Reactive active control system and method for wind power plant

Abstract

The invention discloses a reactive active control system and method for a wind power plant. In the control system, an acquisition device is used for acquiring electricity consumption power of all electricity users connected with a power grid at a plurality of preset time points in a preset time period and acquiring active power output by a plurality of wind driven generators connected with the power grid at the plurality of preset time points in the preset time period; the structuring device is used for obtaining a corresponding electricity utilization input matrix and an active power input matrix based on the electricity utilization power and the active power; the feature processing device is used for obtaining a corresponding power consumption feature map and an active power feature map by using a first convolutional neural network model and a non-local neural network model based on the power consumption input matrix and the active power input matrix; and the calculation device is used for obtaining a difference characteristic pattern based on the power consumption characteristic pattern and the active power characteristic pattern, and inputting the difference characteristic pattern into the decoder for decoding regression to output reactive voltage so as to realize reactive control. According to the control system, the grid connection stability of the wind power plant can be improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent control technology, and in particular to a reactive active control system and method for a wind farm. Background Art

[0002] The strong randomness of wind energy causes the active power of wind turbines to change significantly when tracking the maximum wind power generation, which will cause the charging power on the transmission line to fluctuate accordingly. In order to ensure the stability of the power grid, the system needs to provide sufficient reactive power and require the joint work of conventional power sources.

[0003] Taking into account the instability of wind resources, it is impossible to make an accurate prediction of the active power of wind turbines in wind farms. The power grid system adjusts the power generation output according to the user's electricity load. This leads to a contradiction between wind power output and grid regulation output, which in turn affects the normal scheduling and operation of the power grid system.

[0004] Therefore, there is an urgent need for a wind farm reactive power voltage control technology to ensure the stability of wind farm grid connection. Summary of the invention

[0005] The present invention aims to solve one of the technical problems in the related art at least to a certain extent. To this end, the present invention provides a reactive active control system and method for a wind farm, the main purpose of which is to improve the stability of the wind farm grid connection.

[0006] According to a first aspect of the present invention, there is provided a reactive power active control system for a wind farm, comprising:

[0007] A collection device, used to obtain the power consumption of all electricity users connected to the power grid at multiple predetermined time points within a predetermined time period and to obtain the active power output by multiple wind turbines connected to the power grid at multiple predetermined time points within the predetermined time period;

[0008] A structured device for obtaining a corresponding power consumption input matrix and an active power input matrix based on the power consumption and the active power;

[0009] A feature processing device, used to obtain a corresponding power consumption feature map and an active power feature map based on the power consumption input matrix and the active power input matrix using a first convolutional neural network model and a non-local neural network model;

[0010] The computing device is used to obtain a differential characteristic diagram based on the power consumption characteristic diagram and the active power characteristic diagram, and input the differential characteristic diagram into a decoder for decoding and regression to output reactive voltage, so as to realize reactive power control.

[0011] In one embodiment of the present invention, the structuring device includes: the electricity consumption data structuring module, which is used to arrange the electricity consumption of all electricity users connected to the power grid at multiple predetermined time points within a predetermined time period into an electricity consumption input matrix according to the time dimension and the electricity user sample dimension; the active power structuring module, which is used to arrange the active power output by the multiple wind turbines at multiple predetermined time points within the predetermined time period into an active power input matrix according to the time dimension and the wind turbine sample dimension.

[0012] In one embodiment of the present invention, the electricity consumption data structuring module includes: a row vector arrangement unit, which is used to arrange the electricity consumption of each electricity user at multiple predetermined time points within a predetermined time period into electricity consumption input row vectors according to the time dimension; and a two-dimensional matrixing unit, which is used to arrange the electricity consumption input row vectors in two dimensions according to the electricity user sample dimension to obtain the electricity consumption input matrix.

[0013] In one embodiment of the present invention, the feature processing device includes: a local correlation feature extraction module, which is used to pass the power consumption input matrix through a first convolutional neural network model as a feature extractor to obtain a local power consumption correlation feature map; a global correlation feature extraction module, which is used to pass the local power consumption correlation feature map through a non-local neural network model to obtain a global power consumption correlation feature map; a power consumption feature fusion module, which is used to fuse the local power consumption correlation feature map and the global power consumption correlation feature map to obtain a power consumption feature map; an active power feature extraction module, which is used to obtain an active power feature map from the active power input matrix through the first convolutional neural network model and the non-local neural network model.

[0014] In one embodiment of the present invention, the local correlation feature extraction module is specifically used to: use each layer of the first convolutional neural network model as a feature extractor to perform convolution processing, mean pooling processing and nonlinear activation processing on the input data in the forward pass of the layer so that the output of the last layer of the first convolutional neural network model as a feature extractor is the power consumption local correlation feature map, wherein the input of the first layer of the first convolutional neural network model as a feature extractor is the power consumption input matrix.

[0015] In one embodiment of the present invention, the global correlation feature extraction module includes: a point convolution operation unit, which is used to perform three different point convolution operations on the local correlation feature map of power consumption to obtain a first feature map, a second feature map and a third feature map, wherein the first to the third feature maps have the same number of channels; a first fusion unit, which is used to calculate the position point multiplication between the first feature map and the second feature map to obtain a fused feature map; an activation unit, which is used to input the fused feature map into a Softmax activation function for probabilistic activation to obtain a weighted feature map; a second fusion unit, which is used to calculate the weighted feature map and the The third feature map is multiplied by the position points to obtain a re-fused feature map; a global weight feature map generation unit is used to calculate the similarity between any two pixels in the re-fused feature map by embedding a Gaussian similarity function to obtain a global weight feature map; a channel correction unit is used to perform a point convolution operation on the global weight feature map to adjust the number of channels of the global weight feature map to be consistent with the power consumption global correlation feature map to obtain a channel-corrected global weight feature map; a third fusion unit is used to calculate the positional sum of the channel-corrected global weight feature map and the power consumption global correlation feature map to obtain the power consumption global correlation feature map.

[0016] In one embodiment of the present invention, the computing device comprises: a characteristic distribution correction module, used to perform characteristic distribution correction on the power consumption characteristic diagram and the active power characteristic diagram to obtain a corrected power consumption characteristic diagram and a corrected active power characteristic diagram; a difference module, used to calculate a differential characteristic diagram between the corrected power consumption characteristic diagram and the corrected active power characteristic diagram; and a reactive active control result generation module, used to decode and regress the differential characteristic diagram through a decoder to obtain a decoded value, and the decoded value is used to represent the reactive voltage that the multiple wind turbines should provide at the current time point.

[0017] In one embodiment of the present invention, the feature distribution correction module includes: a first weight calculation unit, which is used to calculate the high-frequency enhancement distillation factor of the wavelet-like function family energy aggregation of the power consumption feature map through formula () as the weighted weight of the power consumption feature map; wherein the formula () is:

[0018]

[0019] Among them, f i,j,k Represents the characteristic value of each position of the power consumption characteristic diagram, σ i,j,l (f i,j,k ) represents the eigenvalue set f i,j,k∈F1, and W1, H1 and C1 are the width, height and number of channels of the feature graph F1 respectively, log represents the logarithmic function with base 2, w1 represents the weighted weight of the power consumption feature graph; a first weighted correction unit is used to weight the power consumption feature graph with the weighted weight of the power consumption feature graph to obtain the corrected power consumption feature graph.

[0020] In one embodiment of the present invention, the reactive active control result generation module is specifically used to: use the decoder to decode and regress the differential characteristic map through formula () to obtain the decoded value; wherein formula () is:

[0021]

[0022] Among them, F d represents the differential feature map, Y represents the decoded value, W represents the weight matrix, B represents the bias vector, Represents matrix multiplication.

