Active power distribution network state estimation method and system based on GCN and Transform fusion

Abstract

The invention discloses an active power distribution network state estimation method and system based on GCN and Transform fusion. The method comprises the following steps: firstly, establishing a power distribution network graph model, and abstracting nodes in a power distribution network and lines connecting the nodes into a graph structure; secondly, constructing a photovoltaic power station physical model, and describing a relation between terminal voltage and current by adopting a five-parameter model; then constructing a state prediction equation based on Transform, and predicting a system state at a future moment by using historical measurement data; a GCN-based measurement equation is constructed, and relation modeling between the node state and the neighbor node state is realized through graph convolution operation; and finally, performing feature fusion on state prediction and measurement results by using a deep neural network, and outputting more accurate state estimation of the power distribution network. According to the method, the GCN is used for conversion of the measurement data, and spatial dependence between nodes and neighbors is fully mined; and the Transform is responsible for state prediction of long sequence time dependence, so that the prediction precision and robustness of the model are effectively improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of distribution network, and specifically relates to a method and system for estimating the state of an active distribution network based on the fusion of GCN and Transformer. Background Art

[0002] The large-scale access of distributed energy, the widespread application of flexible loads, and the introduction of emerging technologies such as edge computing have enabled distribution networks to have a higher level of intelligence, but have also brought new challenges. Compared with traditional distribution networks, new active distribution networks have significant characteristics such as high multi-energy penetration, complex dynamic loads, and highly controllable equipment. These characteristics require the power grid to have stronger adaptability and real-time response capabilities.

[0003] In this context, the operational constraints faced by distribution networks are more complex. Problems such as equipment capacity limitations, overload, aging, reverse flow, voltage / frequency instability, and dual uncertainty of source and load all pose a threat to the safe operation of distribution networks. At the same time, factors such as sparse data collection points, equipment failures, data noise, and communication delays further weaken the accuracy of distribution network status monitoring and limit its accurate perception of future operating status. With the increase in the number of controllable devices and the development of smart grid models, distribution system state estimation (DSSE) plays an increasingly critical role in the operation, protection, and control of distribution networks.

[0004] DSSE uses power system measurements such as line flow, node voltage (amplitude and phase angle) and node injection (obtained from the supervisory control and data acquisition SCADA system) to estimate the power system status such as voltage amplitude and phase angle of the system nodes. DSSE has attracted much attention as a core component of the distribution network energy management system. It obtains real-time data through remote metering and acquisition terminals, and performs necessary monitoring, simulation and verification of the real-time operating status of the distribution network.

[0005] In the field of DSSE, scholars have actively explored and made some important progress. In actual power grid operation, the presence of non-Gaussian noise and gross errors may lead to too many iterations or even failure to converge, and the traditional weighted least square (WLS) estimation has low accuracy. In order to solve this problem, some scholars have proposed some non-quadratic estimation criteria, such as weighted minimum absolute value state estimation, Huber-M estimation, and exponential objective function estimation. However, traditional state estimation methods are difficult to fully cope with the dynamic changes of complex loads and data quality issues, and the computational complexity is high and the solution time is long, which can no longer meet the needs of new active distribution networks.

[0006] Many scholars have introduced artificial intelligence algorithms to improve the efficiency and accuracy of power system state estimation. Compared with traditional methods, deep neural networks (DNNs) have shown higher time efficiency and robustness in power system state estimation. For the state estimation of transmission networks, an online estimation method based on deep neural networks can use the high-frequency input of phasor measurement unit data to estimate the observable state of the system.

[0007] The prior art proposes a Bayesian network method for unobservable states of distribution systems, wherein the Bayesian network is a neural network trained offline for predicting the probability of unobservable states. The prior art is also based on a method of generating virtual measurements using offline trained neural networks, estimating unobservable states through mixed data of real measurements and virtual measurements. However, the neural networks in these methods are all trained offline, and their accuracy cannot be guaranteed without feedback. The prior art also proposes a hybrid learning mechanism that incorporates the system model into the back propagation process of the neural network, but unobservable states are not considered for the time being. The above data-driven method still does not consider the new active distribution network scenario, does not model distributed new energy units, and it is difficult to consider the spatiotemporal characteristics of the distribution network topology and the impact of bidirectional power flow on traditional state estimation methods. Therefore, it is urgent to study a more robust, real-time and accurate distribution network state estimation method to effectively process the measurement data while accurately identifying the physical characteristics and dynamic behavior of the source and load objects. Summary of the invention

[0008] In view of the deficiencies in the prior art, the present invention provides an active distribution network state estimation method and system based on the fusion of deep convolutional neural networks (Graph Convolutional Neural Networks, GCN) and Transformer.

