Wind farm grid-connection performance evaluation system and method

Abstract

A wind farm grid-connected performance evaluation system and method are disclosed. The system obtains multiple third-level index data, processes the data through a trained context encoder containing an embedding layer to obtain multiple third-level index semantic feature vectors, and concatenates the vectors to obtain a second-level index feature vector. After arranging the second-level index feature vector into a two-dimensional feature matrix and a one-dimensional feature vector, the system processes the vectors through a trained first convolutional neural network and a trained multi-scale neighborhood feature extraction unit to obtain a two-dimensional second-level index correlation feature vector and a multi-scale second-level index correlation feature vector, fuses the vectors to obtain a classification feature vector, and processes the vector through a classifier to obtain a classification result representing a wind farm grid-connected performance grade label. In particular, the system uses a convolutional neural network and a multi-scale field feature extraction module to extract correlation features between different levels of indexes using different receptive fields, thereby improving the accuracy of index evaluation.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of new energy, and more particularly to a wind farm grid-connection performance evaluation system and method thereof. BACKGROUND

[0002] In recent years, the wind power industry in China has developed rapidly. The demand for new energy grid-connection is still strong, and the grid-connection appeal is relatively strong. The proportion of wind farms in the installed capacity of power grids is increasing, and the influence of wind farm grid-connection technical performance on power systems is increasingly valued by people. The current wind power grid-connection performance evaluation index is only limited to single index evaluation, and there is no comprehensive and systematic evaluation index system.

[0003] Therefore, it is urgent to establish an objective and scientific evaluation index system and evaluation method to evaluate the wind farm grid-connection technology, promote the understanding of the grid-connection technical performance of power grid dispatching and wind power enterprises, and improve and enhance the overall grid-connection technical level of wind farms.

[0004] In terms of evaluation index, the wind power operation mode has obvious randomness, volatility and intermittency, and it is difficult to reveal the whole link power generation characteristics of wind power only by using conventional evaluation indexes such as power generation and utilization hours.

[0005] Therefore, an optimized wind farm grid-connection performance evaluation system is expected. SUMMARY

[0006] In order to solve the above technical problems, the present application is proposed. The embodiments of the present application provide a wind farm grid-connection performance evaluation system and method thereof, which obtains a plurality of third-level index data through a trained context encoder containing an embedding layer to obtain a plurality of third-level index semantic feature vectors and concatenates them to obtain a second-level index feature vector. After arranging the second-level index feature vector into a two-dimensional feature matrix and a one-dimensional feature vector, a trained first convolutional neural network and a trained multi-scale neighborhood feature extraction unit are used to obtain a two-dimensional second-level index correlation feature vector and a multi-scale second-level index correlation feature vector, and then fuse them to obtain a classification feature vector. The classifier can obtain a classification result for representing the wind farm grid-connection performance grade label. In particular, the present application uses convolutional neural networks and multi-scale field feature extraction modules to extract the correlation features between different levels of indexes with different receptive fields, thereby improving the classification accuracy.

[0007] According to an aspect of the present application, a wind farm grid-connection performance evaluation system is provided, which comprises:

[0008] a third-level index data acquisition unit for acquiring a plurality of third-level index data under each second-level index;

[0009] a third-level index data encoding unit, configured to obtain a plurality of third-level index semantic feature vectors by inputting a plurality of third-level index data under each second-level index into a context encoder comprising an embedding layer which is trained, and concatenate the plurality of third-level index semantic feature vectors to obtain a second-level index feature vector corresponding to each second-level index;

[0010] a second-level index two-dimensional correlation encoding unit, configured to arrange the second-level index feature vector corresponding to each second-level index into a two-dimensional feature matrix, and then input the two-dimensional feature matrix into a first convolutional neural network which is trained as a filter to obtain a two-dimensional second-level index correlation feature vector;

[0011] a second-level index one-dimensional multi-scale encoding unit, configured to arrange the second-level index feature vector corresponding to each second-level index into a one-dimensional feature vector, and then input the one-dimensional feature vector into a multi-scale neighborhood feature extraction unit which is trained to obtain a multi-scale second-level index correlation feature vector;

[0012] a fusion unit, configured to fuse the two-dimensional second-level index correlation feature vector and the multi-scale second-level index correlation feature vector to obtain a classification feature vector; and

[0013] an evaluation index generation unit, configured to input the classification feature vector into a classifier to obtain a classification result, wherein the classification result is used to represent a wind farm grid-connected performance level label.

[0014] In the wind farm grid-connected performance evaluation system, the third-level index data encoding unit comprises: an embedding encoding subunit, configured to use an embedding layer of the context encoder comprising an embedding layer which is trained to respectively convert the plurality of third-level index data under each second-level index into an input vector to obtain a sequence of input vectors; a context encoding subunit, configured to use a transformer of the context encoder comprising an embedding layer which is trained to perform global context semantic encoding on the sequence of input vectors to obtain the plurality of third-level index semantic feature vectors; and a concatenation subunit, configured to concatenate the plurality of third-level index semantic feature vectors to obtain the second-level index feature vector corresponding to each second-level index.

[0015] In the wind farm grid-connected performance evaluation system, the second-level index two-dimensional correlation encoding unit is further configured to: each layer of the first convolutional neural network which is trained as a filter respectively performs, in a forward propagation of the layer, convolution processing on input data to obtain a convolution feature map, mean pooling based on a local feature matrix on the convolution feature map to obtain a pooling feature map, and nonlinear activation on the pooling feature map to obtain an activated feature map; wherein an output of a last layer of the first convolutional neural network is the two-dimensional second-level index correlation feature vector, and an input of a first layer of the first convolutional neural network is the two-dimensional feature matrix.

[0016] In the wind farm grid-connected performance evaluation system, the one-dimensional multi-scale encoding unit of the secondary index comprises: a first convolution subunit configured to input the one-dimensional feature vector into a first convolution layer of the trained multi-scale neighborhood feature extraction unit to obtain a first neighborhood scale secondary index associated feature vector, wherein the first convolution layer has a first one-dimensional convolution kernel with a first length; a second convolution subunit configured to input the one-dimensional feature vector into a second convolution layer of the trained multi-scale neighborhood feature extraction unit to obtain a second neighborhood scale secondary index associated feature vector, wherein the second convolution layer has a second one-dimensional convolution kernel with a second length, and the first length is different from the second length; and a concatenation subunit configured to concatenate the first neighborhood scale secondary index associated feature vector and the second neighborhood scale secondary index associated feature vector to obtain the multi-scale secondary index associated feature vector.

