Distributed photovoltaic output stage identification method and system based on graph neural network

Abstract

The present invention provides a distributed photovoltaic output phase identification method and system based on a graph neural network. The method includes: constructing a feature matrix for each user based on the net load during the daytime period; obtaining a first Euclidean distance and a second Euclidean distance, constructing an objective function with the goal of minimizing the sum of the first and second Euclidean distances, and obtaining a minimum target value; performing feature scaling on the net load during the daytime period based on the minimum target value to obtain the scaled net load; inputting the scaled net load into an improved graph neural network model to obtain a unit matrix; performing a Fourier operation on the unit matrix to obtain a first feature; obtaining a photovoltaic load identification value based on the first feature; and obtaining an actual photovoltaic load value based on the photovoltaic load prediction value. Embodiments of the present invention can improve the accuracy of photovoltaic output phase identification.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic output identification, and in particular to a distributed photovoltaic output stage identification method and system based on graph neural network. Background Art

[0002] Traditional photovoltaic identification methods typically rely on accurate net load data and measured data from photovoltaic reference stations within the same area. This method utilizes photovoltaic base station output data to construct a photovoltaic system sample library based on clustering technology, selects optimal samples through game theory, and employs a semi-supervised source separation model to decompose photovoltaic power generation. However, this method is limited by its high reliance on measured data from photovoltaic reference stations. If the power generation pattern of the target household photovoltaic system does not match that of the photovoltaic reference station, the identification effect will be significantly reduced. Therefore, there is an urgent need to develop a photovoltaic output identification method that focuses more on the characteristics of the net load itself, thereby reducing reliance on external data. Summary of the Invention

[0003] The purpose of the present invention is to provide a distributed photovoltaic output stage identification method and system based on graph neural network, aiming to solve the problems of low accuracy and excessive dependence on external data in traditional photovoltaic identification methods.

[0004] In a first aspect, the present invention provides a distributed photovoltaic output phase identification method based on a graph neural network, the method comprising:

[0005] Obtaining historical load information of users, and obtaining the net load of each user during the daytime and nighttime periods based on the historical load information, and constructing a feature matrix for each user based on the net load during the daytime period;

[0006] The first Euclidean distance is obtained based on the net load of user i during the daytime period and the net load of user j during the nighttime period. The second Euclidean distance is obtained based on the actual load of user i during the daytime period and the actual load of user j during the nighttime period. The objective function is constructed with the goal of minimizing the sum of the first Euclidean distance and the second Euclidean distance to obtain the minimum target value.

[0007] Performing feature scaling on the net load within the daytime period according to the minimum target value to obtain the feature-scaled net load, and inputting the feature-scaled net load into the improved graph neural network model to obtain a unit matrix;

[0008] Performing a Fourier operation on the identity matrix to obtain a first feature, obtaining a photovoltaic load identification value based on the first feature, and obtaining an actual photovoltaic load value based on the photovoltaic load prediction value;

[0009] An identification evaluation index is obtained according to the photovoltaic load prediction value and the photovoltaic load actual value.

[0010] Furthermore, the step of obtaining the historical load conditions of the users and obtaining the net load of each user in the daytime and nighttime periods according to the historical load conditions includes:

[0011] definition Represents the daytime and nighttime periods respectively. 、 、 Respectively represent the start time of the user's photovoltaic daytime phase, the end time of the daytime phase, and the total amount of data measurement throughout the day, then:

[0012] ;

[0013] The net load of users during the daytime and nighttime periods is obtained using the following formula:

[0014] ;

[0015] in, 、 、 、 、 They represent the net load during the daytime period, the net load during the nighttime period, the actual load during the daytime period, the actual load during the nighttime period, and the photovoltaic output during the daytime period.

[0016] Furthermore, the step of constructing a characteristic matrix for each user according to the net load in the daytime period includes:

[0017] The feature matrix is constructed according to the following formula:

[0018] ;

[0019] in, represents the feature matrix of user i, D represents the total number of days covered by the daily net load sample, is the feature weight factor, represents the net load of user i during the daytime period.

