Wind power plant intelligent equivalent modeling method, electronic equipment and storage medium

Abstract

The invention relates to the technical field of power system modeling and simulation, in particular to an intelligent equivalent modeling method for a wind power plant, electronic equipment and a storage medium, and the method comprises the steps: obtaining the dynamic response data of a plurality of doubly-fed fans in the wind power plant in a typical fault scene of a power grid; performing first-layer grouping on the doubly-fed fans according to whether a crowbar circuit is put into the doubly-fed fans or not; respectively extracting low-voltage ride-through fault response characteristics of the first type of input units and the second type of non-input units to serve as second-layer grouping indexes; obtaining an initial mass center of an MI-GSA-K-means algorithm by using a gravitational search algorithm; iteratively calculating DBI indexes corresponding to different clustering numbers, and selecting the clustering number corresponding to the minimum value of the DBI indexes as the optimal clustering number; and aggregating the grouped doubly-fed fans in the same group into an equivalent unit, and determining a wind power plant dynamic equivalent model based on LVRT-GSA dual-stage clustering. According to the method, the consistency of the equivalent model and the system fault process is improved, and the model precision can be effectively improved.

Description

Technical Field

[0001] The present application relates to the technical field of power system modeling and simulation, and in particular to a wind farm intelligent equivalent modeling method, electronic equipment, and storage medium. Background Art

[0002] As wind power penetration continues to increase, the dynamic characteristics of wind farm grid connections are increasingly impacting power system stability analysis. Traditional wind farm equivalent modeling methods are primarily categorized into single-unit aggregation and multi-unit clustering. Single-unit equivalent methods, while simple in structure, fail to characterize unit dynamic differences. Multi-unit equivalent methods, drawing on the "synchronization" concept of synchronous generators, use steady-state parameters such as wind speed and power as the basis for clustering.

[0003] However, the current wind farm equivalent modeling methods constructed using the multi-machine equivalent method are usually based on a single-layer clustering strategy, ignoring the dominant role of fault characteristics on the dynamic characteristics of the unit during voltage sag, resulting in a single clustering dimension. In addition, commonly used clustering indicators (such as wind speed and pitch angle) fail to reflect the transient response differences of the unit during low voltage ride-through (LVRT). When a deep voltage sag occurs in the power grid, the doubly fed wind turbine (DFIG) achieves fault ride-through by inputting a crowbar circuit. Its transient characteristics are significantly different from those when the crowbar is not input. The constructed equivalent model is difficult to accurately describe such dynamic behavior, resulting in insufficient dynamic characteristic representation. In addition, the K-means method uses the Euclidean distance or similarity function between the unit sample and the centroid to obtain the unit clustering results. The centroid usually uses the sample mean, which is difficult to accurately represent the true dynamic characteristics of the unit, affecting the model accuracy. Summary of the Invention

[0004] In order to solve the above-mentioned problems in the prior art, the present application provides an intelligent equivalent modeling method, electronic device and storage medium for a wind farm, which takes the crowbar circuit response behavior and low voltage ride-through characteristics during a fault as key clustering indicators, and combines the improved clustering algorithm to accurately divide the units, thereby improving the consistency between the equivalent model and the system fault process, and can effectively improve the model accuracy.

[0005] In a first aspect, an embodiment of the present application provides a method for intelligent equivalent modeling of a wind farm, the method comprising:

[0006] Obtain dynamic response data of multiple doubly-fed wind turbines in a wind farm under typical grid fault scenarios, the dynamic response data including electrical observations, control signals, and time series. The control signals include the operating state of the crowbar circuit, and the electrical observations include rotor current, grid-connection point active power, and grid-connection point reactive power.

[0007] determining whether the crowbar circuit is put into operation of the doubly-fed wind turbine according to the magnitude of the rotor current;

[0008] According to whether the crowbar circuit is put into use of the doubly fed wind turbine, the doubly fed wind turbine is grouped into a first layer to obtain a first type of put-in-use units and a second type of non-put-in-use units;

[0009] Extracting low voltage ride-through fault response features of the first type of put-in-service units and the second type of non-put-in-service units, respectively, to obtain low voltage ride-through feature target datasets corresponding to the first type of put-in-service units and the second type of non-put-in-service units, respectively, and using the low voltage ride-through feature target datasets as second-layer clustering indicators;

[0010] According to the low voltage ride-through feature target dataset, a gravitational search algorithm is used to obtain the initial centroid of the MI-GSA-K-means algorithm;

[0011] Weight calculation is performed on the low voltage ride through feature target data set using a mutual information coefficient to obtain a low voltage ride through feature weighted data set;

[0012] The low voltage ride-through feature target dataset is fed into the MI-GSA-K-means algorithm, the DBI index corresponding to different cluster numbers is iteratively calculated, and the cluster number corresponding to the minimum DBI index is selected as the optimal cluster number;

[0013] According to the low voltage ride-through feature weighted data set, the initial centroid and the optimal number of clusters, the doubly fed wind turbines in the same group after clustering are clustered into equivalent units in sequence, and a dynamic equivalent model of the wind farm based on LVRT-GSA two-stage clustering is determined.

[0014] According to some embodiments of the first aspect of the present application, obtaining dynamic response data of multiple doubly-fed wind turbines in a wind farm under typical grid fault scenarios caused by different voltage drop degrees and durations includes:

[0015] Based on the hardware-in-the-loop simulation platform, typical grid fault scenarios with different voltage drop levels, durations, and power conditions are constructed;

[0016] The dynamic response data of multiple doubly-fed wind turbines in the wind farm under the typical fault scenario of the power grid are collected.

[0017] According to some embodiments of the first aspect of the present application, the low voltage ride-through fault response features of the first-category put-in-service units and the second-category non-put-in-service units are extracted respectively to obtain low voltage ride-through feature target datasets corresponding to the first-category put-in-service units and the second-category non-put-in-service units, respectively, including:

[0018] Extracting low voltage ride-through fault response features of the first type of put-in-service units and the second type of non-put-in-service units respectively, to obtain initial low voltage ride-through feature data sets corresponding to the first type of put-in-service units and the second type of non-put-in-service units respectively;

[0019] Matrix construction processing is performed on the low voltage ride through feature initial data set to obtain a low voltage ride through feature target data set.

[0020] According to some embodiments of the first aspect of the present application, obtaining an initial centroid of the MI-GSA-K-means algorithm using a gravitational search algorithm based on the low voltage ride-through feature target dataset includes:

[0021] Randomly initializing a group of centroid positions in the low voltage ride-through feature target data set, each group of centroids including multiple centroids;

[0022] Calculating the gravitational masses between the plurality of mass centers respectively to determine the corresponding gravitational masses between the plurality of mass centers;

[0023] The positions of the multiple centroids are iteratively updated through the gravitational search algorithm until the preset maximum number of iterations is reached or the position change of the centroid is less than the preset position change threshold, and the optimization result is output as the initial centroid of the MI-GSA-K-means algorithm.

