A wind farm random model predictive control method considering time delay

Abstract

This invention discloses a stochastic model predictive optimization control method for wind farms that considers delay, comprising: calculating the steady-state background flow field distribution of the wind farm using a steady-state model; calculating the delay time based on the spatial location of the turbine and the background flow velocity, and constructing a quasi-steady-state power prediction model for the wind farm considering delay by combining the turbine power characteristic curve; measuring the natural incoming wind speed at different locations at the wind farm boundary to determine the multi-scenario spatiotemporal characteristics of the incoming wind speed; based on the multi-scenario spatiotemporal characteristics of the incoming wind speed, using artificial intelligence modeling methods to perform ultra-short-term multi-scenario predictions of the natural incoming wind speed at different locations; and designing a stochastic model predictive (SMPC) stationary controller for the wind farm by combining the quasi-steady-state prediction model considering delay and taking into account the multi-scenario ultra-short-term predictions of wind speed uncertainty. This invention reduces the power output fluctuation of the wind farm and improves the wind farm's friendly operation capability by compensating for wind speed uncertainty through feedforward and handling wake delay characteristics.

Description

Technical Field

[0001] This invention belongs to the field of wind farm automatic control, and relates to a wind farm control method that considers delay and uncertainty, specifically a wind farm stochastic model prediction optimization control method that considers delay. Background Technology

[0002] Wind farms are a primary form of wind energy utilization and play a crucial role in my country's energy transition. In recent years, wind farm scale has continuously expanded, application scenarios have become more diverse, commercialization has matured, and grid penetration has gradually increased. This has made optimal control—achieving grid-friendly operation and improving the economics of wind farms—an important research area. The large latency and strong uncertainties inherent in wind farms present many challenges to optimal control.

[0003] In terms of delay, in order to improve the efficiency of wind energy utilization, wind turbines must maintain a certain distance to reduce the impact of wake effect, which results in a certain delay time for the wake of the upstream unit to flow to the downstream unit; in terms of uncertainty, affected by meteorological conditions, topographic factors and thermal stability, the complex turbulence within the atmospheric boundary leads to strong uncertainty in the output of wind farms.

[0004] The wake space delay characteristics and strong wind speed uncertainty of wind farms exacerbate the fluctuations in turbine power, which is detrimental to the safe and stable operation of the power grid. Therefore, how to smoothly control the power output of wind farms under delay conditions and strong uncertainty has become a key point in the optimal control of wind farms. Summary of the Invention

[0005] Therefore, research can be conducted from two aspects: establishing a more accurate control model and adopting more advanced control methods.

[0006] Regarding control models, current wind farm models are mainly divided into three categories: high-precision numerical simulation, empirical formulas, and data-driven models, such as Jensen, Gaussian, and Frandson. However, most of them are steady-state empirical models. Steady-state models cannot consider the delay characteristics of the flow field, thus reducing the accuracy of wind farm models.

[0007] Besides the control model, control methods capable of handling delay characteristics and uncertainties are equally important. Among numerous control methods, Model Predictive Control (MPC) has the advantage of handling delays, multivariables, and constraints, and its rolling optimization and feedback correction concepts improve disturbance suppression capabilities, making it one of the most important methods in engineering control. Stochastic Model Predictive Control (SMPC) adds an uncertainty suppression module to the traditional MPC, improving its uncertainty suppression capabilities. However, SMPC is currently rarely used in wind farms, therefore, it is necessary to further study how to use SMPC methods to suppress wind farm power output fluctuations and improve the stability of wind farm power output.

[0008] Purpose of the invention: In view of the problems and shortcomings of the existing technology, the purpose of this invention is to provide a wind farm stochastic model prediction optimization control method that takes into account delay, which can simultaneously handle wake delay and wind speed uncertainty, and effectively suppress wind farm output fluctuations while ensuring solution efficiency and accuracy, thereby improving grid-friendly operation capability.

[0009] Technical solution: To solve the above technical problems, the technical solution adopted by the present invention is as follows:

[0010] Firstly, a method for predictive optimization control of wind farms based on stochastic models that consider delays is provided, including:

[0011] Step 1) Obtain historical measured natural inflow wind speeds at the boundary of wind farms, analyze the spatiotemporal characteristics of natural inflow wind speeds in multiple scenarios, and determine the number of typical scenarios N. s and the probability θ corresponding to each typical scenario s ;

[0012] Step 2) Based on the historical measured wind speed at the boundary of the wind farm and the wind speed at the current moment, the wind speed at the boundary of the wind farm is predicted in the future using a prediction model pre-built with artificial intelligence modeling methods, so as to obtain the wind speed at the boundary of the wind farm for all typical scenarios.

