CN114583767A - Data-driven wind power plant frequency modulation response characteristic modeling method and system - Google Patents

Abstract

The invention relates to a data-driven wind power plant frequency modulation response characteristic modeling method and system, wherein the method comprises the following steps: analyzing and preprocessing the measured data of the wind power plant frequency modulation response characteristic based on a step response dynamic performance index solving algorithm to obtain processed data under each working condition; establishing a transfer function model for each working condition according to the processed data, and measuring a gap value between the models by using a gap measurement method to determine a working condition area represented by a nonlinear autoregressive neural network model; and merging the frequency modulation data of the working condition according to the gap value, and training the nonlinear autoregressive neural network model according to the merged data to obtain the trained nonlinear autoregressive neural network model. According to the wind power plant frequency modulation response data evaluation method, the actual frequency modulation response data of the wind power plant are utilized, a modeling scheme capable of well representing the wind power plant grid-connected point frequency modulation response characteristics under different working conditions is designed, and the accuracy and the frequency modulation effect of the wind power plant frequency response evaluation can be improved.

Description

A data-driven modeling method and system for frequency modulation response characteristics of wind farms

technical field

The invention relates to the technical field of frequency modulation response characteristics of wind farms, in particular to a data-driven method and system for modeling frequency modulation response characteristics of wind farms.

Background technique

In the research of frequency stability in wind farms, an important issue is to establish a model of the frequency modulation response characteristics of the wind farm. Generally speaking, there are two modeling methods, one is component-based modeling, and the other is data-based modeling. mold. Component-based modeling often requires the parameters of the internal components of the wind turbine and the dynamic mathematical model during operation. In fact, due to the existence of the confidentiality agreement, the wind turbine manufacturer cannot provide detailed wind turbine information, which makes the component modeling have certain limitations. sex. The modeling based on wind farm operation data focuses on the input and output of the research object, avoiding the physical description of the wind turbine components and obtaining the corresponding parameters, and the model can also achieve satisfactory accuracy. Therefore, data-driven modeling methods are more and more applied to the modeling process of wind turbines and wind farms.

The nonlinear autoregressive neural network model with external input can well characterize the nonlinear time series system. The frequency modulation response of wind farm grid connection point is a nonlinear process of time series. According to the existing relevant literature, the most similar to the present invention is the existing wind farm frequency modulation response characteristic modeling method, the transfer function method is often based on components, and considers many single working conditions, nonlinear autoregressive neural network It is mostly used for wind speed and power prediction of wind farms, and there are very few applications of wind farm frequency modulation modeling. We know that when the frequency regulation command from the grid connection point is issued, the wind farm is not necessarily in the maximum power operating state. If the frequency regulation is performed in other active power operating states, a single model may not be able to characterize it. This may lead to inaccurate evaluation of the frequency response of the wind farm based on a single model, and unsatisfactory frequency modulation effect.

SUMMARY OF THE INVENTION

In order to overcome the deficiencies of the prior art, the purpose of the present invention is to provide a data-driven method and system for modeling the frequency modulation response characteristics of a wind farm.

For achieving the above object, the present invention provides following scheme:

A data-driven modeling method for frequency modulation response characteristics of wind farms, comprising:

Based on the step response dynamic performance index solution algorithm, the measured data of the frequency modulation response characteristics of the wind farm are analyzed and preprocessed, and the processed data under each working condition are obtained;

Establish a transfer function model for each working condition according to the processed data, and measure the gap value between the models by using the gap measurement method to determine the working condition area represented by the nonlinear autoregressive neural network model;

The frequency modulation data of the working conditions are combined according to the gap value, and the nonlinear autoregressive neural network model is trained according to the combined data to obtain a trained nonlinear autoregressive neural network model.

Preferably, the step-response-based dynamic performance index solution algorithm analyzes and preprocesses the measured data of the frequency modulation response characteristics of the wind farm to obtain processed data under various operating conditions, including:

Divide the measured data of the frequency modulation response characteristic of the wind farm according to the initial active power during the frequency modulation of the wind farm to obtain a plurality of working conditions;

Obtaining the power curve of the data set of the measured data of the frequency modulation response characteristic of the wind farm, and calculating the upper peak set and the lower peak set of each data point of the power curve;

calculating the maximum value of the upper peak set and the minimum value of the lower peak set;

based on the data points in the data set, calculating an upper error band and a lower error band;

Judging whether the upper bound of the error band is greater than or equal to the maximum value of the upper peak set, and whether the lower bound of the error band is greater than or equal to the minimum value of the lower peak set, if the judgment results are all yes, according to the sampling interval determine the adjustment time;

Determine whether the measured data of the frequency modulation response characteristic of the wind farm is a frequency step underdisturbance, and if so, determine the overshoot amount according to the maximum value of the upper peak set;

Rise time or fall time is determined from the data set.

