A frequency modulation energy assessment method considering the temporal and spatial distribution differences of wind speed in wind farms

Abstract

A frequency modulation energy assessment method that considers the differences in the spatiotemporal distribution of wind speed in a wind farm relates to the technical field of wind farms and is used to solve the problem that only the kinetic energy of the wind turbine blade rotor is considered in the frequency modulation energy assessment of a wind farm, while the loss of mechanical power captured during the frequency modulation process is ignored, resulting in large errors in the assessment results. It includes: (1) constructing a minute-level wind speed probability prediction model for each wind farm group based on numerical weather forecasts; (2) constructing a typical wind speed prediction scenario; (3) evaluating the frequency modulation energy of a wind farm based on a typical wind speed prediction scenario. The frequency modulation energy assessment algorithm for a wind farm proposed in this patent can be achieved by analyzing the differences in the spatiotemporal distribution of wind speed and generating a typical wind speed prediction scenario using a probability prediction distribution function, which can effectively reduce estimation errors.

Description

Technical Field

[0001] The invention relates to the technical field of wind farms, and in particular to a frequency modulation energy evaluation method taking into account the temporal and spatial distribution differences of wind speeds in wind farms. Background Art

[0002] Renewable energy power generation is connected to the power system through power electronic equipment. Under conventional control strategies, it does not have the inertia and primary frequency regulation capability of traditional synchronous machines. As the proportion of renewable energy power generation increases, the system inertia level and frequency regulation response capability will decrease, and the frequency safety characteristics will deteriorate.

[0003] At present, it has become a common requirement for some European and American countries that wind power actively undertakes grid frequency regulation. Domestic and foreign scholars have focused their research on wind turbine frequency regulation strategies. Widely used strategies for wind turbine frequency regulation include adding a power loop to the rotor-side inverter control, introducing virtual inertia control, droop control, or a combination of both to simulate the frequency regulation response process of synchronous machines. In this formula, the energy source for wind turbine frequency regulation is mainly the blade kinetic energy stored in the wind turbine, and its size is affected by the wind speed at the height of the wind turbine hub.

[0004] In comparison, there are relatively few studies on the evaluation of wind turbines or wind farms' ability to participate in power system frequency regulation services. The wind turbine frequency regulation parameter setting methods proposed in existing studies have not established the coupling relationship between the frequency regulation parameters and the available frequency regulation energy behind them, and cannot reflect the actual frequency regulation capacity that wind turbines or wind farms can bear.

[0005] In order to study the frequency regulation capability of wind turbines or wind farms, it is necessary to level out the available frequency regulation energy of wind farms. The literature [T.Liu, W.Pan, R.Quan and M.Liu,"A variabledroop frequency control strategy for wind farms that considers optimal rotor kinetic energy", IEEE Access, vol.7, pp.68636-68645, 2019.] evaluated the maximum obtainable rotor kinetic energy of wind turbines. However, in a wind farm, due to the influence of the wake effect, there are differences in the spatial distribution of wind speed, and the wind speed of each row of wind turbines in the windward direction of the farm gradually decreases. Therefore, there are differences in the blade rotor kinetic energy of wind turbines at different positions in the farm. The literature [S.Kuenzel, L.P.Kunjumuhammed, B.C.Pal, and I.Erlich, “Impact of wakes on wind farm inertial response,” IEEE Trans.Sustain.Energy, vol.5, no.1, pp.237–245, Jan.2014.] points out that the wake effect significantly affects the inertial response of the wind farm. If each wind turbine in the wind farm uses a single wind tower wind speed, there will be a significant error between the blade rotor kinetic energy and the actual wind speed. For this reason, related research [J.Lee, E.Muljadi, P. and Y.C. Kang,"Releasable kinetic energy-based inertial control of a DFIG wind power plant",IEEE Trans. Sustain. Energy, vol. 7, no. 1, pp. 279-288, Jan. 2016.] and [J. Lee, G. Jang, E. Muljadi, F. Blaabjerg, Z. Chen and Y.C. Kang,"Stable short-term frequency support using adaptive gains for a DFIG-based wind power plant",IEEE Trans. Energy Convers., vol. 31, no. 3, pp. 1068-1079, Sep. 2016.] considered the spatial distribution differences of wind speed affected by the wake effect in the wind farm, evaluated the releasable rotor kinetic energy of each wind turbine in the wind farm, and proposed a method for setting the frequency regulation parameters of wind turbines considering the rotor kinetic energy. The above studies only considered the kinetic energy of the wind turbine blade rotor in the wind farm frequency modulation energy assessment, while ignoring the loss of mechanical power capture during the frequency modulation process. In addition, the existing studies only considered the spatial distribution differences of wind speed in wind farms, ignoring the time-varying nature of wind farm frequency modulation energy caused by the time series fluctuations of wind speed affected by atmospheric turbulence, resulting in large errors in the assessment results. Summary of the invention