[0013] Step 3) Based on the predicted natural incoming wind speed v0 at the wind farm boundary for each typical scenario, and using the steady-state flow field model, calculate the effective incoming wind speed v at the rotor of each turbine, considering the delay. τ,j The maximum operating power P of each unit was calculated using Bates' theorem. MPPT,j ;

[0014] Step 4) Based on P MPPT,j Predict the actual maximum available power of each unit under various typical scenarios.

[0015] Step 5) Based on the number of typical scenarios N s The probability θ corresponding to each typical scenario s and the actual maximum available power of each unit under various typical scenarios The rolling optimization solution is performed using a pre-constructed wind farm rolling optimization control objective function to obtain the power setpoints of each turbine in scenario s.

[0016] Step 6) Based on the probability θ corresponding to each typical scenario s Power settings for various typical scenarios Determine the final power setpoint for the j-th unit.

[0017] In some embodiments, step 3) includes:

[0018] Based on the steady-state flow field model, considering the superposition of multiple wake effects, the equivalent inlet wind speed v at the downstream turbine rotor is calculated. j :

[0019]

[0020] In the formula, v i v is the effective wind speed of the i-th unit. ij A is the wake velocity from unit i to unit j. ij Let v0 be the wake overlap area from the i-th turbine to the j-th turbine, v0 be the natural incoming wind speed at the wind farm boundary, and r be the wake overlap area. d The radius of the turbine rotor;

[0021] Calculate the wake delay time τ from unit i to unit j. ij :

[0022]

[0023] In the formula, d ij Let v0 be the distance from the i-th unit to the j-th unit, and v0 be the natural inflow wind speed at the boundary of the wind farm.

[0024] Calculate the effective incoming wind speed v at the rotor of each unit considering the delay. τ,j :

[0025]

[0026] In the formula, v i v is the effective wind speed of the i-th unit. ij A is the wake velocity from unit i to unit j. ij Let v0 be the wake overlap area from the i-th turbine to the j-th turbine, v0 be the natural incoming wind speed at the wind farm boundary, and r be the wake overlap area. dLet be the radius of the turbine rotor, and t represent time t.

[0027] Calculate the maximum operating power P of each unit. MPPT,j :

[0028]

[0029] In the formula, ρ is the air density, and a j Let A be the axial induction factor of the j-th unit, and let A be the swept area of ​​the wind turbine.

[0030] In some embodiments, a is preferred. j It is 1 / 3.

[0031] In some embodiments, step 4), based on P MPPT,j Predict the actual maximum available power of each unit under various typical scenarios. include:

[0032] When the power generation of a wind turbine is lower than its minimum output, the turbine needs to be shut down. If the theoretical power generation is higher than the maximum power, it will continue to operate at maximum power, which is the actual maximum available power after considering constraints. for:

[0033]

[0034] In the formula, P min P max These are the minimum and maximum power constraints set to ensure the safety of the unit.

[0035] In some embodiments, step 1) includes:

[0036] ① Measure the natural inflow wind speed at different locations at the wind farm boundary;

[0037] ② Determine the time fluctuation characteristics of wind speed magnitude at the same location and in the same wind direction;

[0038] ③ Determine the time fluctuation characteristics of wind speed magnitude under different wind directions at the same location;

[0039] ④ Change the spatial location, repeat ② and ③, and analyze the spatiotemporal characteristics of natural wind speed in multiple scenarios;

[0040] ⑤ Determine the number N of typical scenarios s and the probability θ corresponding to this scenario s .

[0041] In some embodiments, step 2) includes:

[0042] ① Construct initial training samples V = {v} of the natural incoming wind speed at the boundary of a wind farm under a typical scenario. in ,vout}:

[0043]

[0044] In the formula, N p For ultra-short-term forecast duration, N w The training sample length is defined by sample V, which contains the magnitude and direction of the wind speed. in v out These represent the input wind speed sample and the output wind speed sample, respectively; k is the sampling time.

[0045] ② Update training samples:

[0046] Update the training sample V based on the current natural wind speed v0(k) at time k. new (k)={v in,new ,v out,new}

[0047]

[0048] In the formula, v in,new v out,new These are the updated input wind speed samples and output wind speed samples, respectively. For the updated input sample vector, The updated output sample;

[0049] ③ Based on training sample V new (k), the parameters and structure f of the prediction model fitted based on artificial intelligence modeling methods. pred,k (·);

[0050] ④ Recursively predict the future N based on the prediction model p Natural wind speed at any given moment:

[0051] remember Given a new test sample, predict the natural incoming wind speed at time k+1.