Preferably, establishing a transfer function model for each operating condition according to the processed data includes:

According to the frequency variation in the processed data under each working condition as the input, and the power variation as the output, an initial function model is constructed;

Set the numerator order and denominator order of the transfer function;

Perform model identification on the initial transfer function model, and adjust the numerator order and the denominator order to obtain the optimal transfer function model.

Preferably, the use of the gap measurement method to measure the gap value between the models to determine the working condition area represented by the nonlinear autoregressive neural network model includes:

Determine the gap measurement formula according to the orthogonal projection matrices of the different transfer function models;

The gap value is calculated according to the gap metric formula; the gap value is used to determine the operating condition area.

Preferably, the merging of the frequency modulation data of the working conditions according to the gap value includes:

Calculate the distance between the gap value and 0 between the transfer function models of different working conditions to obtain the first distance;

Calculate the distance between the gap value and 1 between the transfer function models of different working conditions to obtain the second distance;

Determine whether the first distance is less than or equal to the second distance, and if so, combine the FM data of the two working conditions according to the time series to obtain the combined data; if not, select the FM data of any working condition and The existing FM data is time-series merged to obtain merged data.

Preferably, the nonlinear autoregressive neural network model is trained according to the combined data to obtain a trained nonlinear autoregressive neural network model, including:

Build the initial neural network;

Based on the Levenberg-Marquark algorithm, the initial neural network is trained according to the combined data;

The trained neural network is tested by using the frequency modulation data of each working condition, and the nonlinear autoregressive neural network model is determined according to the test results.

A data-driven wind farm frequency modulation response characteristic modeling system, comprising:

The data processing module is used to analyze and preprocess the measured data of the frequency modulation response characteristics of the wind farm based on the step response dynamic performance index solution algorithm, and obtain the processed data under each working condition;

The transfer function building module is used to establish a transfer function model for each working condition according to the processed data, and use the gap measurement method to measure the gap value between the models to determine the working condition represented by the nonlinear autoregressive neural network model area;

A neural network modeling module is used to merge the frequency modulation data of the working condition according to the gap value, and train the nonlinear autoregressive neural network model according to the merged data to obtain a trained nonlinear autoregressive neural network. network model.

Preferably, the data processing module specifically includes:

A working condition dividing unit, configured to divide the measured data of the frequency modulation response characteristic of the wind farm according to the initial active power during frequency modulation of the wind farm to obtain a plurality of working conditions;

a peak value calculation unit, configured to obtain the power curve of the data set of the measured data of the frequency modulation response characteristic of the wind farm, and calculate the upper peak value set and the lower peak value set of each data point of the power curve;

a maximum value calculation unit, configured to calculate the maximum value of the upper peak set and the minimum value of the lower peak set;

an upper and lower bound calculation unit, configured to calculate the upper bound of the error band and the lower bound of the error band based on the data points in the data set;

A first judging unit, configured to judge whether the upper bound of the error band is greater than or equal to the maximum value of the upper peak set, and whether the lower bound of the error band is greater than or equal to the minimum value of the lower peak set, if the judgment results are all If yes, determine the adjustment time according to the sampling interval;

a second judging unit, configured to judge whether the measured data of the frequency modulation response characteristic of the wind farm is a frequency step underdisturbance, and if so, determine the overshoot amount according to the maximum value of the upper peak set;

A time determination unit, configured to determine a rise time or a fall time according to the data set.

Preferably, the transfer function building module specifically includes:

an initial model building unit, configured to construct an initial function model according to the frequency variation in the processed data under each working condition as the input and the power variation as the output;

Order setting unit, used to set the numerator order and denominator order of the transfer function;

A model determining unit, configured to perform model identification on the initial transfer function model, and adjust the numerator order and the denominator order to obtain the optimal transfer function model.