[0009] In a first aspect, the present invention provides an active distribution network state estimation method based on the fusion of GCN and Transformer, comprising: Step 1: Establish a distribution network graph model, in which each node in the distribution network and the lines connecting these nodes are abstracted into a graph structure; Step 2: construct a physical model of a photovoltaic power station, wherein the physical model of the photovoltaic power station includes a photovoltaic array, a DC / DC converter and a DC / AC converter, and uses a five-parameter model to describe the relationship between terminal voltage and current; Step 3: Construct a state prediction equation based on Transformer and use historical measurement data to predict the system state at future times; Step 4: Construct a measurement equation based on GCN, input the topological structure and measurement data of the distribution network into GCN, and model the relationship between the node state and the state of its neighboring nodes through graph convolution operation; Step 5: Use a deep neural network to fuse the state prediction in step 3 and the measurement results in step 4, integrate the spatial features extracted by GCN and the temporal features captured by Transformer, and output a more accurate distribution network state estimate.

[0010] In a second aspect, the present invention provides an active distribution network state estimation system based on the fusion of GCN and Transformer, comprising: A graph model building unit, used to build a distribution network graph model, in which each node in the distribution network and the lines connecting these nodes are abstracted into a graph structure; A photovoltaic power station physical model building unit is used to build a photovoltaic power station physical model, wherein the photovoltaic power station physical model includes a photovoltaic array, a DC / DC converter and a DC / AC converter, and uses a five-parameter model to describe the relationship between terminal voltage and current; Transformer model unit, used to construct a Transformer-based state prediction equation and use historical measurement data to predict the system state at future times; The GCN model unit is used to construct the measurement equation based on GCN, input the topological structure and measurement data of the distribution network into GCN, and realize the relationship modeling between the node state and the state of its neighboring nodes through graph convolution operation; The feature fusion and state estimation unit is used to use deep neural networks to fuse the state prediction and measurement results, integrate the spatial features extracted by GCN and the temporal features captured by Transformer, and output a more accurate distribution network state estimate.

[0011] In a third aspect, the present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, an active distribution network state estimation method based on the fusion of GCN and Transformer is implemented.

[0012] In a fourth aspect, the present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements an active distribution network state estimation method based on the fusion of GCN and Transformer.

[0013] Beneficial effects of the present invention: While effectively processing complex measurement data and state prediction of the distribution network, the present invention takes into account the spatiotemporal connection between the states of each node in the distribution network. In addition, the photovoltaic power station, a distributed energy source, is physically modeled, and its dynamic physical characteristics are incorporated into the scope of state estimation. Compared with the traditional weighted WLS state estimation method, the present invention can improve prediction accuracy, noise resistance and dynamic adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 Schematic diagram of electrical quantities estimated for the active distribution network state; Figure 2 It is a schematic diagram of the two-stage structure of a photovoltaic power station; Figure 3 This is a GCN-Transformer algorithm framework diagram of an embodiment of the present application; Figure 4 This is the Transformer model unit architecture diagram; Figure 5 This is the GCN model unit architecture diagram; Figure 6 This is the architecture diagram of the feature fusion and state estimation unit. DETAILED DESCRIPTION

[0015] The specific implementation of the embodiment of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the embodiment of the present invention, and is not used to limit the embodiment of the present invention.

[0016] The embodiment of the present application provides an active distribution network state estimation method based on the fusion of GCN and Transformer, including: For the new active distribution network scenario, this application first gives a specific modeling of active distribution network state estimation considering the photovoltaic power station model: A distribution network graph model is established, in which each node in the distribution network and the lines connecting these nodes are abstracted into a graph structure to provide a basic data structure for subsequent steps, as follows: For a power distribution network consisting of N nodes, it is modeled as a graph ,in Represents all nodes, Indicates all lines.

[0017] like Figure 1 As shown, for each node ,make represents its corresponding complex voltage, is the voltage amplitude, is the voltage phase angle, , are the active and reactive power injected into the node respectively. ,make , Indicates the active and reactive power at the sending end, , Indicates the active and reactive power of the receiving end.