[0023] According to a second aspect of the present invention, there is also provided a reactive power active control method for a wind farm, comprising:

[0024] Acquire the power consumption of all electricity users connected to the power grid at multiple predetermined time points within a predetermined time period and acquire the active power output by multiple wind turbines connected to the power grid at multiple predetermined time points within the predetermined time period;

[0025] Obtaining a corresponding power consumption input matrix and an active power input matrix based on the power consumption and the active power;

[0026] Based on the power input matrix and the active power input matrix, using a first convolutional neural network model and a non-local neural network model to obtain a corresponding power characteristic graph and an active power characteristic graph;

[0027] A differential characteristic diagram is obtained based on the power consumption characteristic diagram and the active power characteristic diagram, and the differential characteristic diagram is input into a decoder for decoding and regression to output a reactive voltage, so as to realize reactive power control.

[0028] In one or more embodiments of the present invention, a collection device is used to obtain the power consumption of all electricity users connected to the power grid at multiple predetermined time points within a predetermined time period and to obtain the active power output by multiple wind turbines connected to the power grid at multiple predetermined time points within the predetermined time period; a structuring device is used to obtain a corresponding power input matrix and an active power input matrix based on the power consumption and the active power; a feature processing device is used to obtain a corresponding power consumption feature map and an active power feature map based on the power consumption input matrix and the active power input matrix using a first convolutional neural network model and a non-local neural network model; a calculation device is used to obtain a differential feature map based on the power consumption feature map and the active power feature map, and input the differential feature map into a decoder for decoding and regression to output reactive voltage, so as to achieve reactive control. In this case, the power input matrix and the active power input matrix are obtained based on the power consumption and the output active power at multiple predetermined time points within the predetermined time period, and then the corresponding power consumption feature map and the active power feature map are obtained, so as to obtain a differential feature map, and reactive voltage is obtained based on the differential feature map to achieve reactive control, thereby improving the stability of wind farm grid connection.

[0029] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The above and other purposes, features and advantages of the present invention will become more apparent by describing the embodiments of the present invention in more detail in conjunction with the accompanying drawings. The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings, the same reference numerals generally represent the same components or steps. Among them:

[0031] Figure 1 This is an application scenario diagram of a reactive power active control system for a wind farm according to an embodiment of the present invention;

[0032] Figure 2 A block diagram of a reactive active control system for a wind farm according to an embodiment of the present invention;

[0033] Figure 3 is a block diagram of another reactive power active control system for a wind farm according to an embodiment of the present invention;

[0034] Figure 4 A block diagram of a power consumption data structuring module in a reactive power active control system for a wind farm according to an embodiment of the present invention;

[0035] Figure 5A block diagram of a global correlation feature extraction module in a reactive active control system for a wind farm according to an embodiment of the present invention;

[0036] Figure 6 is a flow chart of a reactive power active control method for a wind farm according to an embodiment of the present invention;

[0037] Figure 7 Schematic diagram of the architecture operation of the reactive power active control method for a wind farm according to an embodiment of the present invention. DETAILED DESCRIPTION

[0038] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the embodiments of the present invention. Instead, they are merely examples of devices and methods consistent with some aspects of the embodiments of the present invention as detailed in the appended claims.

[0039] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0040] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "multiple" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined. It should also be understood that the term "and / or" used in the present invention refers to and includes any or all possible combinations of one or more associated listed items.

[0041] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0042] Considering the instability of wind resources, it is impossible to make an accurate prediction of the active power of wind turbines in wind farms. The power grid system adjusts the power output according to the user's power load, which leads to a contradiction between wind power output and power grid regulation output, thus affecting the normal dispatching and operation of the power grid system. Therefore, a wind farm reactive voltage control scheme is expected to ensure the stability of wind farm grid connection.

[0043] In view of the above technical problems, the present invention provides a reactive power active control system and method for wind farms, the main purpose of which is to improve the stability of wind farm grid connection. The reactive power active control system and method of the present invention are key to the control of reactive power of wind farms in two aspects: on the one hand, they can accurately analyze and predict the power demand of all electricity users connected to the grid; on the other hand, they can accurately analyze and predict the active power output of multiple wind turbines in the wind farm connected to the grid.

[0044] In a first embodiment, Figure 1 FIG. 1 is an application scenario diagram of a reactive power active control system for a wind farm according to an embodiment of the present invention. Figure 1 As shown, in this application scenario, firstly, the electric power C of all electricity users T connected to the power grid at multiple predetermined time points within a predetermined time period, and the active power M output by multiple wind turbines F connected to the power grid at multiple predetermined time points within a predetermined time period are obtained; then, the obtained electric power and active power are input into a server S in which a reactive active control system of a wind farm is deployed, wherein the server can process the electric power and active power using the reactive active control algorithm of the wind farm to generate a decoded value, which is used to represent the reactive voltage value that the multiple wind turbines should provide at the current time point.

[0045] Figure 2 FIG. 1 is a block diagram of a reactive power active control system for a wind farm according to an embodiment of the present invention. Figure 2As shown, a reactive active control system 10 for a wind farm includes a collection device 11, a structuring device 12, a feature processing device 13 and a computing device 14, wherein the collection device 11 is used to obtain the power consumption of all electricity users connected to the power grid at multiple predetermined time points within a predetermined time period and to obtain the active power output of multiple wind turbines connected to the power grid at multiple predetermined time points within a predetermined time period; the structuring device 12 is used to obtain the corresponding power input matrix and active power input matrix based on the power consumption and active power; the feature processing device 13 is used to obtain the corresponding power consumption feature map and active power feature map based on the power input matrix and the active power input matrix using a first convolutional neural network model and a non-local neural network model; the computing device 14 is used to obtain a differential feature map based on the power consumption feature map and the active power feature map, and input the differential feature map into a decoder for decoding and regression to output reactive voltage to achieve reactive control.

[0046] In this embodiment, the power consumption of all electricity users connected to the power grid at multiple predetermined time points within a predetermined time period is first obtained by using the acquisition device 11. It is easy to understand that among all electricity users connected to the power grid, each electricity user is not an independent electricity user. For example, in the office area, the electricity demand of each electricity user is consistent in time distribution. For example, in the working area, some electricity users have a cooperative relationship, and their electricity demand will be complementary in time distribution. By making full use of the implicit relationship between electricity users, the accuracy of electricity demand prediction and analysis of electricity users can be improved.

[0047] In this embodiment, the structuring device 12 includes: a power consumption data structuring module, which is used to arrange the power consumption of all power users connected to the power grid at multiple predetermined time points within a predetermined time period into a power consumption input matrix according to the time dimension and the power user sample dimension; an active power structuring module, which is used to arrange the active power output by multiple wind turbines at multiple predetermined time points within a predetermined time period into an active power input matrix according to the time dimension and the wind turbine sample dimension. The power consumption data structuring module includes: a row vector arranging unit, which is used to arrange the power consumption of each power user at multiple predetermined time points within a predetermined time period into power consumption input row vectors according to the time dimension; and a two-dimensional matrixing unit, which is used to arrange the power consumption input row vectors in two dimensions according to the power user sample dimension to obtain the power consumption input matrix.

[0048] Among them, the power consumption data structuring module is used to arrange the power consumption of all power users connected to the power grid at multiple predetermined time points within a predetermined time period into a power consumption input matrix according to the time dimension and the power user sample dimension. That is, the power consumption of all power users at multiple predetermined time points within a predetermined time period is two-dimensionally structured, wherein the value of each position in the power consumption input matrix is ​​the power consumption value of the corresponding power user at the corresponding time point. In some embodiments, the two-dimensional structuring process includes: firstly arranging the power consumption of each power user at multiple predetermined time points within a predetermined time period into a power consumption input row vector according to the time dimension, and then arranging the power consumption input row vector in two dimensions according to the power user sample dimension to obtain the power consumption input matrix.