[0006] The purpose of the present invention is to provide a frequency modulation energy evaluation method taking into account the differences in the temporal and spatial distribution of wind speed in wind farms, so as to solve the problems that in the frequency modulation energy evaluation of wind farms, only the kinetic energy of wind turbine blade rotors is considered while the loss of mechanical power capture in the frequency modulation process is ignored, and only the spatial distribution differences of wind speed in wind farms are considered while ignoring the time-varying nature of the frequency modulation energy of wind farms caused by the temporal fluctuations of wind speed affected by atmospheric turbulence, resulting in large errors in the evaluation results.

[0007] The technical solution adopted by the present invention to solve the technical problem is: a frequency modulation energy evaluation method considering the differences in the temporal and spatial distribution of wind speed in a wind farm, comprising the following steps:

[0008] 1. Construct a minute-level wind speed probability prediction model S1-1 for each wind farm group based on numerical weather forecast. Using mat1ab, based on the historical wind speed data of each unit in the wind farm, first use the CalinskiHarabasz index to determine the optimal classification number K, and then use the kmeans algorithm to cluster to obtain K groups, with the unit closest to the cluster center as the central unit;

[0009] S1-2. Use the kernel density estimation method to establish the marginal probability density function of wind speed for each central unit;

[0010] S1-3. Based on the marginal probability distribution function of each cluster, different copula functions are established to describe the correlation between the wind speed of the central unit of each cluster and the numerical weather forecast of the wind farm. The maximum likelihood estimation method is used to obtain the relevant parameters. The joint copula function of wind speed is expressed as follows:

[0011]

[0012] In the formula, is the numerical weather forecast wind speed, v i is the minute-level wind speed of each central unit, θ is the parameter of the copula function, u v =F(v),

[0013] S1-4. Use the AIC information criterion to evaluate the fitting quality of different copula functions and select the appropriate copula function. The calculation formula is as follows:

[0014] A=2k-2ln(L)

[0015] Where k is the number of parameters in the Copula function; L is the value of the maximum likelihood function of the Copula function;

[0016] S1-5. Calculate the conditional probability model of the minute-level wind speed of each central unit in the cluster

[0017]

[0018] In the formula,

[0019] 2. Construct a typical wind speed forecast scenario

[0020] S2-1. Sampling

[0021] The probability distribution function is divided into N intervals using the Latin hypercube sampling method. The midpoint of each interval is selected as the sampling point, and the sample value is mapped to the wind speed sampling value v through the inverse function. tn , v tn It is expressed as follows:

[0022]

[0023] The sample values ​​obtained by the above method are randomly arranged into a row of the matrix, and the sampling is repeated T times to form a T*N order initial sampling matrix V, which is expressed as

[0024]

[0025] In the formula, T is the sample time series dimension;

[0026] S2-2. Sorting

[0027] The Cholesky decomposition method is used to reconstruct the sampling matrix time series to obtain N wind speed scene sets;

[0028] S2-3, wind speed scene reduction

[0029] The rapid previous generation elimination method is used to reduce the wind speed scenarios to S typical scenarios with larger probabilities representing the original distribution and their corresponding probabilities p s , and verify the coverage of the generated scenario on the measured wind speed value P B :

[0030]

[0031] Where Bt is a binary variable. When the wind speed measured value in the current period is between the maximum and minimum values ​​of the generated scenario, it takes 1, otherwise it takes 0. T To generate a time series of scenes.

