Zone area load component identification method based on image coding technology

Abstract

The invention discloses a transformer area load component identification method based on an image coding technology, and the method comprises the steps: firstly, analyzing the cluster characteristics of different types of load components, screening and determining a characteristic power curve of each type of component, and coding the power curve into a two-dimensional image through employing a Gramer angle field method; then, based on measurement data of a small number of observable users, similar power curves of various load components are generated by adopting a synthetic minority class oversampling technology, and then a training data set of a load component identification model is synthesized; on the basis, a load component image is input, the model is trained by using the synthetic data set, and the power proportions of various load components in the transformer area are identified through the trained model. According to the method, external factor data such as weather and environment are not needed, under the actual measurement condition of the current user power data, the identification precision of the transformer area load components is effectively improved, and accurate information support is provided for measures such as load prediction and demand response of a power distribution network.

Description

Technical Field

[0001] The present invention relates to the field of electrical engineering technology, and in particular to a method for identifying load components in a substation based on image coding technology. Background Art

[0002] With the proposal of the "dual carbon" goal and the development of smart grids, the penetration rate of distributed power sources in distribution networks continues to increase, and the randomness and volatility of their output have brought a series of challenges to the stable operation of the power system. Since most distributed power sources are installed behind the meter, the measuring device can only monitor and obtain net load data, resulting in the insignificant specific power share of various typical load components in the distribution station area, making it difficult to provide accurate information support for load forecasting, demand-side response, voltage control and other measures. At the same time, it is not practical to install separate metering equipment for all dispersed load component resources, which is not only costly, but also involves user privacy issues. Therefore, it is necessary to solve this problem through a feasible algorithm, directly from the net load at the low-voltage distribution station level, accurately identify the power share of different types of load components, and improve the observability of the distribution network.

[0003] Existing load component identification methods can be mainly divided into two categories: one is based on the idea of ​​physical model, which establishes a power characteristic model through parameters such as meteorological data and capacity information, and then calculates the specific power of the load component. However, the physical models under different distribution systems may not be consistent, and the model parameters in practice are difficult to obtain accurately, which makes this type of method difficult to implement; the other is based on the idea of ​​data-driven, which analyzes the cluster characteristics of load components based on a large amount of measurement data to help identify the power proportion of load components from the net load. However, the above methods all directly process the measured time series, but models with excellent performance such as convolutional neural networks are more suitable for processing image features. The features extracted from the time series are relatively limited, which is not conducive to further improving the identification accuracy of load components. Based on the idea of ​​image coding, the power time series is converted into an image, and the advanced models in the field of image processing can be used to solve the problem of load component identification and realize accurate identification of load components in the substation area. Summary of the invention

[0004] In view of the problems existing in the prior art, the present invention utilizes image coding technology and improves the image processing model, and proposes a method for identifying the load composition of the distribution network based on image coding technology, so as to provide accurate information support for load forecasting, demand response and other measures of the distribution network.

[0005] To achieve the above object, the present invention adopts the following technical solution: a method for identifying load components in a substation based on image coding technology, comprising the following steps:

[0006] (1) Taking the day as the unit, the characteristic power curve of each load component is determined according to the cluster characteristics of the load components; the characteristic curves of the net load and each load component are encoded into a two-dimensional grayscale image using the Gram angle and field method;

[0007] (2) Based on the known measured data of each component, the synthetic minority oversampling technique is used to generate similar load component power curves; the generated power data sets are combined as the power curves of the load components in the same substation, and then the net load curve and the power proportion of each load component are superimposed to obtain the training data set of the synthetic substation identification model;

[0008] (3) The net load grayscale image and the load component grayscale image are input into the recognition model, and the image features are extracted through the encoder composed of residual blocks and convolutional layers; the features of the net load and various load components are respectively fused into multiple sets of composite features, and the images corresponding to various components are respectively identified using the separation module composed of convolutional layers and deconvolutional layers; and the authenticity of the identified image is judged by the discriminator using the framework of the generative adversarial network. In the adversarial game between the load component identification detector and the discriminator, the weighted sum of the adversarial loss, L1 loss, and perceptual loss is used as the loss function, and the model parameters are updated by back propagation according to the real images of the training data set; until a trained load component identification model is obtained;

[0009] (4) For the day to be identified, the actual measured data of the net load of the substation and various components are encoded into images and input into the trained load component identification model to obtain the image corresponding to the total power of the load components in the substation; then, the image is inversely encoded into a power curve to complete the load component identification task of the substation.