[0052]

[0053] renew Then predict based on the above formula And so on, until predictions are made. Let the wind speed prediction sequence be... ⑤ Repeat steps ①-④ to predict the natural inflow wind speed at the boundary of the wind farm for all typical scenarios.

[0054] In some embodiments, step 5) includes:

[0055] Considering both grid command tracking and generator output fluctuation characteristics under typical scenarios, the objective function for the rolling optimization control of the wind farm is:

[0056]

[0057]

[0058]

[0059]

[0060] In the formula, the number of typical scenarios is N. s and the probability θ corresponding to each typical scenario s ;q i,s P represents the weight coefficients for scenario s. grid Set values ​​for the power grid. Let f be the power setpoint of the j-th unit in scenario s, and let f be the variable to be optimized. 1,s f 2,s These are the targets for tracking power grid commands and the targets for fluctuations in generator output; N p For ultra-short-term forecasts; N t r is the number of wind turbine units. j,s Let be the weight coefficients for scenario s, and k+i|k be the predicted value at time k for time k+i.

[0061] In some embodiments, step 5) includes:

[0062]

[0063] In the formula, N is the power setpoint for the j-th unit. s θ represents the number of typical scenarios. s These represent the probabilities corresponding to each typical scenario. This is the power setting value for the j-th unit in scenario s.

[0064] In a second aspect, the present invention provides a wind farm stochastic model prediction optimization control device that takes into account delay, including a processor and a storage medium;

[0065] The storage medium is used to store instructions;

[0066] The processor is configured to operate according to the instructions to perform the steps of the method according to the first aspect.

[0067] Thirdly, the present invention provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect.

[0068] Beneficial effects: Compared with the prior art, the present invention has the following advantages: The control-oriented quasi-steady-state model of wind farm established in the present invention is based on the steady-state background flow field with added delay processing, which can further improve the accuracy of the wake model and realize the rapid estimation of the dynamic flow field of wind farm; In addition, the proposed wind farm stationary control method based on SMPC can simultaneously handle wind speed uncertainty and wake delay characteristics, thereby reducing the fluctuation of wind farm output and improving the grid-friendly operation capability while ensuring solution efficiency and accuracy.

[0069] Furthermore, the invention 2) uses the background natural incoming wind speed to calculate the delay time, and combines it with the steady-state flow field model to obtain the quasi-steady-state flow field model. This model retains the convenient wake coupling relationship of the traditional steady-state model and considers the dynamic characteristics of the wake delay, thus enabling it to approximately simulate the main dynamic characteristics of the wind farm. It is also applicable to the rapid online calculation of the wake of large-scale wind farms or even wind power bases.

[0070] Furthermore, the spatiotemporal characteristic analysis of natural incoming wind speed in the invention (3) can obtain the fluctuation characteristics of incoming wind speed at different spatial locations at the boundary of the wind farm, so as to more comprehensively simulate the actual complex wind conditions of the wind farm.

[0071] Furthermore, in invention 4), because wind speed exhibits strong nonlinearity and requires rapid online solution, an artificial intelligence modeling method is employed to perform ultra-short-term wind speed prediction for wind farms. This method belongs to the data-driven modeling approach, which can fully exploit the nonlinear information in the data and solve the problem efficiently and rapidly online.

[0072] Furthermore, invention 5) employs an indirect method to handle the strongly nonlinear problem of wind farm steady-state control. This method does not directly optimize the axial induction factor in the formula, but instead uses the constructed power prediction model to first calculate the maximum power value of each wind turbine. Then, the power generation of each turbine is used as the optimization variable, and this maximum power value is used to constrain the optimization variable. Through this indirect transformation, the strongly nonlinear optimization control problem is transformed into a weakly nonlinear optimization problem. This type of optimization problem can be directly solved by calling the Gorubi solver, thereby greatly improving the control solution speed and ensuring the practicality of the proposed method. Attached Figure Description

[0073] Figure 1 This is a schematic diagram of the SMPC-based wind farm optimization control of the present invention;

[0074] Figure 2 This is a structural diagram of the wind farm turbine layout in a specific embodiment of the present invention;

[0075] Figure 3 The step wind speed is the one input in the specific embodiment of this invention;