[0049] In this embodiment, the feature processing device 13 includes: a local correlation feature extraction module, which is used to pass the power consumption input matrix through the first convolutional neural network model as a feature extractor to obtain a local power consumption correlation feature map; a global correlation feature extraction module, which is used to pass the local power consumption correlation feature map through the non-local neural network model to obtain a global power consumption correlation feature map; a power consumption feature fusion module, which is used to fuse the local power consumption correlation feature map and the global power consumption correlation feature map to obtain a power consumption feature map; an active power feature extraction module, which is used to obtain an active power feature map from the active power input matrix through the first convolutional neural network model and the non-local neural network model.

[0050] Specifically, the local correlation feature extraction module is used to pass the power input matrix through the first convolutional neural network model as a feature extractor to obtain a power local correlation feature map. That is, a convolutional neural network model with excellent performance in the field of local feature extraction is used as a feature extractor to extract high-dimensional local implicit correlation features in the power input matrix, that is, high-dimensional implicit correlation features associated with the power consumption of different power users at the same time point, high-dimensional implicit correlation features associated with the power consumption of the same power user at different time points, and high-dimensional implicit correlation features associated with the power consumption of different power users at different time points.

[0051] In this embodiment, the local correlation feature extraction module is specifically used to: use each layer of the first convolutional neural network model as a feature extractor to perform convolution processing, mean pooling processing and nonlinear activation processing on the input data in the forward pass of the layer so that the output of the last layer of the first convolutional neural network model as a feature extractor is the local correlation feature map of electricity consumption, wherein the input of the first layer of the first convolutional neural network model as a feature extractor is the electricity consumption input matrix.

[0052] Considering that although the first convolutional neural network model can use its convolution kernel as a local feature extraction operator to extract high-dimensional local implicit correlation features, the convolutional neural network model has certain limitations. For example, in a 5×5 convolution, the convolution filter (i.e., the convolution kernel) has 25 pixels, and the value of the target pixel is calculated with reference to itself and the surrounding 24 pixels, which means that the convolution can only use local information to calculate the target pixel, and the local convolution operation often causes some deviations because it cannot refer to global information. There are some technical means to alleviate this problem, such as using a larger convolution filter or a deep network with more convolution layers. However, these methods often bring a large amount of parameters, and the improvement of the results is very limited. The embodiment of the present invention uses a global correlation feature extraction module to make the relationship between the electricity demand of all electricity users global, avoiding the inability to effectively obtain all the features of the electricity association between electricity users locally like the traditional convolution extraction method. The embodiment of the present invention adds global features to better extract the association between the electricity demand of all electricity users, thereby improving the accuracy of electricity demand estimation.

[0053] In order to obtain the global characteristics of electricity demand of electricity users, a non-local neural network model is introduced. In this embodiment, after obtaining the global correlation feature map of electricity consumption from the electricity input matrix using the first convolutional neural network, the local correlation feature map of electricity consumption is further passed through the non-local neural network model to obtain the global correlation feature map of electricity consumption.

[0054] In this embodiment, the encoding process of the non-local neural network model includes first performing point convolution on the local associated feature map of electricity consumption to adjust the number of channels of the feature map. Specifically, in the non-local neural network model, three different point convolution operations are performed on the local associated feature map of electricity consumption to obtain the first feature map, the second feature map and the third feature map, wherein the first to third feature maps have the same number of channels; then, the positional point multiplication between the first feature map and the second feature map is calculated to obtain a fusion feature map, and the fusion feature map is input into the Softmax activation function for probabilistic activation to obtain a weight feature map; then, the positional point multiplication between the weight feature map and the third feature map is calculated to obtain a re-fusion feature map; then, the similarity between any two pixels in the re-fusion feature map is calculated by embedding a Gaussian similarity function to obtain a global weight feature map; and the number of channels of the global weight feature map is adjusted to be consistent with that of the global associated feature map of electricity consumption by performing point convolution operation on the global weight feature map; finally, the positional sum of the global weight feature map and the global associated feature map of electricity consumption is calculated to obtain the global associated feature map of electricity consumption.

[0055] In this embodiment, the local power consumption correlation feature map and the global power consumption correlation feature map are fused to obtain the power consumption feature map. In some embodiments, the local power consumption correlation feature map and the global power consumption correlation feature map are fused by calculating the position-weighted sum between the two to obtain the power consumption feature map.

[0056] In this embodiment, the active power output of multiple wind turbines in a wind farm connected to the power grid is accurately analyzed and predicted, and the active power output of multiple wind turbines connected to the power grid at multiple predetermined time points within a predetermined time period is encoded in the same sampling manner to obtain an active power characteristic diagram. That is, the active power output of multiple wind turbines connected to the power grid at multiple predetermined time points within a predetermined time period is first two-dimensionally matrixed, and then local features are extracted through a convolutional neural network model, and global features are extracted with a non-local neural network model to obtain an active power characteristic diagram. Then, the difference between the power consumption characteristic diagram and the active power characteristic diagram is calculated to obtain a feature representation for representing the reactive power control demand at the current point. That is, the differential feature diagram between the power consumption characteristic diagram and the active power characteristic diagram is calculated, and the differential feature diagram is decoded and regressed through a decoder to obtain a decoded value of the reactive voltage that should be provided by multiple wind turbines at the current time point.

[0057] In this embodiment, the computing device 14 includes: a characteristic distribution correction module, which is used to perform characteristic distribution correction on the power consumption characteristic diagram and the active power characteristic diagram to obtain a corrected power consumption characteristic diagram and a corrected active power characteristic diagram; a difference module, which is used to calculate the differential characteristic diagram between the corrected power consumption characteristic diagram and the corrected active power characteristic diagram; and a reactive active control result generation module, which is used to decode and regress the differential characteristic diagram through a decoder to obtain a decoded value, and the decoded value is used to represent the reactive voltage that a plurality of wind turbines should provide at the current time point.

[0058] In this embodiment, when calculating the differential feature map between the power consumption feature map and the active power feature map using the computing device 14, since the power consumption feature map and the active power feature map respectively fuse the global and local features, the power consumption feature map and the active power feature map can achieve global and local feature aggregation. However, for the aggregated global and local features, when calculating their differences, it is still necessary to suppress the difference in information expression between the global features and the local features in the full-scale space as much as possible. Therefore, the high-frequency enhancement distillation factor of the wavelet-like energy aggregation is calculated for the power consumption feature map and the active power feature map, wherein the general formula of the high-frequency enhancement distillation factor of the wavelet-like energy aggregation is:

[0059]

[0060] In the formula, f i,j,k,1Represents the eigenvalues ​​of each position of the corresponding feature map, σ i,j,k (f i,j,k,1 ) represents the eigenvalue set f i,j,k,1 ∈F, and W, H and C are the width, height and number of channels of the feature map F respectively, log represents the logarithmic function with base 2, and w represents the weighted weight of the corresponding feature map.

[0061] In this embodiment, considering that the information representation of feature distribution tends to be concentrated on high-frequency components, that is, information tends to be distributed on the edge of the manifold of the high-dimensional manifold of the feature graph, the high-frequency enhancement distillation method of wavelet-like energy aggregation can be used to enhance the high-frequency components of the hidden state features through the distillation of the set variance, and constrain its low-frequency components, so as to strengthen the basic information of the feature graph in the full-scale spatial expression as much as possible, and by weighting the power consumption feature graph and the active power feature graph with it, the degree of information expression fusion of the global features and local features of the power consumption feature graph and the active power feature graph in the full-scale space can be improved. In this case, by calculating its differential feature graph, the global information expression ability of the differential feature graph can be further improved, so as to improve the accuracy of the decoded value obtained by decoding and regressing the differential feature graph through the decoder, thereby improving the accuracy and adaptability of the reactive active control of the wind farm, so as to maintain the stability of the power grid.

[0062] Figure 3 A block diagram of a reactive active control system for a wind farm according to an embodiment of the present invention; Figure 4 A block diagram of a power consumption data structuring module in a reactive power active control system for a wind farm according to an embodiment of the present invention; Figure 5 4 is a block diagram of a global correlation feature extraction module in a reactive power active control system for a wind farm according to an embodiment of the present invention.