[0017] In the wind farm grid-connected performance evaluation system, the first convolution subunit is configured to use the first convolution layer of the trained multi-scale neighborhood feature extraction unit to perform one-dimensional convolution encoding on the one-dimensional feature vector according to the following formula to obtain the first neighborhood scale secondary index associated feature vector; wherein the formula is:

[0018]

[0019] wherein a is the width of the first convolution kernel in the x direction, F(a) is the first convolution kernel parameter vector, G(x-a) is a local vector matrix operated with the convolution kernel function, w is the size of the first convolution kernel, and X represents the one-dimensional feature vector; the second convolution subunit is configured to use the second convolution layer of the trained multi-scale neighborhood feature extraction unit to perform one-dimensional convolution encoding on the one-dimensional feature vector according to the following formula to obtain the second neighborhood scale secondary index associated feature vector; wherein the formula is:

[0020]

[0021] wherein b is the width of the second convolution kernel in the x direction, F(b) is the second convolution kernel parameter vector, G(x-b) is a local vector matrix operated with the convolution kernel function, m is the size of the second convolution kernel, and X represents the one-dimensional feature vector.

[0022] In the wind farm grid-connected performance evaluation system, the fusion unit is further configured to concatenate the two-dimensional secondary index associated feature vector and the multi-scale secondary index associated feature vector to obtain the classification feature vector.

[0023] In the wind farm grid-connected performance evaluation system, the evaluation index generation unit is further configured to use the classifier to process the classification feature vector according to the following formula to obtain the classification result.

[0024] The formula is softmax{(M2, B2):(M1, B1)|X}, where M1 and M2 are weight matrices, B1 and B2 are bias vectors, and X is the classification feature vector.

[0025] In the wind farm grid-connected performance evaluation system, the wind farm grid-connected performance evaluation system further comprises a training module for training the context encoder comprising an embedding layer, the multi-scale neighborhood feature extraction module, and the first convolutional neural network as a filter; wherein the training module comprises: a training data acquisition unit configured to acquire a plurality of third-level indicator data under each second-level indicator; a training third-level indicator data encoding unit configured to obtain a plurality of training third-level indicator semantic feature vectors by inputting the plurality of third-level indicator data under each second-level indicator into the context encoder comprising an embedding layer, and concatenate the plurality of training third-level indicator semantic feature vectors to obtain a training second-level indicator feature vector corresponding to each second-level indicator; a training second-level indicator two-dimensional correlation encoding unit configured to arrange the training second-level indicator feature vector corresponding to each second-level indicator into a two-dimensional feature matrix, and input the two-dimensional feature matrix into the first convolutional neural network as a filter to obtain a training two-dimensional second-level indicator correlation feature vector; a training second-level indicator one-dimensional multi-scale encoding unit configured to arrange the training second-level indicator feature vector corresponding to each second-level indicator into a one-dimensional feature vector, and input the one-dimensional feature vector into the multi-scale neighborhood feature extraction module to obtain a training multi-scale second-level indicator correlation feature vector; a training fusion unit configured to fuse the training two-dimensional second-level indicator correlation feature vector and the training multi-scale second-level indicator correlation feature vector to obtain a training classification feature vector; a classification loss unit configured to input the training classification feature vector into a classifier to obtain a classification loss function value; a classification mode resolution suppression loss function value calculation unit configured to calculate the classification mode resolution suppression loss function value based on the two-norm of a differential feature vector between the training two-dimensional second-level indicator correlation feature vector and the training multi-scale second-level indicator correlation feature vector; and a training unit configured to train the context encoder comprising an embedding layer, the multi-scale neighborhood feature extraction module, and the first convolutional neural network as a filter using the weighted sum of the classification mode resolution suppression loss function value and the classification loss function value as a loss function value.

[0026] In the wind farm grid-connected performance evaluation system, the classification mode elimination inhibition loss function value calculation unit is further configured to calculate the classification mode elimination inhibition loss function value based on a two-norm of a difference feature vector between the training two-dimensional secondary index correlation feature vector and the training multi-scale secondary index correlation feature vector according to the following formula: wherein the formula is:

[0027]

[0028] wherein V1 and V2 represent the training two-dimensional secondary index correlation feature vector and the training multi-scale secondary index correlation feature vector respectively, M1 and M2 are weight matrices of the classifier for the training two-dimensional secondary index correlation feature vector and the training multi-scale secondary index correlation feature vector respectively, ||·||F represents a Frobenius norm of a matrix, and ||·||2 represents a two-norm of a vector, F represents a position difference, and exp(·) represents an exponential operation of a matrix and an exponential operation of a vector, the exponential operation of the matrix representing calculation of a natural exponential function value with a feature value at each position in the matrix as a power, and the exponential operation of the vector representing calculation of a natural exponential function value with a feature value at each position in the vector as a power.

[0029] According to another aspect of the present application, a wind farm grid-connected performance evaluation method is also provided, which comprises:

[0030] obtaining a plurality of tertiary index data under each secondary index;

[0031] obtaining a plurality of tertiary index semantic feature vectors by passing the plurality of tertiary index data under each secondary index through a context encoder comprising an embedding layer trained to be complete, and concatenating the plurality of tertiary index semantic feature vectors to obtain a secondary index feature vector corresponding to each secondary index;

[0032] obtaining a two-dimensional secondary index correlation feature vector by passing the secondary index feature vector corresponding to each secondary index through a first convolutional neural network trained to be complete as a filter after arranging the secondary index feature vector corresponding to each secondary index into a two-dimensional feature matrix;

[0033] obtaining a multi-scale secondary index correlation feature vector by passing the secondary index feature vector corresponding to each secondary index through a multi-scale neighborhood feature extraction unit trained to be complete after arranging the secondary index feature vector corresponding to each secondary index into a one-dimensional feature vector;

[0034] fusing the two-dimensional secondary index correlation feature vector and the multi-scale secondary index correlation feature vector to obtain a classification feature vector; and

[0035] ​The classification feature vector is input into a classifier to obtain a classification result, which is used to represent a wind farm grid-connection performance level label.

[0036] Compared with the prior art, the wind farm grid-connection performance evaluation system and method provided by the application obtain a plurality of third-level index semantic feature vectors through a trained context encoder comprising an embedding layer from the plurality of third-level index data, and concatenate the third-level index semantic feature vectors to obtain a second-level index feature vector. After arranging the second-level index feature vector into a two-dimensional feature matrix and a one-dimensional feature vector, respectively, a trained first convolutional neural network and a trained multi-scale neighborhood feature extraction unit are used to obtain a two-dimensional second-level index correlation feature vector and a multi-scale second-level index correlation feature vector, which are fused to obtain a classification feature vector. Then, a classifier is used to obtain a classification result used to represent a wind farm grid-connection performance level label. In particular, the convolutional neural network and the multi-scale field feature extraction module are used to extract the correlation features between different levels of indexes by using different receptive fields, thereby improving the classification accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0037] The above and other objects, features and advantages of the present application will become more apparent from the following detailed description of embodiments of the present application, when taken in conjunction with the accompanying drawings. The drawings provided in the present application are used to provide further understanding of the embodiments of the present application, and constitute a part of the specification, and are used to explain the present application together with the embodiments of the present application, and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally indicate the same components or steps.