Preferably, the transfer function building module specifically includes:

a formula determination unit, used for determining a gap measurement formula according to the orthogonal projection matrices of the different transfer function models;

A clearance calculation unit, configured to calculate the clearance value according to the clearance measurement formula; the clearance value is used to determine the working condition area.

According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

The present invention provides a data-driven method and system for modeling frequency modulation response characteristics of wind farms. The method includes: analyzing and preprocessing the measured data of frequency modulation response characteristics of wind farms based on a step response dynamic performance index solution algorithm to obtain each The processed data under the working conditions; the transfer function model is established for each working condition according to the processed data, and the gap value between the models is measured by the gap measurement method to determine the working conditions represented by the nonlinear autoregressive neural network model. The frequency modulation data of the working condition is combined according to the gap value, and the nonlinear autoregressive neural network model is trained according to the combined data to obtain a trained nonlinear autoregressive neural network model. The invention uses the actual frequency modulation response data of the wind farm to design a modeling scheme that can well characterize the frequency modulation response characteristics of the grid connection point of the wind farm under different working conditions, and can improve the accuracy of the frequency response evaluation of the wind farm and the frequency modulation effect.

Description of drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings required in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some of the present invention. In the embodiments, for those of ordinary skill in the art, other drawings can also be obtained according to these drawings without creative labor.

1 is a method flowchart of a method for modeling a frequency modulation response characteristic of a wind farm in an embodiment provided by the present invention;

2 is a comparison diagram of the output of the transfer function model in the embodiment provided by the present invention and the first output of actual data;

3 is a comparison diagram of the output of the transfer function model in the embodiment provided by the present invention and the second output of actual data;

Fig. 4 is the output of the transfer function model in the embodiment provided by the present invention and the third output comparison diagram of actual data;

Fig. 5 is the output of the transfer function model in the embodiment provided by the present invention and the fourth output comparison diagram of actual data;

6 is a structural diagram of a nonlinear autoregressive neural network with external input in an embodiment provided by the present invention;

7 is a first result diagram of a nonlinear autoregressive neural network model trained with multi-operating condition data in an embodiment provided by the present invention tested by frequency interference data of a single operating condition;

8 is a second result diagram of a nonlinear autoregressive neural network model trained with multi-operating condition data in an embodiment provided by the present invention tested by frequency interference data of a single operating condition;

9 is a third result diagram of a nonlinear autoregressive neural network model trained on multi-operating condition data in an embodiment provided by the present invention tested by frequency interference data of a single operating condition;

FIG. 10 is a fourth result diagram of a nonlinear autoregressive neural network model trained with data of multiple operating conditions tested by frequency interference data of a single operating condition in the embodiment provided by the present invention.

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

Reference herein to an "embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the present application. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment, nor a separate or alternative embodiment that is mutually exclusive of other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

The terms "first", "second", "third" and "fourth" in the description and claims of the present application and the drawings are used to distinguish different objects, rather than to describe a specific order . Furthermore, the terms "comprising" and "having", and any variations thereof, are intended to cover non-exclusive inclusion. For example, including a series of steps, processes, methods, etc. is not limited to the listed steps, but optionally also includes unlisted steps, or optionally also includes inherent to these processes, methods, products or devices. other steps.

The purpose of the present invention is to provide a data-driven wind farm frequency modulation response characteristic modeling method and system, which can improve the accuracy of wind farm frequency response evaluation and the frequency modulation effect.

In order to make the above objects, features and advantages of the present invention more clearly understood, the present invention will be described in further detail below with reference to the accompanying drawings and specific embodiments.

FIG. 1 is a method flowchart of a method for modeling frequency modulation response characteristics of a wind farm in an embodiment provided by the present invention. As shown in FIG. 1 , the present invention provides a data-driven method for modeling frequency modulation response characteristics of wind farms, including:

Step 100: analyze and preprocess the measured data of the frequency modulation response characteristic of the wind farm based on the step response dynamic performance index solution algorithm, and obtain the processed data under each working condition;

Step 200: establishing a transfer function model for each operating condition according to the processed data, and measuring the gap value between the models by using the gap measurement method to determine the operating condition area represented by the nonlinear autoregressive neural network model;

Step 300: Combine the frequency modulation data of the working conditions according to the gap value, and train the nonlinear autoregressive neural network model according to the combined data to obtain a trained nonlinear autoregressive neural network model.

