A model incomplete wind farm frequency modulation capability online evaluation method

Abstract

The present application relates to a kind of model incomplete wind farm frequency modulation capability online evaluation method.It includes: step S1: obtaining final speed ω final And wind farm overall droop coefficient k And corresponding wind speed data;Step S2: stored k and ω final Data is classified according to wind speed;Step S3;In each wind speed classification, respectively, input variable data is mapped and transformed in state space, and linear model is constructed under the classification data;Step S4: respectively assesses and calculates wind farm overall maximum droop slope k max ;Step S5: fitting wind speed-maximum droop slope curve;Step S6: wind farm frequency modulation capability evaluation is carried out;Step S7: the evaluation result is reported to power grid.Model incomplete wind farm frequency modulation capability online evaluation method based on state space mapping directly utilizes wind farm operation data to train linear model, and then assesses frequency modulation capability.The method avoids dependence on model parameters, and can be evaluated with high precision in the scene that wind farm model is incomplete or parameters are not accurate.

Description

Technical Field

[0001] This invention relates to an online evaluation method for the frequency regulation capability of wind farms with incomplete models. Background Technology

[0002] As the main body of grid connection, the wind farm can simulate the droop characteristics of conventional power plants as a whole during the first frequency regulation period, meet the grid connection guidelines, and coordinate the control of large-scale wind turbines. The output regulation of power electronic converters has the advantages of rapid response and low maintenance cost, but it will cause changes in wind turbine speed and affect safe operation. Therefore, it is necessary to reasonably adjust the active power set value of each wind turbine, optimize the power distribution, and ensure the safe operation of each wind turbine. Droop control is the most commonly used method for wind turbine power control. Reference [1] added a wind turbine speed protection module to the traditional droop control to ensure the safe operation of the wind turbine. Reference [2] divided the operating wind speed into low, medium and high ranges and adopted different droop coefficient setting methods according to the wind speed range. Reference [3] set an adaptive droop control coefficient according to the wind turbine's own operating conditions. However, the above methods ignore the coordinated control between wind turbines. Each wind turbine cannot reach the optimal output synchronously, which limits the frequency regulation capability of the wind farm. Therefore, in order to fully utilize the frequency regulation capability of each wind turbine and achieve coordinated regulation, the article [6-7] used a wind turbine energy state index based on rotational speed.

[0003] Regardless of whether the wind farm operates in a passive response dispatch command mode or an active participation market mode, in order to ensure the safety of the rotational speed of each wind turbine, before participating in the primary frequency regulation, it is necessary to first assess the frequency regulation capability, calculate its own limit droop slope and report it, and then execute the adjustment according to the instructions fed back by the power grid. Theoretically, based on the dynamic model of the wind turbine and the energy state index allocation method, the limit slope can be accurately calculated to realize the frequency regulation capability assessment. Reference [8] adds the rotor kinetic energy assessment factor and the converter capacity limit factor on the basis of virtual inertia control to assess the frequency regulation capability of wind turbine units under different wind speeds. Reference [9] proposes an economic dispatch method with synchronous inertia constraint to meet the minimum synchronous inertia required for frequency control. Reference

[10] uses the lowest frequency in the dynamic response as a constraint to calculate the minimum power frequency static characteristic coefficient of the wind farm and realize the minimum inertia response estimation. Reference

[11] proposes a wind farm frequency regulation capability coefficient based on the frequency regulation capability assessment of a single wind turbine to realize the collaborative virtual inertia control and rotational speed collaborative recovery between different wind farm units. However, due to the large number of wind turbines and complex dynamic characteristics in wind farms, the evaluation models in the aforementioned literature are high-dimensional nonlinear differential-algebraic equation systems, which are difficult to solve analytically with precision. Simulation calculations need to be performed again under different wind speed conditions to determine the frequency regulation capability, which is time-consuming. Furthermore, time-domain analysis methods heavily rely on model parameters; if the model is incomplete or has poor accuracy, the calculation accuracy cannot be guaranteed.