[0076] Figure 4 This is a schematic diagram of the quasi-steady-state model of a wind farm according to the present invention;

[0077] Figure 5 This is a schematic diagram of the ultra-short-term multi-scenario wind speed prediction based on LS-SVM of the present invention;

[0078] Figure 6 This is a comparison chart of the power output characteristics of wind farms using different control methods under typical conditions in specific embodiments of the present invention. Detailed Implementation

[0079] The present invention will be further illustrated below with reference to the accompanying drawings and specific examples. It should be understood that these examples are for illustrative purposes only and are not intended to limit the scope of the invention. After reading this invention, any modifications of the invention in various equivalent forms by those skilled in the art will fall within the scope defined by the appended claims.

[0080] In the description of this invention, the terms "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0081] Example 1

[0082] like Figure 1 As shown, a stochastic model predictive optimization control method for wind farms considering delay includes:

[0083] Step 1) Obtain historical measured natural inflow wind speeds at the boundary of wind farms, analyze the spatiotemporal characteristics of natural inflow wind speeds in multiple scenarios, and determine the number of typical scenarios N. s and the probability θ corresponding to each typical scenario s ;

[0084] Step 2) Based on the historical measured wind speed at the boundary of the wind farm and the wind speed at the current moment, the wind speed at the boundary of the wind farm is predicted in the future using a prediction model pre-built with artificial intelligence modeling methods, so as to obtain the wind speed at the boundary of the wind farm for all typical scenarios.

[0085] Step 3) Based on the predicted natural incoming wind speed v0 at the wind farm boundary for each typical scenario, and using the steady-state flow field model, calculate the effective incoming wind speed v at the rotor of each turbine, considering the delay. τ,j The maximum operating power P of each unit was calculated using Bates' theorem.MPPT,j ;

[0086] Step 4) Based on P MPPT,j Predict the actual maximum available power of each unit under various typical scenarios.

[0087] Step 5) Based on the number of typical scenarios N s The probability θ corresponding to each typical scenario s and the actual maximum available power of each unit under various typical scenarios The rolling optimization solution is performed using a pre-constructed wind farm rolling optimization control objective function to obtain the power setpoints of each turbine in scenario s.

[0088] Step 6) Based on the probability θ corresponding to each typical scenario s Power settings for various typical scenarios Determine the final power setpoint for the j-th unit.

[0089] Preferably, in step 3), the steady-state flow field model includes:

[0090] ① A steady-state flow field model is used to calculate the wake parameters:

[0091] Choose a steady-state flow field model to calculate the wake parameters: the wake radius r at x downstream of the unit. x and wake wind speed v x ;

[0092] ② Calculate the background flow field distribution:

[0093] Based on the steady-state flow field model, considering the superposition of multiple wake effects, the equivalent inlet wind velocity at the downstream turbine rotor is calculated:

[0094]

[0095] In the formula, v j It is the effective wind speed of the j-th unit, v ij and A ij These represent the wake wind speed and wake overlap area from the i-th turbine to the j-th turbine, respectively; v0 is the natural incoming wind speed at the wind farm boundary; and r... d The radius of the turbine rotor.

[0096] In some embodiments, step 3) calculates the effective incoming wind speed v at the turbine rotor of each unit, considering the delay, based on the predicted natural incoming wind speed v0 at the wind farm boundary for each typical scenario and the steady-state flow field model. τ,j The maximum operating power P of each unit was calculated using Bates' theorem. MPPT,j ,include:

[0097] ① Calculate the wake delay time from unit i to unit j:

[0098]

[0099] In the formula, d ij Let be the distance between the i-th generating unit and the j-th generating unit;

[0100] ② Calculate the effective incoming air velocity at the rotor of each unit considering the delay:

[0101]

[0102] ③ Calculate the maximum operating power of each unit:

[0103]

[0104] In the formula, ρ is the air density, and a j Let be the axial induction factor of the j-th unit.

[0105] In some embodiments, step 4), based on P MPPT,j Predict the actual maximum available power of each unit under various typical scenarios. include:

[0106] When the power generation of a wind turbine is lower than its minimum output, the turbine needs to be shut down. If the theoretical power generation is higher than the maximum power, it will continue to operate at maximum power. That is, the actual maximum available power after considering constraints is:

[0107]

[0108] In the formula, P min P max These are the minimum and maximum power constraints set to ensure the safety of the unit.