[0010] Furthermore, the step (1) is specifically as follows: as the number of users increases, the load component cluster shows strong regularity and is highly correlated with the sum of the load components in the substation area. Therefore, the sum of the power curves of the observable load components is selected as the characteristic power curve of each component, and the expression is as follows:

[0011]

[0012] Among them, P fl is the characteristic curve of the load component; is the power curve of the i-th group of observable load components, N l The total number of curves included for the observable load components;

[0013] Normalize the net load curve and the characteristic curves of the load components to the range [0,1]:

[0014]

[0015] in, is the data of the tth time point on the dth day after the power curve is normalized; P d,t is the data of the power curve at the t-th time point on the d-th day, and P is the data set consisting of all power curves;

[0016] The normalized power curve is used as the angle cosine value, the timestamp is the radius, and converted to the polar coordinate system:

[0017]

[0018] in, is the power value of the normalized daily power curve at the i-th point, They are Polar angle and polar diameter in polar coordinate system, t i is the timestamp, N is the total number of sampling points of the daily power curve;

[0019] Calculate the cosine value of the sum of the polar angles corresponding to each point on the daily power curve to obtain the Gram angle and field:

[0020]

[0021] Where I is a unit row vector; the diagonal elements of the GASF matrix retain the original power information of the load components, and the remaining elements reflect the time correlation between different sampling points;

[0022] According to the pixel value range, the range of GASF matrix elements is converted from [0,1] to [0,255] to represent the pixel value of the load component image, and then the grayscale image of the net load and load component is drawn.

[0023] Furthermore, the step (2) is specifically as follows: calculating the similarity between each curve in the historical data set of the observable load component and the other curves in the feature space, and finding the K nearest neighbors of each type of component;

[0024] Among them, the characteristics and changing laws of the load components are characterized by the daily characteristic indicators of the power curve, so the daily characteristic indicators are selected as the daily power average value, maximum value, minimum value, mean square error, skewness, and kurtosis to measure the similarity between the curves;

[0025] Generate similar load component power curves using minority class oversampling techniques:

[0026]

[0027] in, represents the power data of the qth group of load components generated by the observable user i on the dth day, is the power data of the load component measured by observable user i on day d, is a similar power curve selected from the K nearest neighbors, rand(0,1) is a random number between [0,1], and Q is the number of generated power curves;

[0028] Based on the maximum value of the load component, set the capacity factor to determine the power level of the generated load component:

[0029]

[0030] Among them, r l represents the capacity coefficient of the load component. Since the capacity difference between user load components is relatively small, the coefficient range is selected as [0.5, 1.5]; P nl The composite load component curve set is obtained by normalizing the generated curve to its maximum value and then multiplying it by the set capacity value;

[0031] Combine the generated data sets of various components as the load components in the same substation; superimpose to obtain the total power of the load components in the substation and the net load data to synthesize the simulation data set for model training:

[0032]

[0033] in, They are the power values ​​of the comprehensive power of load components and net load at the t-th time point on the d-th day in the simulation training set; is the power value of the generated power curve of the i-th group at the t-th time point on the d-th day, N l is the number of generated curves; is the power value of the sum of the power of the j-th load component at the t-th time point on the d-th day, and M is the number of load component categories.