[0024] According to some embodiments of the first aspect of the present application, after iteratively calculating the DBI indices corresponding to different numbers of clusters and selecting the number of clusters corresponding to the minimum value of the DBI index as the optimal number of clusters, the method includes:

[0025] Dividing the plurality of doubly fed wind turbines into a plurality of groups according to the number of groups corresponding to the optimal number of groups;

[0026] According to the weighted Euclidean distance, the doubly fed wind turbines in each group are assigned to the cluster corresponding to the initial centroid with the smallest distance;

[0027] The weighted mean of each sample in the cluster is iteratively calculated until the initial centroid stops changing or the maximum number of iterations is reached, and the target centroid and the optimal number of clusters are output.

[0028] According to some embodiments of the first aspect of the present application, after dividing the plurality of doubly fed wind turbines into a plurality of groups according to the number of groups corresponding to the optimal number of groups, the method includes:

[0029] Calculating a silhouette coefficient of each doubly-fed wind turbine in the group, and calculating an average value of a plurality of the silhouette coefficients to obtain a silhouette coefficient mean;

[0030] The rationality of the grouping of the doubly fed wind turbines is determined according to the mean value of the silhouette coefficient.

[0031] According to some embodiments of the first aspect of the present application, aggregating the doubly-fed wind turbines in the same group after grouping into equal-value units includes:

[0032] The electrical parameters of the doubly fed wind turbines in the same group are aggregated, the control parameters are equivalent, and the dynamic characteristics are matched to determine the dynamic equivalent model of the wind farm based on the LVRT-GSA two-stage clustering.

[0033] According to some embodiments of the first aspect of the present application, after determining the wind farm dynamic equivalent model based on LVRT-GSA two-stage clustering, the method includes:

[0034] A detailed wind farm model is constructed, and the consistency of the dynamic response curve of the detailed wind farm model is compared with the dynamic response curve of the wind farm dynamic equivalent model based on the LVRT-GSA two-stage clustering to determine whether the wind farm dynamic equivalent model based on the LVRT-GSA two-stage clustering is qualified.

[0035] In a second aspect, an embodiment of the present application provides an electronic device, comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein when the processor executes the computer program, the intelligent equivalent modeling method for a wind farm as described in the first aspect above is implemented.

[0036] In a third aspect, an embodiment of the present application further provides a computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions are used to execute the wind farm intelligent equivalent modeling method as described in the first aspect above.

[0037] The beneficial effects of the present application are embodied in that, by acquiring the dynamic response data of multiple doubly fed wind turbines in a wind farm under a typical grid fault scenario, the dynamic response data includes electrical observation quantities, control signals and time series, the control signal includes the action state of the crowbar circuit, and the electrical observation quantities include rotor current, grid-connected point active power and reactive power; according to the size of the rotor current, it is judged whether the crowbar circuit is put into the doubly fed wind turbine; according to whether the crowbar circuit is put into the doubly fed wind turbine, the doubly fed wind turbines are grouped in the first layer to obtain the first type of put-in-place units and the second type of non-put-in-place units; the low voltage ride-through fault response features of the first type of put-in-place units and the second type of non-put-in-place units are extracted respectively, and the low voltage ride-through feature target data sets corresponding to the first type of put-in-place units and the second type of non-put-in-place units are obtained, and the low voltage ride-through feature target data sets corresponding to the first type of put-in-place units and the second type of non-put-in-place units are obtained, and the low voltage The crossing characteristic target data set is used as the second-layer clustering indicator; based on the low voltage crossing characteristic target data set, the gravitational search algorithm is used to obtain the initial centroid of the MI-GSA-K-means algorithm; the low voltage crossing characteristic target data set is weighted by the mutual information coefficient to obtain the low voltage crossing characteristic weighted data set; the low voltage crossing characteristic target data set is transmitted to the MI-GSA-K-means algorithm, the DBI index corresponding to different clustering numbers is iteratively calculated, and the cluster number corresponding to the minimum DBI index is selected as the optimal clustering number; based on the low voltage crossing characteristic weighted data set, the initial centroid and the optimal clustering number, the doubly fed wind turbines in the same group after clustering are successively aggregated into equivalent units, and the dynamic equivalent model of the wind farm based on the LVRT-GSA two-stage clustering is determined. Through this method, the present application takes the crowbar circuit response behavior and low voltage crossing characteristics during the fault as key clustering indicators, and combines the improved clustering algorithm to accurately divide the units, thereby improving the consistency between the equivalent model and the system fault process, which can effectively improve the model accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 A schematic flow chart of a wind farm intelligent equivalent modeling method provided in an embodiment of the first aspect of the present application;

[0039] Figure 2 This is a schematic diagram of the process of obtaining dynamic response data provided by the embodiment of the first aspect of the present application;

[0040] Figure 3 This is a schematic diagram of a process for obtaining a low voltage ride-through feature target data set according to an embodiment of the first aspect of the present application;

[0041] Figure 4 This is a schematic diagram of a process for obtaining an optimized initial centroid provided in an embodiment of the first aspect of the present application;

[0042] Figure 5 This is a schematic diagram of a process for obtaining a target centroid provided in an embodiment of the first aspect of the present application;

[0043] Figure 6 This is a flow chart of determining the rationality of grouping of doubly-fed wind turbines provided by an embodiment of the first aspect of the present application;

[0044] Figure 7 This is a schematic diagram of a process for aggregating grouped doubly-fed wind turbines according to an embodiment of the first aspect of the present application;

[0045] Figure 8 This is a schematic diagram of testing whether the wind farm dynamic equivalent model based on LVRT-GSA two-stage clustering is qualified, provided by the embodiment of the first aspect of the present application;

[0046] Figure 9 This is a flow chart of a wind farm intelligent equivalent modeling method based on an improved K-means algorithm based on LVRT characteristics provided by another embodiment of the first aspect of the present application;

[0047] Figure 10 This is a schematic diagram of the topological structure of a doubly-fed wind turbine provided in an embodiment of the first aspect of the present application;

[0048] Figure 11 This is a schematic diagram of the gravitational effect between individuals provided by the embodiment of the first aspect of the present application;

[0049] Figure 12 This is a schematic diagram of active power comparison provided by the embodiment of the first aspect of the present application;

[0050] Figure 13 This is a schematic diagram of active power comparison provided by the embodiment of the first aspect of the present application;

[0051] Figure 14 It is a structural diagram of the electronic device provided in the embodiment of the second aspect of the present application. DETAILED DESCRIPTION

[0052] The following describes in detail embodiments of the present application. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application and are not to be construed as limiting the present application.

[0053] In the description of this application, it should be understood that descriptions involving orientations, such as up, down, front, back, left, right, etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings. They are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, they cannot be understood as limitations on this application.

[0054] In the description of this application, if there is a description of first or second, it is only for the purpose of distinguishing technical features, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features or implicitly indicating the order of the indicated technical features.

[0055] In the description of this application, unless otherwise clearly defined, terms such as setting, installing, and connecting should be understood in a broad sense, and technicians in the relevant technical field can reasonably determine the specific meanings of the above terms in this application based on the specific content of the technical solution.