[0004] Traditional model methods are difficult to guarantee the accuracy of the dynamic frequency regulation process of wind farms, while data-driven algorithms have the advantage of solving complex problems. The load reduction scheme when the system frequency decreases is optimized by using neural network methods to ensure the safety of system frequency. Reference

[11] constructed a long short-term neural network model and used historical data to train the relationship between system load and scheduling decision, which improved the accuracy of unit combination decision. However, the above neural network method is detached from the mathematical model, which makes it difficult to reflect the internal relationship of the system and is not applicable to all scenarios. In the actual frequency regulation process of wind farms, it is difficult to have the extreme speed scenario of wind turbines, and sufficient data samples cannot be provided. The above model is difficult to guarantee the accuracy of evaluation under extreme operating scenarios.

[0005] In summary, existing methods for assessing the primary frequency regulation capability of wind farms still have certain shortcomings and deficiencies. Traditional model analysis methods rely on the setting of model parameters, and if the parameters are inaccurate, the accuracy of the assessment results is difficult to guarantee. Furthermore, the model needs to be recalculated in the time domain under different wind speeds. Due to the large number of wind turbines in a wind farm and the complex dynamic characteristics of frequency regulation, the model calculation time is very long, making this method unsuitable for online assessment requirements.

[0006] Data-driven modeling methods, such as neural networks, which are widely used, depart from mathematical mechanistic models. To ensure the accuracy of evaluation results, the training set needs to cover as many typical scenarios as possible. However, in actual engineering, data on extreme operating scenarios of wind farms are difficult to obtain, and sufficient data samples cannot be provided, making it difficult to guarantee the evaluation accuracy of neural network models. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to provide an online evaluation method for the frequency regulation capability of wind farms with incomplete models based on state-space mapping. This method constructs a wind power linearity evaluation model through data-driven training, which has a fast evaluation speed and high evaluation accuracy.

[0008] The present invention adopts the following technical solution:

[0009] This invention includes the following steps:

[0010] Step S1: Obtain the final rotational speed ω of all wind turbines in the wind farm after primary frequency regulation. final And the overall sag coefficient k of the wind farm and the corresponding wind speed data;

[0011] Step S2: For the stored k and ω final The data is categorized according to wind speed;

[0012] Step S3: Perform state-space mapping transformation on the input variable data for each wind speed category, and construct a linear model for that category of data;

[0013] Step S4: Evaluate and calculate the overall maximum droop slope k of the wind farm. max ;

[0014] Step S5: Fit the wind speed-maximum droop slope curve;

[0015] Step S6: Conduct a frequency regulation capability assessment of the wind farm;

[0016] Step S7: Report the assessment results to the power grid.

[0017] In this invention, steps S1-S5 are offline training processes, and steps S6-S7 are online evaluation processes. When the training time period in steps S1-S5 changes or the wind farm's own parameters change, the offline training results need to be updated and maintained to ensure training accuracy, and the above steps S1-S5 are repeated.

[0018] In step S1 of this invention, the final rotational speed ω of all wind turbines in the wind farm after primary frequency regulation is obtained through a Supervisory Control and Data Acquisition (SCADA) system. final The overall sag coefficient k of the wind farm and the corresponding wind speed data are stored on the server.

[0019] The frequency regulation process of a wind farm can be approximated as a nonlinear model with the overall droop coefficient of the wind farm as the input variable and the final rotational speed of all wind turbines after primary frequency regulation as the output variable, as shown in formula (1):

[0020] ω final =φ(k) (1)

[0021] In equation (1) ω final denoted as , where is the final rotational speed of all wind turbines in the wind farm after one frequency adjustment, and k is the overall droop coefficient of the wind farm.

[0022] In step S2 of this invention, the stored k and ω final The data is categorized according to wind speed, with each category covering as much of the common operating wind speed ranges in wind farms as possible. The method divides the common operating wind speed ranges in wind farms into equal-step intervals (e.g., setting the wind speed range as [7,11] m / s with a step size of 0.5 m / s), obtaining multiple sampled wind speeds. Data at the same wind speed are grouped together. If the accumulated data for each sampled wind speed exceeds 1000 sets, the data-driven training condition is met, and the method begins training. If the data-driven condition is not met, the method continues to acquire data.

