Incremental learning CNN (Convolutional Neural Network)-based broadband impedance online identification method for network-constructed wind power plant

Abstract

The invention relates to an incremental learning CNN-Attention-based broadband impedance on-line identification method for a network construction type wind power plant, and belongs to the technical field of new energy grid connection, and the method comprises the following steps: S1, building a parallel collection type offshore wind power plant multi-machine dynamic equivalent model; s2, collecting and preprocessing parameters based on a multi-machine equivalent model of the network-building type wind power plant; s3, a CNN-Attention model is constructed and trained; s4, constructing an equivalent impedance model of the current collection submarine cable line between the wind turbine generators; s5, on the basis of the actual topology of the network construction type wind power plant station, connecting the broadband impedance models of the wind turbine generator and the current collection circuit into a frequency domain impedance network, and calculating the impedance characteristics of the current collection port of the multi-machine equivalent wind power plant; s6, a CNN-Attention model containing an incremental learning updating mechanism is constructed; and S7, performing broadband impedance on-line identification on the network construction type wind power plant.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of new energy grid connection, and relates to a wide-frequency impedance online identification method for a grid-connected wind farm based on incremental learning CNN. BACKGROUND

[0002] The grid-connected wind turbine has the ability to actively support the voltage and frequency of the power grid, and has good stability when connected to a weak power grid. However, the grid-connected wind turbine also has stability problems when connected to a strong power grid, and the stability margin is closely related to the grid-connected control parameters, the wind turbine tie line and the operating conditions. Therefore, it is of great significance to invent a wide-frequency impedance online identification method for a grid-connected wind farm to identify the operating stability margin of the grid-connected wind farm.

[0003] At present, the impedance method is mostly used for modeling of the wind farm grid-connected system. However, due to the black-box nature of commercial wind turbines, it is difficult to obtain all the structures and parameters of the wind turbine in actual engineering, and it is difficult to establish an accurate impedance model. Moreover, the wind turbine has random and uncertain inputs, making it difficult to obtain the output impedance characteristics of the wind turbine, and unable to analyze the stability of the wind farm grid connection in a timely manner. In recent years, many scholars have used impedance frequency response measurement and artificial intelligence methods to identify the impedance of black-box wind turbines. The measurement method has a large lag and is difficult to determine the stability of the wind farm grid connection in real time. The artificial intelligence method relies too much on data samples when facing large-scale wind farm grid-connected systems, and the training sample size is huge, which cannot guarantee the training accuracy and generalization ability of the data model, and it is difficult to accurately identify the impedance of the wind farm. At the same time, it cannot adaptively learn new sample data, which may lead to a decrease in the accuracy of the identification model. SUMMARY

[0004] Therefore, the purpose of the present application is to provide a wide-frequency impedance online identification method for a grid-connected wind farm based on incremental learning CNN-Attention.

[0005] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0006] A wide-frequency impedance online identification method for a grid-connected wind farm based on incremental learning CNN-Attention, comprising the following steps:

[0007] S1: Based on the K-means clustering grouping method of the dominant variable and the unit operating characteristics, a multi-machine dynamic equivalent model of a parallel collection type offshore wind farm is established;

[0008] S2: Based on the multi-machine equivalent model of the grid-connected wind farm, different operating power conditions, control parameters and line length ranges are set, and each parameter is collected and preprocessed;

[0009] S3: Constructing a basic convolutional neural network structure, then introducing a squeeze-and-excitation module SE to provide a channel attention mechanism, obtaining a CNN-Attention model, and then training and verifying the CNN-Attention model using a data set;

[0010] S4: Constructing an equivalent impedance model of a power collection submarine cable line between wind turbines;

[0011] S5: Based on the actual topology of the network-based wind power station, connecting the broadband impedance models of the wind turbines and the power collection line into a frequency domain impedance network, and calculating the impedance characteristics of the power collection port of the multi-turbine equivalent wind power station;

[0012] S6: Constructing a CNN-Attention model containing an incremental learning update mechanism;

[0013] S7: Multi-level equivalent modeling is performed on real-time data of wind turbines, and then the equivalent wind turbine model data is input into the trained CNN-Attention model containing an incremental learning update mechanism, to obtain the real-time broadband impedance characteristics of each wind turbine; finally, the broadband impedance of the large-scale wind power station is obtained through an impedance network aggregation method; during the identification process, when the model output error rises, an automatic update mechanism is started to realize real-time adaptation to new samples.

