Wind farm active power allocation method considering turbulent wind speed fluctuations

Abstract

The application discloses a wind farm active power distribution method considering turbulent wind speed fluctuation. In view of the problem that the existing distribution mode is difficult to maintain the output of the wind turbine in some time periods due to the neglect of the fluctuation characteristics of the actual turbulent wind speed, the method first depicts the power instruction change range of the wind turbine that can continuously respond under the turbulent wind speed, and adopts a convolutional neural network to construct the mapping relationship between the turbulent wind speed characteristics and the power instruction change range; on this basis, the wind farm active power distribution strategy is designed by taking the minimum output deviation and the minimum instruction change amplitude of the wind farm as the optimization objectives and taking the power instruction change range as the constraint. The application fully considers and quantifies the influence of the turbulent wind speed fluctuation in the distribution process, so that the wind farm can accurately respond to the dispatching instruction of the power grid.

Description

Technical Field

[0001] This invention belongs to the field of wind farm optimization scheduling, specifically involving a wind farm active power allocation method that considers turbulent wind speed fluctuations. Background Technology

[0002] As the scale of wind power integration continues to expand, the randomness and volatility of wind energy will significantly impact the safe and stable operation of the power system. Therefore, wind power is required to possess a certain level of active power regulation capability and participate in the grid's Automatic Generation Control (AGC). In this context, wind turbines no longer employ the traditional maximum power point tracking (MPPT) control strategy but need to actively adjust their operating states to respond to power commands, maintaining constant and controllable active power output. Furthermore, compared to conventional thermal power units, individual wind turbines have smaller capacities, and the grid's AGC focuses more on the power response at the wind farm level. Therefore, coordinating wind turbines operating in different states within a wind farm and developing a reasonable active power allocation strategy is crucial to ensuring the wind farm's accurate response to grid AGC commands.

[0003] Currently, engineering practices typically employ methods such as average allocation, capacity allocation, or wind speed allocation. Among these, wind speed allocation is widely used because it reflects the actual operating status of wind turbines. However, this method is limited to a relatively fixed allocation principle and is difficult to apply well to complex and fluctuating wind conditions. Building on this, further research proposes an optimization model aimed at accurately responding to power commands, combining various optimization algorithms to achieve optimal allocation of active power in wind farms. However, this type of research focuses more on the design of the optimization objective and the improvement of the optimization algorithm, only considering the average wind speed obtained through measurement or prediction when utilizing wind speed information.

[0004] Typically, average wind speed only reflects the average intensity of wind speed over a period of time and does not reflect the fluctuating characteristics of actual turbulent wind speed. When turbulent wind speed fluctuates significantly, if wind farms still allocate power commands based on average wind speed information, some wind turbines may struggle to maintain the issued power commands during periods of lower-than-average wind speed, leading to power drops and affecting the accuracy of wind farm's active power control (AGC) response to the grid. Therefore, it is necessary to consider the fluctuating characteristics of turbulent wind speed during the allocation process to improve the reliability of active power dispatching at wind farms. Summary of the Invention

[0005] The purpose of this invention is to provide a method for active power allocation in wind farms that considers turbulent wind speed fluctuations. It introduces an index for the power command variation range of wind turbine units, uses a convolutional neural network to construct a mapping relationship between turbulent wind speed characteristics and command variation range, and designs an optimization model that considers the power command variation range constraint, so that wind farms can more accurately and reliably track and respond to grid power commands.

[0006] The technical solution to achieve the objective of this invention is as follows: Firstly, this invention provides a method for active power allocation in wind farms that considers turbulent wind speed fluctuations, comprising the following steps:

[0007] Step 1: Generate a set of turbulent wind speed sequences;

[0008] Step 2: Calculate three turbulence characteristic indices based on turbulent wind speed: average wind speed, turbulence intensity, and turbulence frequency; and calculate the power fluctuation coefficient through simulation.

[0009] Step 3: Train the evaluation model for power fluctuation coefficient;

[0010] Step 4: Obtain power grid dispatch instructions and real-time wind speed;

[0011] Step 5: Determine the power command variation range based on the evaluation model;

[0012] Step 6: Optimize the allocation of power commands to each generator unit.

[0013] In a second aspect, the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the method described in the first aspect.

[0014] Thirdly, the present invention provides a computer-readable 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.

[0015] Fourthly, the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the method described in the first aspect.