[0056] As wind power penetration continues to increase, the dynamic characteristics of wind farm grid connections are increasingly impacting power system stability analysis. Traditional wind farm equivalent modeling methods are primarily categorized into single-unit aggregation and multi-unit clustering. Single-unit equivalent methods, while simple in structure, fail to characterize unit dynamic differences. Multi-unit equivalent methods, drawing on the "synchronization" concept of synchronous generators, use steady-state parameters such as wind speed and power as the basis for clustering.

[0057] However, the current wind farm equivalent modeling methods constructed using the multi-machine equivalent method are usually based on a single-layer clustering strategy, ignoring the dominant role of fault characteristics on the dynamic characteristics of the unit during voltage sag, resulting in a single clustering dimension. In addition, commonly used clustering indicators (such as wind speed and pitch angle) fail to reflect the transient response differences of the unit during low voltage ride-through (LVRT). When a deep voltage sag occurs in the power grid, the doubly fed wind turbine (DFIG) achieves fault ride-through by inputting a crowbar circuit. Its transient characteristics are significantly different from those when the crowbar is not input. The constructed equivalent model is difficult to accurately describe such dynamic behavior, resulting in insufficient dynamic characteristic representation. In addition, the K-means method uses the Euclidean distance or similarity function between the unit sample and the centroid to obtain the unit clustering results. The centroid usually uses the sample mean, which is difficult to accurately represent the true dynamic characteristics of the unit, affecting the model accuracy.

[0058] In order to solve the above problems, the present application proposes a wind farm intelligent equivalent modeling method. The embodiments of the present application are further described below with reference to the accompanying drawings.

[0059] Reference Figure 1 , Figure 1 A method for intelligent equivalent modeling of a wind farm provided by an embodiment of the first aspect of the present application is shown. This method is also applied to and executed by an electronic device. In other words, the method can be executed by software or hardware installed in the device. The method includes the following steps:

[0060] Step S100: obtaining dynamic response data of multiple doubly-fed wind turbines in a wind farm under a typical grid fault scenario.

[0061] In this step, the dynamic response data includes electrical observations, control signals and time series. The control signals include the action state of the crowbar circuit. The electrical observations include rotor current, grid-connected point active power and reactive power.

[0062] It should be noted that the full name of the doubly fed wind turbine is the doubly fed asynchronous wind turbine generator, also called DFIG. The full name of DFIG is Doubly Fed Induction Generator.

[0063] It should be noted that the dynamic response data also includes stator current, stator voltage, and rotor voltage; the control signal includes the crowbar circuit status and the converter reference current; and the time series includes the time when the fault occurs, the time when the crowbar circuit is put into operation / exited, and the time when the voltage recovers.

[0064] Specifically, the topology of DFIG (doubly fed wind turbine) is as follows: Figure 10 As shown in the figure, L represents the grid-side filter inductance, u dr and u qr Represents the dq axis component of the rotor side voltage, i r represents the rotor current, ω r represents the rotor angular velocity, u dg and u qg represents the dq-axis component of the grid-side converter voltage, u g and i g Indicates the grid-side converter voltage and current, u a 、u b and u c Indicates the grid voltage, u s and i s Represent the stator voltage and stator current respectively, u ds and u qs Represents the dq axis component of the stator voltage. The mechanical power of a wind turbine comes from the conversion of wind energy by the wind turbine, and its calculation formula is as follows:

[0065]

[0066] Among them, C p The DFIG (doubly fed wind turbine) rotor side converter adopts u-based ds The directional power-current dual closed-loop control strategy realizes the independent regulation of active and reactive power. The grid-side converter adopts the voltage-current dual closed-loop control strategy based on the grid voltage orientation to achieve stable control of DC bus voltage and reactive power support function, and ensure the power quality and fault ride-through capability of the unit during grid-connected operation. DFIG (doubly fed wind turbine) stator voltage vector u s Expressed as:

[0067]

[0068] The calculation formula of the dq-axis component of the stator current is as follows:

[0069]

[0070] Among them, L s Indicates stator leakage inductance, L m represents mutual inductance, i ds and i qs represents the dq axis components of the stator current, i dr and i qr Represents the dq-axis components of the rotor-side current. The calculation formulas for the current reference values ​​of the generator rotor-side converter and the grid-side converter are as follows:

[0071]

[0072] in, and The reference value of the dq axis component of the rotor side current, k p,P and k i,P represents the active power control coefficient of the rotor-side converter, k p,Q and k i,Q Represents the reactive power control coefficient of the rotor-side converter, P s and Q s represent the active power and reactive power of the stator respectively, and are the active power and reactive power reference values ​​of the stator, k p,u and k i,u Grid-side converter voltage control coefficient, u dc and Represent the DC side voltage and the reference value of the DC side voltage respectively. According to the instantaneous reactive power theory, the expressions of the active power and reactive power on the stator side of the DFIG (doubly fed wind turbine) are:

[0073]

[0074] The rotor voltage control equation is:

[0075]

[0076] Where k p,qr and k p,dr represents the control coefficient of the q-axis component of the rotor-side converter current, k p,dr and k i,dr represents the control coefficient of the d-axis component of the rotor-side converter current, and Represents the reference value of the dq axis component of the rotor side voltage.

[0077]

[0078] in,

[0079]

[0080] Among them, ω s represents the synchronous angular velocity, L r represents the rotor leakage inductance, and s0 represents the slip.

[0081] According to the control strategies shown in equations (3) to (7), the transient response characteristics of the doubly fed wind turbine during low voltage ride-through are mainly characterized by key physical quantities such as voltage, current, and power. When a deep voltage drop occurs in the power grid, in order to prevent the rotor current from being too large, the unit cuts off the rotor-side converter and starts the Crowbar circuit to achieve protection control.

[0082] Step S200, judging whether the crowbar circuit is put into the doubly fed wind turbine according to the magnitude of the rotor current;

[0083] Step S300 , based on whether the crowbar circuit is put into use, the doubly fed wind turbines are grouped into a first layer to obtain a first type of put-in-use units and a second type of non-put-in-use units.

[0084] It's important to note that a crowbar circuit, also known as a crowbar circuit, is a hardware protection device used to protect the rotor-side converter (RSC) of a doubly-fed wind turbine. When a deep voltage sag occurs in the power grid, the rotor current may exceed the safety threshold. At this point, the crowbar circuit automatically activates, shorting the rotor windings or connecting a current-limiting resistor to absorb the excessive transient current and prevent converter damage. Its operating state (activated / deactivated) directly affects the dynamic response characteristics of the DFIG during a fault and is therefore a key factor in cluster modeling.