[0014] Further, step S1 specifically comprises the following steps:

[0015] S11: Using a wind farm supervisory control and data acquisition system SCADA to obtain the wind speed, output power of each unit, and the length of the power collection section submarine cable;

[0016] S12: Initializing the parameters of each unit controller, the sample number n, the cluster number C, and the iteration number t;

[0017] S13: Calculating the system impedance characteristics and the absolute value of the impedance sensitivity of the power collection submarine cable length, to form a sample data set D n×k ;

[0018] S14: Randomly selecting m objects as initial cluster centers from the data set D n×k ;

[0019] S15: Iterating through all the operating data, and calculating the Euclidean distance d ij ;

[0020] S16: According to the minimum distance center clustering principle, re-clustering each unit, that is, assigning the unit to the cluster closest to it;

[0021] S17: Screening out d ijThe maximum value, update the cluster center of each cluster;

[0022] S18: repeating steps S15-S17, judging whether the cluster center point and the distance square sum of each cluster change or whether the maximum iteration number is reached, if the iteration termination condition is met, outputting the wind farm clustering result.

[0023] Further, step S2 specifically comprises the following steps:

[0024] Based on the multi-machine equivalent model of the network-constructed wind farm, the port voltage, current, wind farm power, line impedance and filter parameter information of the grid-connected point are collected and recorded;

[0025] The output broadband impedance characteristics of the multi-machine equivalent model of the wind farm under each working condition are obtained through frequency sweeping;

[0026] The collected data is normalized so that the input data is normalized to the interval (0, 1), and the normalization formula is as follows:

[0027]

[0028] In the formula, x i is the current data, x min is the minimum value in a group of data, and x max is the maximum value in a group of data.

[0029] Further, the basic convolutional neural network structure of step S3 comprises two convolutional layers, two filter groups, a linear rectifier function ReLU, a maximum pooling, and a fully connected layer; the first convolutional layer contains 32 convolutional kernels with a step of 1, the second convolutional layer uses 128 convolutional kernels, and all convolutional layers are padded; the pool size and step of the maximum pooling are 1x2 and 2 respectively; the feature vector processed by the second maximum pooling operation is data-expanded into a larger vector, which is transmitted to the first fully connected layer with 40 neurons, the second connection layer is a real-valued regression of a neuron, and when the network is trained using the back propagation algorithm, the neuron weight parameters in the CNN are updated.

[0030] Further, the squeeze excitation module SE of step S3 is composed of squeeze, excitation, and fouling operations;

[0031] Squeeze is to compress the one-dimensional global vector of each channel into a scalar, which is realized by using global average pooling;

[0032] The multi-channel information obtained through the squeeze operation is used in the excitation operation to simulate the correlation between different channels by using two fully connected layers, and the weights of them are estimated by using the sigmoid activation function;

[0033] Finally, the normalized weights of the previous multi-channel feature vectors are processed through calculation.

[0034] Furthermore, step S4, which involves constructing the equivalent impedance model for the submarine power collection cable between wind turbine units, specifically includes:

[0035] Equivalent admittance of submarine cable π-type equivalent circuit Equivalent impedance The expression is:

[0036]

[0037] The π-type equivalent circuit consists of a resistor R1 and an inductor L1 connected in series, and a grounding capacitor C1, respectively.