[0032] 3. Wind farm frequency regulation energy assessment based on typical wind speed forecast scenarios

[0033] The following formula is used to calculate the available frequency modulation energy for each typical wind speed scenario of each fleet:

[0034]

[0035] In the formula, ω max is the maximum speed of the fan rotor, ω1 is the initial speed when entering the constant speed zone, v n is the rated wind speed, v1 is the lower limit of the wind speed in the constant speed zone, v is the current wind speed, and v0 is the minimum grid-connected wind speed limit;

[0036] The frequency modulation energy in each scenario is calculated by probability weighting, which is the expected value of the available frequency modulation energy of each wind turbine. Then the energy results of each wind turbine are superimposed to obtain the evaluation results of the available frequency modulation energy of the wind farm as follows:

[0037]

[0038] Where M is the number of wind turbines in the wind farm, and S is the number of wind speed prediction scenarios.

[0039] Furthermore, the specific steps of step S2-2 are:

[0040] First, construct an approximately orthogonal permutation matrix L, where each row element represents the arrangement position of the corresponding row element of the initial sampling matrix V, and the correlation coefficient matrix between each row uses a T*T Pearson correlation coefficient matrix ρ LTo measure, it is expressed as follows:

[0041] ρ L ={ρ i，k i=1, 2,...T; j=1, 2,...T}

[0042] Where ρ i,j is the element in the i-th row and j-th column of the autocorrelation matrix, expressed as:

[0043]

[0044] By decomposing ρ using the Cholesky decomposition method, we can obtain a real lower triangular matrix D, and combine it with the permutation matrix L to construct a matrix W with less correlation:

[0045] ρ L =DD T

[0046] W=D -1 .L

[0047] Target correlation coefficient matrix ρ based on historical wind speed data time series obj , perform Cholesky decomposition on it, and further combine it with matrix W, so that the random permutation of the permutation matrix L produces corresponding correlation, and obtain the matrix W with target correlation n ;

[0048] ρ obj =PP T

[0049] W n =PW

[0050] Where P is a lower triangular matrix;

[0051] The permutation matrix W obtained by the above method n The elements in the initial sampling matrix V are rearranged to obtain a new sample matrix Xn, that is, N wind speed scene sets containing T time series are obtained.

[0052] Furthermore, the wind speed marginal probability density function of each central unit is established using the following formula:

[0053] In the formula, K() is the kernel function, h is the window width or sliding parameter, and X i is the sample point data.

[0054] The beneficial effect of the present invention is that the wind farm proposed in this patent can be realized by using the frequency modulation energy assessment algorithm to analyze the differences in the spatiotemporal distribution of wind speeds and generate typical wind speed prediction scenarios using the probability prediction distribution function, which can effectively reduce estimation errors. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 This is the wind turbine distribution map of the wind farm;

[0056] Figure 2 is the cluster number map;

[0057] Figure 3 This is a typical scene diagram of wind speed;

[0058] Figure 4 This is a typical scene diagram without considering the time series correlation of wind speed;

[0059] Figure 5 It is the frequency modulation energy distribution diagram of each wind turbine in the wind farm;

[0060] Figure 6 This is the energy fluctuation curve of wind power frequency regulation under variable wind speed. DETAILED DESCRIPTION