[0020] Furthermore, the step of obtaining a first Euclidean distance based on the net load of user i during the daytime period and the net load of user j during the nighttime period, obtaining a second Euclidean distance based on the actual load of user i during the daytime period and the actual load of user j during the nighttime period, and constructing an objective function with the goal of minimizing the sum of the first Euclidean distance and the second Euclidean distance to obtain the minimum target value includes:

[0021] The minimum target value is obtained according to the following formula:

[0022] ;

[0023] in, represents the objective function, represents the minimum target value, 、 Respectively represent and Other loads, 、 They represent the actual daytime load and the actual nighttime load after removing the remaining load. represents the actual daytime load of user i after removing the remaining load, represents the actual nighttime load of user j after removing the remaining load, represents the net load of user j during the night time period.

[0024] Furthermore, the step of performing characteristic scaling on the net load in the daytime period according to the minimum target value to obtain the net load after characteristic scaling includes:

[0025] Feature scaling is performed according to the following formula:

[0026] ;

[0027] in, represents the net load after feature scaling, Indicates the total number of night time points, represents the average value of nighttime net load, Indicates the minimum value of net load during the daytime period. Indicates the maximum value of net load during the daytime period;

[0028] The step of inputting the feature-scaled net load into the improved graph neural network model to obtain the unit matrix includes:

[0029] The identity matrix is obtained according to the following formula:

[0030] ;

[0031] in, represents the identity matrix, T represents the transpose, represents the scaled feature matrix.

[0032] Furthermore, the steps of performing a Fourier operation on the unit matrix to obtain a first feature, obtaining a photovoltaic load identification value according to the first feature, and obtaining an actual photovoltaic load value according to the photovoltaic load prediction value include:

[0033] The Fourier operation is performed according to the following formula:

[0034] ;

[0035] in, represents the Fourier operation, represents the graph convolution operation, represents the Hadamard product operation, represents graph matrix operations, Indicates the first feature;

[0036] The photovoltaic load identification value is obtained according to the following formula:

[0037] ;

[0038] in, Indicates the photovoltaic load identification value, Represents the graph neural network model The feature matrix of the layer;

[0039] The actual value of the photovoltaic load is obtained according to the following formula:

[0040] ;

[0041] in, Indicates the actual value of photovoltaic load, Indicates the adjustable amount of photovoltaic power generation by the user, Indicates the household photovoltaic capacity installed by the user;

[0042] According to the following formula, we can get the first The feature matrix of the layer:

[0043] ;

[0044] in, Represents the graph neural network model The feature matrix of the layer, represents the weight matrix, is the activation function of the graph neural network model, It is expressed as a function to prevent overfitting.

[0045] Furthermore, the step of obtaining an identification evaluation index according to the photovoltaic load prediction value and the photovoltaic load actual value includes:

[0046] The identification evaluation index is obtained according to the following formula:

[0047] ;

[0048] in, represents the mean absolute difference, represents the mean absolute percentage difference, Indicates the accuracy of the recognition result, Represents the total number of points in the time series.

[0049] In a second aspect, the present invention provides a distributed photovoltaic output phase identification system based on a graph neural network, the system comprising:

[0050] A feature matrix construction module is used to obtain the historical load conditions of users, and obtain the net load of each user in the daytime and nighttime time periods based on the historical load conditions, and construct a feature matrix for each user based on the net load in the daytime time period;

[0051] An objective function construction module is used to obtain a first Euclidean distance based on the net load of user i during the daytime period and the net load of user j during the nighttime period, obtain a second Euclidean distance based on the actual load of user i during the daytime period and the actual load of user j during the nighttime period, and construct an objective function with the goal of minimizing the sum of the first Euclidean distance and the second Euclidean distance to obtain a minimum target value;

[0052] a feature scaling module, configured to perform feature scaling on the net load within the daytime period according to the minimum target value to obtain the feature-scaled net load, and input the feature-scaled net load into the improved graph neural network model to obtain a unit matrix;

[0053] an identification execution module, configured to perform a Fourier operation on the unit matrix to obtain a first feature, obtain a photovoltaic load identification value based on the first feature, and obtain an actual photovoltaic load value based on the photovoltaic load prediction value;

[0054] An evaluation module is used to obtain an identification evaluation index based on the photovoltaic load prediction value and the photovoltaic load actual value.

[0055] In a third aspect, the present invention provides a storage medium storing one or more programs, which, when executed by a processor, implement the above-mentioned distributed photovoltaic output stage identification method based on graph neural network.