Preferably, the step 100 includes:

Divide the measured data of the frequency modulation response characteristic of the wind farm according to the initial active power during the frequency modulation of the wind farm to obtain a plurality of working conditions;

Obtaining the power curve of the data set of the measured data of the frequency modulation response characteristic of the wind farm, and calculating the upper peak set and the lower peak set of each data point of the power curve;

calculating the maximum value of the upper peak set and the minimum value of the lower peak set;

based on the data points in the data set, calculating an upper error band and a lower error band;

Judging whether the upper bound of the error band is greater than or equal to the maximum value of the upper peak set, and whether the lower bound of the error band is greater than or equal to the minimum value of the lower peak set, if the judgment results are all yes, according to the sampling interval determine the adjustment time;

Determine whether the measured data of the frequency modulation response characteristic of the wind farm is a frequency step underdisturbance, and if so, determine the overshoot amount according to the maximum value of the upper peak set;

Rise time or fall time is determined from the data set.

The first step in this embodiment is to analyze and preprocess the measured data of the frequency modulation response characteristics of the wind farm. In practice, the indicators describing the dynamic performance of a system are generally the rise (fall) time _tr , the adjustment time _ts , and the overrun time. The adjustment amount is expressed as δ%. where t _r is the time it takes for the response value to reach the final value p _∞ from zero, t _s is the time it takes for the response value to change from zero to the time it is stabilized within the artificially specified error band, and δ% is obtained from the following formula

where p _ex is the maximum offset of the response value within the adjustment time.

Since the frequency step response of the wind farm cannot be described by a simple first-order or second-order mathematical model, the dynamic performance index of the response cannot be accurately obtained by analytical expressions. In order to solve this problem, a step based on the measured data is proposed The response dynamic performance index solution algorithm, for the convenience of understanding, the algorithm is embodied in the form of pseudo code, the steps of the algorithm are as follows:

The data in the adjustment time in the index is selected for modeling, and the data is processed as follows: take the time t ₀ of the frequency disturbance of the wind farm as the initial time, and select the active power p ₀ corresponding to the time t ₀ , then the wind farm frequency adjustment process The active power change corresponding to each moment is

_Δpi = _pi -p ₀ (2)

where i=0,1,...,n,n is the number of data with complete frequency step events. p _i is the active power value of the wind farm in the measured data corresponding to time t _i . Taking 50Hz as the grid-connected standard frequency, the frequency deviation corresponding to each moment in the frequency step process is:

Δf _i =f _i -50 (3)

where f _i is the frequency value of the wind farm in the measured data corresponding to time t _i . For the convenience of calculation, Δf=(Δf ₀ ,Δf ₁ ,...,Δf _n ) ^T is defined, and Δp=(Δp ₀ ,Δp ₁ ,...,Δp _n ) ^T .

Further, taking the frequency up-disturbance as an example, the initial active power of the wind farm during frequency modulation is divided into different working conditions, and according to the measured data of the wind farm, it is specifically divided into, working condition 1: frequency step up-disturbance at 25% of the limited power. ; working condition 2: frequency step up disturbance under 50% power limit; working condition 3: frequency step up disturbance under unlimited power; working condition 4: compound frequency up disturbance; The frequency modulation response characteristic indexes of the wind farm under different working conditions are shown in Table 1. Table 1 is the frequency step up disturbance response characteristic index of the wind farm under various working conditions. According to the fall time in Table 1, the data required for modeling is intercepted , It is worth noting that working condition 4 is a composite frequency interference, not a typical step response, so the performance index has no practical value, so Table 1 does not give the index of working condition 4, and for working condition 4, Modeling data selects data within the full FM response time. After selecting the data required for modeling, preprocess the data according to the preprocessing method in the first step of the plan.

Table 1

Preferably, the step 200 includes:

According to the frequency variation in the processed data under each working condition as the input, and the power variation as the output, an initial function model is constructed;

Set the numerator order and denominator order of the transfer function;

Perform model identification on the initial transfer function model, and adjust the numerator order and the denominator order to obtain the optimal transfer function model.

Preferably, the step 200 further includes:

Determine the gap measurement formula according to the orthogonal projection matrices of the different transfer function models;

The gap value is calculated according to the gap metric formula; the gap value is used to determine the operating condition area.