[0061] In order to grasp the frequency regulation capability of wind farms in real time and realize the frequency regulation parameter setting from wind farm layer to single machine layer that matches the frequency regulation capability, it is necessary to evaluate the frequency regulation energy of wind farms. It is known that the available frequency regulation energy of wind turbines is affected by the wind speed of the input wind turbines. In wind farms, affected by the distribution of wind turbines and terrain, there are differences in the spatial distribution of wind speeds of various wind turbines. The use of a single numerical weather forecast or average wind speed data of a wind tower cannot reflect the actual wind speed distribution of the wind farm. In addition, the frequency regulation scale of wind turbines is at the minute level, and the time scale of the average wind speed of numerical weather forecasts is generally 15 minutes, which cannot reflect the minute-level wind speed fluctuation characteristics. This spatiotemporal distribution characteristic of wind speed will inevitably cause differences in the available frequency regulation energy of various wind turbines. Therefore, this difference must be considered when evaluating the frequency regulation energy of wind farms.

[0062] This patent analyzes the spatiotemporal correlation of minute-level wind speed in a wind farm, and proposes a method for evaluating the frequency modulation energy of a wind farm in a minute-level wind speed prediction scenario. First, a probabilistic prediction model of minute-level wind speed for each wind farm group based on numerical weather forecast wind speed is constructed using the copula function, in order to obtain the distribution characteristics of minute-level wind speed. Further, Latin hypercube sampling that considers the autocorrelation of wind speed time series is used to generate a typical wind speed prediction scenario. Finally, the frequency modulation energy evaluation of the wind farm is realized based on the prediction scenario.

[0063] 1. Construct a minute-level wind speed probability prediction model for each wind farm group based on numerical weather forecast

[0064] In order to obtain the spatiotemporal distribution characteristics of minute-level wind speed in a wind farm for more accurate frequency regulation energy assessment of the wind farm, this patent uses a Copula function to analyze the correlation between the numerical weather forecast of the wind farm and the minute-level wind speed of each wind turbine group, and constructs a conditional probability prediction model for the wind speed of each wind turbine group under a single numerical weather forecast wind speed.

[0065] According to Sklar's theorem, let the random variables x1, x2, ..., x n The marginal distribution functions of are F1(x1), F2(x2)..., F n (x n ), if F n (x n ) is continuous, then the joint distribution function F(x1, x2, ..., x n ) can be constructed by the following copula probability distribution function C():

[0066] F(x1, x2, ..., x n )=C(F1(x1), F2(x2),...,F n (x n )) (1)

[0067] Its joint probability density function is as shown in formula (2):

[0068]

[0069] Where c() is the n-dimensional Copula function F(x1, x2, ..., x n ), f i (x i ) is the marginal probability density function of the random variable.

[0070] The marginal distribution of each variable is determined by using the kernel density estimation method to determine the sample. This method can adapt to various types of data, does not require assumptions, and can be more flexibly applied to practical problems. The kernel density estimation method is as follows:

[0071]

[0072] In the formula, K() is the kernel function, h is the window width or sliding parameter, and X i is the sample point data. The marginal density distribution function is obtained from formula (3): f(v), then integrate to obtain the marginal distribution function of the variable F(v).

[0073] The steps for establishing the wind speed conditional probability prediction model for each fleet include:

[0074] S1-1. Using Matlab, based on the historical wind speed data of each unit in the wind farm, the Calinski-Harabasz index is used to determine the optimal classification number K, and then the kmeans algorithm is used for clustering to obtain K groups, with the unit closest to the cluster center as the central unit;

[0075] S1-2, according to formula (3), use the kernel density estimation method to establish the wind speed marginal probability density function of each central unit;

[0076] S1-3. Based on the marginal probability distribution function of each cluster, different copula functions are established by equations (1) and (2) to describe the correlation between the wind speed of the central unit of each cluster and the numerical weather forecast of the wind farm. The maximum likelihood estimation method is used to obtain the relevant parameters. The joint copula function of wind speed is expressed as follows:

[0077]

[0078] In the formula, is the numerical weather forecast wind speed, v i is the minute-level wind speed of each central unit, θ is the parameter of the copula function, u v =F(v),

[0079] S1-4. Use the AIC information criterion to evaluate the fitting quality of different copula functions and select the appropriate copula function. The calculation formula is as follows:

[0080] A=2k-2ln(L)

[0081] Where k is the number of parameters in the Copula function; L is the value of the maximum likelihood function of the Copula function.