[0018] Distribution System State Estimation (DSSE) aims to obtain the state of power distribution systems from the measured values ​​which are usually noisy. , calculate a more accurate system state , the process involves the dynamic state equation and measurement equation of the system, the formula is as follows: (1) (2) in, For the system The state variables at the moment include voltage amplitude and voltage phase angle; The active and reactive power injected into the node, the active and reactive power of the branch, and the node voltage amplitude are Real-time measurement value at the moment; for The state transition function at time ; is the process error; is the state-measurement contact function; The key to state estimation is to measure the error. processing.

[0019] On the other hand, establishing an accurate photovoltaic power generation system model can improve the state estimation accuracy and computational efficiency of the active distribution network.

[0020] A physical model of a photovoltaic power station is constructed, wherein the model includes a photovoltaic array, a DC / DC converter and a DC / AC converter, and a five-parameter model is used to describe the relationship between the terminal voltage and the current, providing a theoretical basis for the explicit fitting function.

[0021] Photovoltaic power stations are usually divided into two structures: one is a two-stage structure with a DC / DC converter and a DC / AC converter; the other is a single-stage structure with only a DC / AC converter. Figure 2 A two-stage structure is shown, including a photovoltaic array, DC / DC Converters and DC / AC Converter.

[0022] Specifically, given the light intensity and temperature , the output current is defined as , the output voltage is U PVS , the value is recorded as For duty cycle ,definition DC / DC The output voltage and current of the converter are and . yes DC / DC and DC / AC The wiring resistance between the converters. DC / AC converter, the modulation ratio is , the input voltage and current are and , the output voltage amplitude and phase angle are and , P ac and Q ac are the active power and reactive power on the AC side respectively. The equivalent impedance is expressed as , the transformer ratio is assumed to be 1: . From the photovoltaic system to the common coupling point ( Point of Common Coupling,PCC ) are the active power and reactive power respectively. and . PCC The voltage amplitude and phase angle are and , PCC The active and reactive power injection at and For the state estimation of a single PV power station, the state variable .

[0023] Furthermore, the five-parameter model is one of the most commonly used models for photovoltaic arrays, and its mathematical expression is a complex implicit transcendental equation that describes the relationship between terminal voltage and current. The embodiment of the present application uses an explicit fitting function to approximate the accuracy of the voltage-current relationship to the five-parameter model, reducing the computational time for obtaining the measurement equation and the Jacobian matrix, and its expression is as follows: (3) This model uses a logarithmic term to simulate the voltage-current relationship of the diode, and the remaining terms are used to simulate the effects of internal resistance and the effects of light and temperature changes.

[0024] Furthermore, based on the given five-parameter model of the photovoltaic array, a random k The light intensity, temperature, output voltage and output current of the group are recorded as , , and .vector , , and The corresponding entries in satisfy the relationship given by the five-parameter model of the photovoltaic array. The fitting parameter vector is denoted as .

[0025] After setting its initial value, , and Substituting into equation (3), we get the estimated voltage vector . The optimal fitting parameter vector is then found by solving the following least squares problem: (4) Since the model has many parameters, an improved differential algorithm that does not rely on the selection of initial values ​​is used for fitting. Therefore, the proposed explicit fitting function greatly simplifies the state estimation computation while maintaining a high approximation to the widely used five-parameter model.

[0026] As mentioned above, the state variable vector of the active distribution network is set as , where the vector is the voltage amplitude of each node, vector is the voltage phase angle of each node. The measurement vector of the distribution network is expressed as ,in, is the measurement vector of the node voltage amplitude, and are the measurement vectors of branch active power and reactive power respectively, and are the measurement vectors of active power and reactive power injected into the node respectively. Assuming that i The node is connected to k If there are 1 photovoltaic power station, the measurement equation of power injection is as follows: (5) (6) in, G ij is the node conductance, B ij is the node susceptance; Assume that the active distribution network consists of m The state vector of a PV power station should be corrected to , and the measurement vector should be corrected to .

