Topological parameter joint identification method based on tidal current diagram convolutional network and related equipment

Abstract

The invention discloses a topological parameter joint identification method based on a tidal current diagram convolutional network and related equipment, and the method comprises the steps: building a node correlation matrix through employing a node correlation principle, proposing a minimum distance iterative algorithm to generate a candidate topological set, and enabling the candidate topological set to serve as the diagram input of a parameter estimation model; based on topological dependence and convolution characteristics of an alternating current power flow equation, designing a PFGCN network for line parameter estimation, and realizing parameter refining through an alternating iteration process of pseudo load flow calculation and neural network training; according to the method, training convergence and a loss function value are used as feedback, candidate topologies are screened and verified, finally, joint accurate estimation of the topologies and parameters is achieved, and the problem that the limitation of relying on high-cost voltage phase angle measurement still exists in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network topology and parameter identification, and in particular to a topology parameter joint identification method based on a power flow graph convolutional network and related equipment. Background Art

[0002] With the integration of new elements such as distributed energy and energy storage systems, the structure of the distribution network has become increasingly complex, and users' electricity demand has continued to expand. This trend has put forward higher requirements for the planning and management, safe operation and status monitoring of the distribution network. Accurate topology information and line parameters are the key prerequisites for efficient planning and operation of the distribution system. However, due to insufficient investment by distribution system operators, the distribution system lacks the necessary status monitoring equipment, resulting in low observability. In addition, due to factors such as the operator's extensive management and the untimely data update during line transformation and maintenance, the problem of missing topology information or line parameters in the distribution network and the difficulty in accurate estimation has become increasingly prominent.

[0003] With the continuous improvement of advanced metering infrastructure, the observability problem of distribution systems has been effectively alleviated. At the same time, these infrastructures also provide massive measurement data, laying a solid foundation for the application of data-driven methods. Data-driven methods have become a research hotspot in the field of topology identification and parameter estimation of distribution systems, which are mainly divided into model-based methods and neural network methods.

[0004] Model-based methods can be divided into single attribute research and joint estimation research according to the research scope. Single attribute research focuses on topology identification or line parameter estimation. Topology identification is usually based on node correlation, while the basic principle of line parameter estimation is the power flow equation.

[0005] In view of the above problems, the existing technology proposes the following methods: detecting the topology change of the distribution network by analyzing the characteristics of PMU data; inferring the topology of the distribution network by energy conservation and principal component analysis based on energy data; using node voltage data, proposing a topology identification algorithm based on graph recovery; using Markov random field theory to analyze the correlation of node voltages, and using the maximum likelihood estimation method to solve the topology; for real-time topology prediction, a two-stage topology identification method based on split expectation maximization is proposed, but this method has high requirements on data integrity. The above-mentioned topology identification research usually only uses a single correlation evaluation, and the identification results are easily interfered by specific factors. Furthermore, a weighted least squares method (WLS) is proposed for line parameter estimation; a weighted nonlinear least squares scheme is proposed. However, the regression algorithm is sensitive to data errors, and its accuracy still needs to be improved. It is worth noting that in actual distribution networks, the coverage of PMU devices is limited and the cost is high, and the method based on voltage phase angle data faces challenges in practical applications.

[0006] In actual distribution networks, the missing topology and line parameters often occur simultaneously. The joint estimation research of topology identification and line parameter estimation can better meet the needs of practical applications. The existing technology solves the network topology based on the linear regression algorithm, and then estimates the line parameters in the order from leaf nodes to root nodes. Or, the resistance distance between nodes is calculated using the linear power flow model, and then the topology and line impedance are restored using the recursive grouping algorithm. Or, a topology label matrix is ​​constructed based on the linear power flow model, the topology is identified by feature clustering, and the line parameters are estimated by linear regression. Or, a small number of PMUs are deployed, and the topology and line parameters are iteratively solved by the unscented Kalman filter and the Newton-Raphson method. Or, the joint estimation of topology and line parameters is realized by using pseudo power flow and Newton-Raphson's iterative method. Unfortunately, the approximate method based on the linear power flow model has limited accuracy, and the convergence of the Newton-Raphson iterative method is unstable and easily affected by the initial value.

[0007] Compared with traditional methods, neural networks have significant advantages in processing complex nonlinear relationships, so they have attracted widespread attention. At present, there is a technology that applies CNN to topology recognition problems, while another technology proposes a distribution network topology recognition method based on attentional and convolutional neural networks (ACNN). Or, based on convolutional neural networks, PMU data features are extracted to build a line parameter identification model. The above technologies model topology identification or line parameter estimation as classification or screening problems, which cannot be applied to distribution networks with missing topology information and line parameters. Or, the joint graph convolutional network and full convolutional network perform the parameter estimation task of the transmission network, which cannot adapt to the distribution network with a lack of measurement devices. Or, a topology and parameter estimation method based on deep and shallow neural networks is proposed, but this method requires regional division based on the observability level, which faces challenges in practical applications. Or, the distribution network topology is embedded in the graph neural network, which proves the feasibility of line parameter estimation by gradient descent. In the above studies, the neural network lacks clear physical information and lacks interpretability. In addition, the non-Euclidean data processing advantages of the graph neural network have not been effectively utilized.

[0008] In summary, the existing technology still has the limitation of relying on high-cost voltage phase angle measurement. Summary of the invention

[0009] Based on the problems raised by the above background technology, the purpose of the present invention is to provide a topology parameter joint identification method based on a power flow graph convolutional network and related equipment. The present invention constructs a node correlation matrix based on the node correlation principle, and proposes a minimum distance iteration algorithm to generate a candidate topology set, and uses the candidate topology set as the graph input of the parameter estimation model; based on the topological dependence and convolution characteristics of the AC power flow equation, a PFGCN network is designed for line parameter estimation, and parameter refinement is achieved through the alternating iterative process of pseudo power flow calculation and neural network training; the method uses training convergence and loss function value as feedback to screen and verify candidate topologies, and finally realizes the joint accurate estimation of topology and parameters, which solves the problem that the prior art still relies on high-cost voltage phase angle measurement.