[0038] R1=r0L section L1 = l0L section C1 = c0L section r0, l0, and c0 represent the resistance, inductance, and capacitance per unit length of the collector section of the submarine cable, respectively. section This refers to the length of the submarine cable in the collector section.

[0039] Furthermore, step S5 specifically includes the following steps:

[0040] S51: Based on the actual topology of the grid-type wind farm, the broadband impedance models of wind turbines and collector lines are connected into a frequency domain impedance network.

[0041] S52: The equivalent impedance of the i-th feeder in a medium-voltage power collection system containing m-1 wind turbine units is obtained using iterative calculation:

[0042]

[0043] Among them, Z WT_ij Z represents the impedance of the j-th wind turbine on the i-th feeder. WT_i1 Let be the impedance of the first wind turbine unit on the i-th feeder. Let be the π-type impedance and admittance of a single section of the current collector cable between the j-th and (j+1)-th generator units in the i-th feeder. and Z represents the π-type impedance and admittance of a single section of the submarine current collector cable between the first and second generator units in the i-th feeder. U_ij Z is the equivalent impedance of the port of the ij-th wind turbine unit after node elimination. U_i1 Z' is the equivalent impedance of the first wind turbine unit port of the i-th feeder. WT_ij is the equivalent impedance after connecting one wind turbine in parallel at the port of the ij-th wind power unit; m is the number of units on feeder i, j is the number of each unit on feeder i, and n is the total number of feeders in the wind farm.

[0044] S53: The submarine cable line close to the medium-voltage collection bus is equivalent to a series π-type circuit, and the impedance aggregation model Z of the high-voltage collection system under the 220kV voltage level is converted WC For:

[0045]

[0046] wherein Z U_im is the equivalent impedance of the mth wind power unit port of the ith feeder line, Z' WT_im is the equivalent impedance of the m-1th wind power unit port of the ith feeder line after connecting 1 wind turbine in parallel, are the π-type impedance and admittance of the collection submarine cable between the mth unit in the ith feeder line and the medium-voltage collection bus, respectively; k TC is the transformation ratio of the offshore booster transformer.

[0047] Further, the step S6 specifically comprises: in the future acquired new data, the CNN-Attention model is incrementally updated in a periodic form, and the incremental CNN-Attention model freezes part of the parameters in the incremental learning stage, and only updates the parameters of the output part of the CNN-Attention model.

[0048] The beneficial effects of the application are that the patent is based on the CNN-Attention model of incremental learning, and realizes the online identification of the wideband impedance of the network-type wind power plant.

[0049] (1) The clustering impedance modeling method can effectively deal with the "black / gray box" problem of wind turbines, and at the same time avoid the combination of a large number of wind turbine steady-state operating points and station topologies in large wind power plants, so as to ensure the convergence and training accuracy of the incremental learning CNN-Attention model.

[0050] (2) The CNN-Attention deep learning structure based on incremental learning combines incremental learning and wind power plant impedance identification model, constantly absorbs new information on the basis of retaining old knowledge, optimizes the identification ability of the model, and thus improves the robustness and generalization ability of the incremental CNN-Attention model.

[0051] Other advantages, objects, and features of the application will be set forth in part in the following specification, and in part will become apparent to those skilled in the art from the following, or can be learned from the practice of the application. The objectives and other advantages of the application can be realized and obtained by the following description. BRIEF DESCRIPTION OF DRAWINGS

[0052] In order to make the objects, technical solutions and advantages of the present application clearer, the preferred embodiments of the present application will be described in detail below with reference to the drawings, in which:

[0053] Figure 1 Flow chart for clustering of offshore wind farm by K-means algorithm;

[0054] Figure 2 Structure diagram of basic convolutional neural network;

[0055] Figure 3 Schematic diagram of adding CNN to SE module;

[0056] Figure 4 Process diagram for constructing impedance identification model of wind turbine generator;

[0057] Figure 5 Schematic diagram of frequency domain impedance network;