[0038] Figure 1 FIG. 1 illustrates a block diagram of a wind farm grid-connection performance evaluation system according to an embodiment of the present application.

[0039] Figure 2 FIG. 1 illustrates a block diagram of a wind farm grid-connection performance evaluation system according to an embodiment of the present application.

[0040] Figure 3 FIG. 1 illustrates a block diagram of a wind farm grid-connection performance evaluation system according to an embodiment of the present application.

[0041] Figure 4 FIG. 1 illustrates a block diagram of a wind farm grid-connection performance evaluation system according to an embodiment of the present application.

[0042] Figure 5 FIG. 1 illustrates a block diagram of a wind farm grid-connection performance evaluation system according to an embodiment of the present application.

[0043] Figure 6 FIG. 1 illustrates a block diagram of a wind farm grid-connection performance evaluation system according to an embodiment of the present application. DETAILED DESCRIPTION

[0044] Hereinafter, the example embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part but not all of the embodiments of the present application, and the present application can be implemented in other embodiments without departing from the spirit of the present application. Therefore, the present application is not limited to the described embodiments.

[0045] SUMMARY

[0046] It is found through research that when evaluating the grid-connection performance of a wind farm, since the first-level indicators for evaluation have different second-level indicators, each second-level indicator also has different third-level indicators, and there are mutual implicit correlation relationships between these indicators. Therefore, in order to accurately evaluate the whole-link power generation characteristics of wind power, it is necessary to extract the correlation characteristics between the indicators of different levels. Therefore, in the technical solution of the present application, a convolutional neural network and a multi-scale domain feature extraction module are selected to utilize different receptive fields, that is, to extract correlation features of different scales, so as to more fully mine the correlation features of different span dependencies between the indicators, thereby improving the rationality and fairness of grid-connection performance evaluation.

[0047] Specifically, in the technical solution of the present application, first, a plurality of third-level indicator data under each second-level indicator is obtained. Here, the second-level indicators include active power control indicators, reactive power regulation indicators, low-high voltage ride-through indicators, power quality indicators, wind farm operation reliability indicators, and wind power prediction error indicators. The third-level indicators under the active power control indicators include 1 min active power variation and 10 min active power variation; the third-level indicators under the reactive power regulation indicators include wind turbine power factor regulation capability, wind farm reactive power regulation capability, and wind farm voltage control capability; the third-level indicators under the low-high voltage ride-through indicators include low voltage ride-through capability and active power recovery rate, dynamic reactive power regulation capability, and high voltage ride-through capability; the third-level indicators under the power quality indicators include frequency deviation, voltage deviation, voltage fluctuation and flicker, harmonic voltage, three-phase voltage allowable unbalance, and inter-harmonic; the third-level indicators under the wind farm operation reliability indicators include wind farm unplanned outage coefficient and wind farm availability coefficient; and the third-level indicators under the wind power prediction error indicators include short-term prediction monthly root mean square error and ultra-short-term prediction 4th hour prediction value monthly root mean square error.

[0048] It should be understood that, in order to accurately evaluate the grid-connected performance of the wind farm, the global implicit correlation characteristics of the indicators are needed to be considered for the hidden correlation between the multiple third-level indicator data under each second-level indicator, and the power generation characteristics of the wind farm are evaluated. Specifically, the context encoder containing the embedding layer is used to encode the multiple third-level indicator data under each second-level indicator to extract the global high-dimensional semantic features between the multiple third-level indicator data, so as to obtain the multiple third-level indicator semantic feature vectors, which are more suitable for representing the essential characteristics of the grid-connected performance of the wind farm.

[0049] As each third-level indicator data belongs to different second-level indicator data, the correlation characteristics between the second-level indicator data need to be further considered after considering the global correlation characteristics of each third-level indicator. Based on this, the multiple third-level indicator semantic feature vectors are further concatenated to obtain the second-level indicator feature vectors corresponding to each second-level indicator.

[0050] Further, after arranging the second-level indicator feature vectors corresponding to each second-level indicator into a two-dimensional feature matrix to integrate the global implicit correlation characteristics of the third-level indicators under each second-level indicator, the local correlation feature distribution of each second-level indicator is extracted by the first convolutional neural network as a filter to obtain the two-dimensional second-level indicator correlation feature vectors. In this way, the feature mining is performed by the first convolutional neural network as a filter, which can not only mine deep correlation feature information, but also reduce the number of parameters to prevent overfitting, thereby improving the accuracy of classification.

[0051] It should be understood that the first convolutional neural network has excellent performance in local feature extraction, but the receptive field of the first convolutional neural network is low, and the convolutional neural network and the multi-scale domain feature extraction module have different receptive fields, that is, different scales of correlation features are extracted. Therefore, in order to more fully mine the correlation features between the indicators with different span dependencies, thereby improving the rationality and fairness of the grid-connected performance evaluation, a multi-scale domain feature extraction unit is further used to perform multi-scale domain feature extraction on the one-dimensional feature vectors arranged by the second-level indicator feature vectors corresponding to each second-level indicator, thereby obtaining the multi-scale second-level indicator correlation feature vectors.

[0052] Then, the two-dimensional secondary index correlation feature vector and the multi-scale secondary index correlation feature vector can be fused to obtain a classification feature vector, and a classification result for representing a wind farm grid-connected performance level label can be obtained through a classifier. However, for the two-dimensional secondary index correlation feature vector, for example, denoted as V1, and the multi-scale secondary index correlation feature vector, for example, denoted as V2, since they contain correlation features of different scales, they have different feature mode expressions. After fusion in a cascading manner, for example, abnormal gradient expansion in the back propagation process of the classifier during the training process can cause the mode expressed by the features to be eliminated, thereby affecting the classification accuracy.

[0053] Therefore, a classification mode elimination inhibition loss function is introduced, denoted as:

[0054]

[0055] where V1 and V2 represent the training two-dimensional secondary index correlation feature vector and the training multi-scale secondary index correlation feature vector, respectively, M1 and M2 are weight matrices of the classifier for the training two-dimensional secondary index correlation feature vector and the training multi-scale secondary index correlation feature vector, respectively, ||·||F represents the Frobenius norm of a matrix, and ||·||2 represents the two-norm of a vector. F represents the position difference, exp(·) represents the matrix exponential operation and the vector exponential operation, the matrix exponential operation represents calculating the natural exponential function value with the eigenvalue of each position in the matrix as the power, and the vector exponential operation represents calculating the natural exponential function value with the eigenvalue of each position in the vector as the power.