[0023] In step S3 of this invention, after the data meets the training requirements, the method of this invention performs state-space mapping transformation on the input variable data under each wind speed classification, and constructs a linear model under the classification data.

[0024] Nonlinear models can be transformed into linear models using the Koopman dimension-up transformation method, as shown in equations (2) and (3):

[0025] ω final =Mk lift (2)

[0026]

[0027] In equation (2), k lift Let represent the matrix after the dimension-up transformation of k; M represents the coefficient matrix. In equation (3), ψ(k) represents the dimension-up function matrix of k. If the dimension-up transformation of the method is n (n can be any positive integer, such as 500, 1000, 1500, etc.), the dimension-up function ψ(x) can be regarded as consisting of n scalar functions ψ i (x) is composed of the following formula:

[0028] ψ i (k)=f lift (k+c i (4)

[0029] Where c i Let c represent the basis vector of the i-th scalar function. i Any value within the interval [0, 40] can be selected. The scalar function ψ in this invention... i (x) is specifically expressed in polyharmonic form as follows:

[0030]

[0031] The coefficient matrix M of the linear model of the wind farm can be estimated using the least squares method, and the historical training data of the wind farm can be represented by formula (6):

[0032]

[0033] In equation (6), n represents the number of historical data points for a wind farm at the same wind speed; k represents the sample set of the overall droop coefficient of the wind farm; W final This represents the sample set of final operating speeds of all wind turbines after a single frequency adjustment. Based on the above data, the coefficient matrix of the wind farm linear model can be estimated using the least squares method, as shown in formula (7):

[0034]

[0035] In equation (7), K lift The matrix representing the matrix after performing a dimension-up transformation on K; Representation matrix K lift Transpose of; Representative matrix The Moore-Penrose inverse. After calculating M according to formula (7), the frequency modulation evaluation linear model described in formula (2) can be constructed.

[0036] In step S4 of this invention, for each sampled wind speed, the linear model uses a bisection method to evaluate and calculate the overall maximum droop slope k of the wind farm. max k max This serves as an assessment result of the wind farm's frequency regulation capability at this wind speed.

[0037] Since the polyharmonic type increasing-dimensional function has monotonicity, the linear model (2) also has monotonicity, and the bisection method can be used to solve the model. If the safe speed limit for primary frequency regulation of all fans is known, the bisection method can be used to calculate ω. final When the safe operating speed limit is reached, the corresponding maximum overall droop slope k of the wind farm is... max k max This is the assessment result of the primary frequency regulation capability of the wind farm under the current wind speed.

[0038] In step S5 of this invention, a wind speed-maximum droop slope curve is fitted using a linear interpolation method, and the fitting result is stored in a server.

[0039] In step S6 of this invention, wind speed data of each wind turbine in the wind farm is obtained through a wind measurement tower device. The method can use the obtained current wind speed to evaluate the frequency regulation capability of the wind farm in real time, or it can be combined with wind speed prediction technology such as ARMA method to predict the future wind speed based on the current wind speed data. The method can also use short-term and ultra-short-term predicted wind speeds to predict and evaluate the frequency regulation capability of the wind farm.

[0040] In step S7 of this invention, the offline training result k under the corresponding wind speed condition is retrieved from the server. max Complete the online assessment of the frequency regulation capability of the wind farm and report the results to the power grid.

[0041] When the power grid in steps S6-S7 has a frequency regulation assessment requirement for wind farms, steps S6-S7 are repeated.

[0042] The positive effects of this invention are as follows:

[0043] The online evaluation method for frequency regulation capability of wind farms with incomplete models based on state-space mapping directly trains a linear model using wind farm operation data to evaluate its frequency regulation capability. This method avoids dependence on model parameters and enables high-precision evaluation even in scenarios where the wind farm model is incomplete or the parameters are inaccurate.