[0056] In a fourth aspect, the present invention provides an electronic device, comprising a memory and a processor, wherein:

[0057] The memory is used to store computer programs;

[0058] When the processor is used to execute the computer program stored in the memory, it implements the above-mentioned distributed photovoltaic output stage identification method based on graph neural network.

[0059] Compared with the prior art, the present invention has the following advantages:

[0060] 1. Traditional distributed photovoltaic output identification methods can be affected by a variety of factors, such as missing and inaccurate data and complex photovoltaic output variations, which can limit the accuracy of identification results. The method proposed in this paper leverages the characteristics of graph-structured data to capture the complex relationships and interactions between photovoltaic power plants, thereby more accurately predicting and identifying photovoltaic output. By introducing a graph neural network model, it can fully consider factors such as the geographical location, environmental conditions, and power generation history of distributed photovoltaic power plants, improving the accuracy and reliability of identification.

[0061] 2. Existing semi-supervised identification methods often rely on measured data from PV reference stations. If the target PV power generation pattern differs from that of the reference station, the identification effect will be significantly reduced. The method proposed in this paper focuses on using the historical power generation data and real-time environmental parameters of the distributed PV power station itself for identification, thereby reducing reliance on external data. This method can more flexibly adapt to the needs of PV output identification in different regions and climate conditions.

[0062] 3. By accurately identifying distributed photovoltaic output, the present invention can provide a theoretical basis for the access location, access method and operation mode of photovoltaic power sources, and provide a reliable basis for optimizing the scheduling of power systems and the rational configuration of energy storage devices. This helps to improve the overall operating efficiency and stability of photovoltaic energy systems, reduce the rotating reserve capacity of power systems, make full use of solar energy resources, and obtain greater economic and social benefits. This method is not only suitable for the output identification of distributed photovoltaic power stations, but can also be expanded to other renewable energy fields such as wind power generation, hydropower generation, etc., providing more accurate and reliable technical support for the grid connection and scheduling of renewable energy. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 This is a flow chart of a distributed photovoltaic output phase identification method based on a graph neural network proposed in one embodiment of the present invention;

[0064] Figure 2 This is a structural diagram of a distributed photovoltaic output phase identification system based on graph neural network according to an embodiment of the present invention.

[0065] The following specific embodiments will further illustrate the present invention in conjunction with the above-mentioned drawings. DETAILED DESCRIPTION

[0066] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. Unless otherwise defined, the technical terms or scientific terms used herein should be the common meanings understood by people with ordinary skills in the field to which the invention belongs. The words "including" and similar words used in this article mean that the elements or objects appearing before the word cover the elements or objects listed after the word and their equivalents, without excluding other elements or objects.

[0067] like Figure 1 As shown, an embodiment of the present invention proposes a distributed photovoltaic output phase identification method based on a graph neural network, the method comprising steps S101 to S105, wherein:

[0068] Step S101: Obtain historical load information of users, and obtain the net load of each user in the daytime and nighttime periods based on the historical load information, and construct a feature matrix for each user based on the net load in the daytime period;

[0069] It should be pointed out that in this step, the user's smart meter data is first collected, and the user's daily photovoltaic power consumption pattern is classified through the K-means clustering algorithm to determine the specific time point or period of photovoltaic installation and obtain the user's historical load situation.

[0070] Then, the user's photovoltaic load is used for analysis to determine the relationship between the photovoltaic net load and the actual load, so as to eliminate the actual load from the net load.

[0071] Specifically, define Represents the daytime and nighttime periods respectively. 、 、 Respectively represent the start time of the user's photovoltaic daytime phase, the end time of the daytime phase, and the total amount of data measurement throughout the day, then:

[0072] ;

[0073] The net load of users during the daytime and nighttime periods is obtained using the following formula:

[0074] ;

[0075] in, 、 、 、 、 They represent the net load during the daytime period, the net load during the nighttime period, the actual load during the daytime period, the actual load during the nighttime period, and the photovoltaic output during the daytime period.

[0076] Furthermore, in some embodiments, the feature matrix is constructed according to the following formula:

[0077] ;

[0078] in, represents the feature matrix of user i, D represents the total number of days covered by the daily net load sample, is a feature weight factor, usually between 0 and 1. This feature matrix helps provide in-depth information about user electricity usage behavior in further analysis and promotes more accurate identification of photovoltaic output.