[0109] In some embodiments, step 1) obtains the historical measured natural inflow wind speed at the boundary of the wind farm, analyzes the spatiotemporal characteristics of the natural inflow wind speed in multiple scenarios, and determines the number of typical scenarios N. s and the probability θ corresponding to each typical scenario s ,include:

[0110] ① Measure the natural inflow wind speed at different locations at the wind farm boundary;

[0111] ② Determine the time fluctuation characteristics of wind speed magnitude at the same location and in the same wind direction;

[0112] ③ Determine the time fluctuation characteristics of wind speed magnitude under different wind directions at the same location;

[0113] ④ Change the spatial location, repeat ② and ③, and analyze the spatiotemporal characteristics of natural wind speed in multiple scenarios;

[0114] ⑤ Determine the number N of typical location scenarios. s and the probability θ corresponding to this scenario s .

[0115] In some embodiments, step 2) includes:

[0116] ① Construct initial training samples V = {v} of the natural incoming wind speed at the boundary of a wind farm under a typical scenario. in ,v out}:

[0117]

[0118] In the formula, N p For ultra-short-term forecast duration, N w The training sample length is defined by sample V, which contains the magnitude and direction of the wind speed. in v out These represent the input wind speed sample and the output wind speed sample, respectively; k is the sampling time.

[0119] ② Update training samples:

[0120] Update the training sample V based on the current natural wind speed v0(k) at time k. new (k)={v in,new ,v out,new}

[0121]

[0122] In the formula, v in,new、 v out,new These are the updated input wind speed samples and output wind speed samples, respectively. For the updated input sample vector, The updated output sample;

[0123] ③ Based on training sample V new (k), the parameters and structure f of the prediction model fitted based on artificial intelligence modeling methods. pred,k (·);

[0124] ④ Recursively predict the future N based on the prediction model p Natural wind speed at any given moment:

[0125] remember Given a new test sample, predict the natural incoming wind speed at time k+1.

[0126]

[0127] renew Then predict based on the above formula And so on, until predictions are made. Let the wind speed prediction sequence be...

[0128] ⑤ Repeat steps ①-④ to predict the natural inflow wind speed at the boundary of the wind farm for all typical scenarios.

[0129] In some embodiments, the design steps of the wind farm stochastic model prediction stationary controller specifically include:

[0130] ①Predict the actual maximum output of a single wind turbine unit under various typical scenarios:

[0131] Based on the prediction results of the ultra-short-term boundary natural inflow wind speed under various typical scenarios and Equations (1)-(3), the equivalent wind speed v at the wind turbine of different units after considering the delay is predicted. τ,j Based on Bates' theorem, the maximum output operating state P of each wind turbine unit when the axial induction factor a is 1 / 3 is determined. MPPT,j Based on equations (4)-(5), the actual maximum available power of each unit under each typical scenario is predicted.

[0132] ② Determine the target for power grid command tracking:

[0133] The power output of a wind farm should be as close as possible to the grid setpoint to ensure its tracking performance; therefore, the optimization control objectives include:

[0134]

[0135] In the formula, q i,s P represents the weight coefficients for scenario s. grid Set values ​​for the power grid. Let be the power setpoint of the j-th unit in scenario s, and be the variable to be optimized;

[0136] ③ Determine the target for output fluctuation:

[0137] The significant uncertainty in wind speed leads to large fluctuations in wind turbine output. To improve power generation stability, the output changes of each unit should not be too drastic.

[0138]

[0139] In the formula, N t r is the number of wind turbine units. j,s Let be the weight coefficients for scenario s, and k+i|k be the predicted value at time k for time k+i.

[0140] ④ Output fluctuation target normalization processing:

[0141] Due to the significant uncertainty in wind speed, the power output of wind turbines inevitably fluctuates considerably. To reduce the impact of these fluctuations, the future power output fluctuation characteristics of wind turbines are normalized using historical data on wind turbine fluctuations.

[0142]

[0143] ⑤ Determine the rolling optimization control objective:

[0144] Considering both grid command tracking and generator output fluctuation characteristics under typical scenarios, the rolling optimization control objective of the wind farm is:

[0145]

[0146] ⑥ Calculate the instructions issued by the j-th generating unit And distributed to the wind turbine units:

[0147]

[0148] In some specific embodiments, the wind farm stochastic model predictive optimization control method considering delay includes: 1) calculating the steady-state background flow field of the wind farm:

[0149] According to the appendix Figure 2 The wind farm shown is 2 kilometers long and 1 kilometer wide. The spacing between the first and second rows, and between the second and third rows, is 11D (where D is the diameter of the wind turbine rotor) and 15D, respectively. The spacing within each group is 1.5D. The air density is assumed to be 1.25 kg / m³. 3 The wind farm turbine model is NREL 5MW (NERL stands for National Renewable Energy Laboratory, and 5MW is the rated power of the turbine). This wind farm model is built based on WFSim (a medium-fidelity simulation model of a control-oriented wind farm). (Input attached) Figure 3 The step wind speed shown is initially 8 m / s, then suddenly increases to 10 m / s at 11 s, then suddenly increases to 12 m / s at 28 s, and then suddenly decreases to 10 m / s and 8 m / s at 40 s and 57 s respectively, with a total duration of 70 s.