[0063] like Figure 3 As shown, the reactive power active control system 100 for a wind farm comprises:

[0064] The power consumption data collection module 101 is used to obtain the power consumption of all power users connected to the power grid at multiple predetermined time points within a predetermined time period;

[0065] The power consumption data structuring module 102 is used to arrange the power consumption of all power users connected to the power grid at multiple predetermined time points within a predetermined time period into a power consumption input matrix according to the time dimension and the power user sample dimension;

[0066] A local correlation feature extraction module 103, used to pass the power consumption input matrix through a first convolutional neural network model as a feature extractor to obtain a power consumption local correlation feature map;

[0067] A global correlation feature extraction module 104 is used to obtain a global correlation feature map of electricity consumption by passing the local correlation feature map of electricity consumption through a non-local neural network model;

[0068] The power consumption feature fusion module 105 is used to fuse the power consumption local correlation feature map and the power consumption global correlation feature map to obtain the power consumption feature map;

[0069] The active power output data acquisition module 106 is used to obtain the active power outputted by a plurality of wind turbines connected to the power grid at a plurality of predetermined time points within a predetermined time period;

[0070] The active power structuring module 107 is used to arrange the active powers output by the plurality of wind turbines at a plurality of predetermined time points within a predetermined time period into an active power input matrix according to the time dimension and the wind turbine sample dimension;

[0071] An active power feature extraction module 108, configured to obtain an active power feature graph from an active power input matrix through a first convolutional neural network model and a non-local neural network model;

[0072] The characteristic distribution correction module 109 is used to perform characteristic distribution correction on the power consumption characteristic diagram and the active power characteristic diagram to obtain a corrected power consumption characteristic diagram and a corrected active power characteristic diagram;

[0073] The difference module 110 is used to calculate a difference characteristic diagram between the corrected power consumption characteristic diagram and the corrected active power characteristic diagram;

[0074] The reactive active control result generating module 111 is used to decode and regress the differential characteristic diagram through a decoder to obtain a decoded value, and the decoded value is used to represent the reactive voltage that the multiple wind turbines should provide at the current time point.

[0075] In some embodiments, the power consumption data collection module 101 is first used to obtain the power consumption of all power users connected to the power grid at multiple predetermined time points within a predetermined time period. It is easy to understand that among all power users connected to the power grid, each power user is not an independent power user. For example, in the office area, the power consumption demand of each power user is consistent in time distribution. For example, in the working area, some power users have a cooperative relationship, and their power consumption demand will be complementary in time distribution. By making full use of the implicit power consumption relationship between power users, the accuracy of power user power demand prediction and analysis can be improved.

[0076] In some embodiments, the power consumption data structuring module 102 is used to arrange the power consumption of all power users connected to the power grid at multiple predetermined time points within a predetermined time period into a power consumption input matrix according to the time dimension and the power user sample dimension. That is, the power consumption of all power users at multiple predetermined time points within a predetermined time period is two-dimensionally structured, wherein the value of each position in the power consumption input matrix is ​​the power consumption value of the corresponding power user at the corresponding time point. In a specific example, the two-dimensional structuring process includes: firstly arranging the power consumption of each power user at multiple predetermined time points within a predetermined time period into a power consumption input row vector according to the time dimension, and then arranging the power consumption input row vector in two dimensions according to the power user sample dimension to obtain the power consumption input matrix.

[0077] Specifically, Figure 4 As shown, the electricity consumption data structuring module 102 includes a row vector arrangement unit 201 and a two-dimensional matrixing unit 202, wherein the row vector arrangement unit 201 is used to arrange the electricity consumption of each electricity user at multiple predetermined time points within a predetermined time period into electricity consumption input row vectors according to the time dimension; the two-dimensional matrixing unit 202 is used to arrange the electricity consumption input row vectors in two dimensions according to the electricity user sample dimension to obtain an electricity consumption input matrix.

[0078] In some embodiments, the local correlation feature extraction module 103 is used to pass the power input matrix through the first convolutional neural network model as a feature extractor to obtain a local power correlation feature map. Next, the power input matrix is ​​passed through the first convolutional neural network model as a feature extractor to obtain a local power correlation feature map. That is, a convolutional neural network model with excellent performance in the field of local feature extraction is used as a feature extractor to extract high-dimensional local implicit correlation features in the power input matrix, that is, high-dimensional implicit correlation features associated with the power consumption of different electricity users at the same time point, high-dimensional implicit correlation features associated with the power consumption of the same electricity user at different time points, and high-dimensional implicit correlation features associated with the power consumption of different electricity users at different time points.

[0079] In some embodiments, the local correlation feature extraction module 103 is further used to: use each layer of the first convolutional neural network model as a feature extractor to perform convolution processing, mean pooling processing and nonlinear activation processing on the input data in the forward pass of the layer so that the output of the last layer of the first convolutional neural network model as a feature extractor is a local correlation feature map of electricity consumption, wherein the input of the first layer of the first convolutional neural network model as a feature extractor is a power consumption input matrix.

[0080] Considering that although the first convolutional neural network model can use its convolution kernel as a local feature extraction operator to extract high-dimensional local implicit correlation features, the convolutional neural network model has certain limitations. Specifically, for example, in a 5×5 convolution, the convolution filter (i.e., the convolution kernel) has 25 pixels, and the value of the target pixel is calculated with reference to itself and the surrounding 24 pixels, which means that the convolution can only use local information to calculate the target pixel, and the local convolution operation often causes some deviations because it cannot refer to global information. There are some technical means to alleviate this problem, such as using larger convolution filters or deep networks with more convolution layers. However, these methods often bring a large amount of parameters, and the improvement of the results is very limited. Therefore, the global correlation feature extraction module 104 is used to obtain globality. Specifically, the global correlation feature extraction module 104 is used to obtain a global correlation feature map of electricity consumption through a non-local neural network model, so as to make the relationship between the electricity demands of all electricity users global, thereby avoiding the problem that the traditional convolution extraction method cannot effectively obtain all the characteristics of the electricity consumption association between electricity users locally. The addition of global features can better extract the association between the electricity demands of all electricity users, thereby improving the accuracy of electricity demand estimation.

[0081] In addition, in order to obtain the global characteristics of the electricity demand of the electricity user, a non-local neural network model is introduced. Therefore, in an embodiment of the present invention, after obtaining the global associated feature map of electricity consumption from the electricity input matrix using the first convolutional neural network, the local associated feature map of electricity consumption is further passed through the non-local neural network model to obtain the global associated feature map of electricity consumption.

[0082] In some embodiments, Figure 5As shown, the global correlation feature extraction module 104 includes: a point convolution operation unit 301, which is used to perform three different point convolution operations on the local correlation feature map of power consumption to obtain a first feature map, a second feature map and a third feature map, wherein the first to third feature maps have the same number of channels; a first fusion unit 302, which is used to calculate the position point multiplication between the first feature map and the second feature map to obtain a fused feature map; an activation unit 303, which is used to input the fused feature map into the Softmax activation function for probabilistic activation to obtain a weighted feature map; a second fusion unit 304, which is used to calculate the weighted feature map and the third feature map. The feature maps are multiplied point by point to obtain a re-fused feature map; a global weight feature map generation unit 305 is used to calculate the similarity between any two pixels in the re-fused feature map by embedding a Gaussian similarity function to obtain a global weight feature map; a channel correction unit 306 is used to perform a point convolution operation on the global weight feature map to adjust the number of channels of the global weight feature map to be consistent with the global power consumption correlation feature map to obtain a channel-corrected global weight feature map; a third fusion unit 307 is used to calculate the positional sum of the channel-corrected global weight feature map and the global power consumption correlation feature map to obtain the global power consumption correlation feature map.