[0079] Step S102: A first Euclidean distance is obtained based on the net load of user i during the daytime period and the net load of user j during the nighttime period. A second Euclidean distance is obtained based on the actual load of user i during the daytime period and the actual load of user j during the nighttime period. An objective function is constructed with the goal of minimizing the sum of the first and second Euclidean distances to obtain a minimum target value.

[0080] It's important to note that, based on the assumption that nighttime PV output is zero and combined with the results of PV output identification, nighttime data containing only actual load and identified daytime actual load data are used to identify net load samples of users with similar electricity usage behavior to the target user but different PV capacities. By comparing these two samples and taking the difference, the actual load component in the net load data can be minimized, highlighting the differences between users due to differences in PV power generation.

[0081] In addition, in order to find user net load samples with similar electricity consumption behavior characteristics on the same date, and to reduce the impact of the initial identification error of the actual load on the matching accuracy, the sum of the Euclidean distances between the target user i and the candidate matching user j at each time point is first calculated based on the obtained data, and then the sum of the Euclidean distances between the two on the actual daytime load data is calculated, and the latter is assigned a weight to obtain the minimum target value.

[0082] Specifically, the minimum target value is obtained according to the following formula:

[0083] ;

[0084] in, represents the objective function, represents the minimum target value, 、 Respectively represent and Other loads, 、 They represent the actual daytime load and the actual nighttime load after removing the remaining load. represents the actual daytime load of user i after removing the remaining load, represents the actual nighttime load of user j after removing the remaining load, represents the net load of user i during the daytime period, represents the net load of user j during the night time period.

[0085] Step S103: performing feature scaling on the net load in the daytime period according to the minimum target value to obtain the feature-scaled net load, and inputting the feature-scaled net load into the improved graph neural network model to obtain a unit matrix;

[0086] In this step, the actual power load of each user is adjusted to the same order of magnitude, given that the actual power demand of each user is different. Scaling the actual load is equivalent to performing the same operation on the net load. During the nighttime period, the net load is equal to the actual load, and its characteristic scaling can be achieved by dividing the net load by the average nighttime net load:

[0087] ;

[0088] in, represents the net load after feature scaling, Indicates the total number of night time points, represents the average value of nighttime net load, Indicates the minimum value of net load during the daytime period. Indicates the maximum value of net load during the daytime period;

[0089] In addition, it should be noted that the graph convolution in the improved graph neural network model directly extracts features from the graph structure data, uses the convolution operation to fuse the topological connection and attribute information of the nodes, and standardizes the structure of the graph by calculating the normalized Laplace matrix in the frequency domain. This ensures that the neighbor information of each node is evenly weighted during the convolution process, ensuring that the graph convolution operation can work smoothly in the frequency domain. At this time, the net load after feature scaling is input into the improved graph neural network model. The formula is as follows:

[0090] ;

[0091] in, represents the identity matrix, T represents the transpose, represents the scaled feature matrix.

[0092] Step S104: performing a Fourier operation on the identity matrix to obtain a first feature, obtaining a photovoltaic load identification value based on the first feature, and obtaining an actual photovoltaic load value based on the photovoltaic load prediction value;

[0093] It should be noted that the graph Fourier transform and its inverse transform theory can be converted into easy-to-use mathematical formulas, thereby improving the execution efficiency of tasks such as signal analysis, feature extraction, or network status prediction. The Fourier operation is performed according to the following formula:

[0094] ;

[0095] in, represents the Fourier operation, represents the graph convolution operation, represents the Hadamard product operation, represents graph matrix operations, Indicates the first feature;

[0096] In addition, in order to further optimize the output of the graph neural network model to achieve the purpose of optimizing photovoltaic power generation and improving recognition accuracy, it is assumed that the household photovoltaic capacity installed by the user is The discretization step size is set to 0.001 based on the meter's accuracy. A comprehensive search is performed on the preselected subset. If the correlation with the grid increment and the user's PV power generation reaches the highest value, the resulting PV power generation is highly correlated with the initial estimate. Once this correlation is confirmed, the optimal coefficient after the comprehensive search is determined to obtain the user's PV power generation.

[0097] During the initial identification of distributed PV output using a graph neural network, only nighttime net load data is used to construct a contextual similarity loss. To achieve more accurate identification results, more contextual information is needed. Given that the output difference that reflects the target user's PV characteristics has been determined using the difference method, the PV output ratio, which is approximately equal to the PV capacity ratio, can be used to predict the target PV output during the daytime. This allows the calculation of the similarity loss of the actual daytime load and combines it with the contextual loss for the nighttime period, achieving refined identification of the actual load.