[0150] The wake effect of a single unit is calculated using the classic Jensen wake model:

[0151]

[0152] To account for the effects of multiple wake effects, the sum of squares method described in the above formula is used to calculate the equivalent inlet wind speed v at the rotor of the j-th downstream unit. j Thus, the distribution of the entire background flow field is obtained.

[0153] 2) Construct a quasi-steady-state power prediction model for wind farms that considers delays:

[0154] ① Calculate the wake delay time τ from the i-th unit to the j-th unit according to formula (2). ij ;

[0155] ② Combining the delay time and background wind speed, the effective incoming wind speed v at the turbine rotor of each unit is calculated using formula (3). τ,j ;

[0156] ③ Calculate the maximum operating power P of each unit according to formula (4). MPPT,j ;

[0157] ④ Calculate the actual maximum available power of each wind turbine after considering constraints using formula (5).

[0158] 3) Randomly select 100 different locations at the boundary of the wind farm, measure the natural incoming wind speed at these 100 locations, and analyze the spatiotemporal characteristics of wind speed direction and magnitude at each location. Based on this, use the backward scene reduction method to determine the number of typical location scenes N. s and the probability θ corresponding to this scenario s To handle wind speed uncertainty.

[0159] 4) Construct a short-term wind speed prediction model for typical scenarios based on the least squares support vector machine (LS-SVM) method:

[0160] ① Construct the initial training samples V = {v} according to formula (6) in ,v out};

[0161] ②Based on the current natural wind speed v at time k 0,k Update the training sample V using formula (7) new,k ={v in,new ,v out,new};

[0162] ③ Based on the latest training samples V new,k Fitting the LS-SVM prediction model (14):

[0163]

[0164] In the formula, N is the number of samples, K(·) is the Gaussian kernel function, and α=[α1,…,α N ] and b are the model parameters to be fitted;

[0165] ④ Based on the prediction model (14), recursively predict the future N p The wind speed prediction vector at time k is obtained by taking the natural incoming wind speed at time k. The specific prediction process is shown in the attached document. Figure 5 As shown;

[0166] ⑤ Repeat steps ①-④ to predict the natural inflow wind speed at the wind farm boundary for all typical scenarios. It should be noted that the wind speed obtained by LS-SVM is an average wind speed. To account for the uncertainty of the model wind speed, a Gaussian function is typically added to the deterministic wind speed to simulate instantaneous wind speed fluctuations. However, in reality, wind speed fluctuations also have spatial characteristics, and the wind speed at different boundaries can vary significantly at the same time. Therefore, this invention predicts the boundary wind speed at different spatial locations within the wind farm separately to more realistically simulate the fluctuation characteristics of actual wind speed.

[0167] 5) Design a wind farm stationary control model based on stochastic model predictive control (SMPC):

[0168] ①Predict the actual maximum available power of a single wind turbine:

[0169] Based on the prediction results of ultra-short-term boundary natural inflow wind speed under various typical scenarios Formulas (1)-(3) predict the equivalent wind speed v at the rotor of different units after considering the delay. τ,j Based on Bates' theorem, the maximum output operating state P of each wind turbine unit when the axial induction factor a is 1 / 3 is determined. MPPT,j Based on equations (4)-(5), the actual maximum available power of each unit under each typical scenario is predicted.

[0170] ② Determine the target for tracking power grid commands according to formula (9);

[0171] ③ Determine the output fluctuation target according to formula (10);

[0172] ④ The power output fluctuation target is normalized according to formula (11). The control time domain and prediction time domain of SMPC are both 5 seconds, the control period is 1 minute, and the wind farm sampling period is 30 seconds.

[0173] ⑤ Optimize the power generation of each wind turbine according to the objective function (12). This optimization problem can be solved by calling the Gorubio solver.