[0058] Figure 6 Architecture diagram of incremental CNN model;

[0059] Figure 7 Fig. (a) is a comparison result diagram of predicted and actual data of positive sequence impedance real part, (b) is a comparison result diagram of predicted and actual data of positive sequence impedance imaginary part, (c) is a comparison result diagram of predicted and actual data of negative sequence impedance real part, and (d) is a comparison result diagram of predicted and actual data of negative sequence impedance imaginary part. DETAILED DESCRIPTION

[0060] The present application will be described in greater detail by way of specific embodiments, from which its advantages and effects will be apparent to those skilled in the art. The present application can also be implemented or applied in other different specific embodiments, and the details in the present specification can be modified or changed based on different views and applications without departing from the spirit of the present application. It should be noted that the diagrams provided in the following embodiments only illustrate the basic concepts of the present application in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0061] It should be noted that the diagrams provided in the following embodiments only illustrate the basic concepts of the present application in a schematic manner, and only the components related to the present application are shown in the diagrams, not the number, shape and size of the components when actually implemented. The shapes, numbers and proportions of the components when actually implemented can be arbitrarily changed, and the layout pattern of the components can also be more complex.

[0062] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.

[0063] Example 1:

[0064] This invention provides a broadband impedance online identification method for networked wind farms based on incremental learning CNN-Attention. The method includes the following steps:

[0065] S1: Based on the K-means clustering method using dominant variables and unit operating characteristics, a dynamic equivalent model of multiple units in a parallel-aggregated offshore wind farm is established. For example... Figure 1 As shown, the wind speed, output power, and submarine cable length of each turbine are obtained using the wind farm SCADA (Supervisory and Data Acquisition) system. The controller parameters of each turbine are initialized, with a sample size n and a cluster size C. The iteration count t = 1. The system impedance characteristics and the absolute value of the impedance sensitivity along the submarine cable length are calculated, forming a sample data set D. n×k From dataset D n×k Randomly select m objects as the initial cluster centers; traverse all running data and calculate the Euclidean distance d from each other unit to each initial cluster center. ij Based on the classification principle of "minimum distance center clustering," the units are re-clustered, that is, the units are assigned to the groups of units closest to them in distance. Then, d units in each new group are filtered out. ij The maximum value is used to update the cluster centers of each turbine cluster. The sum of squared distances from all operating data in each turbine cluster to the new cluster centers is recalculated. It is then determined whether the sum of squared distances between the turbine cluster center and each turbine cluster has changed, or whether the maximum number of iterations has been reached. If the iteration termination condition is met, the wind farm clustering results are output.

[0066] S2: Data Acquisition. Based on the multi-machine equivalent model of a grid-connected wind farm, different operating power conditions, control parameters, and line length ranges are set, and the simulation model is run. At the same time, the voltage and current at the grid connection point, wind farm power, line impedance, and filter parameter information are collected and recorded; then, by frequency sweeping, the output broadband impedance characteristics of the multi-machine equivalent model of the wind farm under each operating condition are obtained.

[0067] S3: Data Preprocessing. Normalize the data to bring the input data to the (0,1) interval. The normalization formula is as follows:

[0068]

[0069] In the formula, xi is the current data, x min is the minimum value in a set of data, x max is the maximum value in a set of data.

[0070] S4: Construct the basic convolutional neural network structure. As shown in Figure 2 , the shallow CNN (convolutional neural network) structure includes two convolutional layers, using two filter sets, 1x8 and 1x32, respectively, followed by a linear rectifier function (ReLU) and a maximum pooling, and a fully connected layer, the first convolutional layer contains 32 convolutional kernels with a step size of 1, the second convolutional layer is the same as the first convolutional layer, using 128 convolutional kernels. All convolutional layers need to be padded, and the pool size and step size of the maximum pooling are 1x2 and 2, respectively. The feature vector processed by the second maximum pooling operation is flattened into a larger vector by data, which is transmitted to the first fully connected layer with 40 neurons, and the second connection layer is a real-valued regression of a neuron. When training the network using the backpropagation algorithm, the neural weight parameters in the CNN will be updated.