[0083] That is, the encoding process of the non-local neural network model includes first performing point convolution on the local associated feature map of electricity consumption to adjust the number of channels of the feature map. Specifically, in the non-local neural network model, three different point convolution operations are performed on the local associated feature map of electricity consumption to obtain the first feature map, the second feature map and the third feature map, wherein the first to third feature maps have the same number of channels; then, the positional point multiplication between the first feature map and the second feature map is calculated to obtain a fused feature map, and the fused feature map is input into the Softmax activation function for probabilistic activation to obtain a weighted feature map; then, the positional point multiplication between the weighted feature map and the third feature map is calculated to obtain a re-fused feature map; then, the similarity between any two pixels in the re-fused feature map is calculated by embedding a Gaussian similarity function to obtain a global weighted feature map; and the number of channels of the global weighted feature map is adjusted to be consistent with that of the global associated feature map of electricity consumption by performing point convolution operations on the global weighted feature map; finally, the positional sum of the global weighted feature map and the global associated feature map of electricity consumption is calculated to obtain the global associated feature map of electricity consumption.

[0084] In some embodiments, the power consumption feature fusion module 105 is specifically used to: fuse the power consumption local correlation feature map and the power consumption global correlation feature map by calculating the position-weighted sum between the two to obtain the power consumption feature map. That is, the power consumption local correlation feature map and the power consumption global correlation feature map are fused by formula (1) to obtain the power consumption feature map; wherein formula (1) is:

[0085] F s =λFa +βF g (1)

[0086] Where F s Represents the power consumption characteristic diagram, F a represents the local correlation feature map of electricity consumption, F g represents the global power consumption correlation feature map, "+" represents the addition of the elements at corresponding positions of the local power consumption correlation feature map and the global power consumption correlation feature map, and λ and β represent weighting parameters for controlling the balance between the local power consumption correlation feature map and the global power consumption correlation feature map.

[0087] In some embodiments, after using the active power output data acquisition module 106 to obtain the active power output of multiple wind turbines connected to the power grid at multiple predetermined time points within a predetermined time period, the active power output of multiple wind turbines in the wind farm connected to the power grid is accurately analyzed and predicted, and the active power output of multiple wind turbines connected to the power grid at multiple predetermined time points within a predetermined time period is encoded in the same sampling manner to obtain an active power characteristic diagram.

[0088] In some embodiments, the active power structuring module 107 is used to arrange the active power output by multiple wind turbines at multiple predetermined time points within a predetermined time period into an active power input matrix according to the time dimension and the wind turbine sample dimension. Then, the active power output by multiple wind turbines connected to the power grid at multiple predetermined time points within a predetermined time period is two-dimensionally matrixed. That is, the active power output by multiple wind turbines at multiple predetermined time points within a predetermined time period is two-dimensionally structured, wherein the value of each position in the active power input matrix corresponds to the active power output by multiple wind turbines at multiple predetermined time points within a predetermined time period. In a specific example, the two-dimensional matrixing process includes: firstly arranging the active power output by multiple wind turbines at multiple predetermined time points within a predetermined time period into active power input row vectors according to the time dimension, and then arranging the active power input row vectors in two dimensions according to the wind turbine sample dimension to obtain the active power input matrix.

[0089] In some embodiments, in the active power feature extraction module 108, local feature extraction is performed by a convolutional neural network model, and global feature extraction is performed by a non-local neural network model to obtain an active power feature map. That is, a convolutional neural network model with excellent performance in the field of local feature extraction is used as a feature extractor to extract high-dimensional local implicit correlation features in the active power input matrix, that is, high-dimensional implicit correlation features associated with the active power of different wind turbines at the same time point, high-dimensional implicit correlation features associated with the active power of the same wind turbine at different time points, and high-dimensional implicit correlation features associated with the active power of different wind turbines at different time points.

[0090] Considering that although the first convolutional neural network model can use its convolution kernel as a local feature extraction operator to extract high-dimensional local implicit correlation features, the convolutional neural network model has certain limitations. In an embodiment of the present invention, the relationship between the active power of multiple wind turbines is global, and using the traditional convolution extraction method, it is impossible to effectively obtain all the features of the active power correlation between multiple wind turbines locally, and adding global features can better extract the correlation between the active power of multiple wind turbines, thereby improving the accuracy of active power estimation. Therefore, in an embodiment of the present invention, in order to obtain the global features of the active power of multiple wind turbines, a non-local neural network model is introduced. That is, through the first convolutional neural network model and the non-local neural network model, the active power feature map is obtained from the active power input matrix.

[0091] In some embodiments, the feature distribution correction module 109 is used to perform feature distribution correction on the power consumption feature graph and the active power feature graph to obtain a corrected power consumption feature graph and a corrected active power feature graph. In particular, in an embodiment of the present invention, when calculating the differential feature graph between the power consumption feature graph and the active power feature graph, since the power consumption feature graph and the active power feature graph respectively integrate global and local features, the power consumption feature graph and the active power feature graph can achieve global and local feature aggregation, but for the aggregated global and local features, when calculating their differences, it is still necessary to suppress the difference in information expression between the global features and the local features in the full-scale space as much as possible.

[0092] In some embodiments, the feature distribution correction module includes a first weight calculation unit and a first weight correction unit. The first weight calculation unit is used to calculate the high-frequency enhancement distillation factor of the wavelet-like function family energy aggregation of the power consumption feature map by formula (2) as the weighted weight of the power consumption feature map; wherein formula (2) is:

[0093]

[0094] Among them, f i,j,k Represents the characteristic value of each position of the power consumption characteristic diagram, σ i,j,k (f i,j,k ) represents the eigenvalue set f i,j,k ∈F1, and W1, H1 and C1 are the width, height and number of channels of the feature map F1 respectively, log represents the logarithmic function with base 2, w1 represents the weighted weight of the power consumption feature map; the first weighted correction unit is used to weight the power consumption feature map with the weighted weight of the power consumption feature map to obtain the corrected power consumption feature map.

[0095] In some embodiments, the feature distribution correction module includes a second weight calculation unit and a second weighted correction unit. The second weight calculation unit is used to calculate the high-frequency enhancement distillation factor of the wavelet-like function family energy aggregation of the active power feature graph as the weighted weight of the active power feature graph through formula (3); wherein formula (3) is:

[0096]

[0097] Among them, f l,m,n Represents the characteristic value of each position of the active power characteristic diagram, σ l,m,n (f l,m,n ) represents the eigenvalue set f l,m,n ∈F2, and W2, H2 and C2 are the width, height and number of channels of the feature graph F2 respectively, w2 represents the weighted weight of the active power feature graph; the second weighted correction unit is used to weight the active power feature graph with the weighted weight of the active power feature graph to obtain the corrected active power feature graph.

[0098] In some embodiments, the high-frequency enhanced distillation factor of the wavelet-like energy aggregation is calculated for the power consumption feature graph and the active power feature graph. Considering that the information representation of the feature distribution tends to be concentrated on the high-frequency component, that is, the information tends to be distributed on the edge of the manifold of the high-dimensional manifold of the feature graph, the high-frequency enhanced distillation method of the wavelet-like energy aggregation can be used to enhance the high-frequency component of the hidden state feature by the distillation of the set variance, and constrain its low-frequency component, so as to strengthen the basic information of the feature graph in the full-scale spatial expression as much as possible, and by weighting the power consumption feature graph and the active power feature graph with it, the degree of information expression fusion of the global features and local features of the power consumption feature graph and the active power feature graph in the full-scale space can be improved. In this case, by calculating the differential feature graph, the global information expression ability of the differential feature graph can be further improved to improve the accuracy of the decoded value obtained by decoding and regressing the differential feature graph through the decoder. Thus, the accuracy and adaptability of the reactive active control of the wind farm can be improved, so as to maintain the stability of the power grid.