[0056] Here, the classification mode elimination inhibition loss function pushes the pseudo difference of the classifier weights to the real feature distribution difference, so as to ensure that the derivative during gradient back propagation is regularly obtained near the gradient branch point, thereby over-weighting the gradient between modes to inhibit the classification mode elimination of the features, and further improving the classification accuracy.

[0057] ​Based on this, the application provides a wind farm grid-connected performance evaluation system, which comprises: a three-level index data acquisition unit, configured to acquire a plurality of three-level index data under each two-level index; a three-level index data encoding unit, configured to obtain a plurality of three-level index semantic feature vectors by inputting the plurality of three-level index data under each two-level index into a trained context encoder comprising an embedding layer, and concatenate the plurality of three-level index semantic feature vectors to obtain a two-level index feature vector corresponding to each two-level index; a two-level index two-dimensional correlation encoding unit, configured to arrange the two-level index feature vector corresponding to each two-level index into a two-dimensional feature matrix, and then input the two-dimensional feature matrix into a trained first convolutional neural network serving as a filter to obtain a two-dimensional two-level index correlation feature vector; a two-level index one-dimensional multi-scale encoding unit, configured to arrange the two-level index feature vector corresponding to each two-level index into a one-dimensional feature vector, and then input the one-dimensional feature vector into a trained multi-scale neighborhood feature extraction unit to obtain a multi-scale two-level index correlation feature vector; a fusion unit, configured to fuse the two-dimensional two-level index correlation feature vector and the multi-scale two-level index correlation feature vector to obtain a classification feature vector; and an evaluation index generation unit, configured to input the classification feature vector into a classifier to obtain a classification result, wherein the classification result is used to represent a wind farm grid-connected performance grade label.

[0058] After introducing the basic principles of the application, various non-limiting embodiments of the application will be specifically introduced below with reference to the accompanying drawings.

[0059] Exemplary System

[0060] Figure 1 Fig. 1 illustrates a block diagram of a wind farm grid-connected performance evaluation system according to an embodiment of the application. As shown in Fig. 1, the wind farm grid-connected performance evaluation system comprises a three-level index data acquisition unit 1, a three-level index data encoding unit 2, a two-level index two-dimensional correlation encoding unit 3, a two-level index one-dimensional multi-scale encoding unit 4, a fusion unit 5, and an evaluation index generation unit 6. Figure 1As shown, the wind farm grid-connected performance evaluation system 100 according to the embodiment of the present application comprises: a three-level index data acquisition unit 110, configured to acquire a plurality of three-level index data under each two-level index; a three-level index data encoding unit 120, configured to obtain a plurality of three-level index semantic feature vectors by using a trained context encoder comprising an embedding layer on the plurality of three-level index data under each two-level index, and concatenate the plurality of three-level index semantic feature vectors to obtain a two-level index feature vector corresponding to each two-level index; a two-level index two-dimensional correlation encoding unit 130, configured to arrange the two-level index feature vector corresponding to each two-level index into a two-dimensional feature matrix, and then pass the two-dimensional feature matrix through a trained first convolutional neural network serving as a filter to obtain a two-dimensional two-level index correlation feature vector; a two-level index one-dimensional multi-scale encoding unit 140, configured to arrange the two-level index feature vector corresponding to each two-level index into a one-dimensional feature vector, and then pass the one-dimensional feature vector through a trained multi-scale neighborhood feature extraction unit to obtain a multi-scale two-level index correlation feature vector; a fusion unit 150, configured to fuse the two-dimensional two-level index correlation feature vector and the multi-scale two-level index correlation feature vector to obtain a classification feature vector; and an evaluation index generation unit 160, configured to pass the classification feature vector through a classifier to obtain a classification result, wherein the classification result is used to represent a wind farm grid-connected performance level label.

[0061] Figure 2 FIG. 1 illustrates a system architecture diagram of a wind farm grid-connected performance evaluation system 100 according to an embodiment of the present application. As shown in the system architecture of the wind farm grid-connected performance evaluation system 100, first, a plurality of three-level index data under each two-level index is acquired by an index data acquisition device. Then, a plurality of three-level index semantic feature vectors are obtained by using a trained context encoder comprising an embedding layer on the plurality of three-level index data under each two-level index, and the plurality of three-level index semantic feature vectors are concatenated to obtain a two-level index feature vector corresponding to each two-level index. Then, the two-level index feature vector corresponding to each two-level index is arranged into a two-dimensional feature matrix, and then passed through a trained first convolutional neural network serving as a filter to obtain a two-dimensional two-level index correlation feature vector. Then, the two-level index feature vector corresponding to each two-level index is arranged into a one-dimensional feature vector, and then passed through a trained multi-scale neighborhood feature extraction unit to obtain a multi-scale two-level index correlation feature vector, and the multi-scale two-level index correlation feature vector is fused with the two-dimensional two-level index correlation feature vector to obtain a classification feature vector. Then, the classification feature vector is passed through a classifier to obtain a classification result, wherein the classification result is used to represent a wind farm grid-connected performance level label. Figure 2

[0062] ​In the wind farm grid-connected performance evaluation system 100, the third-level index data acquisition unit 110 is configured to acquire a plurality of third-level index data under each second-level index. The second-level indexes include active power control indexes, reactive power regulation indexes, low-voltage ride-through indexes, power quality indexes, wind farm operation reliability indexes, and wind power prediction error indexes. The third-level indexes under the active power control indexes include 1-minute active power variation and 10-minute active power variation. The third-level indexes under the reactive power regulation indexes include wind turbine power factor regulation capability, wind farm reactive power regulation capability, and wind farm voltage control capability. The third-level indexes under the low-voltage ride-through indexes include low-voltage ride-through capability, active power recovery rate, dynamic reactive power regulation capability, and high-voltage ride-through capability. The third-level indexes under the power quality indexes include frequency deviation, voltage deviation, voltage fluctuation and flicker, harmonic voltage, three-phase voltage allowable imbalance, and inter-harmonic. The third-level indexes under the wind farm operation reliability indexes include wind farm unplanned outage coefficient and wind farm availability coefficient. The third-level indexes under the wind power prediction error indexes include short-term prediction monthly root mean square error and ultra-short-term prediction 4-hour prediction value monthly root mean square error. In addition, the plurality of third-level index data can be acquired by an electric energy quality analyzer, a same control electric quantity (waveform) recording analyzer, and the like.