Specifically, the second step in this embodiment is to establish a transfer function model for each working condition according to the processed data under each working condition, and use the gap measurement method to measure the gap value between the models, so as to determine the modeled working condition area, The specific method is as follows:

Taking the frequency change as the input and the corresponding power change as the output, the following transfer function model is given

where m and k represent the order of the numerator and denominator of the transfer function, and assuming k≥m, j=1, 2...n, represents the jth frequency fluctuation event, θ _j =[b _1,j ,...,b _{k, j} ,a _0,j ,…,am _,j ] ^T is a vector of discrete transfer function parameters.

Then (4) can be transformed into:

ΔP _j (z)(1+b _1,j z ^-1 +...+b _k,j z ^-k )=Δf _j (z)(a _0,j +a _{1,j z} ^-1 +...+am _,j z ^-m ) (5)

According to the properties of the z-transform, (5) can be directly transformed into the difference equation as follows:

ΔP _j (t)=-b _1,j ΔP _j (t-1)-…-b _k,j ΔP _j (tk)+a _0,j Δf _j (t)+…+am _,j Δf _j ( tm) (6)

Assume that the jth frequency fluctuation event has N measured sample points, and let

Y _j =[ΔP _j (t),...,ΔP _j (N)] ^T ,

Then equation (6) can be expressed by the above matrix in the following compact form:

Y _j = Ξ _j θ _j (7)

From this, we can use the linear least squares method to solve θ _j , namely

In the modeling process, it can be assumed that the order m and k of the numerator and denominator of the transfer function are known. From the properties of the transfer function, it can be known that the values of m and k can reflect the extreme value and the number of zero points of the output vector curve. In the process of analyzing the frequency regulation response characteristics of the wind farm, the present invention firstly judges the number of poles and zeros of the transfer function according to the curve of the active power change in the frequency regulation event process, and gives the values of m and k, and then calculates the model parameters. Identify, and finally obtain the optimal model by continuously adjusting the values of m and k.

After the transfer function model is obtained, the gap measurement method is used to compare the dynamic characteristics difference between each model, so as to achieve the purpose of using a model to characterize the frequency modulation response within a certain range of operating conditions. If the transfer functions of the _two systems are Q1 and _Q2 , respectively, then the gap between the two systems can be expressed as:

定义为：In the formula, G(Q _i ) (i=1,2) represents the graph of the system corresponding to Q _i , and the orthogonal projection matrix

defined as:

(I为单位矩阵，*表示共轭)，由(9)、(10)可得两个系统的间隙度量公式为：Among them, i=1,2,N _i ,D _i ∈RH _∞ can be obtained from the normative right coprime decomposition of Qi _i , that is, Qi _i =N _i D _i ^-1 , and satisfy

(I is the identity matrix, * means conjugation), the gap metric formula of the two systems can be obtained from (9) and (10):

Among them, P is an arbitrary Hilbert matrix, 0≤g≤1, which means that the greater the difference in the dynamic characteristics of the two systems, the closer the g value is to 0, the smaller the difference in the dynamic characteristics of the two systems. The closer the g value is to 1, the greater the difference in dynamic characteristics.

Further, using the second step in this embodiment, the transfer function models under different working conditions are obtained, and the fitting diagrams thereof are shown in Figs. Well, then according to the gap measurement method, the gap values between the models are all close to 1, so we merge the data of all working conditions. At the same time, the gap value also shows that the transfer function model needs to be modeled separately for different working conditions. When the working conditions increase, the workload will also become huge. According to Fig. 5, it can be seen that the transfer function model has poor ability to characterize complex frequency response events.

Preferably, the step 300 includes:

Calculate the distance between the gap value and 0 between the transfer function models of different working conditions to obtain the first distance;

Calculate the distance between the gap value and 1 between the transfer function models of different working conditions to obtain the second distance;

Determine whether the first distance is less than or equal to the second distance, and if so, combine the FM data of the two working conditions according to the time series to obtain the combined data; if not, select the FM data of any working condition and The existing FM data is time-series merged to obtain merged data.

Preferably, the step 300 further includes:

Build the initial neural network;

Based on the Levenberg-Marquark algorithm, the initial neural network is trained according to the combined data;

The trained neural network is tested by using the frequency modulation data of each working condition, and the nonlinear autoregressive neural network model is determined according to the test results.