[0010] The present invention is achieved through the following technical solutions:

[0011] The first aspect of the present invention provides a method for joint identification of topological parameters based on a power flow graph convolutional network, comprising the following steps:

[0012] Acquiring power and voltage amplitude measurement data, and constructing a node correlation matrix using the power and voltage amplitude measurement data;

[0013] Iteratively calculating the node correlation matrix using a minimum distance iteration algorithm to generate a candidate topology set;

[0014] A line parameter estimation model based on a power flow graph convolutional network is established, and the candidate topology set is input into the line parameter estimation model for screening to obtain a final topology structure.

[0015] In the above technical solution, the power and voltage amplitude measurement data are obtained and grouped, which reduces the computational complexity for subsequent joint identification and improves the fault tolerance of topology identification results. The node correlation matrix is ​​constructed by calculating the Euclidean distance and dynamic time warping distance between the grouped data, and the candidate topology set is generated using the minimum distance iteration algorithm.

[0016] The line parameter estimation problem is modeled as an optimization problem of minimizing the power estimation error, and a line parameter estimation model is established based on the topological dependence and complex-valued convolution characteristics of the AC power flow equation. Considering the lack of voltage phase angle data in actual measurement, the pseudo power flow calculation method is used in the line parameter estimation model to supplement the missing data. The line parameter estimation model iteratively calculates the pseudo power flow and admittance estimation to accurately estimate the line parameters.

[0017] In an optional embodiment, constructing a node correlation matrix using the power and voltage amplitude measurement data comprises the following steps:

[0018] Extracting any two sets of node voltage data from the power and voltage amplitude measurement data;

[0019] Calculating the Euclidean distance and the DDTW distance between the two groups of node voltage data, and establishing a Euclidean distance matrix and a DDTW distance matrix according to the Euclidean distance and the DDTW distance;

[0020] The Euclidean distance matrix and the DDTW distance matrix are normalized, and a node correlation matrix is ​​constructed using the normalized Euclidean distance matrix and the DDTW distance matrix.

[0021] In an optional embodiment, the node correlation matrix is ​​iteratively calculated using a minimum distance iteration algorithm, including the following steps:

[0022] Initialize the node pool, line pool and node adjacency matrix;

[0023] Arrange the node correlation matrix in ascending order to obtain an index matrix;

[0024] Traversing the index matrix, checking the connection conditions between nodes, if node i and node j meet the connection conditions, establishing a connection between the node i and the node j, and updating the line pool;

[0025] Checking a minimum distance condition, establishing a connection between corresponding nodes that satisfy the minimum distance condition, and updating the line pool;

[0026] Removing nodes with established connections from the initialization node pool based on the updated line pool;

[0027] When the initialization node pool after removing the nodes with established connections is not empty, traverse the initialization node pool, establish connections between the remaining nodes in the initialization node pool and the corresponding nodes in the index matrix, and update the line pool;

[0028] The node adjacency matrix is ​​updated using the updated line pool to obtain a candidate topology set.

[0029] In an optional embodiment, a line parameter estimation model based on a power flow graph convolutional network is established, and the candidate topology set is input into the line parameter estimation model for screening, including:

[0030] Constructing an AC power flow equation, mapping the AC power flow equation into a complex-valued graph convolution framework, and generating an AC power flow equation in a complex-valued convolution form;

[0031] Constructing an AC power flow equation, mapping the AC power flow equation into a complex-valued graph convolution framework, and generating an AC power flow equation in a complex-valued convolution form;

[0032] Using the phase angle estimation layer to perform pseudo power flow estimation on the candidate topology set to obtain phase angle data;

[0033] Acquiring topological power voltage amplitude data, performing regression estimation on the topological power voltage amplitude data, and obtaining a regression estimation initial value;

[0034] Inputting the regression estimation initial value and the phase angle data into the parameter estimation layer for iterative calculation to obtain estimated values ​​of conductance and susceptance matrices, and performing admittance matrix property correction on the estimated values ​​of conductance and susceptance matrices;

[0035] The estimated values ​​of the conductance and susceptance matrices after the admittance matrix properties are corrected are input into the phase angle estimation layer for the next iteration until a stop iteration condition is met.

[0036] In an optional embodiment, the parameter estimation layer is composed of a power flow graph convolutional network, wherein the neurons of the power flow graph convolutional network and the connection relationship between the neurons correspond one-to-one to the calculation process of the AC power flow equation based on the complex-valued convolution form.

[0037] In an optional embodiment, mapping the AC power flow equation into a complex-valued graph convolution framework includes:

[0038] Establishing a distribution network topology model and preprocessing the distribution network topology model;

[0039] defining an intermediate variable function, bringing the intermediate variable function and the preprocessed distribution network topology model into the AC power flow equation to obtain an AC power flow equation in a complex-valued convolution form;

[0040] Among them, establishing the distribution network topology model includes: encoding the distribution network topology into a node adjacency matrix, expressing the voltage amplitude and phase angle with a diagonal matrix, and expressing the line parameters in complex form of an admittance matrix.

[0041] In an optional embodiment, performing regression estimation on the topological power voltage amplitude data includes:

[0042] The topological power voltage amplitude data is calculated using an AC power flow equation to obtain an active power injection-voltage amplitude ratio and a reactive power injection-voltage amplitude ratio;

[0043] The active power injection-voltage amplitude ratio and the reactive power injection-voltage amplitude ratio are estimated and calculated to obtain conductance estimation and susceptance matrix estimation.