[0098] Specifically, the photovoltaic load identification value is obtained according to the following formula:

[0099] ;

[0100] in, Indicates the photovoltaic load identification value, Represents the graph neural network model The feature matrix of the layer;

[0101] The actual value of the photovoltaic load is obtained according to the following formula:

[0102] ;

[0103] in, Indicates the actual value of photovoltaic load, Indicates the adjustable amount of photovoltaic power generation by the user, Indicates the household photovoltaic capacity installed by the user;

[0104] In addition, this step also uses sparse and efficient Chebyshev polynomials to simplify the graph convolution process and speed up the information transfer between graph convolution layers. Specifically, the following formula is used to obtain the first The feature matrix of the layer:

[0105] ;

[0106] in, Represents the graph neural network model The feature matrix of the layer, represents the weight matrix, is the activation function of the graph neural network model, It is expressed as a function to prevent overfitting.

[0107] Step S105: obtaining an identification evaluation index according to the photovoltaic load prediction value and the photovoltaic load actual value.

[0108] In this step, MAE and MAPE indicators are used to evaluate the difference between the identification results and the measured data. The specific formula is as follows:

[0109] ;

[0110] in, represents the mean absolute difference, represents the mean absolute percentage difference, Indicates the accuracy of the recognition result, Represents the total number of points in the time series.

[0111] It should be noted that and Used to evaluate the difference between PV output identification results and measured values. For the accuracy of identification, and The smaller the value of The larger the value of , the closer the identification result of the photovoltaic output is to the measured value, that is, the higher the accuracy of the photovoltaic output identification.

[0112] In addition, in some embodiments, in order to verify the model effect, the evaluation index of the graph neural network model is obtained according to the following formula:

[0113] ;

[0114] in, Represents the evaluation index of the graph neural network model. The larger the evaluation index, the better the performance of the model.

[0115] In summary, the distributed PV output phase identification method based on a graph neural network, proposed in this embodiment of the present invention, innovatively transforms the distributed PV output identification problem into the reconstruction problem of missing actual load data, indirectly enabling PV output identification. In the rough identification phase, the proposed method achieves comparable results to traditional unsupervised PV decomposition algorithms based on multi-constrained optimization, fully validating the effectiveness and feasibility of the proposed modeling approach. Building on the preliminary identification, the proposed method further explores PV features implicit in the net load data and uses these features to refine the preliminary identification results, achieving refined PV output identification. The proposed method outperforms two unsupervised decomposition identification methods in terms of identification accuracy. It does not rely on external data or historical PV power generation data for the target household, nor does it require explicit modeling. Even in the absence of a PV reference station in the same area, the proposed method can still accurately identify the target PV power by constructing an identification model, demonstrating its strong practicality and applicability. These advantages make this method promising for the application of distributed PV output identification.

[0116] like Figure 2 As shown, one embodiment of the present invention proposes a distributed photovoltaic output phase identification system based on a graph neural network, the system comprising:

[0117] The characteristic matrix construction module 10 is used to obtain the historical load conditions of users, and obtain the net load of each user in the daytime and nighttime time periods based on the historical load conditions, and construct the characteristic matrix of each user based on the net load in the daytime time period;

[0118] An objective function construction module 20 is configured to obtain a first Euclidean distance based on the net load of user i during the daytime period and the net load of user j during the nighttime period, obtain a second Euclidean distance based on the actual load of user i during the daytime period and the actual load of user j during the nighttime period, construct an objective function with the goal of minimizing the sum of the first Euclidean distance and the second Euclidean distance, and obtain a minimum target value;

[0119] a feature scaling module 30 for performing feature scaling on the net load during the daytime period according to the minimum target value to obtain the feature-scaled net load, and inputting the feature-scaled net load into the improved graph neural network model to obtain a unit matrix;

[0120] an identification execution module 40, configured to perform a Fourier operation on the identity matrix to obtain a first feature, obtain a photovoltaic load identification value based on the first feature, and obtain an actual photovoltaic load value based on the photovoltaic load prediction value;

[0121] The evaluation module 50 is configured to obtain an identification evaluation index based on the photovoltaic load prediction value and the photovoltaic load actual value.