[0027] In this embodiment, based on Figure 2, the derivation process of the measurement equation in the photovoltaic power plant is as follows: (7) (8) (9) (10) (11) (12) (13) (14) (15) (16) (17) (18) (19) Among them, the superscript is The quantity is the measured value, and the superscript is The quantity is the measurement error, is the fitting function of the photovoltaic array, R A 、R A-IGBT 、R A-dio They are the three types of equivalent resistances of photovoltaic plants. V dc2 、V A-IGBT1 、V A-dio1 、V A-dio2 、V ac is the equivalent voltage inside the photovoltaic plant model, k A-IGBT2 、k A-dio2 、M pv is the voltage coefficient. In addition, according to the law of conservation of power, the following five virtual measurement equations can be obtained: (20) (twenty one) (twenty two) (twenty three) (twenty four) in, They are virtual measurement values ​​in the photovoltaic system, and their values ​​are all 0; R D 、R D-IGBT 、 R D-dio They are the three types of equivalent resistances of photovoltaic plants. V D-IGBT1 、V D-dio1 、V D-dio2 、V pvs is the equivalent voltage inside the photovoltaic plant model, k D-IGBT2 、k D-dio2 is the voltage coefficient.

[0028] Therefore, the available measurement data for photovoltaic power generation systems are: (25) in, are virtual measured values ​​in the photovoltaic system, respectively.

[0029] After completing the above active distribution network state estimation modeling considering the photovoltaic power station model, the next step is to enter the active distribution network state estimation: In view of the high dynamic load and multi-energy penetration characteristics of active distribution networks, traditional state estimation methods have problems such as low computational efficiency, poor flexibility, and insufficient robustness. The present application embodiment proposes a data-driven state estimation method, which transforms the Transformer prediction model (Transformer model unit 1, see Figure 4 ) and GCN fusion (GCN model unit 2, see Figure 5 ) to obtain the GCN-Transformer method, replacing the traditional state prediction equation and measurement equation, through DNN learning (feature fusion and state estimation unit 3, see Figure 6 ) The feature values ​​extracted by the first two methods are mapped to the final system state value. The specific algorithm framework is as follows Figure 3 shown.

[0030] A Transformer-based state prediction equation is constructed to use historical measurement data to predict the system state at future moments. The optimal fitting parameter vector obtained by the least squares problem is used to improve the accuracy of the prediction.

[0031] In one example, in the state estimation of the distribution network, the state prediction equation is usually based on the power mechanism model and the load curve to jointly infer the system state at the future moment. However, the traditional mechanism model-based method is difficult to consider the real-time dynamic changes of multi-loads and cannot accurately capture the time series correlation and long-term dependence of the system state. Therefore, this embodiment uses the Transformer model to replace the traditional state prediction equation.

[0032] The Transformer model is an efficient model designed for long sequence prediction. Based on the sparse self-attention mechanism, it can extract global time dependencies from historical data and effectively capture the long-term dependency characteristics in the operation of the distribution network. As input, express The measurement vector at each moment, such as node voltage, power injection, etc. The above measurement data is processed by the encoder-decoder architecture of Transformer.

[0033] Specifically, given the historical measurement data, the encoder part of the Transformer first converts it into a high-dimensional time series feature representation. Through the self-attention mechanism, the model can select key features according to the importance weights in the time series. The role of the self-attention mechanism can be expressed as: (26) in, , and They are query matrix, key matrix and value matrix respectively. is the dimension of the key matrix. This mechanism enables the model to assign weights to each historical time step, focusing on the most important time step for the current state prediction. Through the multi-head attention mechanism, Transformer can capture multiple levels of temporal dependencies at the same time.

[0034] After obtaining the time series feature representation, the decoder part of the Transformer predicts the state estimate at the future moment based on these features. ,in is the predicted time step. The predicted state vector Including key system states such as voltage amplitude and phase angle; (27) Construct a measurement equation based on GCN (graph neural network), input the topological structure and measurement data of the distribution network into GCN, and model the relationship between the node state and the state of its neighboring nodes through graph convolution operations to provide spatial features for feature fusion in step 8; specifically: In traditional state estimation, the measurement equation uses the nonlinear equation (2) to convert the physical measurement data With the state vector To associate, Represents the state vector With the measurement vector The physical mapping relationship between them.

[0035] The embodiment of the present application uses GCN to represent the measurement equation, and its input includes the topological structure of the distribution network and the measurement data. The injected active and reactive power and node voltage amplitude data in are taken as the node set, and the active and reactive power and branch impedance data transmitted by the branch are taken as the edge set. The initial state matrix It is expressed as: (28) in, is the number of nodes, each Representative Node The node collection data.