[0044] The second aspect of the present invention provides a topological parameter joint identification system based on a power flow graph convolutional network, comprising:

[0045] A node correlation module, used to obtain power and voltage amplitude measurement data, and construct a node correlation matrix using the power and voltage amplitude measurement data;

[0046] A candidate topology module is used to iteratively calculate the node correlation matrix using a minimum distance iteration algorithm to generate a candidate topology set;

[0047] The parameter estimation model module is used to establish a line parameter estimation model based on a power flow graph convolutional network, input the candidate topology set into the line parameter estimation model for screening, and obtain the final topology structure.

[0048] The third aspect of the present invention 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, a method for joint identification of topological parameters based on a power flow graph convolutional network is implemented.

[0049] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a method for joint identification of topological parameters based on a power flow graph convolutional network.

[0050] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0051] The joint identification method proposed in the present invention has good robustness. Even when there are certain random errors in the measured data, it still shows high identification accuracy and robustness, which solves the problem that the existing technology still relies on high-cost voltage phase angle measurement. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other relevant drawings can be obtained based on these drawings without creative work. In the drawings:

[0053] Figure 1 A schematic diagram of a topology and parameter joint estimation framework provided in Example 1 of the present invention;

[0054] Figure 2 A schematic diagram of the structure of a single feeder topology provided in Embodiment 1 of the present invention;

[0055] Figure 3 A schematic diagram of the structure of a parameter estimation model based on PV-GCN provided in Example 1 of the present invention;

[0056] Figure 4A schematic diagram of the structure of a power flow graph convolutional network provided in Example 1 of the present invention;

[0057] Figure 5 A schematic diagram of the structure of a 33-node system topology provided in Example 1 of the present invention;

[0058] Figure 6 A schematic diagram of the structure of a 69-node system topology provided in Example 1 of the present invention;

[0059] Figure 7 In the C33 scenario provided in Example 1 of the present invention, α1=0.6, 6 groups of correlation matrix K1-K6 results of the 33-node system are obtained;

[0060] Figure 8 The training process loss function value F of topology 1 and topology 2 in the C33 scenario provided in Example 1 of the present invention;

[0061] Fig. 9 The initial value of the convolution kernel obtained by regression estimation in the C33 scenario provided in Example 1 of the present invention and the final result of the convolution kernel after training;

[0062] Fig.10 Comparison between the theoretical value and the estimated result of the line impedance of the C33 scenario provided in Example 1 of the present invention;

[0063] Fig.11 The IEEE-33 system α1=0.2, 0.4, 0.6, 0.8 provided in Example 1 of the present invention obtains the average value of the accuracy A1 of the candidate topology (C33 scenario);

[0064] Fig.12 The regression estimation initial values ​​and final estimation results of the line parameters in the IC69 scenario provided in Example 1 of the present invention;

[0065] Fig.13 Comparison between the reference value of the line parameter of the IC69 scenario provided in Example 1 of the present invention and the estimation result;

[0066] Fig.14 The line parameter estimation accuracy of the PFGCN model and the ARR model for the C33 scenario provided in Example 1 of the present invention;

[0067] Fig.15 The line parameter estimation accuracy of the PFGCN model and the ARR model for the IC69 scenario provided in Example 1 of the present invention;

[0068] Fig.16 The topology identification results of IEEE-33 system and improved IEEE-69 system with different error levels (ε=0, 0.1%, 0.2%) provided in Example 1 of the present invention;

[0069] Fig.17 A schematic diagram of the structure of an electronic device provided in Example 3 of the present invention. DETAILED DESCRIPTION

[0070] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0071] Example 1

[0072] Embodiment 1 of the present invention proposes a method for joint identification of topological parameters based on a power flow graph convolutional network, comprising the following steps:

[0073] Acquiring power and voltage amplitude measurement data, and constructing a node correlation matrix using the power and voltage amplitude measurement data;

[0074] Iteratively calculating the node correlation matrix using a minimum distance iteration algorithm to generate a candidate topology set;

[0075] A line parameter estimation model based on a power flow graph convolutional network is established, and the candidate topology set is input into the line parameter estimation model for screening to obtain a final topology structure.

[0076] like Figure 1 As shown in the figure, the power and voltage amplitude measurement data are obtained and the obtained data are grouped, which reduces the computational complexity for subsequent joint identification and improves the fault tolerance of the topology identification results. The grouped data is used to calculate the Euclidean distance and dynamic time warping distance (DDTW) between nodes to construct a node correlation matrix, and then the minimum distance iteration algorithm is used to generate a candidate topology set. The purpose of constructing a candidate topology set rather than a single topology is to improve the fault tolerance of topology identification errors. These candidate topology information is encoded in the node correlation matrix and used as the graph structure input of the subsequent line parameter estimation model.

[0077] The line parameter estimation problem is modeled as an optimization problem of minimizing the power estimation error. Based on the topological dependence and complex-valued convolution characteristics of the AC power flow equation, a line parameter estimation model is established. Different from the traditional graph convolutional network, the power flow graph convolutional network (PFGCN) is an integrated network. The neuron connection and operation process correspond to the AC power flow equation, and the network convolution kernel is equivalent to the admittance matrix of the AC power flow equation. The filter is customized to match the characteristics of the admittance matrix, thus showing excellent model interpretability. The initial value of the convolution kernel is obtained based on the linear estimation method of power and voltage amplitude measurement. The size of the network convolution kernel is determined by the graph input and has topological adaptability. The topology identification link provides candidate topologies for the complex-valued graph convolutional network. The organic combination of the two makes this method have system adaptability. A parameter matrix attribute verification module is embedded in the framework. This module dynamically adjusts the convolution kernel parameters during the iteration process to ensure that it meets the inherent properties of the admittance matrix. Finally, the candidate topology is selected based on the final loss function value and its process convergence.