[0063] In the wind farm grid-connected performance evaluation system 100, the third-level index data encoding unit 120 is configured to obtain a plurality of third-level index semantic feature vectors by encoding the plurality of third-level index data under each second-level index by using a context encoder comprising an embedding layer, and concatenate the plurality of third-level index semantic feature vectors to obtain a second-level index feature vector corresponding to each second-level index. It can be understood that, in order to accurately evaluate the grid-connected performance of the wind farm, the global implicit correlation features of the indexes need to be used to evaluate the power generation characteristics of the wind farm in view of the hidden correlations between the plurality of third-level index data under each second-level index. Specifically, the context encoder comprising the embedding layer is used to encode the plurality of third-level index data under each second-level index to extract the global high-dimensional semantic features between the plurality of third-level index data, so as to obtain the plurality of third-level index semantic feature vectors, which are more suitable for representing the essential features of the grid-connected performance of the wind farm. As for the plurality of third-level index semantic feature vectors, since each third-level index data belongs to different second-level index data, the correlation features between the second-level index data also need to be further considered after considering the global correlation features of each third-level index. Based on this, the plurality of third-level index semantic feature vectors are further concatenated to obtain the second-level index feature vector corresponding to each second-level index.

[0064] Figure 3 The diagram illustrates a block diagram of a three-level index data encoding unit in a wind farm grid-connected performance evaluation system according to an embodiment of this application. For example... Figure 3 As shown, the third-level indicator data encoding unit 120 of the power grid fault system according to an embodiment of this application includes: an embedding encoding subunit 121, used to convert multiple third-level indicator data under each second-level indicator into input vectors using the embedding layer of the trained context encoder containing the embedding layer to obtain a sequence of input vectors; a context encoding subunit 122, used to perform global context semantic encoding on the sequence of input vectors using the converter of the trained context encoder containing the embedding layer to obtain the multiple third-level indicator semantic feature vectors; and a cascading subunit 123, used to cascade the multiple third-level indicator semantic feature vectors to obtain the second-level indicator feature vectors corresponding to each second-level indicator.

[0065] In the aforementioned wind farm grid-connected performance evaluation system 100, the two-dimensional correlation encoding unit 130 for secondary indicators is used to arrange the feature vectors of the secondary indicators corresponding to each secondary indicator into a two-dimensional feature matrix, and then pass it through a trained first convolutional neural network acting as a filter to obtain a two-dimensional secondary indicator correlation feature vector. Further, the feature vectors of the secondary indicators corresponding to each secondary indicator are arranged into a two-dimensional feature matrix to integrate the global correlation implicit features of the tertiary indicators under each secondary indicator. Then, deep implicit correlation feature mining is performed through the first convolutional neural network acting as a filter to extract the local correlation feature distribution of each secondary indicator, thereby obtaining a two-dimensional secondary indicator correlation feature vector. In this way, feature mining through the first convolutional neural network acting as a filter not only uncovers deep-level correlation feature information but also reduces the number of parameters to prevent overfitting, thereby improving the accuracy of classification.

[0066] In one example, in the wind farm grid-connected performance evaluation system 100 described above, the two-dimensional correlation encoding unit 130 for secondary indicators is further configured to: perform the following on each layer of the first convolutional neural network, which has been trained as a filter, during the forward propagation of the layer: perform convolution processing on the input data to obtain a convolutional feature map; perform mean pooling on the convolutional feature map based on the local feature matrix to obtain a pooled feature map; and perform nonlinear activation on the pooled feature map to obtain an activation feature map; wherein the output of the last layer of the first convolutional neural network is the two-dimensional secondary indicator correlation feature vector, and the input of the first layer of the first convolutional neural network is the two-dimensional feature matrix.

[0067] In the wind farm grid-connected performance evaluation system 100, the two-level index one-dimensional multi-scale coding unit 140 is configured to arrange the two-level index feature vectors corresponding to the respective two-level indexes into one-dimensional feature vectors, and then input the one-dimensional feature vectors into the trained multi-scale neighborhood feature extraction unit to obtain multi-scale two-level index correlation feature vectors. It should be understood that the first convolutional neural network has excellent performance in local feature extraction of features, but the receptive field of the first convolutional neural network is low. The convolutional neural network and the multi-scale neighborhood feature extraction module have different receptive fields, that is, they extract correlation features of different scales. Therefore, in order to more fully excavate the correlation features of different span dependencies between the indexes, thereby improving the rationality and fairness of the grid-connected performance evaluation, the multi-scale neighborhood feature extraction unit is further used to perform multi-scale neighborhood feature extraction on the one-dimensional feature vectors arranged by the two-level index feature vectors corresponding to the respective two-level indexes, thereby obtaining the multi-scale two-level index correlation feature vectors.

[0068] Figure 4 FIG. 1 illustrates a block diagram of a two-level index one-dimensional multi-scale coding unit in a wind farm grid-connected performance evaluation system according to an embodiment of the present application. As shown in the figure, the two-level index one-dimensional multi-scale coding unit 140 of the power grid fault system according to an embodiment of the present application includes a first convolutional subunit 141, a second convolutional subunit 142, and a concatenation subunit 143. Figure 4 The first convolutional subunit 141 is configured to input the one-dimensional feature vectors into a first convolutional layer of the trained multi-scale neighborhood feature extraction unit to obtain a first neighborhood scale two-level index correlation feature vector, where the first convolutional layer has a first one-dimensional convolution kernel of a first length. The second convolutional subunit 142 is configured to input the one-dimensional feature vectors into a second convolutional layer of the trained multi-scale neighborhood feature extraction unit to obtain a second neighborhood scale two-level index correlation feature vector, where the second convolutional layer has a second one-dimensional convolution kernel of a second length, and the first length is different from the second length. The concatenation subunit 143 is configured to concatenate the first neighborhood scale two-level index correlation feature vector and the second neighborhood scale two-level index correlation feature vector to obtain the multi-scale two-level index correlation feature vector.

[0069] In one example, in the wind farm grid-connected performance evaluation system 100, the first convolutional subunit 141 is configured to use the first convolutional layer of the trained multi-scale neighborhood feature extraction unit to perform one-dimensional convolution coding on the one-dimensional feature vectors according to the following formula to obtain the first neighborhood scale two-level index correlation feature vector:

[0070]

[0071] wherein a is a width of the first convolution kernel in the x direction, F(a) is a first convolution kernel parameter vector, G(x-a) is a local vector matrix operated with the convolution kernel function, w is a size of the first convolution kernel, and X represents the one-dimensional feature vector; the second convolution subunit 142 is configured to perform one-dimensional convolution coding on the one-dimensional feature vector using a second convolution layer of the trained multi-scale neighborhood feature extraction unit according to the following formula to obtain the second neighborhood scale secondary index correlation feature vector; wherein the formula is:

[0072]

[0073] wherein b is a width of the second convolution kernel in the x direction, F(b) is a second convolution kernel parameter vector, G(x-b) is a local vector matrix operated with the convolution kernel function, m is a size of the second convolution kernel, and X represents the one-dimensional feature vector.

[0074] In the wind farm grid-connected performance evaluation system 100, the fusion unit 150 is configured to fuse the two-dimensional secondary index correlation feature vector and the multi-scale secondary index correlation feature vector to obtain a classification feature vector.