[0099] In some embodiments, the differential module 110 is used to calculate the differential characteristic diagram between the corrected power consumption characteristic diagram and the corrected active power characteristic diagram. Further, by calculating the difference between the power consumption characteristic diagram and the active power characteristic diagram, a characteristic representation for representing the reactive power control demand at the current point can be obtained. That is, by calculating the differential characteristic diagram between the power consumption characteristic diagram and the active power characteristic diagram, and decoding and regressing the differential characteristic diagram through a decoder, a decoded value for representing the reactive voltage that should be provided by multiple wind turbines at the current time point can be obtained.

[0100] In some embodiments, the differential module is further used to: use formula (4) to calculate the differential characteristic diagram between the corrected power consumption characteristic diagram and the corrected active power characteristic diagram; wherein formula (4) is:

[0101]

[0102] Among them, F d represents the differential characteristic diagram, F1 represents the power consumption characteristic diagram after correction, and F2 represents the active power characteristic diagram after correction. Represents positional subtraction.

[0103] In some embodiments, the reactive active control result generation module 111 is specifically used to: use a decoder to perform decoding regression on the differential feature map through formula (5) to obtain a decoding value; wherein formula (5) is:

[0104]

[0105] Among them, F d represents the differential feature map, Y represents the decoded value, W represents the weight matrix, B represents the bias vector, Represents matrix multiplication.

[0106] In the reactive active control system for a wind farm in an embodiment of the present invention, a collection device is used to obtain the power consumption of all electricity users connected to the power grid at multiple predetermined time points within a predetermined time period and to obtain the active power output of multiple wind turbines connected to the power grid at multiple predetermined time points within a predetermined time period; a structuring device is used to obtain a corresponding power input matrix and an active power input matrix based on the power consumption and active power; a feature processing device is used to obtain a corresponding power consumption feature map and an active power feature map based on the power input matrix and the active power input matrix using a first convolutional neural network model and a non-local neural network model; a computing device is used to obtain a differential feature map based on the power consumption feature map and the active power feature map, and input the differential feature map into a decoder for decoding and regression to output reactive voltage to achieve reactive control. In this case, the power input matrix and the active power input matrix are obtained based on the power consumption and the output active power at multiple predetermined time points within a predetermined time period, and then the corresponding power consumption feature map and the active power feature map are obtained, thereby obtaining a differential feature map, and the reactive voltage is obtained based on the differential feature map to achieve reactive control, thereby improving the stability of the wind farm connected to the grid. Specifically, the reactive active control system of the present invention extracts local features of the power consumption of all electricity users connected to the power grid through the first convolutional neural network model, and extracts global features through the non-local neural network model; then, extracts local features of the active power output by multiple wind turbines connected to the power grid through the convolutional neural network model, and extracts global features through the non-local neural network model; then, compares the feature consistency between the power consumption features of the electricity users and the active power features of the wind turbines to obtain a differential feature map, and uses this to obtain the reactive voltage that the multiple wind turbines should provide at the current time point. In this way, the accuracy and adaptability of the reactive active control of the wind farm can be improved, which is conducive to maintaining the stability of the power grid.

[0107] As described above, the reactive active control system for a wind farm according to an embodiment of the present invention can be implemented in various terminal devices, such as a server having a reactive active control algorithm for a wind farm. In some examples, the reactive active control system for a wind farm can be integrated into the terminal device as a software module and / or a hardware module. For example, the reactive active control system for a wind farm can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the reactive active control system for a wind farm can also be one of the many hardware modules of the terminal device. Alternatively, in other examples, the reactive active control system for a wind farm and the terminal device can also be separate devices, and the reactive active control system for a wind farm can be connected to the terminal device via a wired and / or wireless network, and the interactive information is transmitted in accordance with an agreed data format.

[0108] The following is a method embodiment of the present invention, which is used in a reactive power active control system embodiment of a wind farm to perform reactive power control. For details not disclosed in the method embodiment of the present invention, please refer to the system embodiment of the present invention.

[0109] Figure 6 is a flow chart of a reactive power active control method for a wind farm according to an embodiment of the present invention; Figure 7 Schematic diagram of the architecture operation of the reactive power active control method for a wind farm according to an embodiment of the present invention.

[0110] like Figure 6 As shown, the reactive power active control method for a wind farm comprises:

[0111] Step S11, obtaining the power consumption of all electricity users connected to the power grid at multiple predetermined time points within a predetermined time period and obtaining the active power output of multiple wind turbines connected to the power grid at multiple predetermined time points within a predetermined time period.

[0112] Step S12: obtaining a corresponding power consumption input matrix and an active power input matrix based on the power consumption and the active power.

[0113] Specifically, in step S12, the power consumption of all electricity users connected to the power grid at multiple predetermined time points within a predetermined time period is arranged into a power consumption input matrix according to the time dimension and the electricity user sample dimension; the active power output by multiple wind turbines at multiple predetermined time points within a predetermined time period is arranged into an active power input matrix according to the time dimension and the wind turbine sample dimension.

[0114] In step S12, the power consumption of all electricity users connected to the power grid at multiple predetermined time points within a predetermined time period is arranged into a power consumption input matrix according to the time dimension and the electricity user sample dimension, including: arranging the power consumption of each electricity user at multiple predetermined time points within the predetermined time period into power consumption input row vectors according to the time dimension; arranging the power consumption input row vectors in two dimensions according to the electricity user sample dimension to obtain the power consumption input matrix.

[0115] Step S13, based on the power input matrix and the active power input matrix, using the first convolutional neural network model and the non-local neural network model to obtain the corresponding power characteristic graph and active power characteristic graph.

[0116] Specifically, in step S13, the power consumption input matrix is ​​passed through the first convolutional neural network model as a feature extractor to obtain a local power consumption correlation feature map; the local power consumption correlation feature map is passed through the non-local neural network model to obtain a global power consumption correlation feature map; the local power consumption correlation feature map and the global power consumption correlation feature map are fused to obtain a power consumption feature map; and the active power feature map is obtained from the active power input matrix through the first convolutional neural network model and the non-local neural network model.

[0117] In step S13, the electricity consumption input matrix is ​​passed through the first convolutional neural network model as a feature extractor to obtain a local electricity consumption correlation feature map, including: using each layer of the first convolutional neural network model as a feature extractor to perform convolution processing, mean pooling processing and nonlinear activation processing on the input data in the forward pass of the layer so that the output of the last layer of the first convolutional neural network model as a feature extractor is the local electricity consumption correlation feature map, wherein the input of the first layer of the first convolutional neural network model as a feature extractor is the electricity consumption input matrix.

[0118] In step S13, the local power consumption correlation feature map is passed through a non-local neural network model to obtain a global power consumption correlation feature map, including: performing three different point convolution operations on the local power consumption correlation feature map to obtain a first feature map, a second feature map and a third feature map, wherein the first to third feature maps have the same number of channels; calculating the positional point multiplication between the first feature map and the second feature map to obtain a fused feature map; inputting the fused feature map into a Softmax activation function for probabilistic activation to obtain a weighted feature map; calculating the positional point multiplication between the weighted feature map and the third feature map to obtain a re-fused feature map; using an embedded Gaussian similarity function to calculate the similarity between any two pixels in the re-fused feature map to obtain a global weighted feature map; performing a point convolution operation on the global weighted feature map to adjust the number of channels of the global weighted feature map to be consistent with that of the global power consumption correlation feature map to obtain a channel-corrected global weighted feature map; and calculating the positional sum of the channel-corrected global weighted feature map and the global power consumption correlation feature map to obtain the global power consumption correlation feature map.

[0119] In step S13, the local power consumption correlation feature map and the global power consumption correlation feature map are integrated to obtain the power consumption feature map, including: integrating the local power consumption correlation feature map and the global power consumption correlation feature map through formula (1) to obtain the power consumption feature map; wherein formula (1) is:

[0120] F s =λF a +βF g (1)

[0121] Where F s Represents the power consumption characteristic diagram, F arepresents the local correlation feature map of electricity consumption, F g represents the global power consumption correlation feature map, "+" represents the addition of the elements at corresponding positions of the local power consumption correlation feature map and the global power consumption correlation feature map, and λ and β represent weighting parameters for controlling the balance between the local power consumption correlation feature map and the global power consumption correlation feature map.