[0078] Taking into account the lack of voltage phase angle data in actual measurement, the pseudo power flow calculation method is used in the line parameter estimation model to supplement the missing data. The line parameter estimation model iterates the pseudo power flow calculation and admittance estimation to accurately estimate the line parameters.

[0079] In an optional embodiment, constructing a node correlation matrix using the power and voltage amplitude measurement data comprises the following steps:

[0080] Extracting any two sets of node voltage data from the power and voltage amplitude measurement data;

[0081] Calculating the Euclidean distance and the DDTW distance between the two groups of node voltage data, and establishing a Euclidean distance matrix and a DDTW distance matrix according to the Euclidean distance and the DDTW distance;

[0082] The Euclidean distance matrix and the DDTW distance matrix are normalized, and a node correlation matrix is ​​constructed using the normalized Euclidean distance matrix and the DDTW distance matrix.

[0083] The low-voltage distribution network usually presents a radial grid structure, and its topology can be represented by an undirected graph as follows: Where V and B represent the vertex set and edge set respectively. The node adjacency matrix A encodes the undirected graph information and stores it in the matrix, which is defined as:

[0084]

[0085] This embodiment uses Figure 2 Taking the single feeder as an example, node i and node i+1 satisfy:

[0086]

[0087] Node voltage and line current satisfy:

[0088]

[0089] Where: Indicates I i The conjugation of.

[0090] Substituting equation 3 into equation 2 and simplifying it, we can get:

[0091]

[0092] The voltage between any two nodes in the line is related to the load distribution and line parameters. The closer the distance between the two nodes, the smaller the transverse and longitudinal components of the voltage drop, the smaller the difference in voltage amplitude at the same time, and the more similar the voltage change trends at different times.

[0093] Based on the above node correlation principle, this embodiment adopts the DDTW distance and the Euclidean distance to jointly characterize the node voltage correlation, and constructs a node correlation matrix based on this.

[0094] Specifically, assume that the two sets of node voltage data are: V a =[V a,1 V a,2 V a,3 ...V a,m ],V b =[V b,1 V b,2 V b, 3...V b,m ], where m is the number of measurement groups. The Euclidean distance between two nodes is:

[0095]

[0096] Where: V a,i ,V b,i V a ,V b The i-th element in .

[0097] The DDTW distance between two nodes is:

[0098]

[0099] Where: C is the optimal path of DDTW, C={c1,c2,...,c p}; S is the number of elements on the path.

[0100] Furthermore, the DDTW distance is calculated based on the first-order discrete derivative of the original data, which is defined as:

[0101]

[0102] Where: V a ′ is the original data V a The first-order discrete derivative value of , m is the number of measurement groups.

[0103] By V a ′,V b The m×m distance matrix composed of ′ can determine the DDTW optimal path C, where the elements in the distance matrix are:

[0104] d(i,j)=(V a ' ,i -V b ' ,j ) 2

[0105] Among them, i, j are the coordinates of the distance matrix.

[0106] Furthermore, the optimal path satisfies the following conditions:

[0107]

[0108] c1=d(1,1),c p =d(p,p)

[0109] c k =d(i k ,j k ),c k+1 =(i k+1 ,j k+1 )

[0110] i k+1 -i k ≤1,j k+1 -j k ≤1;

[0111] i k ≤i k+1 ,j k ≤j k+1 .

[0112] Through the above calculation, we can get the Euclidean distance matrix of the n-node system and DDTW distance matrix Perform matrix normalization on the Euclidean distance matrix and the DDTW distance matrix:

[0113]

[0114] Where: n is the number of nodes; are the normalized D , T matrix.

[0115] Use the normalized Euclidean distance matrix and DDTW distance matrix to construct the node correlation matrix K:

[0116]

[0117] Where: α is the combination coefficient.

[0118] It should be noted that the K main diagonal elements are 0, and the values ​​of the elements in the remaining positions are inversely proportional to the correlation.

[0119] In an optional embodiment, the node correlation matrix is ​​iteratively calculated using a minimum distance iteration algorithm, including the following steps:

[0120] Initialize the node pool, line pool and node adjacency matrix;

[0121] Arrange the node correlation matrix in ascending order to obtain an index matrix;

[0122] Traversing the index matrix, checking the connection conditions between nodes, if node i and node j meet the connection conditions, establishing a connection between the node i and the node j, and updating the line pool;

[0123] Checking a minimum distance condition, establishing a connection between corresponding nodes that satisfy the minimum distance condition, and updating the line pool;

[0124] Removing nodes with established connections from the initialization node pool based on the updated line pool;

[0125] When the initialization node pool after removing the nodes with established connections is not empty, traverse the initialization node pool, establish connections between the remaining nodes in the initialization node pool and the corresponding nodes in the index matrix, and update the line pool;

[0126] The node adjacency matrix is ​​updated using the updated line pool to obtain a candidate topology set.

[0127] Specifically, the node correlation matrix K contains the node connection relationship. The smaller the element value, the stronger the correlation between the corresponding index nodes. Therefore, this implementation proposes to identify the topological connection relationship from the matrix K and generate a candidate topology. The algorithm flow is as follows:

[0128]

[0129]

[0130] In an optional embodiment, a line parameter estimation model based on a power flow graph convolutional network is established, and the candidate topology set is input into the line parameter estimation model for screening, including:

[0131] Constructing an AC power flow equation, mapping the AC power flow equation into a complex-valued graph convolution framework, and generating an AC power flow equation in a complex-valued convolution form;

[0132] A phase angle estimation layer for pseudo power flow estimation is established, and a parameter estimation layer is established based on the AC power flow equation in complex valued convolution form;

[0133] Using the phase angle estimation layer to perform pseudo power flow estimation on the candidate topology set to obtain phase angle data;

[0134] Acquiring topological power voltage amplitude data, performing regression estimation on the topological power voltage amplitude data, and obtaining a regression estimation initial value;

[0135] Inputting the regression estimation initial value and the phase angle data into the parameter estimation layer for iterative calculation to obtain estimated values ​​of conductance and susceptance matrices, and performing admittance matrix property correction on the estimated values ​​of conductance and susceptance matrices;

[0136] The estimated values ​​of the conductance and susceptance matrices after the admittance matrix properties are corrected are input into the phase angle estimation layer for the next iteration until a stop iteration condition is met.