[0085] Specifically, the Crowbar circuit operation is analyzed: During a grid fault, to ensure safe and stable operation of the power system, wind turbines must remain connected to the grid for a period of time and maintain a certain level of reactive power support capability. When the rotor current amplitude exceeds twice its rated value, a Crowbar circuit is connected in parallel on the rotor side to absorb the excessive transient current and simultaneously lock the rotor converter. Once the current drops to a safe range, the Crowbar circuit is deactivated, and normal system control is restored. Since the activation of the Crowbar circuit depends on whether the rotor current exceeds its activation threshold during the fault, wind turbines can be divided into two categories based on this, representing two types of fleets with significantly different dynamic responses during LVRT (Low Voltage Ride Through). Based on this analysis, the Crowbar circuit response status is introduced as the first-level classification basis. Whether the rotor current exceeds the Crowbar activation threshold during the fault determines whether the turbine will activate Crowbar control. Generally speaking, during mild voltage sags, the turbine maintains normal control without activating Crowbar control. However, during severe voltage sags, the Crowbar circuit automatically activates to limit rotor overcurrent. Among units with consistent Crowbar status, the fault characteristics may still lead to certain differences in the transient response of the wind turbines. To improve the expression accuracy of the equivalent model, the units can be grouped in a second layer based on the LVRT response characteristics to achieve more refined equivalent modeling.

[0086] For example, the first level clustering is based on the Crowbar circuit operation state. The clustering basis is to determine whether the unit is put into the Crowbar circuit. If the rotor current i r If the value exceeds twice the rated value, the Crowbar action is triggered and the unit is classified as a "first-class operational unit". Otherwise, it is classified as a "second-class non-operated unit". This is to preliminarily separate units with significant dynamic differences and narrow the scope of subsequent clustering.

[0087] In step S400, the low voltage ride-through fault response characteristics of the first type of put-in-service units and the second type of non-put-in-service units are extracted respectively to obtain the low voltage ride-through feature target data sets corresponding to the first type of put-in-service units and the second type of non-put-in-service units, and the low voltage ride-through feature target data sets are used as the second-layer clustering indicators.

[0088] It should be noted that the low voltage ride-through characteristic is referred to as the LVRT characteristic. LVRT is the abbreviation of Low Voltage Ride Through.

[0089] It should be noted that among units with similar Crowbar circuit states, the units are grouped in a second layer based on the low voltage ride-through response characteristics (stator voltage, rotor current, grid-side current, active power and reactive power, etc.) to achieve more refined equivalent modeling. The two-layer grouping method based on Crowbar state and low voltage ride-through characteristics can effectively improve the accuracy and representativeness of wind farm aggregation modeling during fault transient processes.

[0090] Step S500 : According to the low voltage ride-through feature target dataset, the initial centroid of the MI-GSA-K-means algorithm is obtained using the gravitational search algorithm.

[0091] It should be noted that in clustering algorithms (such as K-means), the centroid is the data point representing the center position of a cluster, which is usually obtained by calculating the mean value of all samples in the cluster.

[0092] Step S600 : performing weight calculation on the low voltage ride through feature target dataset using the mutual information coefficient to obtain a low voltage ride through feature weighted dataset.

[0093] It should be noted that the K-means algorithm improved by the mutual information coefficient in this application is used to measure the importance of different low voltage ride-through characteristics. Mutual information (MI) is an important indicator in information theory that measures the statistical dependence between two random variables. It can capture linear or nonlinear associations. MI can be used to measure the impact of the grid connection point voltage drop on the unit's LVRT (low voltage ride-through) response:

[0094] I(X;Y)=H(X)-H(X|Y),

[0095] Among them, X and Y represent variables with a certain correlation, H(X) and H(X|Y) represent the information entropy and conditional entropy of variable X respectively. The calculation formulas of H(X) and H(X|Y) are as follows:

[0096]

[0097] Where p(x) is the probability that X=x, p(y) is the probability that Y=y, and p(x|y) is the probability that X=x under the condition that Y=y.

[0098] In step S700 , the low voltage ride-through feature target dataset is fed into the MI-GSA-K-means algorithm, the DBI index corresponding to different cluster numbers is iteratively calculated, and the cluster number corresponding to the minimum DBI index is selected as the optimal cluster number.

[0099] It should be noted that the full name of the MI-GSA-K-means algorithm is Mutual Information and Gravitational Search Optimized K-means Algorithm. The naming analysis is as follows: MI (mutual information coefficient): reflects the dynamic weighting mechanism of feature importance and enhances the role of key LVRT characteristics in clustering; GSA (gravitational search algorithm): reflects the optimization of the initial centroid position of the clustering algorithm through gravitational search, avoiding the traditional K-means' dependence on initial parameters; K-means: retains the core framework of the original algorithm and clarifies the method foundation.

[0100] It should be noted that the full name of the DBI index is the Davies-Bouldin index. The DBI index is an indicator used to evaluate the quality of clustering results. It can comprehensively consider the degree of compactness within the cluster and the degree of separation between clusters. For better clustering results, the higher the compactness within the cluster, the greater the degree of separation between clusters. j The center of mass is μ j , C j The sample in is denoted as w ji (i=1,…,j0), then C j The average distance from all samples to the centroid is defined as the divergence S j for:

[0101]

[0102] Different clusters and The calculation formula of the centroid distance between is as follows:

[0103]

[0104] in, and express and The value of the centroid of on feature t.

[0105] The similarity calculation formula is as follows:

[0106]

[0107] The calculation formula of the DBI index is as follows: If the DBI index value is smaller, it means that the current clustering effect is better, and the value that minimizes DBI is selected as the optimal number of clusters.

[0108] Step S800: According to the low voltage ride-through feature weighted data set, the initial centroid and the optimal number of clusters, the doubly fed wind turbines in the same group after clustering are clustered into equivalent units, and a dynamic equivalent model of the wind farm based on LVRT-GSA two-stage clustering is determined.

[0109] In one possible implementation, refer to Figure 7 The step S800 of aggregating the doubly fed wind turbines in the same group after grouping into equivalent units includes:

[0110] In step S810, electrical parameter aggregation, control parameter equivalence, and dynamic characteristic matching are performed on each doubly-fed wind turbine in the same group after grouping.

[0111] Exemplarily, for each subgroup in the final grouping result: electrical parameter aggregation is performed: rotor resistance and stator inductance of parallel equivalent units; control parameter equivalence is performed: weighted average voltage loop PI coefficient and current loop PI coefficient; dynamic characteristic matching is performed: the consistency of the LVRT characteristics of the equivalent units and subgroups is analyzed.

[0112] It should be noted that, compared with the traditional correlation coefficient, the wind farm intelligent equivalent modeling method provided by this application can identify nonlinear, non-monotonic and even complex functional relationships based on mutual information, and is applicable to a wider range of data feature analysis scenarios. It does not depend on the numerical size or unit scale of the variable, has strong applicability, and can accurately reflect the correlation between the voltage drop degree and the low voltage ride-through characteristics. In the clustering algorithm, the introduction of a feature weighting mechanism helps to highlight the impact of key features on the clustering results, thereby improving the accuracy and stability of clustering. By weighting the importance of features, a more reasonable distance measurement method can be achieved, making the clustering process more sensitive to key features, especially when processing high-dimensional or complex response data, effectively enhancing the model's distinguishing ability and adaptability. The gravitational search algorithm is used to adaptively select cluster centers without presetting the number of clusters or initial centers, and the silhouette coefficient is used as a fitness function to determine the optimal number of clusters, effectively overcoming the problem of traditional K-means' strong dependence on initial parameters. This method has a stronger global search capability and can avoid falling into local optimality. It is suitable for clustering tasks that process complex shape distributions or high-dimensional data.