[0075] In one example, in the wind farm grid-connected performance evaluation system 100, the fusion unit 150 is further configured to concatenate the two-dimensional secondary index correlation feature vector and the multi-scale secondary index correlation feature vector to obtain the classification feature vector.

[0076] In the wind farm grid-connected performance evaluation system 100, the evaluation index generation unit 160 is configured to pass the classification feature vector through a classifier to obtain a classification result, which is used to represent a wind farm grid-connected performance level label.

[0077] In one example, in the wind farm grid-connected performance evaluation system 100, the evaluation index generation unit 160 is further configured to use the classifier to process the classification feature vector according to the following formula to obtain the classification result.

[0078] wherein the formula is: softmax{(M2, B2):(M1, B1)|X}, wherein M1 and M2 are weight matrices, B1 and B2 are bias vectors, and X is the classification feature vector.

[0079] In one example, in the wind farm grid-connected performance evaluation system 100, the wind farm grid-connected performance evaluation system further comprises a training module 200 configured to train the context encoder comprising the embedding layer, the multi-scale neighborhood feature extraction module, and the first convolutional neural network as a filter. Figure 5The diagram illustrates a block diagram of a training module in a wind farm grid-connected performance evaluation system according to an embodiment of this application. Figure 5 As shown, the training module 200 of the power grid fault system according to an embodiment of this application includes: a training data acquisition unit 210, used to acquire multiple tertiary indicator data under each secondary indicator; a training tertiary indicator data encoding unit 220, used to pass the multiple tertiary indicator data under each secondary indicator through the context encoder containing the embedding layer to obtain multiple training tertiary indicator semantic feature vectors, and concatenate the multiple training tertiary indicator semantic feature vectors to obtain training secondary indicator feature vectors corresponding to each secondary indicator; a training secondary indicator two-dimensional association encoding unit 230, used to arrange the training secondary indicator feature vectors corresponding to each secondary indicator into a two-dimensional feature matrix and then pass it through the first convolutional neural network as a filter to obtain training two-dimensional secondary indicator association feature vectors; and a training secondary indicator one-dimensional multi-scale encoding unit 240, used to arrange the training secondary indicator feature vectors corresponding to each secondary indicator into a one-dimensional feature vector and then pass it through the first convolutional neural network as a filter to obtain training two-dimensional secondary indicator association feature vectors. The multi-scale neighborhood feature extraction module obtains a training multi-scale secondary indicator association feature vector; the training fusion unit 250 fuses the training two-dimensional secondary indicator association feature vector and the training multi-scale secondary indicator association feature vector to obtain a training classification feature vector; the classification loss unit 260 passes the training classification feature vector through a classifier to obtain a classification loss function value; the classification pattern elimination suppression loss function value calculation unit 270 calculates the classification pattern elimination suppression loss function value based on the L2 norm of the difference feature vector between the training two-dimensional secondary indicator association feature vector and the training multi-scale secondary indicator association feature vector; and the training unit 280 trains the context encoder containing the embedding layer, the multi-scale neighborhood feature extraction module, and the first convolutional neural network as a filter using the weighted sum of the classification pattern elimination suppression loss function value and the classification loss function value as the loss function value.

[0080] In one example, in the wind farm grid-connected performance evaluation system 100 described above, the classification pattern elimination suppression loss function value calculation unit 270 is further used to: calculate the classification pattern elimination suppression loss function value based on the L2 norm of the difference feature vector between the training two-dimensional secondary index association feature vector and the training multi-scale secondary index association feature vector using the following formula;

[0081] The formula is as follows:

[0082]

[0083] wherein V1 and V2 represent the training two-dimensional secondary index correlation feature vector and the training multi-scale secondary index correlation feature vector respectively, M1 and M2 are the weight matrix of the classifier for the training two-dimensional secondary index correlation feature vector and the training multi-scale secondary index correlation feature vector respectively, ||·||F represents the Frobenius norm of a matrix, ||·||2 represents the two-norm of a vector, F wherein ||·||F represents the Frobenius norm of a matrix, ||·||2 represents the two-norm of a vector, wherein diff represents the difference by position, exp(·) represents the matrix exponential operation and the vector exponential operation, the matrix exponential operation represents calculating the natural exponential function value with the eigenvalue of each position in the matrix as the power, and the vector exponential operation represents calculating the natural exponential function value with the eigenvalue of each position in the vector as the power.

[0084] For the two-dimensional secondary index correlation feature vector, for example, denoted as V1, and the multi-scale secondary index correlation feature vector, for example, denoted as V2, since it contains correlation features of different scales, which makes it have different feature mode expressions, and after fusion in the form of, for example, cascading, it may cause abnormal gradient expansion of the classifier in the back propagation process during the training process, leading to the elimination of the mode expressed by the feature, thereby affecting the classification accuracy. Therefore, a classification mode elimination suppression loss function is introduced.

[0085] Here, the classification mode elimination suppression loss function ensures that the derivative during gradient back propagation is regularly obtained near the gradient branch point in a directional manner, so as to over-weight the gradient between modes, so as to suppress the classification mode elimination of the feature, and further improve the classification accuracy.

[0086] In summary, the wind farm grid connection performance evaluation system 100 according to the embodiments of the present application is illustrated, which obtains a plurality of third-level index data through a trained context encoder containing an embedding layer to obtain a plurality of third-level index semantic feature vectors and performs cascading to obtain a secondary index feature vector. After arranging the secondary index feature vector into a two-dimensional feature matrix and a one-dimensional feature vector, respectively, a trained first convolutional neural network and a trained multi-scale neighborhood feature extraction unit are used to obtain a two-dimensional secondary index correlation feature vector and a multi-scale secondary index correlation feature vector, and the two are fused to obtain a classification feature vector, and then a classifier is used to obtain a classification result representing the wind farm grid connection performance level label. In particular, the present application uses convolutional neural networks and multi-scale field feature extraction modules to extract the correlation features between different levels of indicators with different receptive fields, thereby improving the classification accuracy.

[0087] As described above, the wind farm grid connection performance evaluation system 100 according to the embodiments of this application can be implemented in various terminal devices, such as servers with wind farm grid connection performance evaluation functions. In one example, the wind farm grid connection performance evaluation system 100 according to the embodiments of this application can be integrated into the terminal device as a software module and / or hardware module. For example, the wind farm grid connection performance evaluation system 100 can be a software module in the operating system of the terminal device, or it can be an application developed for the terminal device; of course, the wind farm grid connection performance evaluation system 100 can also be one of many hardware modules of the terminal device.

[0088] Alternatively, in another example, the wind farm grid connection performance evaluation system 100 and the terminal device can also be separate devices, and the wind farm grid connection performance evaluation system 100 can be connected to the terminal device via wired and / or wireless networks, and transmit interactive information in accordance with an agreed data format.