[0082] S1-5. Calculate the conditional probability model of the minute-level wind speed of each central unit in the cluster

[0083]

[0084] In the formula,

[0085] 2. Construct a typical wind speed forecast scenario

[0086] The minute-level wind speed probability prediction model obtained above cannot be directly used for wind farm frequency regulation energy assessment. Therefore, a wind farm frequency regulation energy assessment method based on typical wind speed prediction scenarios is proposed. In the formula, the generation of typical wind speed prediction scenarios is achieved by Latin hypercube sampling and scenario reduction method with fast previous generation elimination considering its time series autocorrelation. The generated typical scenarios better retain the fluctuation characteristics of the original scenarios. This method can realize the frequency regulation energy assessment of wind farms in a specific time period in the future. The specific steps are as follows:

[0087] S2-1. Sampling

[0088] Given the minute-level wind speed conditional probability prediction model F(v) for each wind farm, the probability distribution function is divided into N intervals using the Latin hypercube sampling method. The midpoint of each interval is selected as the sampling point, and the sample value is mapped to the wind speed sampling value v through the inverse function. tn , v tn It is expressed as follows:

[0089]

[0090] The sample values ​​obtained by the above method are randomly arranged into a row of the matrix, and the sampling is repeated T times to form a T*N order initial sampling matrix V, which is expressed as

[0091]

[0092] Where T is the sample time series dimension, T=10.

[0093] S2-2. Sorting Since the initial sampling matrix V cannot reflect the time series correlation of wind speed, the Cholesky decomposition method is used to reconstruct the sampling matrix time series. First, an approximately orthogonal permutation matrix L is constructed, in which each row element represents the arrangement position of the corresponding row element of the initial sampling matrix V. The correlation coefficient matrix between each row uses a T*T Pearson correlation coefficient matrix ρ L To measure, it is expressed as follows:

[0094] ρ L ={ρ i，j i=1, 2,...T; j=1, 2,...T}

[0095] Where ρ i,j is the element in the i-th row and j-th column of the autocorrelation matrix, expressed as:

[0096]

[0097] By decomposing ρ using the Cholesky decomposition method, we can obtain a real lower triangular matrix D, and combine it with the permutation matrix L to construct a matrix W with less correlation:

[0098] ρ L =DD T

[0099] W=D -1 .L

[0100] Target correlation coefficient matrix ρ based on historical wind speed data time series obj , perform Cholesky decomposition on it, and further combine it with matrix W, so that the random permutation of the permutation matrix L produces corresponding correlation, and obtain the matrix W with target correlation n

[0101] ρ obj =PP T

[0102] W n =PW

[0103] Where P is a lower triangular matrix.

[0104] The permutation matrix W obtained by the above method n The elements in the initial sampling matrix V are rearranged to obtain a new sample matrix Xn, that is, N wind speed scene sets containing T time series are obtained.

[0105] S2-3, wind speed scene reduction

[0106] The above process generates N scenarios that can be used to approximate the wind speed fluctuation process. To improve the calculation efficiency, the present invention uses the fast previous generation elimination method to reduce the wind speed scenarios to S typical scenarios with larger probabilities representing the original distribution and their corresponding probabilities p s , and verify the coverage of the generated scenario on the measured wind speed value P B :

[0107]

[0108] In the formula, Bt is a binary variable. When the measured wind speed value in the current period is between the maximum and minimum values ​​of the generated scenario, it takes 1, otherwise it takes 0. That is, the generated scenario covers the measured value of the current period, and the effective value of the generated scenario is 1. B The value range is [0, 1]. The larger the calculated result is, the more effectively the generated scene can take into account its uncertainty.