[0174] ⑥ Determination of evaluation indicators and comparison methods:

[0175] This invention uses the average relative deviation ε MRE Used to evaluate the tracking performance and root mean square deviation ε of the power grid setpoint. RMSE The formulas used to evaluate output fluctuations are as follows:

[0176]

[0177]

[0178] Where N lThis represents the total simulation duration for the wind farm.

[0179] In addition, since wind speed is a non-stationary random process, this paper adopts a dynamic fluctuation index to evaluate the smoothness characteristics of wind farm output at different times. That is, the average fluctuation characteristic within a window is calculated each time as the fluctuation characteristic at the current time, and the root mean square error within the window is used as the rolling root mean square error at the current time.

[0180]

[0181]

[0182] Where, N r P is the width of the scroll window. bar (k) represents N within the rolling window at time k. r The average value at each time point, ε ARMSE This represents the root mean square error value for the window.

[0183] ⑧ Simulation solution:

[0184] In addition, to demonstrate the effectiveness of the method proposed in this invention, the following methods are used as comparative methods:

[0185] 1. SMPC-D: This refers to the stationary control method proposed in this paper that simultaneously considers uncertainty and delay characteristics;

[0186] 2. SMPC-S: A stationary control method that considers uncertainties but not delay characteristics;

[0187] 3. MPC-D: A stationary control method that considers delay but not uncertainty.

[0188] Table 1 presents quantitative comparison indicators for different methods. The table quantitatively shows that the average relative deviations of the different methods are roughly the same, indicating that all methods can track power grid commands well.

[0189] Combination Figure 6As shown in (a), all three methods can meet the grid command tracking requirements well, regardless of whether the operation is power-limited or MPPT. However, the fluctuations of different methods vary greatly, as shown in (b) and Table 1. The MPC method has the largest root mean square variance, because the MPC (Model Predictive Control) method does not consider the influence of wind speed uncertainty, making it difficult for actual equipment to fully suppress wind speed uncertainty. Secondly, the root mean square errors of the SMPC-D and SMPC-S methods are roughly the same, but due to the non-stationarity of wind speed, the root mean square error cannot effectively quantify the uncertainty of wind farm output. From the moving mean square error, it can be seen that the SMPC-D method has the smallest value, indicating that the wind farm output is smoother, proving that considering the dynamic characteristics of wake delay can improve the overall stability of wind farm output.

[0190] The comparison of the three demonstrates the effectiveness of the present invention in suppressing wind farm fluctuations that take into account delays and uncertainties.

[0191] Table 1. Quantitative Comparison Indicators of Different Methods

[0192]

[0193] Example 2

[0194] Secondly, this embodiment provides a wind farm stochastic model prediction optimization control device that takes into account delay, including a processor and a storage medium;

[0195] The storage medium is used to store instructions;

[0196] The processor is configured to operate according to the instructions to perform the steps of the method according to Embodiment 1.

[0197] Example 3

[0198] Thirdly, this embodiment provides a storage medium on which a computer program is stored, which, when executed by a processor, implements the steps of the method described in Embodiment 1.

[0199] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0200] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0201] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0202] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0203] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for predictive optimization control of wind farms using stochastic models that consider delay, characterized in that, include: Step 1), obtain the historically measured wind speed of natural incoming flow at the boundary of the wind farm, analyze the spatiotemporal characteristics of the natural incoming flow wind speed in multiple scenarios, and determine the number N of typical scenarios s and the probability θ corresponding to each typical scenario s ; Step 2) Based on the historical measured wind speed at the boundary of the wind farm and the wind speed at the current moment, the wind speed at the boundary of the wind farm is predicted in the future using a prediction model pre-built with artificial intelligence modeling methods, so as to obtain the wind speed at the boundary of the wind farm for all typical scenarios. Step 3) Based on the predicted natural incoming wind speed v0 at the wind farm boundary for each typical scenario, and using the steady-state flow field model, calculate the effective incoming wind speed v at the rotor of each turbine, considering the delay. τ,j The maximum operating power P of each unit was calculated using Bates' theorem. MPPT,j ; Step 4) Based on P MPPT,j Predict the actual maximum available power of each unit under various typical scenarios. ; Step 5) Based on the number of typical scenarios N s The probability θ corresponding to each typical scenario s and the actual maximum available power of each unit under various typical scenarios The rolling optimization control method for wind farms, which is pre-constructed, is used to solve the rolling optimization problem and obtain the power setpoints of each unit under scenario s. ; Step 6) Based on the probability θ corresponding to each typical scenario s Power settings for various typical scenarios Determine the final power setpoint for the j-th unit. .