[0034] Furthermore, the step (3) is specifically as follows: the load component identification model includes a load component identification module and a discrimination module, the load component identification module includes an encoder submodule and a plurality of heads, the encoder is used to extract the image features of the input net load and the load components, the head is used to perform the identification task of a certain type of load component to obtain the image recognition result of the component, and the discrimination module is used to judge the authenticity of the image recognition result, thereby learning through game with the identification module and training the model parameters; until a trained load component identification model is obtained;

[0035] The encoder submodule includes three branches, each of which includes multiple convolutional layers and residual blocks, and captures multi-scale image features through convolution kernels of different sizes and network layers; the encoder submodule extracts the features of the input image and inputs them into the head, the number of heads is the same as the number of categories of the load components, and is used to output images corresponding to each type of component; each group of heads includes a feature fusion module and an image reconstruction module, the feature fusion module splices the features of the net load and the corresponding load component in the channel dimension, and reduces the dimension to obtain the fusion feature; the image reconstruction module includes multiple convolutional layers and deconvolution layers, and reconstructs the image corresponding to a single type of load component according to the fusion feature;

[0036] The discrimination module receives a load component identification image and an actual image as input, calculates the authenticity of the input image through multiple convolutional layers, and iteratively updates the parameters of the load component identification model in a confrontation game between the recognition module and the discrimination module.

[0037] Furthermore, in the step (3), the weighted sum of adversarial loss, L1 loss and perceptual loss is used as the loss function, and the model parameters are updated by back propagation according to the real image of the training data set. Specifically, during the model training process, the weighted sum of adversarial loss, L1 loss and perceptual loss is used to form a composite loss function, and the load component identification model is updated by back propagation; wherein the adversarial loss is used to measure the overall quality and authenticity of the load component image; the L1 loss and perceptual loss are used to measure the similarity of the content and texture features of the load component image respectively; the calculation formula is as follows:

[0038]

[0039] L S (S,D)=λ1L GAN (S,D)+λ2L L1 (S)+λ3L VGG (S);

[0040] Among them, X is the input load component image, S is the identification module, S(X) is the identification result, and Y is the real load component image; L S is the total loss function of the model, λ1,λ2,λ3 are the weight coefficients of the three types of loss functions, which are initially set to equal weights and dynamically adjusted during the training process. When a certain loss function decreases too fast, its weight coefficient is reduced; L GAN To counter the loss, the model performance can be measured; L L1 L1 loss is the average absolute error calculated pixel by pixel to measure the content loss of the image; L VGG For perceptual loss, the high-level features of image texture information are extracted through the pre-trained VGG-19 network, and then the similarity of images in the feature space is measured; is the feature output of the jth layer of the i-th convolutional module of the VGG network, Wi,j ,H i,j are the height and width of the feature, ω i is the weight corresponding to the i-th group of features, which is selected as the weight coefficient used by the pix2pixHD network.

[0041] Furthermore, in step (4), for the day to be identified, the characteristic power curves of the net load and the observable load components are encoded into images and input into the trained load component identification model to obtain the image corresponding to the sum of the power of the load components in the substation area; then, the identification result of the load components in the substation area is obtained by the inverse encoding technology of the image; the expression is as follows:

[0042]

[0043] in, is the identification value of the load component power curve at point i, is the value of the i-th point on the diagonal of the GASF matrix corresponding to the load component image.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] 1. The present invention takes into account the limitations of processing measurement data from the perspective of time series, and identifies the power proportion of the load components in the substation through the idea of ​​image coding and image separation. It does not require external factors such as meteorology and environment, and effectively integrates the characteristics of the net load and the observable load components, making the identification results of the load components more accurate.

[0046] 2. The Grami angular field algorithm is used to implement image coding, which can effectively retain the time correlation of the load, remove redundant information, and give full play to the powerful feature extraction ability of the model;

[0047] 3. Based on the idea of ​​data enhancement, a small amount of power data of load components is used to generate new samples, and then the training data set is synthesized. This can realize semi-supervised training of the model in the absence of measurement data, so that the load component identification model can be applied to more actual operation scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a framework diagram for implementing the method for identifying the components of load in a substation area of ​​the present invention;

[0049] Figure 2 A schematic diagram of the load component image encoding in the present invention;

[0050] Figure 3 It is a structural schematic diagram of the load component identification module in the present invention. DETAILED DESCRIPTION

[0051] Here, exemplary embodiments are described in detail, and examples thereof are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application.