[0109] 3. Wind farm frequency regulation energy assessment based on typical wind speed forecast scenarios

[0110] The energy source of wind turbines participating in frequency regulation is the kinetic energy of the blade rotor. However, due to the speed limit, it cannot be fully used to provide frequency support. It is known that the actual speed of the wind turbine is w r Since the kinetic energy of the blade rotor cannot be fully released, there is a minimum speed limit w min , the wind turbine blade rotor kinetic energy ΔE is expressed by the available rotor kinetic energy, the available rotor speed change and the inherent moment of inertia J of the wind turbine. k0 :

[0111] ΔE k0 =0.5J(w r 2 -w min 2 ) (4)

[0112] Further, w rExpressed as a function of wind speed, the wind turbine blade rotor kinetic energy ΔE k0 Can represent:

[0113]

[0114] In the formula, λ opt is the tip speed ratio, r is the fan blade radius, ω max is the maximum speed of the fan rotor, ω1 is the initial speed when entering the constant speed zone, v n is the rated wind speed, v1 is the lower limit of the wind speed in the constant speed zone, v is the current wind speed, v0 is the minimum grid-connected wind speed limit, ω min The minimum speed of the fan rotor.

[0115] Combined with the wake model, the wind speed distribution inside each group is obtained, so as to obtain the frequency regulation energy of each unit. Formula (5) is used to calculate the available frequency regulation energy of each group in each typical wind speed scenario. The frequency regulation energy in S scenarios is probability-weighted and calculated to obtain the expected value of the available frequency regulation energy of each wind turbine. Then, the energy results of each wind turbine are superimposed to obtain the evaluation result of the available frequency regulation energy of the wind farm as follows:

[0116]

[0117] In the formula, M is the number of wind turbines in the wind farm, and S is the number of wind speed prediction scenarios. It should be pointed out that based on the above analysis, the available frequency modulation energy of wind turbines is nonlinearly related to wind speed. Under low and medium wind speeds, there will be no available frequency modulation energy due to the loss of mechanical energy, and the available frequency modulation energy remains unchanged under high wind speeds. Therefore, the expected value of the available frequency modulation energy of the wind turbine based on the typical wind speed prediction scenario can effectively estimate the frequency modulation energy of the wind turbine. Repeat the above steps to evaluate the available frequency modulation energy of each wind farm in the power grid, and the total available wind power frequency modulation energy of the entire network can be obtained, which provides a basis for subsequent parameter optimization.

[0118] 4. Wind farm frequency regulation energy assessment and verification

[0119] 4.1. To verify the feasibility of the wind farm frequency modulation energy assessment method of this patent, a wind farm is taken as an example for analysis. The wind farm includes 20 wind turbines, model Vestas V52-852kW, and the wind turbines are arranged as follows: Figure 1 The wind farm monitoring system can measure the wind speed, wind direction and wind power of each wind turbine.

[0120] 4.2 Unit clustering results

[0121] The minute-level measured historical wind speed data of 20 wind turbines in a certain year and month of the wind farm were selected for analysis, and the k-means algorithm was used to cluster the wind turbines. First, the Calinski Harabasz index was used to determine the optimal number of cluster groups, such as Figure 2It can be seen that the best effect is achieved when the number of clusters is 3. After the number of clusters is determined, the wind speed variation curves (time scale) of the 20 wind turbines in the wind farm are clustered, and the results are shown in Table 1.

[0122] Table 1 Clustering results of wind turbine groups

[0123]

[0124] 4.3. Optimal copula function verification

[0125] Based on the historical wind speed data measured by the SCADA system of the central unit and the historical wind speed data of the wind farm numerical weather forecast, the non-parametric kernel density estimation method is used to determine the marginal distribution. The t-copula, gauss-copula, Clayton-Copula, Frank-Copula, and Gumbel-Copula functions are further used to construct the copula function of the wind farm numerical weather forecast. The parameter values ​​are determined by the maximum likelihood method, and the AIC (Akaike Information Criterion) criterion is compared for optimal selection. The results of cluster 1 are shown in Table 2.