Specifically, the third step in this embodiment is to combine the corresponding data according to the gap value of the transfer function model under each working condition to construct the nonlinear autoregressive neural network model. The specific method is as follows:

After the gap value between the models is obtained from the second step, if the gap value between the transfer function models of the two working conditions is close to 1, it means that the dynamic characteristics are very different, and the FM data of the two working conditions are merged according to the time series. , to obtain the data set _Du , if the gap value is close to 0, it means that the two models have very similar dynamic characteristics, then select the frequency modulation data of any working condition and merge the existing _Du to obtain a new _Du . Use the merged dataset _Du to train the neural network. The following describes the nonlinear autoregressive neural network:

The output of the nonlinear autoregressive neural network not only depends on the current input, but also on the past input and output, which makes it have a certain memory function. The frequency modulation process of the wind farm is dynamic and the frequency modulation data is also a time series, so the nonlinear autoregressive neural network model of the frequency modulation process of the wind farm is as follows:

ΔP(t)=h(ΔP(t-1), ΔP(t-1), ..., ΔP(t-d), Δf(t), Δf(t-1), ..., Δf(t-d)) (12)

Among them, ΔP(t) is the active power change corresponding to time t, which is the output of the network, Δf(t) is the frequency change corresponding to time t, which is the input of the network, and h( ) is the network construction about ΔP( t) and non-linear functions of Δf(t). d is the delay number. Its network structure is shown in Figure 6.

The present invention adopts the Levenberg-Marquark algorithm to train the nonlinear autoregressive neural network, the algorithm inherits the advantages of the Newton method and the gradient descent method, and the convergence speed is fast and stable. Its parameter update rules are:

In the formula, q _n is the network weight of the nth iteration, and e( ) is the error vector. H and J are the Jacobian and Hessian matrices, respectively. The algorithm uses the Jacobian to give an approximation of the Hessian matrix:

H≈ηI+J ^T J (14)

where n and I are the learning coefficient and identity matrix, respectively.

The training process of the Levenberg-Marquark algorithm is based on adjusting the value of η, and iteratively finds the smallest error. When the error value is smaller than the error of the previous iteration, the value of η is decreased, otherwise, the value of η is increased. When η is close to 0, the algorithm has the local convergence approximate to the gradient descent method, and when η is large, the algorithm has the global convergence approximate to the Gauss-Newton method.

Further, using the third step of the scheme, the data of the corresponding working conditions are merged, the nonlinear autoregressive neural network is trained, and the trained network model is tested with the data of the different working conditions, and Figures 7 to 10 are obtained. Figures 7 to 10 show that, compared with the transfer function model, the nonlinear autoregressive neural network model can simultaneously represent multiple operating conditions, and also has a good ability to characterize the more nonlinear FM response.

The present invention also provides a data-driven wind farm frequency modulation response characteristic modeling system, comprising:

The data processing module is used to analyze and preprocess the measured data of the frequency modulation response characteristics of the wind farm based on the step response dynamic performance index solution algorithm, and obtain the processed data under each working condition;

The transfer function building module is used to establish a transfer function model for each working condition according to the processed data, and use the gap measurement method to measure the gap value between the models to determine the working condition represented by the nonlinear autoregressive neural network model area;

A neural network modeling module is used to merge the frequency modulation data of the working condition according to the gap value, and train the nonlinear autoregressive neural network model according to the merged data to obtain a trained nonlinear autoregressive neural network. network model.

Preferably, the data processing module specifically includes:

A working condition dividing unit, configured to divide the measured data of the frequency modulation response characteristic of the wind farm according to the initial active power during frequency modulation of the wind farm to obtain a plurality of working conditions;

a peak value calculation unit, configured to obtain the power curve of the data set of the measured data of the frequency modulation response characteristic of the wind farm, and calculate the upper peak value set and the lower peak value set of each data point of the power curve;

a maximum value calculation unit, configured to calculate the maximum value of the upper peak set and the minimum value of the lower peak set;

an upper and lower bound calculation unit, configured to calculate the upper bound of the error band and the lower bound of the error band based on the data points in the data set;

A first judging unit, configured to judge whether the upper bound of the error band is greater than or equal to the maximum value of the upper peak set, and whether the lower bound of the error band is greater than or equal to the minimum value of the lower peak set, if the judgment results are all If yes, determine the adjustment time according to the sampling interval;

a second judging unit, configured to judge whether the measured data of the frequency modulation response characteristic of the wind farm is a frequency step underdisturbance, and if so, determine the overshoot amount according to the maximum value of the upper peak set;

A time determination unit, configured to determine a rise time or a fall time according to the data set.