[0113] It should be noted that this application introduces the Crowbar circuit response state as the basis for the first-level clustering, and combines it with the LVRT dynamic characteristics for secondary clustering; uses the mutual information coefficient to improve the feature weighting mechanism and enhance the clustering sensitivity of key response parameters; combines the gravitational search algorithm (GSA) to optimize the initial center of mass, and adaptively determines the optimal number of clusters through the Davies-Bouldin index (DBI) to avoid manual intervention and local optimal problems.

[0114] For example, refer to Figure 9The provided wind farm intelligent equivalent modeling method based on the improved K-means algorithm based on LVRT characteristics includes the following steps: step 1: start; step 2: obtain the LVRT characteristics of N doubly fed wind turbine units; step 3: set the initial value of the cluster number K=2; step 4: use GSA to optimize the initial centroid of K clusters; step 5: cluster the N units considering the Crowbar circuit response; step 6: perform secondary clustering considering the LVRT characteristics; step 7: determine whether the centroid of the clustered cluster is updated, if updated, return to step 5, if not updated, execute step 8; step 8: clustering is completed and the grouping results of N units are output; step 9: calculate the DBI index of the current grouping result; step 10: determine whether K is equal to N, if K is not equal to N, command K=K+1 and return to step 3, if K=N, execute step 11; step 11: select the K value corresponding to min(DBI); step 12: calculate the multi-machine equivalent parameters; step 13: end.

[0115] That is, the wind farm intelligent equivalent modeling method provided in this application is used for the multi-machine equivalent modeling process of the wind farm. Compared with the unit clustering results obtained by the traditional K-means algorithm, this method introduces the LVRT (low voltage ride through) dynamic response characteristics as the basis for clustering, and effectively reduces the influence of human subjective factors on the clustering results through improved strategies, thereby constructing a wind farm multi-machine equivalent model that can accurately reflect the LVRT (low voltage ride through) behavior of the unit.

[0116] The present application obtains dynamic response data of multiple doubly fed wind turbines in a wind farm under typical grid fault scenarios, the dynamic response data including electrical observation quantities, control signals and time series, the control signals including the action state of the crowbar circuit, the electrical observation quantities including rotor current, grid-connected active power and reactive power; according to the magnitude of the rotor current, it is determined whether the crowbar circuit is put into the doubly fed wind turbine; according to whether the crowbar circuit is put into the doubly fed wind turbine, the doubly fed wind turbines are grouped in the first layer to obtain the first type of put-in units and the second type of non-put-in units; the low voltage ride-through fault response features of the first type of put-in units and the second type of non-put-in units are extracted respectively, and the low voltage ride-through feature target data sets corresponding to the first type of put-in units and the second type of non-put-in units are obtained, and the low voltage ride-through feature target data sets are grouped into the first layer. The target data set is used as the second-layer clustering indicator; based on the low voltage crossing characteristic target data set, the gravitational search algorithm is used to obtain the initial centroid of the MI-GSA-K-means algorithm; the low voltage crossing characteristic target data set is weighted by the mutual information coefficient to obtain the low voltage crossing characteristic weighted data set; the low voltage crossing characteristic target data set is transmitted to the MI-GSA-K-means algorithm, the DBI index corresponding to different clustering numbers is iteratively calculated, and the number of clusters corresponding to the minimum value of the DBI index is selected as the optimal clustering number; based on the low voltage crossing characteristic weighted data set, the initial centroid and the optimal clustering number, the doubly fed wind turbines in the same group after clustering are successively aggregated into equivalent units, and the dynamic equivalent model of the wind farm based on the LVRT-GSA two-stage clustering is determined. Through this method, the present application takes the crowbar circuit response behavior and low voltage crossing characteristics during the fault as key clustering indicators, and combines the improved clustering algorithm to accurately divide the units, thereby improving the consistency between the equivalent model and the system fault process, which can effectively improve the model accuracy.

[0117] It is understandable that, referring to Figure 2 Step S100 includes but is not limited to the following steps:

[0118] Step S110: constructing typical power grid fault scenarios under different voltage drop levels, different durations, and different power conditions based on a hardware-in-the-loop simulation platform;

[0119] Step S120 , collecting dynamic response data of multiple doubly-fed wind turbines in the wind farm under typical grid fault scenarios.

[0120] It should be noted that a DFIG hardware-in-the-loop simulation test platform was built to simulate typical grid fault scenarios (such as different voltage drop levels, different durations, and different power conditions) and collect the following dynamic response data during the fault period: Voltage drop level h: record the instantaneous voltage drop amplitude of the grid connection point (h = voltage at fault / rated voltage); Unit response time: the time delay from voltage drop to unit starting Crowbar or adjusting the control strategy; Rotor current transient peak value and duration: collect the rotor current dq axis component (i dr ,i qr ) and the duration that it exceeds the threshold; active power recovery rate: the slope of the active power P recovering from the minimum value to the steady-state value during the fault recovery period (ΔP / Δt); reactive power support capacity: the maximum instantaneous value and duration of the reactive power Q during the fault period; Crowbar action threshold and operation duration: records the Crowbar triggering conditions (such as rotor current > 2 times the rated current) and the actual operation time window.

[0121] It is understandable that, referring to Figure 3 The step S400 of extracting the low voltage ride through fault response features of the first type of put-in-service units and the second type of non-put-in-service units respectively to obtain the low voltage ride through feature target data sets corresponding to the first type of put-in-service units and the second type of non-put-in-service units respectively includes but is not limited to the following steps:

[0122] Step S410: extracting low voltage ride-through fault response features of the first type of put-in-service units and the second type of non-put-in-service units, respectively, to obtain initial low voltage ride-through feature data sets corresponding to the first type of put-in-service units and the second type of non-put-in-service units, respectively;

[0123] Step S420 : performing matrix construction processing on the initial low voltage ride through feature data set to obtain a low voltage ride through feature target data set.

[0124] For example, different voltage drop degrees (such as 20% to 80% of the rated voltage), different fault durations (such as 100ms to 1s), and different power conditions (such as 0.2 rated active power or 1.0 rated active power)).

[0125] Specifically, at time t0, a fault occurs at the grid connection point, resulting in a voltage drop of h. The transient process of the DFIG is mainly divided into the transient process of the asynchronous induction motor and the transient process of the back-to-back converter:

[0126]

[0127] Among them, ψ s represents the stator flux, ψ rThe main manifestation of the motor in the transient process is the dynamic response of the stator and rotor voltage and current. When a low voltage fault occurs in the power grid, the stator flux is disturbed. Affected by the strong coupling relationship between the stator flux and the rotor, the rotor flux changes synchronously, which in turn causes the rotor current amplitude to increase. The relationship is shown in the formula:

[0128]

[0129] Among them, i s represents the stator current, i r Indicates the rotor current.