[0071] S5: Provide a channel attention mechanism (Attention) by introducing a squeeze and excitation module (SE), as shown in Figure 3 . The SE module mainly consists of squeeze, excitation, and scaling operations. Squeeze is to compress the one-dimensional global vector of each channel into a scalar, which is achieved by using global average pooling. Using the multi-channel information obtained by the squeeze operation, two fully connected layers are used in the excitation operation to simulate the correlation between different channels, and the weights of each channel are estimated by the sigmoid activation function. Finally, the normalized weights of the multi-channel feature vectors before operation are processed.

[0072] S6: Train the basic CNN-Attention deep learning network. Use 80% of the data to train the model, and update the weights and biases in each convolutional network and the weights in the attention mechanism through backpropagation to minimize the prediction error on the test set.

[0073] S7: Verify the accuracy of the basic CNN-Attention deep learning network model. Substitute the validation set into the trained model to obtain the output impedance model under different operating conditions. By comparing the predicted impedance results with the sweep frequency impedance results, the higher the curve fitting degree, the higher the accuracy of the trained model. The process of realizing the wideband impedance of the wind turbine connected to the grid based on the CNN-Attention deep learning model is shown in Figure 4 .

[0074] S8: Construct the equivalent impedance model of the power collection submarine cable line between wind turbines. The equivalent admittance of the submarine cable π-type equivalent circuit Equivalent impedance The expression is:

[0075]

[0076] The π-type equivalent circuit is composed of series resistance R1 and inductance L1 and ground capacitance C1, and R1=r0L section , L1=l0L section , C1=c0L section . r0, l0, c0 represent the resistance, inductance and capacitance per unit length of the collector section cable line, and L section is the length of the collector section cable.

[0077] S9: Based on the actual topology of the networked wind power station, the broadband impedance model of the wind turbine and the collector line is connected as a frequency domain impedance network, as shown in Figure 5 .

[0078] Using the iterative calculation method, the equivalent impedance of the i-th feeder line containing m-1 wind power units in the medium voltage collector system can be derived as:

[0079]

[0080] Where Z WT_ij is the impedance of the j-th wind turbine of the i-th feeder line, Z WT_i1 is the impedance of the first wind turbine of the i-th feeder line, and Y are the single-section type impedance and admittance of the collector cable between the j-th and j+1-th wind turbines in the i-th feeder line, and Y U_ij are the single-section type impedance and admittance of the collector cable between the first and second wind turbines in the i-th feeder line, Z U_i1 is the equivalent impedance of the first wind turbine port of the i-th feeder line, and Z' WT_ij is the equivalent impedance of the i-jth wind turbine port after connecting one wind turbine in parallel. m is the number of wind turbines of the i-th feeder line, j is the number of wind turbines on the i-th feeder line, and n is the total number of feeder lines in the wind farm.

[0081] The cable line close to the medium voltage collection bus is equivalent to a series π-type circuit, and the impedance aggregation model of the high voltage collector system converted to the 220kV voltage level Z WC is:

[0082]

[0083] Where Z U_im is the equivalent impedance of the m-th wind turbine port of the i-th feeder line, and Z' WT_imZm-1(i) is the equivalent impedance of the m-1th wind turbine connected in parallel to the i th feeder, Zm(i) and Ym(i) are the cable impedance and admittance between the mth wind turbine and the medium voltage bus in the i th feeder, respectively. TC K is the transformer ratio of the offshore booster transformer.