[0122] On the other hand, the present invention further proposes a storage medium storing one or more programs, which, when executed by a processor, implement the above-mentioned distributed photovoltaic output stage identification method based on graph neural network.

[0123] On the other hand, the present invention also proposes an electronic device, including a memory and a processor, wherein the memory is used to store computer programs, and the processor is used to execute the computer programs stored in the memory to implement the above-mentioned distributed photovoltaic output stage identification method based on graph neural network.

[0124] Those skilled in the art will appreciate that the logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device), or in conjunction with such instruction execution system, apparatus, or device. For purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transmit a program for use by an instruction execution system, apparatus, or device, or in conjunction with such instruction execution system, apparatus, or device.

[0125] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting, or processing it in another suitable manner as necessary, and then storing it in a computer memory.

[0126] It should be understood that various components of the present invention may be implemented using hardware, software, firmware, or a combination thereof. In the aforementioned embodiments, multiple steps or methods may be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one or a combination of the following technologies known in the art may be used: a discrete logic circuit having logic gate circuits for implementing logic functions on data signals, an application-specific integrated circuit having suitable combinational logic gate circuits, a programmable gate array (PGA), a field-programmable gate array (FPGA), etc.

[0127] While the embodiments of the present invention have been described in detail above, it will be apparent to those skilled in the art that various modifications and variations of these embodiments are possible. However, it should be understood that such modifications and variations are within the scope and spirit of the present invention as set forth in the claims. Furthermore, the invention described herein is susceptible to other embodiments and may be practiced or implemented in a variety of ways.

Claims

1. A distributed photovoltaic output phase identification method based on graph neural network, characterized in that: The method comprises: The K-means clustering algorithm is used to classify the user's daily photovoltaic power consumption pattern to determine the specific time point or time period of photovoltaic installation, and to obtain the user's historical load situation. Based on the historical load situation, the net load of each user during the daytime and nighttime time periods is obtained, and the feature matrix of each user is constructed based on the net load during the daytime period; The feature matrix is constructed according to the following formula: ； in, represents the feature matrix of user i, D represents the total number of days covered by the daily net load sample, is the feature weight factor, ranging from 0 to 1; The first Euclidean distance is obtained based on the net load of user i during the daytime period and the net load of user j during the nighttime period. The second Euclidean distance is obtained based on the actual load of user i during the daytime period and the actual load of user j during the nighttime period. The objective function is constructed with the goal of minimizing the sum of the first Euclidean distance and the second Euclidean distance to obtain the minimum target value. The minimum target value is obtained according to the following formula: ； in, represents the objective function, represents the minimum target value, 、 Respectively represent and Other loads, 、 They represent the actual daytime load and the actual nighttime load after removing the remaining load. represents the actual daytime load of user i after removing the remaining load, represents the actual nighttime load of user j after removing the remaining load, represents the net load of user i during the daytime period, represents the net load of user j during the night time period; Performing feature scaling on the net load within the daytime period according to the minimum target value to obtain the feature-scaled net load, and inputting the feature-scaled net load into the improved graph neural network model to obtain a unit matrix; The characteristic scaling is achieved by dividing the net load by the average nightly net load: ； in, represents the net load after feature scaling, Indicates the total number of night time points, represents the average value of nighttime net load, Indicates the minimum value of net load during the daytime period. Indicates the maximum value of net load during the daytime period; The identity matrix is obtained according to the following formula: ； in, represents the identity matrix, T represents the transpose, represents the scaled feature matrix; Performing a Fourier operation on the identity matrix to obtain a first feature, obtaining a photovoltaic load identification value based on the first feature, and obtaining an actual photovoltaic load value based on the photovoltaic load identification value; An identification evaluation index is obtained according to the photovoltaic load identification value and the photovoltaic load actual value.

2. The distributed photovoltaic output phase identification method based on graph neural network according to claim 1 is characterized in that: The steps of obtaining the historical load conditions of the users and obtaining the net load of each user in the daytime and nighttime time periods according to the historical load conditions include: definition Represents the daytime and nighttime periods respectively. 、 、 Respectively represent the start time of the user's photovoltaic daytime phase, the end time of the daytime phase, and the total amount of data measurement throughout the day, then: ； The net load of users during the daytime and nighttime periods is obtained using the following formula: ； in, 、 、 、 、 They represent the net load during the daytime period, the net load during the nighttime period, the actual load during the daytime period, the actual load during the nighttime period, and the photovoltaic output during the daytime period.