[0130] In the two-phase rotating coordinate system, the reference value of the dq-axis component of the rotor side current is:

[0131]

[0132] Among them, u Lin Indicates the LVRT (low voltage ride-through) voltage threshold, U N Indicates the voltage rating, P0 indicates the active power before the fault, i r_lim Represents the output current limit value of the rotor-side converter, and k represents the reactive current support coefficient. The transient process of the converter is mainly reflected in the coordinated control and power interaction between the rotor-side converter (RSC) and the grid-side converter (GSC). In the transient stage between the occurrence of the fault and the input of the Crowbar circuit, the dynamic behavior of the system is jointly determined by the induction motor, GSC (grid-side converter) and RSC (rotor-side converter). During the low-voltage ride-through process, the core goal of the converter control is to suppress the rotor current fluctuation and control the rotor-side power output. The power of the grid-connected point is determined by the stator voltage, rotor current and GSC (grid-side converter) output current. The transient characteristics during the fault can be reflected by the changes in the active power P and reactive power Q at the grid-connected point:

[0133]

[0134] To identify wind turbines with similar dynamic response characteristics, voltage disturbances of varying amplitude, duration, and drop depth are designed, and multiple LVRT operating conditions are constructed to simulate grid anomalies. Under each fault condition, dynamic response data from each grid-connected turbine is collected, including key parameters such as active power, reactive power, and the d / q components of the rotor and stator currents. This data supports subsequent cluster analysis and equivalent modeling, helping to reveal differences in the turbine's dynamic response under different operating conditions.

[0135] It should be noted that the LVRT dynamic response characteristic modeling: establish a mathematical model of the DFIG during the LVRT process to reveal the transient response mechanism. Steps: Stator-rotor electromagnetic relationship analysis: Based on equations (1)-(8), the stator voltage vector equation, power equation, and converter control equation are established; the dynamic coupling relationship between the stator flux and the rotor flux during the fault is analyzed (Equations 13-14); converter transient control strategy analysis: the rotor-side converter (RSC) switches to the Crowbar protection mode during the LVRT period and locks the converter control; the grid-side converter (GSC) maintains the DC bus voltage stability and provides reactive power support; key response feature extraction: rotor current overshoot (Equation 15) and recovery time; grid-connected power fluctuation amplitude; Crowbar circuit action threshold (e.g., triggered when the rotor current is >2 times the rated current). The extracted features are normalized to eliminate dimensionality effects.

[0136] It is understandable that, referring to Figure 4 Step S500 includes but is not limited to the following steps:

[0137] Step S510 : randomly initializing a set of centroid positions in the low voltage ride-through feature target data set.

[0138] In this step, each group of centroids includes multiple centroids, each of which represents a potential cluster center.

[0139] Step S520, calculating the gravitational masses between the multiple mass centers respectively to determine the corresponding gravitational masses between the multiple mass centers;

[0140] In step S530, the positions of multiple centroids are iteratively updated through the gravitational search algorithm until the preset maximum number of iterations is reached or the position change of the centroid is less than the preset position change threshold, and the optimization result is output as the initial centroid of the MI-GSA-K-means algorithm.

[0141] Specifically, refer to Figure 11 The center of mass optimization method based on GSA (gravitational search algorithm) is as follows: There is a gravitational force between all particles in the universe, and the larger the mass of the particles and the closer the distance between them, the stronger the gravitational force. The Gravitational Search Algorithm (GSA) simulates the interaction between the law of universal gravitation and mass, guiding the individual to gradually approach the optimal solution in the search space. In GSA, the center of mass to be initialized is regarded as an object with mass, the size of its mass reflects its fitness level, and has a position μ, an inertial mass m i , active gravitational mass m a and the passive gravitational mass m pFour characteristics. Individuals interact with each other through gravity, pushing all solutions in the system toward individuals with larger mass. The larger the mass, the better the solution, and the stronger the gravity, which can guide other individuals with smaller mass to move towards it. At the same time, individuals with larger mass move slower, which helps to balance the efficiency of the algorithm in the global exploration and local development stages. Figure 11 As shown in the figure, the increase in the distance between individuals will lead to the weakening of the gravitational force between them, thereby reducing their mutual attraction. 12 and F 13 It is m i,1 Acts on m i,2 and m i,3 The force on the center of mass. Let the number of centers of mass be N, and the calculation formula for its position is as follows:

[0142]

[0143] Where, and Represents the position information of the centroid 1 in dimension 1 and dimension T, and It is the position information of the center of mass N in dimension 1 and dimension T.

[0144] In the gravitational search algorithm, the gravitational mass m a 、m p and inertial mass m i They are all calculated through fitness, and their relationship is usually shown in Equation (9). The larger the mass, the stronger the centroid's attraction in the search space. At the same time, its movement speed is relatively slow, which is conducive to guiding the group to converge to a better solution.

[0145] m a,j =m p,j =m i,j =m j (9)

[0146] Among them, m a,j 、m p,j and m i,j are the active gravitational mass, passive gravitational mass and inertial mass of the center of mass j, m j The calculation formula is as follows:

[0147]

[0148] Among them, f j is the fitness of the centroid j, is the sum of the masses of all centroids in cluster j.

[0149] The gravitational force between mass j1 and mass j2 is:

[0150]

[0151] in,

[0152]

[0153] in, represents the active gravitational mass of the center of mass j2, represents the passive gravitational mass of the mass center j1, ε is a constant, R(j1,j2) represents the Euclidean distance between the mass center j1 and the mass center j2, G is the gravitational coefficient, and Indicates the position information of the centroid j1 and the centroid j2 in dimension t. To enhance the randomness of the algorithm, the centroid Gravity in dimension t The calculation formula is as follows:

[0154]

[0155] GSA needs to focus on global exploration in the early stage of the algorithm. As the iteration proceeds, the local development capability should be gradually enhanced. The centroid of the first N0 groups with the best fitness in each iteration is recorded as μ best , and rewrite the above formula as:

[0156]

[0157] speed The update formula is: in, is the acceleration, and its expression is:

[0158]

[0159] Then the position update result of the center of mass j1 is:

[0160]

[0161] in, represents a random number in the range [0,1], represents the inertial mass of the center of mass j1.

[0162] It is understandable that, referring to Figure 5 After iteratively calculating the DBI indices corresponding to different numbers of clusters in step S700 and selecting the number of clusters corresponding to the minimum DBI index as the optimal number of clusters, the following steps are included but are not limited to:

[0163] Step S710, dividing the plurality of doubly fed wind turbines into a plurality of groups according to the number of groups corresponding to the optimal number of groups;

[0164] Step S720, allocating the DFIGs of each group to the cluster corresponding to the initial centroid with the smallest distance based on the weighted Euclidean distance;

[0165] Step S730: Iteratively calculate the weighted mean of samples in each cluster until the initial centroid stops changing or the maximum number of iterations is reached, and output the target centroid and the optimal number of clusters.