[0084] S10: Construct a CNN-Attention update model based on incremental learning. In the future, new data is obtained, and the CNN-Attention model is updated in a regular manner. The size of the time interval is set according to the actual situation. The process of incremental CNN-Attention learning new data is defined as an incremental training stage. In order to maintain the knowledge learned before, part of the parameters of the incremental CNN-Attention model is frozen, that is, these parameters will not be updated in the i th incremental learning stage, and only the parameters of the output part of the CNN-Attention model are updated. Ensure that the model adapts to new data through incremental learning, without affecting the weights of the basic model. The process is as shown in Figure 6 .

[0085] S11: Online application of impedance identification model. In the face of actual large-scale grid-connected wind farms, multi-stage equivalent modeling is carried out using real-time data of wind turbines, and then the port voltage, current and other real-time data of the equivalent wind turbine model are input into the trained CNN-Attention model containing incremental learning to obtain the real-time wideband impedance characteristics of each wind turbine. Finally, the wideband impedance of the large-scale wind farm station is obtained through the impedance network aggregation method. When the model output error rises in the identification process, the automatic update mechanism is started to realize real-time adaptation to new samples.

[0086] In order to verify the effect of the present application, a wind farm with a rated power of 400 MW composed of 64 wind turbines connected to a flexible AC transmission system (FACTS) is taken as an example. A virtual synchronous control wind farm physical model is established using the Simulink simulation platform to obtain the output impedance characteristics of the wind farm under different internal and external operating conditions. The deep learning algorithm model based on incremental learning proposed in the present application is built using the Pytroch framework of the Pycharm platform, and the verification results are as shown in Figure 7 (a)-(d) of FIG. 8, the prediction effect of the present application is excellent.

[0087] Example 2:

[0088] An electronic device includes a memory and a processor;

[0089] The memory is configured to store a computer program.

[0090] The processor, when executing the computer program, implements the method as described in Embodiment 1.

[0091] Embodiment 3:

[0092] A computer readable storage medium, the storage medium has a computer program stored thereon, when the computer program is executed by a processor, a method as described in Embodiment 1 is implemented.

[0093] Embodiment 4:

[0094] A computer program product, comprising a computer program, when executed by a processor, a method as described in Embodiment 1 is implemented.

[0095] In the above embodiments, the description refers to "this embodiment" means that the specific features, structures or characteristics described in the embodiment are included in at least some embodiments, but not necessarily all embodiments. Multiple occurrences of "this embodiment" do not necessarily all refer to the same embodiment.

[0096] In the above embodiments, although the present application has been described in conjunction with specific embodiments thereof, many alternatives, modifications and variations will be apparent to those skilled in the art in light of the foregoing description. For example, other storage structures (e.g., dynamic RAM (DRAM)) can use the embodiments discussed. Embodiments of the present application are intended to embrace all such alternatives, modifications and variations as can fall within the scope of the appended claims.

[0097] The computer readable storage medium in the embodiments can be understood by those skilled in the art that all or part of the steps of the above-mentioned method embodiments can be completed by a computer program related hardware. The foregoing computer program can be stored in a computer readable storage medium. The program, when executed, executes steps including the above-mentioned method embodiments; and the foregoing storage medium includes ROM, RAM, magnetic disk or optical disk and various media that can store program codes.

[0098] The electronic terminal provided in the embodiments includes a processor, a memory, a transceiver and a communication interface. The memory and the communication interface are connected with the processor and the transceiver and complete communication between each other. The memory is used to store a computer program, the communication interface is used for communication, and the processor and the transceiver are used to run the computer program, so that the electronic terminal executes each step of the above method.

[0099] In the embodiments, the memory can include a random access memory (RAM) and can also include a non-volatile memory, such as at least one disk memory.

[0100] The processor described above can be a general processor, including a central processing unit (CPU), a network processor (NP), etc.; can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component.

[0101] The present application can be used in many general or special computing system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc.

[0102] The present application can be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform particular tasks or implement particular abstract data types. The present application can also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules can be located in both local and remote computer storage media including memory storage devices.