[0122] Step S14, obtaining a differential characteristic diagram based on the power consumption characteristic diagram and the active power characteristic diagram, and inputting the differential characteristic diagram into a decoder for decoding and regression to output reactive voltage, so as to realize reactive power control.

[0123] Specifically, in step S14, characteristic distribution correction is performed on the power consumption characteristic diagram and the active power characteristic diagram to obtain a corrected power consumption characteristic diagram and a corrected active power characteristic diagram; a differential characteristic diagram between the corrected power consumption characteristic diagram and the corrected active power characteristic diagram is calculated; the differential characteristic diagram is decoded and regressed through a decoder to obtain a decoded value, and the decoded value is used to represent the reactive voltage that a plurality of wind turbines should provide at the current time point.

[0124] In step S14, characteristic distribution correction is performed on the power consumption characteristic graph and the active power characteristic graph to obtain a corrected power consumption characteristic graph and a corrected active power characteristic graph, including: a high-frequency enhancement distillation factor for energy aggregation of a wavelet-like function family of the power consumption characteristic graph is calculated by formula (2) as a weighted weight of the power consumption characteristic graph; wherein formula (2) is:

[0125]

[0126] Among them, f i,j,k Represents the characteristic value of each position of the power consumption characteristic diagram, σ i,j,k (f i,j,k ) represents the eigenvalue set f i,j,k ∈F1, and W1, H1 and C1 are the width, height and number of channels of the feature map F1 respectively, log represents the logarithmic function with base 2, and w1 represents the weighted weight of the power consumption feature map; the power consumption feature map is weighted by the weighted weight of the power consumption feature map to obtain the corrected power consumption feature map.

[0127] In step S14, the power consumption characteristic graph and the active power characteristic graph are subjected to characteristic distribution correction to obtain the corrected power consumption characteristic graph and the corrected active power characteristic graph, including: calculating the high-frequency enhancement distillation factor of the wavelet-like function family energy aggregation of the active power characteristic graph by formula (3) as the weighted weight of the active power characteristic graph; wherein formula (3) is:

[0128]

[0129] Among them, f l,m,n Represents the characteristic value of each position of the active power characteristic diagram, σl,m,n (f l,m,n ) represents the eigenvalue set f l,m,n ∈F2, and W2, H2 and C2 are the width, height and number of channels of the feature map F2 respectively, w2 represents the weighted weight of the active power feature map; the active power feature map is weighted by the weighted weight of the active power feature map to obtain the corrected active power feature map.

[0130] In step S14, calculating the difference characteristic diagram between the corrected power consumption characteristic diagram and the corrected active power characteristic diagram includes: using formula (4) to calculate the difference characteristic diagram between the corrected power consumption characteristic diagram and the corrected active power characteristic diagram; wherein formula (4) is:

[0131]

[0132] Among them, F d represents the differential characteristic diagram, F1 represents the power consumption characteristic diagram after correction, and F2 represents the active power characteristic diagram after correction. Represents positional subtraction.

[0133] In step S14, the differential characteristic diagram is decoded and regressed by a decoder to obtain a decoded value, and the decoded value is used to represent the reactive voltage that the multiple wind turbines should provide at the current time point, including: using the decoder to decode and regress the differential characteristic diagram by formula (5) to obtain a decoded value; wherein formula (5) is:

[0134]

[0135] Among them, F d represents the differential feature map, Y represents the decoded value, W represents the weight matrix, B represents the bias vector, Represents matrix multiplication.

[0136] like Figure 7 As shown, in the architecture of the reactive power active control method for wind farms, the operation process is as follows:

[0137] First, the power consumption of all electricity users connected to the power grid at multiple predetermined time points within a predetermined time period is obtained; then, the power consumption of all electricity users connected to the power grid at multiple predetermined time points within the predetermined time period is arranged into a power consumption input matrix according to the time dimension and the power user sample dimension; then, the power consumption input matrix is ​​passed through the first convolutional neural network model as a feature extractor to obtain a local power consumption correlation feature map; then, the local power consumption correlation feature map is passed through the non-local neural network model to obtain a global power consumption correlation feature map; then, the local power consumption correlation feature map and the global power consumption correlation feature map are fused to obtain a power consumption feature map; then, the active power output of multiple wind turbines connected to the power grid at multiple predetermined time points within the predetermined time period is obtained. ; Then, the active power output by multiple wind turbines at multiple predetermined time points within a predetermined time period is arranged into an active power input matrix according to the time dimension and the wind turbine sample dimension; then, the active power characteristic diagram is obtained from the active power input matrix through the first convolutional neural network model and the non-local neural network model; then, the power consumption characteristic diagram and the active power characteristic diagram are subjected to feature distribution correction to obtain a corrected power consumption characteristic diagram and a corrected active power characteristic diagram; then, the differential characteristic diagram between the corrected power consumption characteristic diagram and the corrected active power characteristic diagram is calculated; and, finally, the differential characteristic diagram is decoded and regressed through a decoder to obtain a decoded value, and the decoded value is used to represent the reactive voltage that the multiple wind turbines should provide at the current time point.

[0138] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0139] In the reactive active control method for a wind farm in an embodiment of the present invention, the power consumption of all electricity users connected to the power grid at multiple predetermined time points within a predetermined time period and the active power output of multiple wind turbines connected to the power grid at multiple predetermined time points within a predetermined time period are obtained; based on the power consumption and active power, a corresponding power input matrix and an active power input matrix are obtained; based on the power input matrix and the active power input matrix, a first convolutional neural network model and a non-local neural network model are used to obtain a corresponding power characteristic diagram and an active power characteristic diagram; based on the power characteristic diagram and the active power characteristic diagram, a differential characteristic diagram is obtained, and the differential characteristic diagram is input into a decoder for decoding and regression to output reactive voltage, so as to realize reactive control. In this case, based on the power consumption and the output active power at multiple predetermined time points within a predetermined time period, a power input matrix and an active power input matrix are obtained, and then the corresponding power characteristic diagram and active power characteristic diagram are obtained, so as to obtain a differential characteristic diagram, and reactive voltage is obtained based on the differential characteristic diagram to realize reactive control, thereby improving the stability of the wind farm connected to the grid. Specifically, the reactive active control method of the present invention extracts local features of the power consumption of all electricity users connected to the power grid through a first convolutional neural network model, and extracts global features through a non-local neural network model; then, extracts local features of the active power output by multiple wind turbines connected to the power grid through a convolutional neural network model, and extracts global features through a non-local neural network model; then, compares the feature consistency between the electricity consumption features of the electricity users and the active power features of the wind turbines to obtain a differential feature map, and uses this to obtain the reactive voltage that the multiple wind turbines should provide at the current time point. In this way, the accuracy and adaptability of the reactive active control of the wind farm can be improved, which is conducive to maintaining the stability of the power grid.

[0140] It should be noted that the reactive power active control system for wind farms provided in the above embodiment only uses the division of the above functional modules as an example when executing the reactive power active control method for wind farms. In actual applications, the above functional distribution can be completed by different functional modules as needed, that is, the internal structure of the reactive power active control system for wind farms is divided into different functional modules to complete all or part of the functions described above. In addition, the reactive power active control method for wind farms provided in the above embodiment and the reactive power active control system embodiment for wind farms belong to the same concept, and the implementation process thereof is detailed in the method embodiment, which will not be repeated here.

[0141] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution disclosed in the present invention can be achieved, and the present invention is not limited here.

[0142] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

[0143] The basic principle of the present invention is described above in conjunction with specific embodiments. However, it should be pointed out that the advantages, strengths, effects, etc. mentioned in the present invention are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. must be possessed by each embodiment of the present invention. In addition, the specific details disclosed above are only for the purpose of illustration and facilitation of understanding, rather than limitation, and the above details do not limit the present invention to being implemented by adopting the above specific details.