[0036] In the GCN model, each node Status The status of its neighbor nodes The relationship between is realized through graph convolution operation, which can be expressed as: (29) in, Representation Node In the The state after the layer graph convolution, Is a node The set of neighbor nodes of is the weight matrix of the graph convolutional layer, is the bias, is a nonlinear activation function, such as ReLU. Through such neighborhood aggregation operations, each node can continuously update its state representation, comprehensively consider the measurement information of itself and its neighboring nodes, and finally obtain the system state High-dimensional representation of .

[0037] Furthermore, in order to improve the robustness of measurement data processing, the embodiment of the present application also introduces an attention mechanism in GCN, which assigns different weights between different neighbor nodes, thereby focusing on nodes that have a greater impact on state estimation. The attention mechanism is defined in the following way: (30) in, Representation Node and nodes In between Attention weights in layer graph convolution, is the attention parameter vector. Through such an attention mechanism, GCN can automatically focus on the most critical measurement data for state estimation when there is noise or missing measurement data, thereby improving the accuracy and robustness of the estimation.

[0038] A deep neural network (DNN) is used to fuse the state prediction and measurement results, integrating the spatial features extracted by GCN and the temporal features captured by Transformer to output a more accurate distribution network state estimate. The enhanced robustness in the previous step ensures the accuracy of the final state estimate.

[0039] Specifically, suppose the output of GCN is , represents the high-dimensional spatial features of the node, and the output of Transformer is , reflecting the time series estimation characteristics of the node, the concatenated joint feature vector can be expressed as: (31) This joint feature vector is input into the DNN for further processing.

[0040] DNN consists of several fully connected layers, which gradually map these high-dimensional features to the final state estimate by learning the relationship between different features. Each layer of DNN is connected by a weight matrix and the bias vector To calculate the linear combination, and use the nonlinear activation function for nonlinear transformation. Through this layer-by-layer mapping, DNN can comprehensively process the spatial information from GCN and the temporal information from Transformer, and gradually learn the complex nonlinear relationship between the two.

[0041] The calculation form of the output layer is: (32) in, For the final estimation result, For the The output of the layer, For the The weight matrix of the layer, It is The bias vector for the layer.

[0042] By integrating deep features with models, DNN can automatically learn and extract the dynamic temporal features of Transformer and the spatiotemporal features of GCN output, and has stronger robustness when processing incomplete or noisy data.

[0043] The following is a system embodiment of the present invention, which can be used to implement the method embodiment. For details not disclosed in this embodiment, please refer to the method embodiment.

[0044] This embodiment provides an active distribution network state estimation system based on the fusion of GCN and Transformer, including: A graph model building unit, used to build a distribution network graph model, in which each node in the distribution network and the lines connecting these nodes are abstracted into a graph structure; A photovoltaic power station physical model building unit, used to build a photovoltaic power station physical model, wherein the photovoltaic power station physical model includes a photovoltaic array, a DC / DC converter and a DC / AC converter, and uses a five-parameter model to describe the relationship between terminal voltage and current; Transformer model unit, used to construct a Transformer-based state prediction equation and use historical measurement data to predict the system state at future times; The GCN model unit is used to construct the measurement equation based on GCN, input the topological structure and measurement data of the distribution network into GCN, and realize the relationship modeling between the node state and the state of its neighboring nodes through graph convolution operation; The feature fusion and state estimation unit is used to use deep neural networks to fuse the state prediction and measurement results, integrate the spatial features extracted by GCN and the temporal features captured by Transformer, and output a more accurate distribution network state estimate.

[0045] Similarly, based on the concept of the method embodiment, the embodiment of the present application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, an active distribution network state estimation method based on the fusion of GCN and Transformer is implemented.

[0046] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in a computer storage medium to implement the corresponding method flow or corresponding function.

[0047] Similarly, based on the concept of the method embodiment, the embodiment of the present application also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, an active distribution network state estimation method based on the fusion of GCN and Transformer is implemented.

[0048] It can be understood that the computer-readable storage medium here can include both built-in storage media in the computer device and, of course, extended storage media supported by the computer device. The computer-readable storage medium provides a storage space that stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space. These instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the active distribution network state estimation method based on the fusion of GCN and Transformer in the above embodiment.