[0044] The online evaluation method for the frequency regulation capability of wind farms based on state-space mapping uses a linear model to calculate the evaluation results. Compared with the traditional nonlinear model analysis method, the calculation speed is greatly improved, which can meet the time requirements of online evaluation.

[0045] The online evaluation method for wind farm frequency regulation capability based on state-space mapping proposes a state-space mapping transformation process as shown in formula (2), which is a linear approximation of the nonlinear wind farm frequency regulation physical model. Compared with data-driven modeling methods such as neural networks, the training set of this method does not need to cover the scenario where the wind turbine reaches the safe speed limit. It can build a global model and then calculate the evaluation results under the safe speed limit through the model, resulting in stronger evaluation accuracy and reliability. Attached Figure Description

[0046] Figure 1 This is a flowchart of the method proposed in this invention for evaluating the frequency regulation capability of a wind farm;

[0047] Figure 2 In the scenario of increasing frequency, a comparison of the limiting slope evaluation results of the method proposed in this invention and the method with inaccurate time series model parameters is shown in the figure.

[0048] Figure 3 This is a comparison chart of the error between the limit slope evaluation results of the proposed method and the method with inaccurate time series model parameters relative to the accurate time series model in the context of rising frequency.

[0049] Figure 4 In the scenario of decreasing frequency, a comparison of the limiting slope evaluation results of the method proposed in this invention and the method with inaccurate time series model parameters is shown in the figure.

[0050] Figure 5 This is a comparison chart of the error between the limit slope evaluation results of the proposed method and the method with inaccurate time series model parameters in the context of frequency decline, and the accurate time series model.

[0051] Figure 6 This is a comparison chart of the relative errors of the extreme slope to the accurate time series model under different training sets for the proposed method in the context of rising frequency.

[0052] Figure 7 This is a comparison chart of the relative errors of the limiting slope of the proposed method on different training sets in the context of frequency descent.

[0053] Figure 8 In the scenario of increasing frequency, the method proposed in this invention fits the final global rotational speed map;

[0054] Figure 9 It is a global graph of the final rotational speed that is fitted by an accurate timing model for the frequency increase scenario;

[0055] Figure 10In the scenario of frequency decrease, the method proposed in this invention fits the final global rotational speed map;

[0056] Figure 11 It is a global graph of the final rotational speed that is fitted by an accurate timing model for the frequency reduction scenario;

[0057] Figure 12 This is a comparison chart of the frequency modulation results of the proposed method and the method with inaccurate time series model parameters in the scenario of frequency decrease at a wind speed of 8.2 m / s.

[0058] Figure 13 This is a comparison chart of the evaluation results of the proposed method and the method with inaccurate time series model parameters in the scenario of frequency increase at a wind speed of 8.2 m / s.

[0059] Figure 14 This is a comparison chart of the evaluation results of the proposed method and the method with inaccurate time series model parameters in the frequency descent scenario at a wind speed of 10.3 m / s, showing a primary frequency modulation method.

[0060] Figure 15 This is a comparison chart of the evaluation results of the proposed method and the method with inaccurate time series model parameters in the frequency descent scenario at a wind speed of 10.3 m / s, showing a primary frequency modulation method. Detailed Implementation

[0061] Existing wind farm frequency regulation capability assessment schemes mostly employ traditional model analysis methods. First, a wind farm frequency regulation capability model is constructed, and then the assessment results are obtained through model solving. However, this approach is heavily reliant on the setting of model parameters. If the model parameters are not set accurately or the wind farm model is incomplete, the assessment accuracy will be affected. Furthermore, this approach requires recalculation under different wind speed conditions, resulting in lengthy calculations and making it difficult to meet the requirements for online assessment.

[0062] Therefore, in order to optimize the online evaluation scheme of wind farm frequency regulation capability, a method for online evaluation of wind farm frequency regulation capability based on state-space mapping with incomplete models is proposed.

[0063] The inventors discovered that data-driven methods are well-suited for solving complex problems, achieving high-precision evaluations without relying on model parameter settings, while simultaneously reducing the time cost of evaluation computation. This invention proposes a data-driven method that constructs a linear model through dimensionality-up transformation and estimates the coefficient matrix using the least squares method.