[0052] The terms used in this application are for the purpose of describing specific embodiments only and are not intended to limit this application. The singular forms of "a", "said" and "the" used in this application and the appended claims are also intended to include plural forms unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0053] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0054] In order to make full use of the measured data of net load and various load components, the present invention combines the image coding algorithm and the improved image separation model to propose a method for identifying the load components of the distribution area based on image coding, so as to identify the power proportion of various load components in the distribution area, thereby accurately knowing the load status of the distribution network and supporting the dispatching and operation of the power grid. Figure 1 The method for identifying load components in a substation area of ​​the present invention comprises the following steps:

[0055] (1) Based on the cluster characteristics of load components, the characteristic power curves of each component are determined on a daily basis. The Gram angle and field method is used to encode the net load and the characteristic curves of each load component into a two-dimensional grayscale image.

[0056] First, as the number of users increases, the load component clusters show stronger regularity, and the correlation with the sum of the load components in the substation area also increases significantly. Therefore, the sum of the power curves of the significant load components is selected as the characteristic power curve of each component:

[0057]

[0058] Among them, P fl is the characteristic curve of the load component; is the power curve of the i-th group of observable load components, N l The total number of curves included for the observable load component.

[0059] Secondly, if Figure 2 As shown, image encoding of the load component characteristic curve includes the following steps:

[0060] (1.1) Normalize the net load curve and the characteristic curve of the load component to the range of [0,1]:

[0061]

[0062] in, is the data of the tth time point on the dth day after the power curve is normalized; P d,t is the data of the power curve at the t-th time point on the d-th day, and P is the data set consisting of all power curves.

[0063] (1.2) The normalized power curve is used as the angle cosine value, the timestamp is the radius, and converted to the polar coordinate system:

[0064]

[0065] in, is the power value of the normalized daily power curve at the i-th point, They are Polar angle and polar diameter in polar coordinate system, t i is the timestamp, and N is the total number of sampling points of the daily power curve.

[0066] (1.3) Calculate the cosine value of the sum of the polar angles corresponding to each point on the daily power curve to obtain the Gram angle and field:

[0067]

[0068] Where I is a unit row vector. The diagonal elements of the GASF matrix retain the original power information of the load components, and the remaining elements reflect the time correlation between different sampling points.

[0069] (1.4) According to the pixel value range, the range of GASF matrix elements is converted from [0,1] to [0,255] to represent the pixel value of the load component image, and then the grayscale image of the net load and load component is drawn.

[0070] (2) Based on the known measured data of each component, the synthetic minority oversampling technique is used to generate similar load component power curves. The generated power data sets are combined as the power curves of the load components in the same substation, and then the net load curve and the power proportion of each load component are superimposed to obtain the training data set of the synthetic substation identification model.

[0071] First, the similarity between each curve in the historical data set of the observable load component and the other curves in the feature space is calculated to find the K nearest neighbors of each component. To ensure accuracy, the value of K should be relatively small, and the historical data set can be divided into a training set and a validation set and selected using the cross-validation method.

[0072] Among them, the characteristics and changing laws of the load components are represented by the daily characteristic indicators of the power curve. The daily characteristic indicators are selected as the daily power average, maximum, minimum, mean square error, skewness, and kurtosis to measure the similarity between curves.

[0073] Subsequently, the minority class oversampling technique is used to generate similar load component power curves:

[0074]

[0075] in, represents the power data of the qth group of load components generated by the observable user i on the dth day, is the power data of the load component measured by observable user i on day d, is a similar power curve selected from the K nearest neighbors, rand(0,1) is a random number between [0,1], and Q is the number of generated power curves.

[0076] Then, based on the maximum value of the load component, the capacity factor is set to determine the power level of the generated load component:

[0077] P nl =[P nl / max(P nl )]*r l *max(P ol )r l ∈[0.5,1.5];

[0078] Among them, r l It represents the capacity coefficient of the load component. Since the capacity difference between user load components is relatively small, the coefficient range is selected as [0.5,1.5]. nl The synthetic load component curve set is obtained by normalizing the generated curve to its maximum value and then multiplying it by the set capacity value.

[0079] Then, the generated data sets of various components are combined as the load components in the same substation. The total power of the load components in the substation and the net load data are obtained by superposition to synthesize the simulation data set for model training:

[0080]

[0081]

[0082] in, They are respectively the power values ​​of the comprehensive load components in the simulation training area and the net load at the tth time point on the dth day. is the power value of the generated power curve of the i-th group at the t-th time point on the d-th day, N l is the number of generated curves. is the power value of the sum of the power of the j-th load component at the t-th time point on the d-th day, and M is the number of load component categories.