[0103] Finally, it should be pointed out that the above embodiments are only used to illustrate the technical solutions of the present application and not to limit it. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the purpose and scope of the technical solutions, which should be covered in the scope of the claims of the present application.

Claims

1. A network-based wind farm wide-frequency impedance online identification method based on incremental learning CNN-Attention, characterized in that: The method comprises the following steps: S1: based on the dominant variable and the K-means clustering method of unit operation characteristics, a multi-machine dynamic equivalent model of parallel collection type offshore wind farm is established; S2: based on the multi-machine equivalent model of network type wind farm, different operating power conditions, control parameters and line length ranges are set, and each parameter is collected and preprocessed; S3: a basic convolutional neural network structure is constructed, then a channel attention mechanism is introduced by introducing a squeezing and excitation module SE to obtain a CNN-Attention model, and then the CNN-Attention model is trained and verified using a data set; S4: an equivalent impedance model of the power collection submarine cable line between wind turbines is constructed; S5: based on the actual topology of the network type wind farm station, the broadband impedance model of the wind turbine and the power collection line are connected to form a frequency domain impedance network, and the impedance characteristics of the power collection port of the multi-machine equivalent wind farm are calculated; S6: a CNN-Attention model with an incremental learning update mechanism is constructed; S7: multi-level equivalent modeling is performed on the real-time data of the wind turbine, then the equivalent wind turbine model data is input into the trained CNN-Attention model with an incremental learning update mechanism, the real-time broadband impedance characteristics of each wind turbine are obtained, and finally the broadband impedance of the large-scale wind farm station is obtained through the impedance network aggregation method; during the identification process, when the model output error rises, the automatic update mechanism is started to realize real-time adaptation to new samples.

2. The CNN-Attention-based incremental learning online impedance identification method for a wide-frequency wind farm according to claim 1, characterized in that: Step S1 specifically comprises the following steps: S11: the wind speed, output power and power collection section cable length of each unit are obtained by using the wind farm monitoring and data acquisition system SCADA; S12: the controller parameters of each unit, the sample number n, the cluster number C and the iteration number t are initialized; S13: Calculate the system impedance characteristic and the absolute value of the collector cable length impedance sensitivity, and form a sample data set D n×k ; S14: From the dataset D n×k Randomly select m objects as the initial swarm clustering centers; S15: Traverse all running data, calculate the Euclidean distance d of each initial machine group center to each other machine group point by point ij ; S16: according to the classification principle of minimum distance center clustering, each unit is re-clustered, that is, the unit is assigned to the cluster closest to it; S17: screen out d from each new machine group ij the largest value, update the cluster centers of each machine group; S18: steps S15-S17 are repeated to determine whether the cluster center point and the distance square sum of each cluster change, or whether the maximum iteration number is reached, if the iteration termination condition is met, the wind farm clustering result is output.

3. The CNN-Attention based incremental learning online impedance identification method for wideband impedance of a wind farm with network structure according to claim 2, characterized in that: Step S2 specifically comprises the following steps: Based on the multi-machine equivalent model of the network type wind farm, the grid-connected point port voltage, current, wind farm power, line impedance and filter parameter information are collected and recorded; By sweeping, the output broadband impedance characteristics of the multi-machine equivalent model of the wind farm under each working condition are obtained; The collected data is normalized so that the input data is normalized to the interval (0, 1), and the normalization formula is as follows: where x i is the current data, x min is the minimum value in a set of data, and x max is the maximum value in a set of data.

4. The CNN-Attention-based incremental learning online impedance identification method for a wideband wide-area wind farm according to claim 1, characterized in that: The convolutional neural network structure of the base in step S3 includes two convolutional layers, two filter groups, linear rectifier function ReLU, maximum pooling, and a fully connected layer; the first convolutional layer contains 32 convolutional kernels with a step of 1, and the second convolutional layer uses 128 convolutional kernels, and padding is performed on all convolutional layers; the pool size and step of the maximum pooling are 1x2 and 2, respectively; the feature vector processed by the second maximum pooling operation is data-expanded into a larger vector, which is transmitted to the first fully connected layer with 40 neurons, and the second connection layer is a real-valued regression of a neuron, and when the network is trained using the back propagation algorithm, the neuron weight parameters in the CNN are updated.