[0064] A precise time-series simulation model of a wind farm was built using MATLAB software to replace the actual operation of the wind farm and obtain data-driven training data. The evaluation results of the Koopman linear model and the precise time-series simulation model of the wind farm were compared in a numerical example. The wind farm has a rated power of 32MW, including eight permanent magnet synchronous generators with a rated power of 4MW each. The turbine blade speed range is [0.7, 1.44] rad / s, and the training set selects data within the speed range [0.75, 1.39] rad / s. The training wind speed range is [7, 11] m / s, the wind speed sampling step size is 0.5 m / s, and the input variable of the linear model is k, and the output variable is ω. final To calculate the wind farm k max At the start of a frequency regulation, a large load fluctuation is introduced into the power system. The load fluctuation is jointly regulated by wind farms and conventional power plants to obtain the limit of unbalanced power that wind farms can regulate. The remaining unbalanced power is regulated by conventional power plants, thus obtaining wind farm training data under frequency rise and fall conditions.

[0065] Example 1

[0066] Figures 2-5 The Koopman linear model evaluation results k are given under different wind speeds. max The curve, k obtained from accurate time-series model simulation. max The curves also provide the time series model calculation k under the condition that the three model parameters have errors. max The curves and their relative errors compared to the results of the accurate time series model. The specific errors for the parameters are J... c Decrease by 12%; R increases by 5%; C p,i The coefficient decreases by 10%, while J c The increase is 12%. It can be seen that the Koopman linear model curve and the time-series model curve are very close overall, with a very small relative error, and the two curves highly overlap. If multiple parameters in the wind farm model are inaccurately set simultaneously, or if the setting of a single key parameter is too large, it may cause excessive errors in the evaluation results. If the time-series method is used to analyze the frequency regulation dynamics of a wind farm, if the wind farm model parameters are inaccurate, the evaluation accuracy will be far lower than that of the upgraded linear model.

[0067] Example 2

[0068] Figure 6 and Figure 7The relative errors of the evaluation results of this invention under different training sets were compared. Training set 1 had a wind turbine speed range of [0.75, 1.39] rad / s, training set 2 had a wind turbine speed range of [0.80, 1.34] rad / s, and training set 3 had a wind turbine speed range of [0.85, 1.29] rad / s. It can be seen that the wider the coverage of the training set, the higher the accuracy of the evaluation method, but the difficulty of data acquisition will increase relatively. Therefore, it is necessary to comprehensively consider both accuracy and data acquisition factors, and to reasonably select the training set range while avoiding large errors, taking into account the actual operation of the wind farm. In the simulation case of this invention, training set 1 was selected for data training, which can relatively well balance accuracy and data requirements.

[0069] Example 3

[0070] To more comprehensively verify the effectiveness of the method proposed in this invention, the fitting of the rotational speed of the wind field under different wind speed scenarios was analyzed.

[0071] Figures 8-11 The fitting of ω by the Koopman upscaling linear model and the time series model was compared in different scenarios. fina The two figures are highly similar, indicating that the Koopman linear model not only has a good fitting effect for extreme cases, but also perfectly maps the complex nonlinear relationships of wind farm frequency regulation models. Based on the real-time dynamics of k and wind speed, it issues reasonable adjustment commands to the wind turbines, obtaining high-precision ω. final This demonstrates that the method proposed in this invention has the performance advantages of fast analytical solution and applicability to online evaluation of wind farms.

[0072] Example 4

[0073] exist Figures 12-15 In the process, errors exist in the parameters set for the wind farm model, leading to k max The deviation from the true value of the precise time series model is ±10%. A frequency modulation process is performed to compare the proposed method with that of this invention, using two wind speed scenarios: 8.2 m / s and 10.3 m / s. The proposed method works at a wind speed of 8.2 m / s. max The calculation results show a frequency decrease of 5.951 MW / Hz and a frequency increase of 24.394 MW / Hz; at a wind speed of 10.3 m / s, k max The calculation results show a frequency decrease of 10.672 MW / Hz and a frequency increase of 7.237 MW / Hz.