[0144] The block diagrams of the devices, apparatuses, equipment, and systems involved in the present invention are only illustrative examples and are not intended to require or imply that they must be connected, arranged, and configured in the manner shown in the block diagrams. As will be appreciated by those skilled in the art, these devices, apparatuses, equipment, and systems can be connected, arranged, and configured in any manner. Words such as "including", "comprising", "having", etc. are open words, referring to "including but not limited to", and can be used interchangeably with them. The words "or" and "and" used here refer to the words "and / or" and can be used interchangeably with them, unless the context clearly indicates otherwise. The word "such as" used here refers to the phrase "such as but not limited to", and can be used interchangeably with it.

[0145] It should also be noted that in the device, apparatus and method of the present invention, each component or each step can be decomposed and / or reassembled, and such decomposition and / or reassembly should be regarded as an equivalent solution of the present invention.

[0146] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of the present invention. Therefore, the present invention is not intended to be limited to the aspects shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

[0147] The above description has been given for the purpose of illustration and description. In addition, this description is not intended to limit the embodiments of the present invention to the forms disclosed herein. Although a number of example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, changes, additions and sub-combinations thereof.

Claims

1. A reactive power active control system for a wind farm, characterized in that: include: A collection device, used to obtain the power consumption of all electricity users connected to the power grid at multiple predetermined time points within a predetermined time period and to obtain the active power output by multiple wind turbines connected to the power grid at multiple predetermined time points within the predetermined time period; A structured device for obtaining a corresponding power consumption input matrix and an active power input matrix based on the power consumption and the active power; A feature processing device, used to obtain a corresponding power consumption feature map and an active power feature map based on the power consumption input matrix and the active power input matrix using a first convolutional neural network model and a non-local neural network model; The computing device is used to obtain a differential characteristic diagram based on the power consumption characteristic diagram and the active power characteristic diagram, and input the differential characteristic diagram into a decoder for decoding and regression to output reactive voltage, so as to realize reactive power control.

2. The reactive power active control system for a wind farm according to claim 1, characterized in that: The structural device comprises: The power consumption data structuring module is used to arrange the power consumption of all the power users connected to the power grid at multiple predetermined time points within a predetermined time period into a power consumption input matrix according to the time dimension and the power user sample dimension; The active power structuring module is used to arrange the active power output by the multiple wind turbines at multiple predetermined time points within the predetermined time period into an active power input matrix according to the time dimension and the wind turbine sample dimension.

3. The reactive power active control system for a wind farm according to claim 2, characterized in that: The electricity consumption data structuring module comprises: A row vector arrangement unit, used to arrange the power consumption of each electricity user at a plurality of predetermined time points within a predetermined time period into power consumption input row vectors according to the time dimension; The two-dimensional matrixing unit is used to arrange the power input row vector in two dimensions according to the power user sample dimension to obtain the power input matrix.

4. The reactive power active control system for a wind farm according to claim 3, characterized in that: The feature processing device comprises: A local correlation feature extraction module, used for passing the power consumption input matrix through a first convolutional neural network model as a feature extractor to obtain a power consumption local correlation feature map; A global correlation feature extraction module, used for obtaining a global correlation feature map of electricity consumption by passing the local correlation feature map of electricity consumption through a non-local neural network model; A power consumption feature fusion module, used for fusing the power consumption local correlation feature map and the power consumption global correlation feature map to obtain a power consumption feature map; An active power feature extraction module is used to obtain an active power feature graph from the active power input matrix through the first convolutional neural network model and the non-local neural network model.

5. The reactive power active control system for a wind farm according to claim 4, characterized in that: The local correlation feature extraction module is specifically used for: Using each layer of the first convolutional neural network model as a feature extractor, convolution processing, mean pooling processing and nonlinear activation processing are performed on the input data in the forward pass of the layer so that the output of the last layer of the first convolutional neural network model as a feature extractor is the local correlation feature map of electricity consumption, wherein the input of the first layer of the first convolutional neural network model as a feature extractor is the electricity consumption input matrix.

6. The reactive power active control system for a wind farm according to claim 5, characterized in that: The global correlation feature extraction module comprises: a point convolution operation unit, configured to perform three different point convolution operations on the power consumption local correlation feature map to obtain a first feature map, a second feature map, and a third feature map, wherein the first to third feature maps have the same number of channels; A first fusion unit, configured to calculate the position point multiplication between the first feature map and the second feature map to obtain a fused feature map; An activation unit, used for inputting the fused feature map into a Softmax activation function for probabilistic activation to obtain a weighted feature map; A second fusion unit, used for calculating the position point multiplication between the weight feature map and the third feature map to obtain a re-fused feature map; A global weight feature map generating unit, used to calculate the similarity between any two pixels in the re-fused feature map by embedding a Gaussian similarity function to obtain a global weight feature map; A channel correction unit, configured to perform a point convolution operation on the global weight feature map to adjust the number of channels of the global weight feature map to be consistent with the power consumption global correlation feature map to obtain a channel-corrected global weight feature map; The third fusion unit is used to calculate the positional sum of the channel-corrected global weight feature map and the power consumption global correlation feature map to obtain the power consumption global correlation feature map.

7. The reactive power active control system for a wind farm according to claim 6, characterized in that: The computing device comprises: A characteristic distribution correction module, used for performing characteristic distribution correction on the power consumption characteristic diagram and the active power characteristic diagram to obtain a corrected power consumption characteristic diagram and a corrected active power characteristic diagram; A difference module, used for calculating a difference characteristic diagram between the corrected power consumption characteristic diagram and the corrected active power characteristic diagram; The reactive active control result generation module is used to decode and regress the differential characteristic diagram through a decoder to obtain a decoded value, and the decoded value is used to represent the reactive voltage that the multiple wind turbines should provide at the current time point.

8. The reactive power active control system for a wind farm according to claim 7, characterized in that: The feature distribution correction module comprises: A first weight calculation unit, used for calculating the high-frequency enhancement distillation factor of the wavelet-like function family energy aggregation of the power consumption characteristic diagram through formula () as the weighted weight of the power consumption characteristic diagram; Wherein, the formula () is: Among them, f i,j,k Represents the characteristic value of each position of the power consumption characteristic diagram, σ i,j,k (f i,j,k ) represents the eigenvalue set f i,j,k ∈F1, and W1, H1 and C1 are the width, height and number of channels of the feature map F1 respectively, log represents the logarithmic function with base 2, and w1 represents the weighted weight of the power consumption feature map; The first weighted correction unit is used to weight the power usage characteristic diagram with the weighted weight of the power usage characteristic diagram to obtain the corrected power usage characteristic diagram.

9. The reactive power active control system for a wind farm according to claim 8, characterized in that: The reactive active control result generation module is specifically used for: Using the decoder to perform decoding regression on the differential feature map through formula () to obtain the decoding value; Among them, the formula () is: where F d represents the differential feature map, Y represents the decoded value, W represents the weight matrix, B represents the bias vector, Represents matrix multiplication.

10. A reactive power active control method for a wind farm, characterized in that: include: Acquire the power consumption of all electricity users connected to the power grid at multiple predetermined time points within a predetermined time period and acquire the active power output by multiple wind turbines connected to the power grid at multiple predetermined time points within the predetermined time period; Obtaining a corresponding power consumption input matrix and an active power input matrix based on the power consumption and the active power; Based on the power input matrix and the active power input matrix, using a first convolutional neural network model and a non-local neural network model to obtain a corresponding power characteristic graph and an active power characteristic graph; A differential characteristic diagram is obtained based on the power consumption characteristic diagram and the active power characteristic diagram, and the differential characteristic diagram is input into a decoder for decoding and regression to output a reactive voltage, so as to realize reactive power control.

Images