[0126] Table 2 Copula parameters and evaluation indicators

[0127]

[0128] The smaller the AIC value, the better the fitting effect of the corresponding copula function. It can be seen from the results shown in Table 3 that the Frank-Copula function in the formula has a better effect in describing the correlation of variables. The same method is used to obtain the joint copula function of the central unit of other fleets and the wind speed of the wind farm numerical weather forecast, as shown in Table 3.

[0129] Table 3 Optimal copula functions and parameters for each cluster

[0130]

[0131] 4.4. Generation of typical wind speed scenarios

[0132] Based on the aforementioned typical wind speed prediction scenario generation method, the numerical weather forecast wind speed is known to be 9.6m / s. The spatiotemporal characteristics of the wind speed of a wind turbine in the wind farm are analyzed to generate a set of 1000 minute-level wind speed scenarios. The forward reduction method is then used to reduce the typical scenarios to 6, and the probability of each scenario is calculated. The results are as follows: Figure 3 shown.

[0133] Figure 4 This is a typical wind speed scenario generated without considering the wind speed time series correlation. Figure 3It can be seen that the wind speed fluctuation of the scene generation method proposed in this patent is relatively stable compared with the scene generation method that does not consider the wind speed time series correlation. The quality index P of the scene generated by this patent is B The P value is 0.9583 without considering the timing correlation. B It is 0.3561, which shows that the patented method can better reflect the actual fluctuation characteristics of wind speed.

[0134] 4.5. Verification of wind farm frequency regulation energy assessment algorithm

[0135] To further verify the feasibility of the wind farm frequency modulation energy assessment method of this patent, the wind farm frequency modulation energy assessment method of this patent is used. It is known that the numerical weather forecast wind speed is 9.6m / s. Based on the typical wind speed prediction scenario, the frequency modulation energy distribution of the wind turbines in the wind farm is obtained as follows: Figure 5 As shown, it can be observed that due to the difference in the spatial distribution of wind speed in the wind farm, the frequency modulation energy of each wind turbine is different. The calculation result of the energy evaluation method proposed in this patent is 46.43MJ. The result of the frequency modulation energy available in the entire wind farm using the unified wind speed algorithm is 68.39MJ. Compared with this, the frequency modulation energy evaluation algorithm of this patent is reduced by 32%. When compared with the actual wind speed of each wind turbine at the current moment, the frequency modulation energy evaluation result is 71.48MJ, with an error of 4.6%. This shows the practical application value of the wind farm frequency modulation energy evaluation method proposed in this patent, which can more truly reflect the actual frequency modulation capability of the wind farm.

[0136] Based on the 10-minute numerical weather forecast wind speed data of a wind farm on a certain day in the summer of a certain year, a rolling assessment of the available frequency modulation energy of the wind farm is performed every 1 minute to obtain a wind farm frequency modulation energy assessment curve that changes synchronously with the real-time wind speed, such as Figure 6 This patented algorithm can directly reflect the time-varying fluctuation of the available frequency modulation energy of the wind farm with the wind speed.

[0137] In response to the problem of lack of accurate estimation of wind farm frequency modulation energy, the wind farm frequency modulation energy assessment algorithm proposed in this patent can be achieved by analyzing the differences in the spatiotemporal distribution of wind speed and generating typical wind speed prediction scenarios using the probability prediction distribution function. Example analysis verifies that this method can effectively reduce estimation errors.