Preferably, the transfer function building module specifically includes:

an initial model building unit, configured to construct an initial function model according to the frequency variation in the processed data under each working condition as the input and the power variation as the output;

Order setting unit, used to set the numerator order and denominator order of the transfer function;

A model determining unit, configured to perform model identification on the initial transfer function model, and adjust the numerator order and the denominator order to obtain the optimal transfer function model.

Preferably, the transfer function building module specifically includes:

a formula determination unit, used for determining a gap measurement formula according to the orthogonal projection matrices of the different transfer function models;

A gap calculation unit, configured to calculate the gap value according to the gap metric formula; the gap value is used to determine the working condition area.

The beneficial effects of the present invention are as follows:

(1) The invention proposes a data-based frequency step response characteristic index solution algorithm, which is used to analyze the frequency modulation response characteristic of a wind farm and process the data required for modeling.

(2) The present invention constructs the transfer function model of the frequency modulation response characteristic of the wind farm under different working conditions, and uses a model gap measurement method to measure the dynamic similarity between the models to determine the working condition area.

(3) The present invention constructs a NARX neural network model to characterize the frequency modulation response characteristics of the wind farm under multiple operating conditions. This model overcomes the disadvantage of the cross between the transfer function model's ability to characterize the range of operating conditions and the ability to characterize strong nonlinear responses. After simulation verification, this model has a good application prospect in the field of wind farm frequency modulation response characteristics modeling.

The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same and similar parts between the various embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant part can be referred to the description of the method.

In this paper, specific examples are used to illustrate the principles and implementations of the present invention. The descriptions of the above embodiments are only used to help understand the methods and core ideas of the present invention; meanwhile, for those skilled in the art, according to the present invention There will be changes in the specific implementation and application scope. In conclusion, the contents of this specification should not be construed as limiting the present invention.

Claims (10)