3. The distributed photovoltaic output stage identification method based on graph neural network according to claim 2 is characterized in that: The steps of performing a Fourier operation on the unit matrix to obtain a first feature, obtaining a photovoltaic load identification value according to the first feature, and obtaining an actual photovoltaic load value according to the photovoltaic load identification value include: The Fourier operation is performed according to the following formula: ； in, represents the Fourier operation, represents the graph convolution operation, represents the Hadamard product operation, represents graph matrix operations, Indicates the first feature; The photovoltaic load identification value is obtained according to the following formula: ； in, Indicates the photovoltaic load identification value, Represents the graph neural network model The feature matrix of the layer; The actual value of the photovoltaic load is obtained according to the following formula: ； in, Indicates the actual value of photovoltaic load, Indicates the adjustable amount of photovoltaic power generation by the user, Indicates the household photovoltaic capacity installed by the user; According to the following formula, we can get the first The feature matrix of the layer: ； in, Represents the graph neural network model The feature matrix of the layer, represents the weight matrix, is the activation function of the graph neural network model, It is expressed as a function to prevent overfitting.

4. The distributed photovoltaic output stage identification method based on graph neural network according to claim 3 is characterized in that: The step of obtaining the identification evaluation index according to the photovoltaic load identification value and the photovoltaic load actual value includes: The identification evaluation index is obtained according to the following formula: ； in, represents the mean absolute difference, represents the mean absolute percentage difference, Indicates the accuracy of the recognition result, Represents the total number of points in the time series.

5. A distributed photovoltaic output stage identification system based on graph neural network, characterized in that: The system comprises: A feature matrix construction module is used to classify users' daily photovoltaic power consumption patterns using a K-means clustering algorithm to determine the specific time point or time period of photovoltaic installation, obtain the user's historical load situation, and obtain the net load of each user during the daytime and nighttime time periods based on the historical load situation, and construct a feature matrix for each user based on the net load during the daytime period; The feature matrix is constructed according to the following formula: ； in, represents the feature matrix of user i, D represents the total number of days covered by the daily net load sample, is the feature weight factor, ranging from 0 to 1; An objective function construction module is used to obtain a first Euclidean distance based on the net load of user i during the daytime period and the net load of user j during the nighttime period, obtain a second Euclidean distance based on the actual load of user i during the daytime period and the actual load of user j during the nighttime period, and construct an objective function with the goal of minimizing the sum of the first Euclidean distance and the second Euclidean distance to obtain a minimum target value; The minimum target value is obtained according to the following formula: ； in, represents the objective function, represents the minimum target value, 、 Respectively represent and Other loads, 、 They represent the actual daytime load and the actual nighttime load after removing the remaining load. represents the actual daytime load of user i after removing the remaining load, represents the actual nighttime load of user j after removing the remaining load, represents the net load of user i during the daytime period, represents the net load of user j during the night time period; a feature scaling module, configured to perform feature scaling on the net load within the daytime period according to the minimum target value to obtain the feature-scaled net load, and input the feature-scaled net load into the improved graph neural network model to obtain a unit matrix; The minimum target value is obtained according to the following formula: ； in, represents the objective function, represents the minimum target value, 、 Respectively represent and Other loads, 、 They represent the actual daytime load and the actual nighttime load after removing the remaining load. represents the actual daytime load of user i after removing the remaining load, represents the actual nighttime load of user j after removing the remaining load, represents the net load of user i during the daytime period, represents the net load of user j during the night time period; an identification execution module, configured to perform a Fourier operation on the unit matrix to obtain a first feature, obtain a photovoltaic load identification value based on the first feature, and obtain an actual photovoltaic load value based on the photovoltaic load identification value; An evaluation module is used to obtain an identification evaluation index based on the photovoltaic load identification value and the photovoltaic load actual value.

6. A storage medium, characterized in that The storage medium stores one or more programs, which, when executed by the processor, implement the distributed photovoltaic output stage identification method based on graph neural network as described in any one of claims 1 to 4.

7. An electronic device, characterized in that: The electronic device comprises a memory and a processor, wherein: The memory is used to store computer programs; When the processor is used to execute the computer program stored in the memory, it implements the distributed photovoltaic output stage identification method based on graph neural network as described in any one of claims 1 to 4.

Images