[0016] Compared with the prior art, the significant advantages of this invention are: 1) This invention fully considers and quantifies the influence of turbulent wind speed fluctuations during the allocation process, thereby proposing a wind farm active power allocation strategy that considers turbulent wind speed fluctuations, solving the problem that wind turbines have difficulty maintaining the issued power commands when turbulent wind speed fluctuations are severe; 2) This invention discloses the steps of the wind farm active power allocation strategy that considers turbulent wind speed fluctuations. First, it characterizes the range of power command changes that the wind turbine can continuously respond to under turbulent wind speeds, and uses a convolutional neural network to construct the mapping relationship between turbulent wind speed characteristics and the range of power command changes. Based on this, with the minimum wind farm output deviation and the minimum command change amplitude as optimization objectives and the range of power command changes as constraints, a wind farm active power allocation strategy is designed.

[0017] The present invention will now be described in further detail with reference to the accompanying drawings. Attached Figure Description

[0018] Figure 1 This is a flowchart of the active power allocation strategy for wind farms that takes into account turbulent wind speed fluctuations, as presented in this invention.

[0019] Figure 2(a) and Figure 2(b) show the simulation results for verifying the effectiveness of the present invention. Figure 2(a) shows the accuracy test error of the trained evaluation model, and Figure 2(b) shows the power response curve of the wind farm and the speed change curve of a wind turbine under the proposed method and the traditional allocation method. Detailed Implementation

[0020] like Figure 1 As shown, this invention proposes a method for active power allocation in wind farms that considers turbulent wind speed fluctuations, comprising the following steps:

[0021] Step 1: Generate a set of turbulent wind speed sequences;

[0022] Step 2: Calculate three turbulence characteristic indicators based on turbulent wind speed: average wind speed Turbulence intensity TI and turbulence frequency ω eff The power fluctuation coefficient α is calculated through simulation.

[0023] Step 3: Train the evaluation model for the power fluctuation coefficient α;

[0024] Step 4: Obtain the power grid dispatch instruction P ref and real-time wind speed v;

[0025] Step 5: Determine the power command variation range based on the evaluation model;

[0026] Step 6: Optimize the allocation of power commands to each generator unit;

[0027] The present invention will be further described in detail below with reference to embodiments:

[0028] In this embodiment, the wind farm model consists of nine wind turbine units (3×3), with the spacing between each turbine being five times the rotor diameter. The wind turbine unit model adopts NREL's 5MW model, and the main parameters are shown in Table 1.

[0029] Table 1 Main parameters of wind turbine generators

[0030]

[0031] Average wind speed The turbulent wind speed reflects the average level of change over a period of time. Turbulence intensity TI represents the amplitude of fluctuations in turbulent wind speed around the average wind speed, and turbulence frequency ω... eff This indicates the degree of turbulent wind speed change over time. The specific calculations for the three characteristic indicators are as follows:

[0032]

[0033]

[0034]

[0035] In the formula, v(t) represents the wind speed point sequence sampled within the statistical period T, and in this embodiment, T = 10 min is taken; σ represents the standard deviation of the wind speed sequence v(t); Δt represents the sampling time interval, and in this embodiment, Δt = 1 s is taken; Δv represents the wind speed difference between adjacent sampling points; N represents the total number of times it is taken within the statistical period T, and N = T / Δt = 600.

[0036] The power fluctuation coefficient α is calculated using the active power of the wind turbine obtained from the simulation of the turbulent wind speed sequence, and can be specifically expressed as:

[0037]

[0038] In the formula, n represents the number of samples within 10 minutes; P i P represents the active power of the wind turbine in the i-th sampling period; mean This represents the average active power of the wind turbine over a 10-minute period. First, the fluctuation range of the active power of the wind turbine in each sampling period is calculated based on the average value. Then, the standard deviation of all fluctuation ranges is calculated, and this standard deviation result is the power fluctuation coefficient α.

[0039] Wind power output exhibits a cubic relationship with wind speed. In this case, the active power output of the wind turbine will vary within a certain range as the turbulent wind speed fluctuates. Therefore, the wind turbine's WT can be defined. i Power command variation range:

[0040]

[0041]

[0042]

[0043] In the formula, This represents the average level of the active power output of the wind turbine, derived from the average wind speed. Decide; and Δ P This represents the upper and lower limits of the wind turbine output fluctuation, where α represents the power fluctuation coefficient, which is determined by the turbulence intensity TI and the turbulence frequency ω. eff A decision can be expressed as:

[0044] α=f(TI,ω eff )

[0045] Turbulence intensity TI and turbulence frequency ω are constructed using deep learning algorithms. eff The mapping relationship with the power fluctuation coefficient α is used to evaluate the range of power command variation of the wind turbine under turbulent wind speeds.