1. a data-driven wind farm frequency modulation response characteristic modeling method, is characterized in that, comprises: Based on the step response dynamic performance index solution algorithm, the measured data of the frequency modulation response characteristics of the wind farm are analyzed and preprocessed, and the processed data under each working condition are obtained; Establish a transfer function model for each working condition according to the processed data, and measure the gap value between the models by using the gap measurement method to determine the working condition area represented by the nonlinear autoregressive neural network model; The frequency modulation data of the working conditions are combined according to the gap value, and the nonlinear autoregressive neural network model is trained according to the combined data to obtain a trained nonlinear autoregressive neural network model. 2 . The data-driven method for modeling frequency modulation response characteristics of wind farms according to claim 1 , wherein the step-response-based dynamic performance index solution algorithm analyzes and preprocesses the measured data of frequency modulation response characteristics of wind farms, 2 . Obtain the processed data under each working condition, including: Divide the measured data of the frequency modulation response characteristic of the wind farm according to the initial active power of the wind farm during frequency modulation, and obtain a plurality of working conditions; Obtaining the power curve of the data set of the measured data of the frequency modulation response characteristic of the wind farm, and calculating the upper peak set and the lower peak set of each data point of the power curve; calculating the maximum value of the upper peak set and the minimum value of the lower peak set; based on the data points in the data set, calculating an upper error band and a lower error band; Judging whether the upper bound of the error band is greater than or equal to the maximum value of the upper peak set, and whether the lower bound of the error band is greater than or equal to the minimum value of the lower peak set, if the judgment results are all yes, according to the sampling interval determine the adjustment time; Determine whether the measured data of the frequency modulation response characteristic of the wind farm is a frequency step underdisturbance, and if so, determine the overshoot amount according to the maximum value of the upper peak set; Rise time or fall time is determined from the data set. 3 . The data-driven method for modeling frequency modulation response characteristics of wind farms according to claim 1 , wherein, establishing a transfer function model for each operating condition according to the processed data, comprising: 3 . According to the frequency variation in the processed data under each working condition as the input and the power variation as the output, an initial function model is constructed; Set the numerator order and denominator order of the transfer function; Perform model identification on the initial transfer function model, and adjust the numerator order and the denominator order to obtain the optimal transfer function model. 4. The data-driven method for modeling the frequency modulation response characteristics of wind farms according to claim 1, wherein the gap measurement method is used to measure the gap value between the models to determine the value represented by the nonlinear autoregressive neural network model. Condition area, including: Determine the gap measurement formula according to the orthogonal projection matrices of the different transfer function models; The gap value is calculated according to the gap metric formula; the gap value is used to determine the operating condition area. 5 . The data-driven method for modeling frequency modulation response characteristics of a wind farm according to claim 1 , wherein the combining the frequency modulation data of the working conditions according to the gap value comprises: 6 . Calculate the distance between the gap value and 0 between the transfer function models of different working conditions to obtain the first distance; Calculate the distance between the gap value and 1 between the transfer function models of different working conditions to obtain the second distance; Determine whether the first distance is less than or equal to the second distance, and if so, combine the FM data of the two working conditions according to the time series to obtain the combined data; if not, select the FM data of any working condition and The existing FM data is time-series merged to obtain merged data. 6 . The data-driven method for modeling frequency modulation response characteristics of wind farms according to claim 1 , wherein the nonlinear autoregressive neural network model is trained according to the combined data, and a trained non-linear autoregressive neural network model is obtained. 7 . Linear autoregressive neural network models, including: Build the initial neural network; Based on the Levenberg-Marquark algorithm, the initial neural network is trained according to the combined data; The trained neural network is tested by using the frequency modulation data of each working condition, and the nonlinear autoregressive neural network model is determined according to the test results. 7. A data-driven wind farm frequency modulation response characteristic modeling system, characterized in that, comprising: The data processing module is used to analyze and preprocess the measured data of the frequency modulation response characteristics of the wind farm based on the step response dynamic performance index solution algorithm, and obtain the processed data under each working condition; The transfer function building module is used to establish a transfer function model for each working condition according to the processed data, and use the gap measurement method to measure the gap value between the models to determine the working condition represented by the nonlinear autoregressive neural network model area; A neural network modeling module is used to merge the frequency modulation data of the working condition according to the gap value, and train the nonlinear autoregressive neural network model according to the merged data to obtain a trained nonlinear autoregressive neural network. network model. 8. The data-driven wind farm frequency modulation response characteristic modeling system according to claim 7, wherein the data processing module specifically comprises: A working condition dividing unit, configured to divide the measured data of the frequency modulation response characteristic of the wind farm according to the initial active power during frequency modulation of the wind farm to obtain a plurality of working conditions; a peak value calculation unit, configured to obtain the power curve of the data set of the measured data of the frequency modulation response characteristic of the wind farm, and calculate the upper peak value set and the lower peak value set of each data point of the power curve; a maximum value calculation unit, configured to calculate the maximum value of the upper peak set and the minimum value of the lower peak set; an upper and lower bound calculation unit, configured to calculate the upper bound of the error band and the lower bound of the error band based on the data points in the data set; A first judging unit, configured to judge whether the upper bound of the error band is greater than or equal to the maximum value of the upper peak set, and whether the lower bound of the error band is greater than or equal to the minimum value of the lower peak set, if the judgment results are all If yes, determine the adjustment time according to the sampling interval; a second judging unit, configured to judge whether the measured data of the frequency modulation response characteristic of the wind farm is a frequency step underdisturbance, and if so, determine the overshoot amount according to the maximum value of the upper peak set; A time determination unit, configured to determine a rise time or a fall time according to the data set. 9. The data-driven wind farm frequency modulation response characteristic modeling system according to claim 7, wherein the transfer function building module specifically comprises: an initial model building unit, configured to construct an initial function model according to the frequency variation in the processed data under each working condition as the input and the power variation as the output; Order setting unit, used to set the numerator order and denominator order of the transfer function; A model determining unit, configured to perform model identification on the initial transfer function model, and adjust the numerator order and the denominator order to obtain the optimal transfer function model. 10. The data-driven wind farm frequency modulation response characteristic modeling system according to claim 7, wherein the transfer function building module specifically comprises: a formula determination unit, used for determining a gap measurement formula according to the orthogonal projection matrices of the different transfer function models; A gap calculation unit, configured to calculate the gap value according to the gap metric formula; the gap value is used to determine the working condition area.

Images