[0089] Exemplary Method

[0090] According to another aspect of this application, a method for evaluating the grid connection performance of wind farms is also provided. For example... Figure 6 As shown, the wind farm grid connection performance evaluation method according to the embodiment of this application includes the following steps: S110, acquiring multiple tertiary indicator data under each secondary indicator; S120, passing the multiple tertiary indicator data under each secondary indicator through a trained context encoder including an embedding layer to obtain multiple tertiary indicator semantic feature vectors, and concatenating the multiple tertiary indicator semantic feature vectors to obtain secondary indicator feature vectors corresponding to each secondary indicator; S130, arranging the secondary indicator feature vectors corresponding to each secondary indicator into a two-dimensional feature matrix and passing it through a trained first convolutional neural network as a filter to obtain a two-dimensional secondary indicator association feature vector; S140, arranging the secondary indicator feature vectors corresponding to each secondary indicator into a one-dimensional feature vector and passing it through a trained multi-scale neighborhood feature extraction unit to obtain a multi-scale secondary indicator association feature vector; S150, fusing the two-dimensional secondary indicator association feature vector and the multi-scale secondary indicator association feature vector to obtain a classification feature vector; and S160, passing the classification feature vector through a classifier to obtain a classification result, the classification result being used to represent the wind farm grid connection performance level label.

[0091] In summary, according to the wind farm grid connection performance evaluation method of the embodiment of the present application, the obtained multiple third-level index data is input into the trained context encoder comprising an embedding layer to obtain multiple third-level index semantic feature vectors and then concatenated to obtain a second-level index feature vector. After arranging the second-level index feature vector into a two-dimensional feature matrix and a one-dimensional feature vector respectively, the trained first convolutional neural network and the trained multi-scale neighborhood feature extraction unit are used to obtain a two-dimensional second-level index correlation feature vector and a multi-scale second-level index correlation feature vector, and then the two are fused to obtain a classification feature vector, and finally the classifier is used to obtain a classification result representing the wind farm grid connection performance grade label. In particular, the convolutional neural network and the multi-scale field feature extraction module are used to extract the correlation features between different levels of indexes by using different receptive fields, thereby improving the classification accuracy.

Claims

1. A wind farm grid integration performance evaluation system, characterized in that, The method comprises the following steps: a three-level index data acquisition unit is configured to acquire a plurality of three-level index data under each two-level index; a three-level index data encoding unit is configured to obtain a plurality of three-level index semantic feature vectors by inputting the plurality of three-level index data under each two-level index into a trained context encoder comprising an embedding layer, and concatenate the plurality of three-level index semantic feature vectors to obtain a two-level index feature vector corresponding to each two-level index; a two-level index two-dimensional correlation encoding unit is configured to arrange the two-level index feature vector corresponding to each two-level index into a two-dimensional feature matrix, and then input the two-dimensional feature matrix into a trained first convolutional neural network serving as a filter to obtain a two-dimensional two-level index correlation feature vector; a two-level index one-dimensional multi-scale encoding unit is configured to arrange the two-level index feature vector corresponding to each two-level index into a one-dimensional feature vector, and then input the one-dimensional feature vector into a trained multi-scale neighborhood feature extraction unit to obtain a multi-scale two-level index correlation feature vector; a fusion unit is configured to fuse the two-dimensional two-level index correlation feature vector and the multi-scale two-level index correlation feature vector to obtain a classification feature vector; and an evaluation index generation unit is configured to input the classification feature vector into a classifier to obtain a classification result, wherein the classification result is used to represent a wind farm grid-connected performance level label. The two-level index includes an active power control index, a reactive power regulation index, a low-high voltage ride-through index, a power quality index, a wind farm operation reliability index, and a wind power prediction error index. The three-level index under the active power control index includes a 1 min active power change and a 10 min active power change. The three-level index under the reactive power regulation index includes a wind turbine power factor regulation capability, a wind farm reactive power regulation capability, and a wind farm voltage control capability. The three-level index under the low-high voltage ride-through index includes a low voltage ride-through capability, an active power recovery rate, a dynamic reactive power regulation capability, and a high voltage ride-through capability. The three-level index includes a frequency deviation, a voltage deviation, a voltage fluctuation and flicker, a harmonic voltage, a three-phase voltage allowable imbalance, and an inter-harmonic. The three-level index under the wind farm operation reliability index includes a wind farm unplanned outage coefficient and a wind farm availability coefficient. The three-level index under the wind power prediction error index includes a short-term prediction monthly root mean square error and a super-short-term prediction 4th hour prediction value monthly root mean square error.

2. The wind farm grid integration performance evaluation system of claim 1, wherein, The three-level index data encoding unit comprises: an embedding encoding subunit configured to use an embedding layer of the trained context encoder comprising an embedding layer to respectively convert the plurality of three-level index data under each two-level index into an input vector to obtain a sequence of input vectors; a context encoding subunit configured to use a transformer of the trained context encoder comprising an embedding layer to perform global context semantic encoding on the sequence of input vectors to obtain the plurality of three-level index semantic feature vectors; and a concatenation subunit configured to concatenate the plurality of three-level index semantic feature vectors to obtain the two-level index feature vector corresponding to each two-level index.

3. The wind farm grid integration performance evaluation system of claim 2, wherein, The secondary index two-dimensional correlation coding unit is further configured to respectively perform forward transmission of layers of the trained first convolutional neural network serving as a filter: perform convolution processing on the input data to obtain a convolution feature map; perform mean pooling based on a local feature matrix on the convolution feature map to obtain a pooling feature map; and perform nonlinear activation on the pooling feature map to obtain an activated feature map; wherein an output of a last layer of the first convolutional neural network is the two-dimensional secondary index correlation feature vector, and an input of a first layer of the first convolutional neural network is the two-dimensional feature matrix.

4. The wind farm grid integration performance evaluation system of claim 3, wherein, The secondary index one-dimensional multi-scale coding unit comprises: a first convolutional subunit configured to input the one-dimensional feature vector into a first convolutional layer of the trained multi-scale neighborhood feature extraction unit to obtain a first neighborhood scale secondary index correlation feature vector, wherein the first convolutional layer has a first one-dimensional convolution kernel with a first length; a second convolutional subunit configured to input the one-dimensional feature vector into a second convolutional layer of the trained multi-scale neighborhood feature extraction unit to obtain a second neighborhood scale secondary index correlation feature vector, wherein the second convolutional layer has a second one-dimensional convolution kernel with a second length, and the first length is different from the second length; and a cascading subunit configured to cascade the first neighborhood scale secondary index correlation feature vector and the second neighborhood scale secondary index correlation feature vector to obtain the multi-scale secondary index correlation feature vector.