[0046] A convolutional neural network (CNN) consists of an input layer, convolutional layers, pooling layers, fully connected layers, and an output layer. Based on the turbulent wind speed characteristics of the input, the convolutional layers utilize convolutional kernels from the CNN model for local learning. The convolution operation expression for each kernel is as follows:

[0047] a c,k =f(X*W c,k +b c,k )

[0048] In the formula, a c,k f(x) is the output of the k-th convolutional surface of this convolutional layer; f(x) is the activation function, and in this embodiment, the ReLU function is chosen; X is the input matrix; W c,k b is the weight matrix corresponding to the k-th convolution kernel; c,k This is a bias term.

[0049] In the pooling layer, using max pooling, it can be represented as:

[0050] a p,k =max(a ij i,j=1,2,…,n

[0051] In the formula, a p,k Represents the k-th pooling surface; a ij This represents a sub-block of the output matrix of the previous convolutional layer; n is a... ij Dimensions.

[0052] In the fully connected layer, the calculation formula can be expressed as:

[0053] afc =f(a p *W fc +b fc )

[0054] In the formula: a fc The output is f(x); the activation function is a. p For input; W fc b is the weight matrix; fc This is the bias term. The fully connected layer passes the calculation results to the output layer, ultimately yielding the power fluctuation coefficient α.

[0055] Of the 1800 10-minute turbulent wind speed sequences, 90% were selected as training samples and 10% as test samples. After training the evaluation model, the accuracy of the model was tested using the remaining 10% of samples, and the output error results are shown in Figure 2(a).

[0056] When the power grid AGC issues a power command At that time, the wind farm dispatch center first calculates the turbulence characteristic index (average wind speed v) of the wind speed v based on the wind speed sequence. Turbulence intensity TI and turbulence frequency ω eff Then, based on the constructed evaluation model, the range of power command variation for the current wind turbine is determined, and finally, the power command for each wind turbine is determined through an active power optimization allocation strategy.

[0057] Taking into account both minimizing the response deviation and minimizing the change range of the power command, and using the power command variation range as a constraint, the specific optimization model is as follows:

[0058]

[0059]

[0060] In the formula: This represents the power grid AGC command for the j-th scheduling cycle; represent the power commands of the i-th wind turbine unit in the j-th and (j-1)-th scheduling periods, respectively; k1 and k2 correspond to the weight coefficients of the two optimization objectives, respectively; This represents the lower limit of the output of the i-th wind turbine unit within the j-th scheduling cycle, which is taken as 10% of the rated power in this embodiment; This represents the power command variation range of the i-th wind turbine unit within the j-th scheduling period, where... Determined by average wind speed, and Δ P The turbulence intensity and turbulence frequency are determined by the calculation based on the constructed evaluation model.

[0061] By simplifying and solving the above equation, the power command of each wind turbine unit in each scheduling cycle can be obtained.

[0062] The traditional allocation method allocates power based on a weighted coefficient within the wind speed range. Figure 2(b) shows the overall power output response of the wind farm compared to the traditional method. Furthermore, to more intuitively compare the power response deviations of the two methods, the cumulative response deviation ΔP of the power response for each allocation method is calculated. sum The root mean square error (RMSE) is calculated as follows:

[0063]

[0064]

[0065] In the formula, This indicates the actual power output of the wind farm. This represents the AGC command issued by the power grid to the wind farm, and N represents the number of sampling points. The calculation results are shown in Table 2.

[0066] Table 2 Weighting coefficients for different wind speed ranges

[0067]

[0068] As shown in Figure 2(b), the traditional allocation method exhibits varying degrees of power drop across multiple scheduling cycles, while the allocation method proposed in this invention accurately and continuously responds to the grid AGC commands in each scheduling cycle. This is because the traditional allocation method allocates power commands based solely on average wind speed information, neglecting the fluctuation characteristics of actual wind speed. This results in the wind turbines' actual output failing to respond in real-time to meet the allocated power commands. In contrast, the allocation method proposed in this invention quantifies the turbulence characteristics of wind speed, thereby determining the range of power command variations for each wind turbine under actual turbulent wind speeds and allocating power accordingly. This allows the wind turbines to respond to the issued power commands more accurately and reliably.

[0069] As shown in Figure 2(b), under the traditional allocation method, the actual output of the WT3 wind turbine cannot meet the allocated power command during certain periods, requiring it to release its own kinetic energy to compensate, leading to a continuous decrease in speed. When the speed falls below the minimum speed limit, the wind turbine will switch from active power control back to MPPT operation, unable to continue responding to the power command, resulting in a power drop. In contrast, the allocation method proposed in this invention accurately assesses the active power regulation capability of each wind turbine under actual turbulent wind speed by constraining the range of power command changes. The allocated power command ensures that each unit can respond reliably, and the actual speed is always above the lower speed limit. At the same time, the optimization model comprehensively considers the constraint of the power command amplitude, making the change of the allocated power command smoother during each scheduling cycle, which is conducive to the healthy operation of the wind turbine in the grid AGC.