[0074] Figure 12 The paper presents the primary frequency regulation process of a wind farm under a frequency decrease scenario at a wind speed of 8.2 m / s. It can be seen that the Koopman up-dimensional linear evaluation is superior to the k-mode evaluation. max Compared to the result of increasing ω by 10%, finalThe error is extremely small, effectively solving the safety problem caused by the fan speed exceeding the limit; compared with k max With a 10% reduction, Koopman's linear upscaling assessment can increase the active power output of wind farms, more fully exploit the frequency regulation capabilities of wind turbines, and provide stronger frequency support. Figure 13 The text presents scenarios of increasing frequency, and the analysis process is similar to... Figure 12 They are basically the same.

[0075] Figure 14 and Figure 15 The frequency regulation process at a wind speed of 10.3 m / s is presented. Compared to a wind speed of 8.2 m / s, the speed deviation caused by model parameter errors is larger at this wind speed, resulting in more serious speed limit exceedance problems and wasted wind turbine frequency regulation capacity. The Koopman state-space mapping linear model evaluation result ω... final The relatively small error between the speed and the critical value indicates that the method proposed in this invention maintains good evaluation performance at higher wind speeds. Compared with cases where the model parameters are inaccurate, the method of this invention has more significant advantages at higher wind speeds.

[0076] The frequency regulation capability of wind farms is evaluated by constructing a linear model using the state-space mapping method, a scheme that has not been considered by other technologies.

[0077] This invention utilizes historical data for data-driven training in linear models, specifically employing least squares estimation of the coefficient matrix of the linear model. This method is not considered in other techniques.

[0078] This invention selects a polyharmonic function as the dimension-increasing function in the state-space mapping process, which makes the linear model monotonic and can use the bisection method to solve the wind farm evaluation results. It has the advantages of simple solution and accurate results.

[0079] This invention consists of two parts: offline training and online evaluation. The evaluation results under different wind speeds are calculated in the offline training. The actual online evaluation process only requires calling the evaluation results under the corresponding wind speed in the server to complete the evaluation, which reduces the calculation time and better meets the time requirements of online evaluation.

[0080] This invention achieves high-precision evaluation without relying on wind turbine models. Traditional model analysis is easily affected by parameter accuracy. The data-driven method proposed in this paper directly utilizes actual operating data, does not depend on model parameters, has good scalability, and high computational accuracy.

[0081] This invention offers analytical solutions, rapid computation, and suitability for online applications. Koopman theory projects the evaluation model from a low-dimensional nonlinear equation system to a high-dimensional linear equation system, significantly improving online computation speed.

[0082] This invention integrates data-driven and model-based features, reducing the difficulty of acquiring training data. Compared to neural network methods, the Koopman method is a simplified expression of the physical model, and the training set does not need to cover extreme speed scenarios, ensuring the reliability of the evaluation results.

[0083] Currently, the technical solution of this application has undergone pilot testing, which is a small-scale experiment before the product is mass-produced. After the pilot testing was completed, a user survey was conducted on a small scale, and the survey results showed that user satisfaction was high. Now, preparations are underway for the formal production and industrialization of the product (including intellectual property risk warning surveys).

[0084] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0085] References:

[0086] [1] Ding Lei, Yin Shanyao, Wang Tongxiao, et al. Active speed protection control strategy for doubly fed wind turbines with inertial frequency regulation [J]. Automation of Electric Power Systems, 2015, 39(24):29-34+95.

[0087] [2] Pan Wenxia, ​​Quan Rui, Wang Fei, et al. Variable droop coefficient control strategy based on doubly fed wind turbines [J]. Automation of Electric Power Systems, 2015, 39(11): 126-131+186.

[0088] [3]L Ee J, Jang G, Muljadi E, et al.Stable Short-Te-rm Frequency SupportUsing Adaptive Gains for a DFIG-Based Wind Power Plant[J].IEEE TransactionsonEnergy Conversion, 2016, 31(3):1-12.