5. The wind farm grid integration performance evaluation system of claim 4, wherein, The first convolutional subunit is configured to use the first convolutional layer of the trained multi-scale neighborhood feature extraction unit to perform one-dimensional convolution coding on the one-dimensional feature vector according to the following formula to obtain the first neighborhood scale secondary index correlation feature vector: wherein the formula is: wherein a is a width of the first convolution kernel in an x direction, F(a) is a first convolution kernel parameter vector, G(x-a) is a local vector matrix for convolution kernel function operation, w is a size of the first convolution kernel, and X represents the one-dimensional feature vector; The second convolutional subunit is configured to use the second convolutional layer of the trained multi-scale neighborhood feature extraction unit to perform one-dimensional convolution coding on the one-dimensional feature vector according to the following formula to obtain the second neighborhood scale secondary index correlation feature vector: wherein the formula is: wherein b is a width of the second convolution kernel in the x direction, F(b) is a second convolution kernel parameter vector, G(x-b) is a local vector matrix for convolution kernel function operation, m is a size of the second convolution kernel, and X represents the one-dimensional feature vector.

6. The wind farm grid integration performance evaluation system of claim 5, wherein, The fusion unit is further configured to cascade the two-dimensional secondary index correlation feature vector and the multi-scale secondary index correlation feature vector to obtain the classification feature vector.

7. The wind farm grid integration performance evaluation system of claim 6, wherein, The evaluation index generation unit is further configured to use the classifier to process the classification feature vector according to the following formula to obtain the classification result: The formula is: softmax {(M2, B2):(M1, B1)|X}, wherein M1 and M2 are weight matrices, B1 and B2 are bias vectors, and X is the classification feature vector.

8. The wind farm grid integration performance evaluation system of claim 1, wherein, The wind farm grid-connected performance evaluation system further comprises a training module for training the context encoder comprising an embedding layer, the multi-scale neighborhood feature extraction unit and the first convolutional neural network as a filter. The training module comprises: a training data acquisition unit configured to acquire a plurality of third-level indicator data under each second-level indicator; a training third-level indicator data encoding unit configured to obtain a plurality of training third-level indicator semantic feature vectors by inputting the plurality of third-level indicator data under each second-level indicator into the context encoder comprising an embedding layer, and concatenate the plurality of training third-level indicator semantic feature vectors to obtain a training second-level indicator feature vector corresponding to each second-level indicator; a training second-level indicator two-dimensional correlation encoding unit configured to arrange the training second-level indicator feature vector corresponding to each second-level indicator into a two-dimensional feature matrix, and input the two-dimensional feature matrix into the first convolutional neural network as a filter to obtain a training two-dimensional second-level indicator correlation feature vector; a training second-level indicator one-dimensional multi-scale encoding unit configured to arrange the training second-level indicator feature vector corresponding to each second-level indicator into a one-dimensional feature vector, and input the one-dimensional feature vector into the multi-scale neighborhood feature extraction unit to obtain a training multi-scale second-level indicator correlation feature vector; a training fusion unit configured to fuse the training two-dimensional second-level indicator correlation feature vector and the training multi-scale second-level indicator correlation feature vector to obtain a training classification feature vector; a classification loss unit configured to input the training classification feature vector into a classifier to obtain a classification loss function value; a classification mode resolution inhibition loss function value calculation unit configured to calculate the classification mode resolution inhibition loss function value based on a two-norm of a differential feature vector between the training two-dimensional second-level indicator correlation feature vector and the training multi-scale second-level indicator correlation feature vector; and a training unit configured to train the context encoder comprising an embedding layer, the multi-scale neighborhood feature extraction unit and the first convolutional neural network as a filter by taking a weighted sum of the classification mode resolution inhibition loss function value and the classification loss function value as a loss function value.

9. The wind farm grid integration performance evaluation system of claim 8, wherein, The classification mode resolution inhibition loss function value calculation unit is further configured to calculate the classification mode resolution inhibition loss function value based on a two-norm of a differential feature vector between the training two-dimensional second-level indicator correlation feature vector and the training multi-scale second-level indicator correlation feature vector according to the following formula: The formula is: wherein V1 and V2 represent the training two-dimensional secondary indicator correlation feature vector and the training multi-scale secondary indicator correlation feature vector respectively, M1 and M2 are the weight matrix of the classifier for the training two-dimensional secondary indicator correlation feature vector and the training multi-scale secondary indicator correlation feature vector respectively, ||·||F represents the Frobenius norm of a matrix, ||·||2 represents the two-norm of a vector, F represents the Frobenius norm of a matrix, ||·||2 represents the two-norm of a vector, represents the position difference, exp(·) represents the matrix exponential operation and the vector exponential operation, the matrix exponential operation represents calculating the natural exponential function value with the eigenvalue of each position in the matrix as the power, and the vector exponential operation represents calculating the natural exponential function value with the eigenvalue of each position in the vector as the power.

10. A method of evaluating the grid integration performance of a wind farm, characterized in that, comprises: acquiring a plurality of third-level indicator data under each second-level indicator; inputting the plurality of third-level indicator data under each second-level indicator into the trained context encoder comprising an embedding layer to obtain a plurality of third-level indicator semantic feature vectors, and concatenating the plurality of third-level indicator semantic feature vectors to obtain a second-level indicator feature vector corresponding to each second-level indicator; arranging the secondary index feature vectors corresponding to each secondary index into a two-dimensional feature matrix, and then passing the two-dimensional feature matrix through a first convolutional neural network trained as a filter to obtain a two-dimensional secondary index correlation feature vector; arranging the secondary index feature vectors corresponding to each secondary index into a one-dimensional feature vector, and then passing the one-dimensional feature vector through a multi-scale neighborhood feature extraction unit trained to obtain a multi-scale secondary index correlation feature vector; fusing the two-dimensional secondary index correlation feature vector and the multi-scale secondary index correlation feature vector to obtain a classification feature vector; and passing the classification feature vector through a classifier to obtain a classification result, wherein the classification result is used to represent a wind farm grid-connected performance level label; wherein the secondary indexes include active power control indexes, reactive power regulation indexes, low-voltage ride-through indexes, power quality indexes, wind farm operation reliability indexes, and wind power prediction error indexes; the tertiary indexes under the active power control indexes include 1 min active power variation and 10 min active power variation; the tertiary indexes under the reactive power regulation indexes include wind turbine power factor regulation capability, wind farm reactive power regulation capability, and wind farm voltage control capability; the tertiary indexes under the low-voltage ride-through indexes include low-voltage ride-through capability and active power recovery rate, dynamic reactive power regulation capability, and high-voltage ride-through capability; the tertiary indexes include frequency deviation, voltage deviation, voltage fluctuation and flicker, harmonic voltage, three-phase voltage allowable imbalance, and inter-harmonic; the tertiary indexes under the wind farm operation reliability indexes include wind farm unplanned downtime coefficient and wind farm availability coefficient; and the tertiary indexes under the wind power prediction error indexes include short-term prediction monthly root mean square error and ultra-short-term prediction 4th hour prediction value monthly root mean square error.

Images