Claims

1. A method for active power allocation in a wind farm considering turbulent wind speed fluctuations, characterized in that, Includes the following steps: Step 1: Generate a set of turbulent wind speed sequences; Step 2: Calculate three turbulence characteristic indices based on turbulent wind speed: average wind speed, turbulence intensity, and turbulence frequency; and calculate the power fluctuation coefficient through simulation. Step 3, Training Power Fluctuation Coefficient The evaluation model and specific methods are as follows: Define the fan Power command variation range: ； ； ； In the formula, This represents the average level of the active power output of the wind turbine, derived from the average wind speed. Decide; and This indicates the upper and lower limits of fluctuation in wind turbine output. The power fluctuation coefficient is represented by the turbulence intensity. and turbulence frequency A decision is expressed as: ； Constructing turbulence intensity using a convolutional neural network turbulence frequency With power fluctuation coefficient The mapping relationship is used to evaluate the range of power command variation of the wind turbine under turbulent wind speed; A convolutional neural network consists of an input layer, convolutional layers, pooling layers, fully connected layers, and an output layer. Based on the turbulent wind speed characteristics of the input, the convolutional layers use convolutional kernels in the CNN model for local learning. The convolution operation expression for each kernel is as follows: ； In the formula, For the first convolutional layer The output of each convolutional surface; For activation functions; The input matrix; For the first The weight matrix corresponding to each convolution kernel; For bias terms; In the pooling layer, max pooling is used, as follows: ； In the formula, Indicates the first A pooled surface; This is a sub-block of the output matrix of the previous convolutional layer; for The dimension; In the fully connected layer, the calculation formula is expressed as: ； In the formula: For output; For activation functions; For input; It is a weight matrix; For bias terms; The fully connected layer passes the calculation results to the output layer, ultimately yielding the power fluctuation coefficient. ; Step 4: Obtain power grid dispatch instructions and real-time wind speed; Step 5: Determine the power command variation range based on the evaluation model; Step 6: Optimize the allocation of power commands to each generator unit; The specific form of the optimized allocation model is as follows: ； In the formula: Indicates the first Power grid AGC commands for each scheduling cycle; , They represent the first The scheduling cycle and the first Within the scheduling cycle, the first Power command for typhoon generators; , These correspond to the weight coefficients of the two optimization objectives, respectively. Indicates the first Within the scheduling cycle, the first Lower limit of typhoon generator output; Indicates the first Within the scheduling cycle, the first The maximum output of the typhoon generator unit.

2. The active power allocation method for wind farms considering turbulent wind speed fluctuations according to claim 1, characterized in that, The turbulent characteristic indices and power fluctuation coefficient of the turbulent wind speed in step 2 are calculated using the following methods: Average wind speed The average level of turbulent wind speed variation over a period of time, turbulence intensity The turbulent frequency represents the magnitude of the fluctuation in turbulent wind speed around the average wind speed. This indicates the degree of drastic change in turbulent wind speed over time; The specific calculations for the three characteristic indicators are as follows: ； ； ； In the formula, Represents a statistical period Internally sampled wind speed point sequence; Represents wind speed sequence Standard deviation; Indicates the sampling time interval; This represents the wind speed difference between adjacent sampling points; ; Power fluctuation coefficient The active power of the wind turbine generator is calculated using simulations of turbulent wind speed sequences, and can be specifically expressed as follows: ； In the formula, Indicates the statistical period Number of samplings within; Indicates the first The active power of the wind turbine in each sampling period; Indicates the statistical period The average active power of the internal wind turbine units is calculated. First, the fluctuation range of the active power of the wind turbines in each sampling period is calculated based on the average value. Then, the standard deviation of all fluctuation ranges is calculated. The result is the power fluctuation coefficient. .

3. The active power allocation method for wind farms considering turbulent wind speed fluctuations according to claim 2, characterized in that, Pick .

4. The active power allocation method for wind farms considering turbulent wind speed fluctuations according to claim 1, characterized in that, Choose the ReLU function.

5. The active power allocation method for wind farms considering turbulent wind speed fluctuations according to claim 1, characterized in that, Take 10% of the rated power.

6. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method described in any one of claims 1-5.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the method described in any of claims 1-5.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method described in any one of claims 1-5.

Images