[0049] In summary, this application proposes a new state estimation method based on the combination of GCN and Transformer model to solve the problems of dynamic load, complex topology and data quality in new active distribution networks. By using GCN for the transformation of measurement data, the spatial dependency between nodes and neighbors is fully exploited; while Transformer is responsible for the state prediction of long-series time dependency, which effectively improves the prediction accuracy and robustness of the model. It is significantly superior to traditional state estimation methods in terms of accuracy, anti-interference ability and computational efficiency. This method can be applied to new power system applications with more voltage levels.

[0050] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. Active distribution network state estimation method based on GCN and Transformer fusion, characterized by: Step 1: Establish a distribution network graph model, in which each node in the distribution network and the lines connecting these nodes are abstracted into a graph structure; Step 2: construct a physical model of a photovoltaic power station, wherein the physical model of the photovoltaic power station includes a photovoltaic array, a DC / DC converter and a DC / AC converter, and uses a five-parameter model to describe the relationship between terminal voltage and current; Step 3: Construct a state prediction equation based on Transformer and use historical measurement data to predict the system state at future times; Step 4: Construct a measurement equation based on GCN, input the topological structure and measurement data of the distribution network into GCN, and model the relationship between the node state and the state of its neighboring nodes through graph convolution operation; Step 5: Use a deep neural network to fuse the state prediction in step 3 and the measurement results in step 4, integrate the spatial features extracted by GCN and the temporal features captured by Transformer, and output a more accurate distribution network state estimate.

2. The active distribution network state estimation method based on GCN and Transformer fusion according to claim 1 is characterized by: In step 1, the electrical properties of each node are determined, including voltage amplitude, phase angle, active power and reactive power; the electrical properties of each edge are determined, including line impedance and line flow.

3. The active distribution network state estimation method based on GCN and Transformer fusion according to claim 1 is characterized by: Step 2 also includes approximating the voltage-current relationship of the photovoltaic array using an explicit fitting function.

4. The active distribution network state estimation method based on GCN and Transformer fusion according to claim 3 is characterized by: Based on the five-parameter model, multiple sets of light intensity, temperature, output voltage and output current data are randomly generated in the feasible region; the optimal fitting parameter vector is found by solving the least squares problem to improve the accuracy of state prediction.

5. The method for estimating the state of an active distribution network based on the fusion of GCN and Transformer according to claim 1 is characterized in that: In step 3, the encoder-decoder architecture of the Transformer model is used to process historical measurement data and extract global time dependencies; Multi-level temporal dependencies are captured through self-attention mechanism and multi-head attention mechanism.

6. The method for estimating the state of an active distribution network based on the fusion of GCN and Transformer according to claim 5 is characterized in that: The attention mechanism is introduced into GCN to dynamically adjust the weights according to the correlation between nodes, improve the robustness of measurement data processing, and ensure that the feature fusion in step 5 can focus on key measurement data when there is noise or missing data.

7. The method for estimating the state of an active distribution network based on the fusion of GCN and Transformer according to claim 5 is characterized in that: Step 5 is as follows: Use DNN to effectively integrate the spatial features extracted by GCN and the temporal features captured by Transformer to output a more accurate distribution network status; The fully connected layers of DNN learn the relationship between different features and gradually map these high-dimensional features to the final state estimation.

8. Active distribution network state estimation system based on GCN and Transformer fusion, characterized by: include: A graph model building unit, used to build a distribution network graph model, in which each node in the distribution network and the lines connecting these nodes are abstracted into a graph structure; A photovoltaic power station physical model building unit is used to build a photovoltaic power station physical model, wherein the photovoltaic power station physical model includes a photovoltaic array, a DC / DC converter and a DC / AC converter, and uses a five-parameter model to describe the relationship between terminal voltage and current; Transformer model unit, used to construct a Transformer-based state prediction equation and use historical measurement data to predict the system state at future times; The GCN model unit is used to construct the measurement equation based on GCN, input the topological structure and measurement data of the distribution network into GCN, and realize the relationship modeling between the node state and the state of its neighboring nodes through graph convolution operation; The feature fusion and state estimation unit is used to use deep neural networks to fuse the state prediction and measurement results, integrate the spatial features extracted by GCN and the temporal features captured by Transformer, and output a more accurate distribution network state estimate.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the active distribution network state estimation method based on GCN and Transformer fusion according to any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the active distribution network state estimation method based on GCN and Transformer fusion according to any one of claims 1 to 7 is implemented.

Images