[0137] It should be noted that the graph convolutional neural network (GCN) is a deep learning algorithm based on graph structure learning.

[0138] The core idea is to effectively extract and integrate the local features and global features of nodes and their neighboring nodes by performing convolution operations on the adjacency relationship of the graph. The forward propagation formula is:

[0139]

[0140] Where: H (l) represents the feature matrix of the l-th layer node, H (l+1) is the node feature matrix of the next layer; Represents the adjacency matrix A of the graph plus the identity matrix I, is the adjacency matrix considering node self-loops; for The degree matrix of represents the total number of edges connected to the i-th node; W (l) is the trainable weight matrix of the lth layer; σ(·) is the activation function.

[0141] The complex-valued graph convolutional network (CV-GCN) extends the graph convolution operation from the real domain to the complex domain. It processes graph structure data with amplitude and phase information through complex-valued convolution operations and has strong complex signal modeling capabilities.

[0142] The forward propagation formula is:

[0143]

[0144] Different from traditional GCN, H in CV-GCN (l) ,W (l) are all complex-valued matrices, The subscripts r and i represent the real part and imaginary part respectively.

[0145] Based on the theoretical basis of CV-GCN, the present invention proposes a parameter estimation model of the power flow graph convolutional network. The core of the parameter estimation model proposed in the present invention is the AC power flow equation, which is as follows:

[0146]

[0147] Where: p i ,q i ,v i are the active power injection, reactive power injection and voltage amplitude of the node respectively; G ij ,B ij are the conductance and susceptance of the line respectively; θ ij is the node phase angle difference.

[0148] The AC power flow equation has topological dependence, which means that the power of any node is not only related to the state of the node itself but also to the state of the adjacent nodes. Traditional parameter identification methods often find it difficult to effectively capture and utilize this topological dependence, and face challenges when dealing with complex power grid structures. It is particularly important to explore a method that efficiently integrates power grid topology information.

[0149] The complex-valued characteristics of the AC power flow equation are reflected in the fact that the electrical quantities therein, such as voltage, line parameters, power, etc., can all be expressed in complex form. The complex-valued graph convolutional network not only has excellent graph structure data processing capabilities, but also realizes the processing of complex domain data features, making it an ideal choice for realizing node data feature aggregation based on power grid topology.

[0150] Among them, in the AC power flow equation, the power data of any node is only affected by itself and its neighboring nodes, so a single-layer graph convolutional network can meet the needs of data feature extraction.

[0151] In view of the high cost of obtaining voltage phase angle data in actual distribution networks, this method intentionally avoids the reliance on such measurement data during design. Nevertheless, the lack of voltage phase angle information may have an adverse effect on the accuracy of line parameter estimation. Therefore, the pseudo power flow calculation method is adopted in the present invention to supplement the necessary voltage phase angle information, thereby ensuring the accuracy of parameter estimation.

[0152] In summary, the present invention constructs Figure 3 The parameter estimation model based on PF-GCN is shown in the figure, where the parameter estimation model includes a phase angle estimation layer and a parameter estimation layer. The phase angle estimation layer completes the voltage phase angle estimation task when the line parameters are given through pseudo power flow calculation. The result of topology identification is used as the graph input of the parameter estimation model. The parameter estimation layer completes the line parameter estimation task based on the voltage amplitude, voltage phase angle data estimation value and power measurement.

[0153] After the parameter estimation layer completes the line parameter estimation, the estimation result is verified for the admittance matrix properties. If it does not meet the requirements, further property correction is required before the next iteration. If it meets the admittance matrix property verification, the training loss convergence judgment is performed. If it does not meet the training loss convergence requirements, topology correction is required before iteration.

[0154] The above iteration process is repeated until the termination condition is met. The iteration process sets a clear termination condition, that is, when the change in the loss function of the joint estimate is less than the preset iteration threshold, the iteration is terminated. ξ is the iteration threshold, is the loss function value of the nth iteration of the i-th group.

[0155] In an optional embodiment, the parameter estimation layer is composed of a power flow graph convolutional network, wherein the neurons of the power flow graph convolutional network and the connection relationship between the neurons correspond one-to-one to the calculation process of the AC power flow equation based on the complex-valued convolution form.

[0156] It should be noted that the parameter estimation layer is composed of a power flow graph convolutional network, which is a new network constructed by integrating graph convolution theory and AC power flow equation. The neurons and connections of the power flow graph convolutional network correspond one-to-one to the operation process of the AC power flow equation, and the line parameters to be estimated are modeled as sparse convolution kernels of the power flow graph convolutional network. Different from traditional convolutional neural networks, the power flow graph convolutional network optimizes the convolution kernel through gradient descent during the training process to achieve line parameter estimation.

[0157] In an optional embodiment, mapping the AC power flow equation into a complex-valued graph convolution framework includes:

[0158] Establishing a distribution network topology model and preprocessing the distribution network topology model;

[0159] defining an intermediate variable function, bringing the intermediate variable function and the preprocessed distribution network topology model into the AC power flow equation to obtain an AC power flow equation in a complex-valued convolution form;

[0160] Among them, establishing the distribution network topology model includes: encoding the distribution network topology into a node adjacency matrix, expressing the voltage amplitude and phase angle with a diagonal matrix, and expressing the line parameters in complex form of an admittance matrix.