[0083] (3) The net load grayscale image and the load component grayscale image are input into the recognition model, and the image features are extracted through the encoder composed of residual blocks and convolutional layers. Then, the features of the net load and various load components are fused into multiple sets of composite features, and the images corresponding to each component are respectively identified using the separation module composed of convolutional layers and deconvolutional layers. At the same time, the framework of the generative adversarial network is adopted to judge the authenticity of the identified image through the discriminator. Under the adversarial game between the load component identifier and the discriminator, the weighted sum of adversarial loss, L1 loss, and perceptual loss is used as the loss function, and the model parameters are updated by back propagation based on the real images of the training data set.

[0084] The load component identification model includes a load component identification module and a discrimination module. Figure 3 The schematic diagram of the load component identification module is shown in Figure 1. The identification module includes an encoder and multiple heads. The encoder is used to extract the image features of the input net load and load components. The head is used to perform the identification task of a certain type of load component to obtain the image recognition result of this type of component. The discrimination module is used to judge the authenticity of the image recognition result, and thus to learn from the game with the identification module and train the model parameters.

[0085] like Figure 3 As shown in the figure, the encoder submodule includes three branches, each of which includes multiple convolutional layers and residual blocks, and captures multi-scale image features through convolution kernels of different sizes and network layers. The encoder submodule extracts the features of the input image and inputs them into the head. The number of heads is the same as the number of categories of load components, which are used to output images corresponding to each type of component. Each group of heads includes a feature fusion module and an image reconstruction module. The feature fusion module splices the features of the net load and the corresponding load component in the channel dimension, and reduces the dimension to obtain the fusion feature. The image reconstruction module includes multiple convolutional layers and deconvolution layers, and reconstructs the image corresponding to a single type of load component based on the fusion feature.

[0086] The discrimination module receives the load component identification image and the actual image as input, calculates the authenticity of the input image through multiple convolutional layers, and iteratively updates the parameters of the load component identification model in an adversarial game between the recognition module and the discrimination module.

[0087] During the model training process, the weighted sum of adversarial loss, L1 loss and perceptual loss is used to form a composite loss function, and the load component identification model is updated by back propagation. Among them, adversarial loss is used to measure the overall quality and authenticity of the load component image; L1 loss and perceptual loss are used to measure the similarity of the content and texture features of the load component image respectively. The calculation formula is as follows:

[0088]

[0089]

[0090] L S (S,D)=λ1L GAN (S,D)+λ2L L1 (S)+λ3L VGG (S);

[0091] Among them, X is the input load component image, S is the identification module, S(X) is the identification result, and Y is the real load component image. S is the total loss function of the model, λ1,λ2,λ3 are the weight coefficients of the three types of loss functions, which are initially set to equal weights and dynamically adjusted during the training process. When a certain loss function decreases too fast, its weight coefficient is reduced; L GAN To counter the loss, the model performance can be measured; L L1 L1 loss is the average absolute error calculated pixel by pixel to measure the content loss of the image; L VGG For perceptual loss, the pre-trained VGG-19 network is used to extract high-level features such as image texture information, and then measure the similarity of images in the feature space, which is more consistent with the image differences perceived by human vision. is the feature output of the jth layer of the i-th convolutional module of the VGG network, W i,j ,H i,j are the height and width of the feature, ω i is the weight corresponding to the i-th group of features, which is selected as the weight coefficient used by the pix2pixHD network.

[0092] The synthesized training data set is used as the real substation measurement data, and the load component identification model is used to estimate the power share of each component. The loss function between the identification result and the real data is calculated, and the model parameters are continuously iterated and updated.

[0093] (4) For the day to be identified, the actual measured data of the net load of the substation and various components are encoded into images and input into the trained identification model to obtain the image corresponding to the total power of the load components of the substation. Subsequently, the image is inversely encoded into a power curve to complete the task of identifying the load components of the substation; the identification result of the load components of the substation is obtained through the inverse encoding technology of the image; the expression is as follows:

[0094]

[0095] in, is the identification value of the load component power curve at point i, is the value of the i-th point on the diagonal of the GASF matrix corresponding to the load component image.