[0089] [4]Wang Z, Wu W. Coordinated Control Method for DFIG-Based Wind Farm toProvide Primary Frequency Regulation Service[J]. IEEE Transactions on PowerSystems, 2017: 1-1.

[0090] [5]Guo L, Ren Y, Wang Z, et al. Double-Layer Fee-dback Control Method for Synchronized Frequency Regulation of PMSG-Based Wind Farm[J]. IEEE Tr-ansactionson Sustainable Energy, 2021, 12(4): 2423-2435.

[0091] [6] CHEN Yuehang, WANG Gang, SHI Qiaoming, et al. A novel virtual inertia collaborative control strategy for wind farms [J]. Automation of Electric Power Systems, 2015, 39(5):7.

[0092] [7]Gu H, Yan R, Saha T K. Minimum Synchronous Inertia Requirement of Renewable Power Systems[J]. IEEE Transactions on Power Systems, 2017, PP(99): 1-1.

[0093] [8] Li Dongdong, Sun Yaru, Xu Bo, et al. Evaluation method of minimum inertia and primary frequency regulation capacity of new energy high penetration power system considering frequency stability [J]. Power System Protection and Control, 2021, 49(23):

[0094] [9] Shi Qiaoming, Wang Gang, Li Haiying, et al. Multi-machine cooperative control strategy for virtual inertia of wind farm considering frequency regulation capability [J]. Power System Technology, 2019, 43(11):

[0095]

[10] Padron S, Hernandez M, Falcon A.Reducing Und-er-Frequency LoadShedding in Isolated Power Syste-ms Using Neural Networks.Gran Canaria: A CaseStudy[J].IEEE Transactions on Power Systems, 2015, 31(1): 63-71.

[0096]

[11] Yang Nan, Ye Di, Lin Jie, et al. Research on intelligent decision-making method for unit combination with self-learning ability based on data-driven approach [J]. Proceedings of the CSEE, 2019, 39(10): 2934-2945.

Claims

1. A method for online evaluation of the frequency regulation capability of wind farms with incomplete models, characterized in that: Includes the following steps: Step S1: Obtain the final rotational speed ω of all wind turbines in the wind farm after primary frequency regulation. final And the overall sag coefficient k of the wind farm and the corresponding wind speed data; Step S2: For the stored k and ω final The data is categorized according to wind speed; Step S3; For each wind speed category, a state-space mapping transformation is performed on the input variable data, and a linear model is constructed for that category of data. Step S4: Evaluate and calculate the overall maximum droop slope k of the wind farm. max ; Step S5: Fit the wind speed-maximum droop slope curve; Step S6: Conduct a frequency regulation capability assessment of the wind farm; Step S7: Report the assessment results to the power grid.

2. The online evaluation method for the frequency regulation capability of an incomplete wind farm according to claim 1, characterized in that: Steps S1-S5 are offline training processes, and steps S6-S7 are online evaluation processes. When the training time period in steps S1-S5 changes or the wind farm's own parameters change, the offline training results need to be updated and maintained to ensure training accuracy. Steps S1-S5 are then repeated.

3. The online evaluation method for the frequency regulation capability of an incomplete wind farm according to claim 2, characterized in that: In step S1, the final rotational speed ω of all wind turbines in the wind farm after primary frequency regulation is obtained through the Supervisory Control and Data Acquisition (SCADA) system. final The overall sag coefficient k of the wind farm and the corresponding wind speed data are stored on the server; The frequency regulation process of a wind farm is represented by a nonlinear model with the overall droop coefficient of the wind farm as the input variable and the final rotational speed of all wind turbines after primary frequency regulation as the output variable, as shown in formula (1): oh final =φ(k)(1) In equation (1) ω final denoted as , where is the final rotational speed of all wind turbines in the wind farm after one frequency adjustment, and k is the overall droop coefficient of the wind farm.