[0161] It should be noted that the distribution network topology model is established: the distribution network topology is encoded as the node adjacency matrix A, the voltage amplitude and phase angle are expressed by a diagonal matrix, and the line parameters are expressed in the complex form of the admittance matrix. The distribution network topology model is preprocessed:

[0162]

[0163] Where: A is the node topology association matrix; is the node association matrix considering self-loops; diag(θ i ) is the diagonal matrix of node voltage phase angles, and the head transformer node is taken as the reference node by default; G , B are the conductance and susceptance matrices respectively.

[0164] Define the intermediate variable function:

[0165]

[0166] In the formula: Φ(G,B,N)Γ(G,B,N) is the defined intermediate variable function, which corresponds to the complex-valued convolution operation; ⊙ represents the matrix Hadamard product; cosN, sinN represent the operation of calculating the cosine and sine values ​​of each element of the matrix.

[0167] Substituting equations 13 and 14 into equation 12, we obtain;

[0168]

[0169] Where: diag(v i ) is the diagonal matrix of node voltage amplitude; I n ,I n×1 are the identity matrix and the n×1 identity matrix respectively; P=[p1 p2...p n ] T ,Q=[q1 q2…q n ] T .

[0170] Further combining Equation 15, we get the AC power flow equation in complex valued convolution form:

[0171]

[0172] In an optional embodiment, performing regression estimation on the topological power voltage amplitude data includes:

[0173] The topological power voltage amplitude data is calculated using an AC power flow equation to obtain an active power injection-voltage amplitude ratio and a reactive power injection-voltage amplitude ratio;

[0174] The active power injection-voltage amplitude ratio and the reactive power injection-voltage amplitude ratio are estimated and calculated to obtain conductance estimation and susceptance matrix estimation.

[0175] In this embodiment, the line parameter estimation problem is modeled as an optimization problem of minimizing the power estimation error, where the objective function is constructed as:

[0176]

[0177] Line parameter estimation uses the method of iterative estimation of voltage phase angle and line parameters, that is, alternating training and method.

[0178] The basic architecture of the flow graph convolutional network is as follows Figure 4 As shown in the figure, unlike traditional graph convolutional networks, the power flow graph convolutional network is an integrated structure. Compared with traditional graph convolutional networks, the power flow graph convolutional network also relies on the topological information of external input, which is encoded as a node adjacency matrix containing self-loops. In this model, the node characteristic inputs include voltage amplitude and power measurement data.

[0179] It should be noted that the weight convolution kernel to be trained of the power flow graph convolution network is specially designed according to the admittance matrix, and the position of its non-zero items matches the admittance matrix and has the same dimension.

[0180] It should be noted that the line parameter admittance matrix and the convolution kernel of the power flow diagram convolution network mentioned in the present invention are essentially the same concept, only different terms are used in the fields of power systems and neural networks.

[0181] In the present invention, the joint estimation of topology and parameters is realized through the topology correction link. This link uses the convergence of the parameter estimation process and the loss function value as the state signal to perform topology screening, and finally outputs the topology and its corresponding line parameters with converged and minimum training loss.

[0182] The input of the parameter estimation layer is the topology, power, voltage amplitude data and phase angle data of pseudo power flow estimation. The training goal is to minimize the error of active power and reactive power estimation. The convolution kernel with admittance matrix properties is used to learn and aggregate the characteristics of distribution network neighbor nodes to achieve line parameter estimation.

[0183] Specifically, the forward propagation formula of the parameter estimation layer is:

[0184]

[0185] Where: h (l) ,h (l-1) are the output and input of layer l respectively, W is the convolution kernel, is the node topology adjacency matrix considering self-loops, I n ,I n×1 Used for matrix transformations.

[0186] In order to constrain the numerical characteristics of the convolution kernel, the sum of the rows is 0, and the sum of the columns is 0. The following loss function is designed and applied:

[0187]

[0188] In this embodiment, the line parameter values ​​estimated based on linear regression (i.e., conductance estimation and susceptance matrix estimation) are used as iterative initial values ​​of the convolution kernel in the parameter estimation layer. Good initial values ​​help accelerate convergence and improve operating efficiency.

[0189] When the distribution network is operating safely, the voltage phase angle difference between adjacent bus nodes is within 5°, so cosθ≈1, sinθ≈θ. The AC power flow equation shown in Equation 12 can be rewritten as:

[0190]

[0191] Where: v=[v1v2...v n ] T θ ij =θ i -θ j .

[0192] Further ignoring the influence of phase angle difference, the m group measurements satisfy:

[0193]

[0194] Where: [V]=[[v]1...[v] M ].

[0195] Based on the regression estimation, G ij ′,B ij ′ as G ij ,B ij A rough estimate is valuable. The calculation equation is:

[0196]

[0197] The regression estimation result is often a non-sparse matrix. The regression estimation result is corrected by the properties of the admittance matrix and used as the initial value of the convolution kernel.

[0198] This embodiment is to demonstrate the effect of the topological parameter joint identification method proposed by the present invention. Figure 5 and Figure 6 The algorithm performance test was performed on the IEEE-33 node system and the improved IEEE-69 node system shown in the figure. 600 sets of ideal voltage amplitude and power measurement (random error ε = 0) were simulated for the two test systems to verify the effectiveness and scalability of the method. Since the PFGCN network proposed in this paper is different from the traditional GCN, it completes the corresponding tasks during the training process, so there is no need to distinguish between the training set and the validation set. The 600 sets of simulated data are directly used as the training set. The scene settings are as follows:

[0199] C33: IEEE-33 system, 600 measurements, ε=0.