[0096] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the contents disclosed herein. The present application is intended to cover any variations, uses or adaptations of the present application, which follow the general principles of the present application and include common knowledge or customary technical means in the art that are not disclosed in the present application.

[0097] It will be appreciated that the present application is not limited to the exact construction that has been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof.

Claims

1. A method for identifying load components in a substation area based on image coding technology, characterized in that: The steps include: (1) Taking the day as the unit, the characteristic power curve of each load component is determined according to the cluster characteristics of the load components; the characteristic curves of the net load and each load component are encoded into a two-dimensional grayscale image using the Gram angle and field method; (2) Based on the known measured data of each component, the synthetic minority oversampling technique is used to generate similar load component power curves; the generated power data sets are combined as the power curves of the load components in the same substation, and then the net load curve and the power proportion of each load component are superimposed to obtain the training data set of the synthetic substation identification model; (3) The net load grayscale image and the load component grayscale image are input into the recognition model, and the image features are extracted through the encoder composed of residual blocks and convolutional layers; the features of the net load and various load components are respectively fused into multiple sets of composite features, and the images corresponding to various components are respectively identified using the separation module composed of convolutional layers and deconvolutional layers; and the authenticity of the identified image is judged by the discriminator using the framework of the generative adversarial network. In the adversarial game between the load component identification detector and the discriminator, the weighted sum of the adversarial loss, L1 loss, and perceptual loss is used as the loss function, and the model parameters are updated by back propagation according to the real images of the training data set; until a trained load component identification model is obtained; (4) For the day to be identified, the actual measured data of the net load of the substation and various components are encoded into images and input into the trained load component identification model to obtain the image corresponding to the total power of the load components in the substation; then, the image is inversely encoded into a power curve to complete the load component identification task of the substation.

2. The method for identifying load components in a substation according to claim 1, characterized in that: The step (1) is specifically as follows: as the number of users increases, the load component cluster shows strong regularity and is highly correlated with the sum of the load components in the substation area. Therefore, the sum of the power curves of the observable load components is selected as the characteristic power curve of each component. The expression is as follows: Among them, P fl is the characteristic curve of the load component; is the power curve of the i-th group of observable load components, N l The total number of curves included for the observable load components; Normalize the net load curve and the characteristic curves of the load components to the range [0,1]: in, is the data of the tth time point on the dth day after the power curve is normalized; P d,t is the data of the power curve at the t-th time point on the d-th day, and P is the data set consisting of all power curves; The normalized power curve is used as the angle cosine value, the timestamp is the radius, and converted to the polar coordinate system: in, is the power value of the normalized daily power curve at the i-th point, r i They are Polar angle and polar diameter in polar coordinate system, t i is the timestamp, N is the total number of sampling points of the daily power curve; Calculate the cosine value of the sum of the polar angles corresponding to each point on the daily power curve to obtain the Gram angle and field: Where I is a unit row vector; the diagonal elements of the GASF matrix retain the original power information of the load components, and the remaining elements reflect the time correlation between different sampling points; According to the pixel value range, the range of GASF matrix elements is converted from [0,1] to [0,255] to represent the pixel value of the load component image, and then the grayscale image of the net load and load component is drawn.

3. The method for identifying load components in a substation according to claim 1, characterized in that: The step (2) specifically comprises: calculating the similarity between each curve in the historical data set of the observable load component and the other curves in the feature space, and finding the K nearest neighbors of each component; Among them, the characteristics and changing laws of the load components are characterized by the daily characteristic indicators of the power curve, so the daily characteristic indicators are selected as the daily power average value, maximum value, minimum value, mean square error, skewness, and kurtosis to measure the similarity between the curves; Generate similar load component power curves using minority class oversampling techniques: in, represents the power data of the qth group of load components generated by the observable user i on the dth day, is the power data of the load component measured by observable user i on day d, is a similar power curve selected from the K nearest neighbors, rand(0,1) is a random number between [0,1], and Q is the number of generated power curves; Based on the maximum value of the load component, set the capacity factor to determine the power level of the generated load component: Among them, r l represents the capacity coefficient of the load component. Since the capacity difference between user load components is relatively small, the coefficient range is selected as [0.5, 1.5]; P nl The composite load component curve set is obtained by normalizing the generated curve to its maximum value and then multiplying it by the set capacity value; Combine the generated data sets of various components as the load components in the same substation; superimpose to obtain the total power of the load components in the substation and the net load data to synthesize the simulation data set for model training: in, They are the power values ​​of the comprehensive power of load components and net load at the t-th time point on the d-th day in the simulation training set; is the power value of the generated power curve of the i-th group at the t-th time point on the d-th day, N l is the number of generated curves; is the power value of the sum of the power of the j-th load component at the t-th time point on the d-th day, and M is the number of load component categories.