2. The wind farm stochastic model prediction optimization control method considering delay as described in claim 1, characterized in that, Step 3) includes: Based on the steady-state flow field model, considering the superposition of multiple wake effects, the equivalent inlet wind speed at the downstream turbine rotor is calculated. : ； In the formula, v i v is the effective wind speed of the i-th unit. ij A is the wake velocity from unit i to unit j. ij Let v0 be the wake overlap area from the i-th turbine to the j-th turbine, v0 be the natural incoming wind speed at the wind farm boundary, and r be the wake overlap area. d The radius of the turbine rotor; Calculate the wake delay time from unit i to unit j. : ； In the formula, d ij Let v0 be the distance from the i-th unit to the j-th unit, and v0 be the natural inflow wind speed at the boundary of the wind farm. Calculate the effective incoming air velocity at the rotor of each unit considering the delay. : ； In the formula, express time; Calculate the maximum operating power of each unit. : ； In the formula, It is air density, a j Let be the axial induction factor of the j-th unit. The area swept by the wind turbine.

3. The wind farm stochastic model prediction optimization control method considering delay as described in claim 2, characterized in that, a j It is 1 / 3.

4. The wind farm stochastic model prediction optimization control method considering delay as described in claim 1, characterized in that, Step 4) Based on P MPPT,j Predict the actual maximum available power of each unit under various typical scenarios. ,include: When the power generation of a wind turbine is lower than its minimum output, the turbine needs to be shut down. If the theoretical power generation is higher than the maximum power, it will continue to operate at maximum power, which is the actual maximum available power after considering constraints. for: ； In the formula, P min P max These are the minimum and maximum power constraints set to ensure the safety of the unit.

5. The wind farm stochastic model predictive optimization control method considering delay as described in claim 1, characterized in that, Step 1) includes: ① Measure the natural inflow wind speed at different locations at the wind farm boundary; ② Determine the time-varying characteristics of wind speed magnitude at the same location and in the same wind direction; ③ Determine the time fluctuation characteristics of wind speed magnitude under different wind directions at the same location; ④ Change the spatial location, repeat ② and ③, and analyze the spatiotemporal characteristics of natural wind speed in multiple scenarios; ⑤ Determine the number N of typical scenarios. s and the probability θ corresponding to this scenario s .

6. The wind farm stochastic model predictive optimization control method considering delay according to claim 1, characterized in that, Step 2) includes: ① Construct initial training samples V={v} of the natural incoming wind speed at the boundary of a wind farm under a typical scenario. in , v out }: ； ； In the formula, N p For ultra-short-term forecast duration, N w The training sample length is defined by sample V, which contains the magnitude and direction of the wind speed. in、 v out These represent the input wind speed sample and the output wind speed sample, respectively; k is the sampling time. ② Update training samples: ; Update the training sample V based on the current natural wind speed v0(k) at time k. new (k)={v in,new , v out,new } ； In the formula, v in,new、 v out,new These are the updated input wind speed samples and output wind speed samples, respectively. For the updated input sample vector, The updated output sample; ③ Based on training sample V new (k), the parameters and structure f of the prediction model fitted based on artificial intelligence modeling methods. pred,k (·); ④ Recursively predict the future N based on the prediction model p Natural wind speed at any given moment: remember Given a new test sample, predict the natural incoming wind speed at time k+1. : ； renew Then predict based on the above formula ; and so on, until predictions are made. Let the wind speed prediction sequence be denoted as ; ⑤ Repeat steps ①-④ to predict the natural inflow wind speed at the boundary of the wind farm for all typical scenarios.

7. The wind farm stochastic model predictive optimization control method considering delay according to claim 1, characterized in that, Step 5) includes: Considering both grid command tracking and generator output fluctuation characteristics under typical scenarios, the objective function for the rolling optimization control of the wind farm is: ； ； ； In the formula, the number of typical scenarios is N. s and the probability θ corresponding to each typical scenario s ;q i,s These are the weighting coefficients for scenario s. Let be the power setpoint of the j-th unit in scenario s, and be the variable to be optimized; , These are the targets for tracking power grid commands and the targets for fluctuations in generating unit output; For ultra-short-term forecasts; N t r is the number of wind turbine units. j,s These are the weighting coefficients for scenario s. This represents the power grid prediction value at time k+i for time k.

8. A wind farm stochastic model predictive optimization control device considering delay, characterized in that, Including processor and storage media; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method according to any one of claims 1 to 7.

9. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images