm based on the output power of each wind turbine includes summing the output power of each wind turbine in the target wind farm as the total output power of the wind farm.

8. The wind farm multi-objective yawing wake optimization method of claim 1, wherein, The wind turbine fatigue damage includes blade root synthetic equivalent fatigue damage and tower bottom synthetic equivalent fatigue damage; the calculation of the fatigue damage of the wind turbine using the fatigue damage proxy model includes inputting the local inflow wind speed, inflow turbulence intensity and operating condition parameters into the fatigue damage proxy model to output the blade root synthetic equivalent fatigue damage and the tower bottom synthetic equivalent fatigue damage; and / or The determination of the fatigue damage of the wind farm based on the fatigue damage of the wind turbine includes taking the maximum wind turbine fatigue damage in the target wind farm as the fatigue damage of the wind farm.

9. An electronic device, comprising: It includes: At least one processor; and a memory connected in communication with the at least one processor; wherein the memory has stored therein a computer program, which, when executed by the at least one processor, implements the wind farm multi-objective yaw wake optimization method of any one of claims 1 to 8.

10. A computer readable storage medium having stored therein a plurality of program codes, characterized in that, The program code is adapted to be loaded and run by the processor to execute the wind farm multi-objective yaw wake optimization method of any one of claims 1 to 8.

Images