[0200] IC69: Improved IEEE-69 system, 600 measurements, ε=0.

[0201] Among them, the topological information is encoded in the node association matrix, and the topological accuracy function is defined as follows:

[0202]

[0203] Where n is the number of nodes; A and E are the candidate topology node association matrix and the reference topology node association matrix respectively. The closer A1 is to 1, the greater the proportion of correctly identified lines.

[0204] Let the parameter identification result be the line admittance g,b. This embodiment uses Mean Absolute Percentage Error (MAPE) as the main evaluation index for evaluating the parameter estimation effect. The smaller its value is, the higher the parameter identification accuracy is.

[0205]

[0206] Where: N b is the number of lines; g i ′,b i ′ is the estimated value of line parameters; g i ,b i is the reference value of line parameters.

[0207] Furthermore, this embodiment supplements the integral absolute error (IAE), mean absolute error (MAE) and coefficient of determination (R 2 ) for a more comprehensive evaluation.

[0208] Among them, the integrated absolute error (IAE):

[0209]

[0210] Mean Absolute Error (MAE):

[0211]

[0212] Coefficient of Determination(R 2 ):

[0213]

[0214] To verify the topology identification effect of the C33 scenario, the data grouping is set to 6, and the combination coefficient of the node correlation matrix K is α1 = 0.6. Data grouping reduces the computational complexity of the matrix K and increases the robustness of the topology identification algorithm. Data grouping is divided according to time sequence, and the number of groups is an empirical value, the same below.

[0215] Figure 7 The calculation results of the node correlation matrix K of 6 groups of data in the C33 scenario are shown. The element values ​​of the correlation matrix K are inversely proportional to the correlation between the corresponding index nodes. The experimental results show that the distribution of the correlation matrix data corresponds to the topology of the IEEE 8033 system to a certain extent. This proves that the node correlation matrix defined in the present invention is effective in describing node correlation.

[0216] Table 1 shows the candidate topology identification results of the minimum distance iteration algorithm in the C33 scenario. The candidate topology set consisting of 6 groups of candidate topologies contains the correct topology, which can be considered that the effect of the candidate topology identification method proposed in the present invention has been proven.

[0217] Table 1: 6 sets of candidate topology results obtained for scenario C33, α1 = 0.6

[0218]

[0219] The candidate topology set of the C33 scenario is used as the topology input, and the parameter identification experiment is carried out. In the parameter estimation model, the Adam optimizer is used in the parameter estimation layer, with a learning rate of lr = 0.01 and an iteration number of epoch = 30. The outer iteration threshold is set to ξ = 0.05. Table 2 shows the candidate topologies and their final loss function F.

[0220] Table 2C33 candidate topology and its final loss function

[0221]

[0222] Figure 8The figure shows the changes in the loss function value during the training process in the C33 scenario, where topology 1 and topology 2 are used as graph inputs of the parameter estimation model. The fluctuating nodes indicate that the convolution kernel of PFGCN violated the properties of the admittance matrix during the training process, and the training was resumed after being processed by the property correction module.

[0223] The parameter estimation model proposed in the present invention is designed based on the original AC power flow equation, so the correct topology and more accurate line parameters tend to obtain smaller loss function values ​​during the training process. Compared with topology 2 with an erroneous line, topology 1 has a smaller loss function value, which is also the basis for topology screening in the joint estimation method proposed in the present invention. Figure 8 The training process shown and the final loss function results in Table 2 demonstrate the effectiveness of screening topologies based on loss function values.

[0224] In the C33 scenario, the initial value of the convolution kernel iteration and the final result are as follows Fig. 9 shown. Fig.10 The theoretical value and final estimation result of line impedance for C33 scenario are shown.

[0225] In the C33 scenario, the initial value of the convolution kernel iteration obtained by regression estimation by ignoring the phase angle has a large error, and the average mape of g is 104.8%, and the average mape of b is 102.3%. After the parameter estimation model of the present invention is iteratively solved, the accuracy of the line parameter estimation value is greatly improved, specifically, the average mape of g is 4.6%, and the average mape of b is 5.7%.

[0226] In summary, in the C33 scenario, the effectiveness of the proposed joint estimation method for topology and line parameters is demonstrated. Based on the C33 scenario, the influence of the combination coefficient α1 of the node correlation matrix K is analyzed. The coefficient α1 is used as a weight coefficient to adjust the amplitude dimension represented by the Euclidean distance and the proportion of the change trend dimension represented by the DDTW distance. The experimental settings are as follows: C33: IEEE-33 system, 600 groups of measurements, ε = 0. Set the data grouping to 6, and set α1 = 0.2, 0.4, 0.6, 0.8.

[0227] Fig.11 The C33 scenario is shown, and the average value of the candidate topology recognition accuracy is shown when α1 = 0.2, 0.4, 0.6, and 0.8. As α1 increases, the topology recognition accuracy increases first and then decreases. When α1 is around 0.6, the topology accuracy reaches a peak. This is also the basis for setting the hyperparameter α1 in the present invention. The experimental results show that the method of combining Euclidean distance and DDTW distance is superior to the single-dimensional evaluation method.

[0228] In summary, the experiment of C33 scene proves the effectiveness of the joint estimation method of topological parameters, and verifies that the correlation evaluation method of comprehensive amplitude and change trend proposed in the present invention is better than that of a single angle.

[0229] In order to verify the scalability of the method proposed in this invention, a test was conducted based on the improved IEEE-69 system. This section verifies the topology identification algorithm in the IC69 scenario. Set the data grouping to 6, and the node correlation matrix combination coefficient α1 = 0.6. Table 3 shows the candidate topology identification results of the IC69 scenario. In the IC69 scenario, the method proposed in this invention successfully identifies the correct candidate topology, and its effectiveness is proven.