4. The method for identifying load components in a substation according to claim 1, characterized in that: Specifically, the load component identification model includes a load component identification module and a discrimination module. The load component identification module includes an encoder submodule and multiple heads. The encoder is used to extract image features of the input net load and load components. The head is used to perform an identification task of a certain type of load component to obtain an image recognition result of the component. The discrimination module is used to judge the authenticity of the image recognition result, thereby learning through game with the identification module and training model parameters; until a trained load component identification model is obtained; The encoder submodule includes three branches, each of which includes multiple convolutional layers and residual blocks, and captures multi-scale image features through convolution kernels of different sizes and network layers; the encoder submodule extracts the features of the input image and inputs them into the head, the number of heads is the same as the number of categories of the load components, and is used to output images corresponding to each type of component; each group of heads includes a feature fusion module and an image reconstruction module, the feature fusion module splices the features of the net load and the corresponding load component in the channel dimension, and reduces the dimension to obtain the fusion feature; the image reconstruction module includes multiple convolutional layers and deconvolution layers, and reconstructs the image corresponding to a single type of load component according to the fusion feature; The discrimination module receives a load component identification image and an actual image as input, calculates the authenticity of the input image through multiple convolutional layers, and iteratively updates the parameters of the load component identification model in a confrontation game between the recognition module and the discrimination module.

5. The method for identifying load components in a substation according to claim 1, characterized in that: In the step (3), the weighted sum of adversarial loss, L1 loss and perceptual loss is used as the loss function, and the model parameters are updated by back propagation according to the real image of the training data set. Specifically, during the model training process, the weighted sum of adversarial loss, L1 loss and perceptual loss is used to form a composite loss function, and the load component identification model is updated by back propagation; wherein the adversarial loss is used to measure the overall quality and authenticity of the load component image; the L1 loss and perceptual loss are used to measure the similarity of the content and texture features of the load component image respectively; the calculation formula is as follows: L S (S,D)=λ1L GAN (S,D)+λ2L L1 (S)+λ3L VGG (S); Among them, X is the input load component image, S is the identification module, S(X) is the identification result, and Y is the real load component image; L S is the total loss function of the model, λ1,λ2,λ3 are the weight coefficients of the three types of loss functions, which are initially set to equal weights and dynamically adjusted during the training process. When a certain loss function decreases too fast, its weight coefficient is reduced; L GAN To counter the loss, the model performance can be measured; L L1 L1 loss is the average absolute error calculated pixel by pixel to measure the content loss of the image; L VGG For perceptual loss, the high-level features of image texture information are extracted through the pre-trained VGG-19 network, and then the similarity of images in the feature space is measured; is the feature output of the jth layer of the i-th convolutional module of the VGG network, W i,j ,H i,j are the height and width of the feature, ω i is the weight corresponding to the i-th group of features, which is selected as the weight coefficient used by the pix2pixHD network.

6. The method for identifying load components in a substation according to claim 1, characterized in that: In the step (4), for the day to be identified, the characteristic power curves of the net load and the observable load components are encoded into images and input into the trained load component identification model to obtain the image corresponding to the sum of the power of the load components in the substation area; then, the identification result of the load components in the substation area is obtained by the inverse encoding technology of the image; the expression is as follows: in, is the identification value of the load component power curve at point i, is the value of the i-th point on the diagonal of the GASF matrix corresponding to the load component image.

Images