[0166] It should be noted that the traditional K-means algorithm calculates the Euclidean distance or other distance metrics between each data point and all cluster centers and assigns it to the cluster corresponding to the nearest cluster center. i …, w z ], the number of clusters is K, the centroid of each cluster [μ1,…,μ j …, μ k ]The above process is shown in formula (10).

[0167] ||w i -μ j ||2<||w i -μ j+1 ||2 (10)

[0168] Among them, w i is the i-th unit, μ j and μ j+1 denote the jth and j+1th centroids respectively, ||w i -μ j ||2 means w i to μ j The Euclidean distance, ||w i -μ j+1 ||2 means w i to μ j+1 For each cluster, use formula (11) to calculate the average value of all data points in each feature dimension, and use the mean vector as the new centroid of the cluster to update the position of the next round of clustering center:

[0169]

[0170] Among them, C j represents the jth cluster, w i Indicates C j Unit i, j0 represents C jThe number of units in the MI-based improved K-means algorithm introduces feature importance weights on the basis of formula (10), that is, in the process of calculating the Euclidean distance, different weights are assigned to each feature, thereby enhancing the response sensitivity to key variables, achieving more reasonable data point attribution judgment, and improving the distinguishing ability of clustering results in feature dimensions and the overall clustering accuracy. In the clustering process of the units, the low voltage ride-through feature is used to construct a dynamic feature weighting mechanism to enhance the clustering accuracy of key response features, as shown in formula (12):

[0171]

[0172] in, and Respectively represent w i and μ j The value of LVRT response t, ω t Represents the weight of unit i in response t.

[0173] It is understandable that, referring to Figure 6 After step S710, the following steps are included but not limited to:

[0174] Step S711, calculating the silhouette coefficient of each doubly-fed wind turbine in the group, and calculating the average value of multiple silhouette coefficients to obtain a silhouette coefficient mean;

[0175] Step S712: determining the rationality of the grouping of the doubly fed wind turbines based on the mean value of the silhouette coefficient.

[0176] Specifically, in order to further verify the rationality and effectiveness of the proposed method, the silhouette coefficient with a value range of [-1,1] is introduced as an evaluation of the doubly fed wind turbine clustering results to accurately analyze its rationality. According to the clustering results, the silhouette coefficient k of sample i is i It can be expressed as:

[0177] k i =min(b)-a / max[a,min(b)],

[0178] Where a is the average distance between sample i and other samples in the same group, and b is the average distance between sample i and other samples in different groups. i The larger the value, the better the clustering effect of the proposed method on wind turbines. Conversely, it means that there is room for further improvement in the current clustering method. The closer the mean value of the silhouette coefficient of all samples is to 1, the more reasonable the overall clustering result is. In other words, the silhouette coefficient is used to verify the rationality of the improved K-means clustering results based on LVRT characteristics and ensure that the dynamic response differences of the units are accurately captured.

[0179] That is, the process of the wind farm intelligent equivalent modeling method based on the MI-GSA-K-means algorithm improved by LVRT characteristics in this application is as follows: Overall process: By combining the dynamic characteristics of low voltage ride-through (LVRT) with the improved K-means clustering algorithm, the doubly fed wind turbines (DFIG) in the wind farm are clustered to build a high-precision multi-machine equivalent model. Compared with the traditional method, the improvements include: Clustering basis: Introducing the Crowbar circuit response state and LVRT dynamic characteristics as double-layer clustering indicators; Algorithm improvement: Based on the feature weighting mechanism of mutual information (MI), the influence of key features on clustering is enhanced; The gravitational search algorithm (GSA) is used to optimize the initial centroid to avoid the problem of traditional K-means being sensitive to initial values; The optimal number of clusters is determined adaptively through the DBI index to avoid the limitations of manual presets; Verification mechanism: The silhouette coefficient is used to evaluate the rationality of clustering.

[0180] It is understandable that, referring to Figure 8 After step S800, the following steps are included but not limited to:

[0181] Step S900: construct a detailed wind farm model, compare the dynamic response curve of the detailed wind farm model with the dynamic response curve of the wind farm dynamic equivalent model based on LVRT-GSA two-stage clustering, and determine whether the wind farm dynamic equivalent model based on LVRT-GSA two-stage clustering is qualified.

[0182] It should be noted that, referring to Figures 12 to 13 To verify the effectiveness of the proposed clustering method, we use Figure 9 The equivalent model generated by the process shown in the figure was tested, and the dynamic response curves (active / reactive power and current) of the equivalent model were compared with those of the detailed model. The evaluation metric was the mean square error (MSE). The results showed that the response power curve of the equivalent model constructed based on the proposed method was highly consistent with that of the detailed model, and it could better reflect the dynamic characteristics of the original system.

[0183] It should be noted that the dynamic equivalent model of wind farms based on LVRT-GSA two-stage clustering is a multi-machine equivalent model based on clustering using the improved MI-GSA-K-means algorithm. Each equivalent unit represents a group of actual units with similar LVRT characteristics, and the model complexity is simplified through parameter aggregation.

[0184] It should be noted that the sources of the detailed model are: physical modeling: based on the physical equations of DFIG (such as Park transformation, magnetic flux equation) and control strategies (such as PI regulator parameters), it relies on the equipment parameters provided by the manufacturer; data-driven modeling: dynamic response data during the fault period is obtained through on-site measurement or high-precision offline simulation (such as EMTP, PSCAD), and the model parameters are fitted using system identification methods (such as least squares method).

[0185] Alternatively, as Figure 14 As shown, the second embodiment of the present application further provides an electronic device 10, including a processor 11 and a memory 12, and the memory 12 stores a program or instruction that can be run on the processor 11. When the program or instruction is executed by the processor 11, the various processes of the embodiment of the wind farm intelligent equivalent modeling method of the first aspect above are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0186] It should be noted that the devices in the embodiments of the present application include: servers, terminals, or other devices other than terminals.

[0187] The above device structure does not constitute a limitation of the device. The device may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently. For example, the input unit may include a graphics processing unit (GPU) and a microphone, and the display unit may be configured in the form of a liquid crystal display, an organic light emitting diode, etc. to configure the display panel. The user input unit includes a touch panel and at least one of other input devices. The touch panel is also called a touch screen. Other input devices may include but are not limited to a physical keyboard, function keys (such as volume control buttons, switch buttons, etc.), a trackball, a mouse, and a joystick, which will not be repeated here.

[0188] The memory can be used to store software programs and various data. The memory may mainly include a first storage area for storing programs or instructions and a second storage area for storing data, wherein the first storage area may store an operating system, applications or instructions required for at least one function (such as a sound playback function, an image playback function, etc.), etc. In addition, the memory may include a volatile memory or a non-volatile memory, or the memory may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. Volatile memory can be random access memory (RAM), static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM) and direct rambus random access memory (DRRAM).

[0189] The processor may include one or more processing units; optionally, the processor may integrate an application processor and a modem processor, wherein the application processor primarily handles operations related to the operating system, user interface, and application programs, and the modem processor primarily processes wireless communication signals, such as a baseband processor. It is understood that the modem processor may not be integrated into the processor.