Claims

1. A frequency modulation energy assessment method considering the temporal and spatial distribution differences of wind speed in a wind farm, characterized in that: The following steps are involved: S1. Construct a probability prediction model for minute-level wind speed of each wind farm group based on numerical weather forecast S1-1. Based on the historical wind speed data of each unit in the wind farm, the Calinski-Harabasz index is first used to determine the optimal classification number K, and then the kmeans algorithm is used for clustering to obtain K groups, with the unit closest to the cluster center as the central unit; S1-2. Use the kernel density estimation method to establish the marginal probability density function of wind speed for each central unit; S1-3. Based on the marginal probability density function of wind speed of each cluster, different copula functions are established to describe the correlation between the wind speed of the central unit of each cluster and the numerical weather forecast of the wind farm. The maximum likelihood estimation method is used to obtain the relevant parameters. The joint copula function of wind speed is expressed as follows: In the formula, is the numerical weather forecast wind speed, v i is the minute-level wind speed of each central unit, θ is the parameter of the copula function, S1-4. Use the AIC information criterion to evaluate the fitting quality of different copula functions and select the appropriate copula function. The calculation formula is as follows: A=2k-21n(L) Where k is the number of parameters in the Copula function; L is the value of the maximum likelihood function of the Copula function; S1-5. Calculate the wind speed probability prediction model for the minute-level wind speed of each central unit in the cluster In the formula, S2. Construct a typical wind speed forecast scenario S2-1. Sampling The wind speed probability prediction model is divided into N intervals using the Latin hypercube sampling method. The midpoint of each interval is selected as the sampling point, and the sample value is mapped to the wind speed sampling value v through the inverse function. tn , v tn It is expressed as follows: n=1,2,...,N;t=1,2,...,T; The sample values ​​obtained by the above method are randomly arranged into a row of the matrix, and the sampling is repeated T times to form a T*N order initial sampling matrix V, which is expressed as In the formula, T is the sample time series dimension; S2-2. Sorting The Cholesky decomposition method is used to reconstruct the sampling matrix time series to obtain N wind speed scene sets; S2-3, wind speed scene reduction The rapid previous generation elimination method is used to reduce the wind speed scenarios to S typical scenarios with larger probabilities representing the original distribution and their corresponding probabilities p s , and verify the coverage of the generated scenario on the measured wind speed value P B : Where Bt is a binary variable. When the wind speed measured value in the current period is between the maximum and minimum values ​​of the generated scenario, it takes 1, otherwise it takes 0. T To generate a time series of scenarios; S3. Wind farm frequency regulation energy assessment based on typical wind speed forecast scenarios The following formula is used to calculate the available frequency modulation energy for each typical wind speed scenario of each fleet: In the formula, ω max is the maximum speed of the fan rotor, ω1 is the initial speed when entering the constant speed zone, v n is the rated wind speed, v1 is the lower limit of the wind speed in the constant speed zone, v is the current wind speed, and v0 is the minimum grid-connected wind speed limit; The frequency modulation energy in each scenario is calculated by probability weighting, which is the expected value of the available frequency modulation energy of each wind turbine. Then the energy results of each wind turbine are superimposed to obtain the evaluation results of the available frequency modulation energy of the wind farm as follows: In the formula, M is the number of wind turbines in the wind farm, and S is the number of wind speed prediction scenarios; The specific steps of step S2-2 are: First, construct an approximately orthogonal permutation matrix L, where each row element represents the arrangement position of the corresponding row element of the initial sampling matrix V, and the correlation coefficient matrix between each row uses a T*T Pearson correlation coefficient matrix ρ L To measure, it is expressed as follows: r L ={ρ i，j i=1,2,...T;j=1,2,...T} In the formula, ρ i，j is the element in the i-th row and j-th column of the autocorrelation matrix, expressed as: The Cholesky decomposition method is used to decompose ρ to obtain a real lower triangular matrix D, and then a matrix W with less correlation is constructed by combining the permutation matrix L: r L =DD T W=D -1 .L Target correlation coefficient matrix ρ based on historical wind speed data time series obj , perform Cholesky decomposition on it, and further combine it with matrix W, so that the random permutation of the permutation matrix L produces corresponding correlation, and obtain the matrix W with target correlation n ; r obj =PP T W n =P.W Where P is a lower triangular matrix; The permutation matrix W obtained by the above method n Rearrange the elements in the initial sampling matrix V to obtain a new sample matrix Xn, that is, obtain N wind speed scene sets containing T time series; The marginal probability density function of wind speed of each central unit is established using the following formula: In the formula, K() is the kernel function, h is the window width or sliding parameter, and X i is the sample point data.

Images