5. The CNN-Attention-based incremental learning online impedance identification method for a wideband wide-area wind farm according to claim 4, characterized in that: The squeeze excitation module SE in step S3 is composed of squeezing, excitation, and scaling operations; Squeezing is to compress a one-dimensional global vector of each channel into a scalar, which is realized by using global average pooling; Using the multi-channel information obtained by the squeezing operation, two fully connected layers are used in the excitation operation to simulate the correlation between different channels, and the sigmoid activation function is used to estimate their respective weights; Finally, the normalized weights of the multi-channel feature vectors before operation processing are obtained.

6. The CNN-Attention-based incremental learning online impedance identification method for a wideband wide-area wind farm according to claim 1, characterized in that: The step S4 of constructing the equivalent impedance model of the wind turbine interconnection submarine cable line comprises the following steps: Equivalent admittance of a submarine cable pi-type equivalent circuit Equivalent impedance The expression is: wherein the type π equivalent circuit is composed of a series connection of a resistance R1 and an inductance L1 and a grounded capacitance C1, respectively R1 = r0L section , L1 = l0L section , C1 = c0L section ; r0, l0, c0 respectively represent the resistance, inductance and capacitance per unit length of the collector section submarine cable line, and L section is the length of the collector section submarine cable.

7. The CNN-Attention-based incremental learning online impedance identification method for wideband impedance of a wind farm according to claim 1, characterized in that: Step S5 specifically comprises the following steps: S51: based on the actual topology of the network-constructed wind power station, the broadband impedance models of the wind turbine and the power collection line are connected into a frequency domain impedance network; S52: using the iterative calculation method, the equivalent impedance of the ith feeder line containing m-1 wind power units in the medium voltage power collection system is: wherein Z WT_ij is the impedance of the jth wind turbine of the ith feeder line, Z WT_i1 is the impedance of the 1st wind turbine of the ith feeder line, are the π-type impedance and admittance of the single section of the collection cable between the jth and j+1th wind turbines of the ith feeder line, respectively, and are the π-type impedance and admittance of the single section of the collection cable between the 1st and 2nd wind turbines of the ith feeder line, respectively, Z U_ij is the equivalent impedance of the ijth wind turbine port after node elimination, Z U_i1 is the equivalent impedance of the 1st wind turbine port of the ith feeder line, Z WT_ij is the equivalent impedance of the ijth wind turbine port after connecting 1 wind turbine in parallel; m is the number of wind turbines of the feeder line i, j is the number of each wind turbine on the feeder line i, and n is the total number of feeder lines of the wind farm. S53: The submarine cable line close to the medium voltage collection bus is equivalent to a series π-type circuit, and the impedance aggregation model Z of the high voltage current collection system under the 220 kV voltage level is converted WC is: wherein Z U_im is the equivalent impedance of the mth wind power unit port of the ith feeder line, Z' WT_im is the equivalent impedance of the m-1th wind power unit port of the ith feeder line after connecting one wind turbine in parallel, are the π-type impedance and admittance of the power collection cable between the mth unit and the medium-voltage collection bus in the ith feeder line, respectively; k TC is the transformer ratio of the offshore booster transformer.

8. The CNN-Attention based incremental learning online impedance identification method for wideband impedance of a wind farm according to claim 1, characterized in that: Step S6 specifically includes: in the future obtained new data, the CNN-Attention model is incrementally updated in a regular form, and the incremental CNN-Attention model freezes part of the parameters in the incremental learning stage and only updates the parameters of the output part of the CNN-Attention model.

Images