[0190] The present application also provides a readable storage medium on which a program or instruction is stored. When the program or instruction is executed by a processor, each process of the embodiment of the wind farm intelligent equivalent modeling method of the first aspect is implemented, and the same technical effect is achieved. To avoid repetition, it is not described here. The processor is the processor in the device in the above embodiment. The readable storage medium includes a computer-readable storage medium such as ROM, RAM, magnetic disk or optical disk. It should be noted that, in this article, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "including one..." does not exclude the presence of other identical elements in the process, method, article or device including the element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of the present application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in reverse order depending on the functions involved. For example, the described methods may be performed in an order different from that described. In addition, features described with reference to certain examples may be combined in other examples.

[0191] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a computer software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), including a number of instructions for enabling a terminal (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods in each embodiment of the present application.

[0192] In the description of the embodiments of the present application, the terms "first," "second," "third," and "fourth" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the technical features indicated. Thus, a feature defined as "first," "second," "third," or "fourth" may explicitly or implicitly include one or more of the features. In the description of the present application, unless otherwise specified, "plurality" means two or more.

[0193] In the description of the embodiments of the present application, specific features, structures, materials or characteristics may be combined in an appropriate manner in any one or more embodiments or examples.

[0194] Although the embodiments of the present application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations may be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A wind farm intelligent equivalent modeling method, characterized in that: include: Obtain dynamic response data of multiple doubly-fed wind turbines in a wind farm under typical grid fault scenarios, the dynamic response data including electrical observations, control signals, and time series. The control signals include the operating state of the crowbar circuit, and the electrical observations include rotor current, grid-connection point active power, and grid-connection point reactive power. determining whether the crowbar circuit is put into operation of the doubly-fed wind turbine according to the magnitude of the rotor current; According to whether the crowbar circuit is put into use of the doubly fed wind turbine, the doubly fed wind turbine is grouped into a first layer to obtain a first type of put-in-use units and a second type of non-put-in-use units; Extracting low voltage ride-through fault response features of the first type of put-in-service units and the second type of non-put-in-service units, respectively, to obtain low voltage ride-through feature target datasets corresponding to the first type of put-in-service units and the second type of non-put-in-service units, respectively, and using the low voltage ride-through feature target datasets as second-layer clustering indicators; According to the low voltage ride-through feature target dataset, a gravitational search algorithm is used to obtain the initial centroid of the MI-GSA-K-means algorithm; Weight calculation is performed on the low voltage ride through feature target data set using a mutual information coefficient to obtain a low voltage ride through feature weighted data set; The low voltage ride-through feature target dataset is fed into the MI-GSA-K-means algorithm, the DBI index corresponding to different cluster numbers is iteratively calculated, and the cluster number corresponding to the minimum DBI index is selected as the optimal cluster number; According to the low voltage ride-through feature weighted data set, the initial centroid and the optimal number of clusters, the doubly fed wind turbines in the same group after clustering are clustered into equivalent units in sequence, and a dynamic equivalent model of the wind farm based on LVRT-GSA two-stage clustering is determined.

2. The wind farm intelligent equivalent modeling method according to claim 1, characterized in that: The acquisition of dynamic response data of multiple doubly-fed wind turbines in a wind farm under typical grid fault scenarios caused by different voltage drop degrees and durations includes: Based on the hardware-in-the-loop simulation platform, typical grid fault scenarios with different voltage drop levels, durations, and power conditions are constructed; The dynamic response data of multiple doubly-fed wind turbines in the wind farm under the typical fault scenario of the power grid are collected.

3. The wind farm intelligent equivalent modeling method according to claim 1, characterized in that: The low voltage ride-through fault response features of the first type of put-in-service units and the second type of non-put-in-service units are extracted respectively to obtain low voltage ride-through feature target data sets corresponding to the first type of put-in-service units and the second type of non-put-in-service units, respectively, including: Extracting low voltage ride-through fault response features of the first type of put-in-service units and the second type of non-put-in-service units respectively, to obtain initial low voltage ride-through feature data sets corresponding to the first type of put-in-service units and the second type of non-put-in-service units respectively; Matrix construction processing is performed on the low voltage ride through feature initial data set to obtain a low voltage ride through feature target data set.

4. The wind farm intelligent equivalent modeling method according to claim 1, characterized in that: The method of obtaining the initial centroid of the MI-GSA-K-means algorithm using a gravitational search algorithm based on the low voltage ride-through feature target dataset includes: Randomly initializing a group of centroid positions in the low voltage ride-through feature target data set, each group of centroids including multiple centroids; Calculating the gravitational masses between the plurality of mass centers respectively to determine the corresponding gravitational masses between the plurality of mass centers; The positions of the multiple centroids are iteratively updated through the gravitational search algorithm until the preset maximum number of iterations is reached or the position change of the centroid is less than the preset position change threshold, and the optimization result is output as the initial centroid of the MI-GSA-K-means algorithm.

5. The wind farm intelligent equivalent modeling method according to claim 4, characterized in that: After iteratively calculating the DBI indices corresponding to different numbers of clusters and selecting the number of clusters corresponding to the minimum value of the DBI index as the optimal number of clusters, the method further includes: Dividing the plurality of doubly fed wind turbines into a plurality of groups according to the number of groups corresponding to the optimal number of groups; According to the weighted Euclidean distance, the doubly fed wind turbines in each group are assigned to the cluster corresponding to the initial centroid with the smallest distance; The weighted mean of each sample in the cluster is iteratively calculated until the initial centroid stops changing or the maximum number of iterations is reached, and the target centroid and the optimal number of clusters are output.

6. The wind farm intelligent equivalent modeling method according to claim 5, characterized in that: After dividing the plurality of doubly-fed wind turbines into a plurality of groups according to the number of groups corresponding to the optimal number of groups, the method includes: Calculating a silhouette coefficient of each doubly-fed wind turbine in the group, and calculating an average value of a plurality of the silhouette coefficients to obtain a silhouette coefficient mean; The rationality of the grouping of the doubly fed wind turbines is determined according to the mean value of the silhouette coefficient.

7. The wind farm intelligent equivalent modeling method according to claim 1, characterized in that: The step of aggregating the doubly-fed wind turbines in the same group after grouping into equal-value units includes: The electrical parameters of the doubly fed wind turbines in the same group are aggregated, the control parameters are equivalent, and the dynamic characteristics are matched to determine the dynamic equivalent model of the wind farm based on the LVRT-GSA two-stage clustering.

8. The wind farm intelligent equivalent modeling method according to claim 1, characterized in that: After determining the wind farm dynamic equivalent model based on the LVRT-GSA two-stage clustering, the method includes: A detailed wind farm model is constructed, and the consistency of the dynamic response curve of the detailed wind farm model is compared with the dynamic response curve of the wind farm dynamic equivalent model based on the LVRT-GSA two-stage clustering to determine whether the wind farm dynamic equivalent model based on the LVRT-GSA two-stage clustering is qualified.

9. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method for intelligent equivalent modeling of a wind farm according to any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to: implement the wind farm intelligent equivalent modeling method according to any one of claims 1 to 8.

Images