4. The online evaluation method for the frequency regulation capability of an incomplete wind farm according to claim 3, characterized in that: In step S2, the stored k and ω final The data is categorized according to wind speed, and the wind speed categories should cover as much as possible the common operating wind speed ranges of wind farms; The method divides the common operating wind speed range of wind farms into equal step sizes to obtain multiple sampling wind speeds. Data under the same wind speed will be grouped into one category. If the accumulated data for each sampling wind speed exceeds 1000 sets, it is determined that the data-driven training condition is met, and the method starts training. If the data-driven condition is not met, the method continues to acquire data.

5. The online evaluation method for the frequency regulation capability of an incomplete wind farm according to claim 4, characterized in that: In step S3, after the data meets the training requirements, the input variable data is transformed by state space mapping under each wind speed category, and a linear model is constructed under the data of that category. Nonlinear models can be transformed into linear models using the Koopman dimension-up transformation method, as shown in equations (2) and (3): oh final =Mk lift (2) (3) In equation (2), k lift Let represent the matrix after the dimension-up transformation of k; M represents the coefficient matrix; in equation (3), ψ(k) represents the dimension-up function matrix of k; if the dimension-up transformation of the method is n, where n is any positive integer, the dimension-up function ψ(x) is regarded as being composed of n scalar functions ψ i (x) is composed of the following formula: ψ i (k)=f lift (k+c i (4) Where c i Let c represent the basis vector of the i-th scalar function. i Select any value within the interval [0, 40]; the scalar function ψi(x) specifically adopts the polyharmonic form, as follows: (5) The coefficient matrix M of the linear model of the wind farm is estimated using the least squares method, and the historical training data of the wind farm is represented by formula (6): (6) In equation (6), n represents the number of historical data points for a wind farm at the same wind speed; k represents the sample set of the overall droop coefficient of the wind farm; W final This represents the sample set of final operating speeds of all wind turbines after a single frequency regulation. Based on the above data, the coefficient matrix of the wind farm linear model can be estimated using the least squares method, as shown in formula (7): (7) In equation (7), K lift The matrix representing the matrix after performing a dimension-up transformation on K; Representation matrix K lift Transpose of; Representative matrix The Moore-Penrose inverse; after calculating M according to formula (7), the frequency modulation evaluation linear model can be constructed.

6. The online evaluation method for the frequency regulation capability of an incomplete wind farm according to claim 5, characterized in that: In step S4, for each sampled wind speed, the linear model uses a bisection method to evaluate and calculate the overall maximum droop slope k of the wind farm. max k max This serves as an assessment result of the wind farm's frequency regulation capability at this wind speed. Since the polyharmonic type increasing-dimensional function has monotonicity, the linear model (2) is also monotonic and can be solved using the bisection method; if the primary frequency regulation safe speed limit of all fans is known, the bisection method can be used to calculate ω. final When the safe operating speed limit is reached, the corresponding maximum overall droop slope k of the wind farm is... max k max This is the assessment result of the primary frequency regulation capability of the wind farm under the current wind speed.

7. The online evaluation method for the frequency regulation capability of an incomplete wind farm according to claim 6, characterized in that: In step S5, the wind speed-maximum droop slope curve is fitted using a linear interpolation method, and the fitting result is stored in the server.

8. The online evaluation method for the frequency regulation capability of an incomplete wind farm according to claim 7, characterized in that: In step S6, wind speed data of each wind turbine in the wind farm is obtained through a wind measurement tower device. The method uses the obtained current wind speed to evaluate the frequency regulation capability of the wind farm in real time, or combines the wind speed prediction technology ARMA method to predict the future wind speed based on the current wind speed data, or uses short-term and ultra-short-term predicted wind speeds to predict and evaluate the frequency regulation capability of the wind farm.

9. The online evaluation method for the frequency regulation capability of an incomplete wind farm according to claim 8, characterized in that: In step S7, the offline training result k under the corresponding wind speed condition is retrieved from the server. max Complete the online assessment of the frequency regulation capability of the wind farm and report the results to the power grid.

10. The online evaluation method for the frequency regulation capability of an incomplete wind farm according to claim 9, characterized in that: If the power grid has a frequency regulation assessment requirement for wind farms in steps S6-S7, then steps S6-S7 are repeated.

Images