[0230] Table 3 Candidate topology identification results for IC69 scenario α1=0.6

[0231]

[0232]

[0233] The parameter identification experiment was conducted in the IC69 scenario. The topological input of the parameter identification link is the IC69 candidate topology set. In the parameter estimation model, the Adam optimizer is used in the parameter estimation layer, with a learning rate of lr = 0.01 and an iteration number of epoch = 30. The iteration threshold ξ = 0.05 is set for the outer iteration.

[0234] Fig.12 It is the initial value and final estimation result of the regression estimation of line parameters in the IC69 scenario. Fig.13 The theoretical value and final estimation result of the line impedance for the IC69 scenario are shown.

[0235] Compared with the C33 scenario, the initial values ​​of the line parameters g and b estimated by regression estimation also have similar errors, with the average mape of g being 102.9% and the average mape of b being 112.5%. In the final estimation results of the line parameter estimation model, the average mape of g is 3.0% and the average mape of b is 5.9%.

[0236] In summary, the experiments based on the improved IEEE69 system have demonstrated the scalability of the method proposed in this invention and the transferability of the PF-GCN parameter identification model.

[0237] In order to further analyze the performance of the method proposed in the present invention, this embodiment conducts an algorithm comparison experiment. The above has verified that the topological identification method of the comprehensive data amplitude and change trend proposed in the present invention has higher identification accuracy than a single angle. This section mainly compares different parameter identification methods. In recent studies, there are technologies that also use a cross-iteration estimation method and apply an adaptive ridge regression method in parameter estimation, which is better than traditional linear regression. Therefore, it is appropriate to select the adaptive ridge regression (ARR) method of this technology as the control group of the method proposed in the present invention. We conduct a comparative experiment of algorithm performance based on the C33 and IC69 scenarios, and fix the correct topology to simplify the experimental process. PFGCN method: The parameter estimation layer uses the Adam optimizer, the learning rate lr = 0.01, the number of iterations epoch = 30, and the iteration threshold ξ = 0.05. Adaptive ridge regression method: set the regularization coefficient κ = 0.003, the adaptive coefficient δ = 5e-5, γ = 2, and the iteration threshold ξ = 0.05. Fig.14 and Fig.15 The parameter estimation effects of the PFGCN method and the adaptive ridge regression method in the C33 and IC69 scenarios are shown respectively. Table 4 shows the comparison of various evaluation index results of the PFGCN method and the adaptive ridge regression (ARR) method, including MAPE, IAE, MAE and R 2 .

[0238] Table 4 Performance comparison of PFGCN algorithm and ARR algorithm in C33 and IC69 scenarios

[0239]

[0240]

[0241] In the same scenario, the PFGCN method achieves better results than the ARR algorithm in various indicators, and the line parameter estimation accuracy of the PFGCN method is higher.

[0242] Specifically, in the C33 scenario, the PFGCN method achieves smaller MAPE values, which are 4.6% for MAPE of g and 5.7% for MAPE of b, respectively, which are lower than the ARR algorithm's MAPE of g (7.1%) and MAPE of b (7.9%). The IAE and MAE indicators of the PFGCN method are lower than those of the ARR method. g =2.605, lower than the IAE of ARR g =5.053; IAE of PFGCN b =3.748, lower than the IAE of ARR g =5.223. Among the MAE indicators, the line admittance estimated by PFGCN has better accuracy.2 Among the indicators, the R 2 g =0.999, R 2 b =0.996, both are better than the ARR method.

[0243] In the IC69 scenario, the PFGCN method achieves smaller MAPE values, which are 3.0% for MAPE of g and 5.9% for MAPE of b, respectively, which are lower than the ARR algorithm’s MAPE of g (5.1%) and MAPE of b (8.3%). The IAE and MAE indicators of the PFGCN method are lower than those of the ARR method, while R 2 Among the indicators, the PFGCN method is still better than the ARR method.

[0244] The experimental results of the C33 and IC69 scenarios prove that the PFGCN-based parameter estimation model proposed in the present invention has better line parameter identification accuracy.

[0245] Considering that it is difficult to obtain ideal measurement data (ε=0) in an actual distribution network, this section conducts an error sensitivity experiment to analyze the robustness of the joint estimation method proposed in the present invention.

[0246] Referring to the accuracy level of the actual distribution network data acquisition device, we set two random error levels of ε = 0.1% and ε = 0.2% for the voltage amplitude and power measurement of the IEEE-33 system and the improved IEEE-69 system, simulated 600 sets of measurement data for each scenario, and conducted error sensitivity experiments. The scenario settings are as follows:

[0247] C33_0.1: IEEE-33 system, 600 sets of measurements, ε = 0.1%; C33_0.2: IEEE-33 system, 600 sets of measurements, ε = 0.2%; IC69_0.1: Improved IEEE-69 system, 600 sets of measurements, ε = 0.1%; IC69_0.2: Improved IEEE-69 system, 600 sets of measurements, ε = 0.2%.

[0248] The data grouping is set to 6, and the node correlation matrix combination coefficient α1 = 0.6. The topological identification results of six scenarios, C33, C33_0.1, C33_0.2, IC69, IC69_0.1, and IC69_0.2, are obtained by repeated experiments. Fig.11As shown. The increase in the error level will cause the topology recognition accuracy of the IEEE-33 system and the improved IEEE-69 system to show a slow downward trend. Although random errors will endanger the average accuracy of topology identification, due to the grouping strategy, when ε=0.1% and 0.2%, the candidate topology sets of the IEEE-33 system and the improved IEEE-69 system still contain the correct topology. Experiments show that the topology identification method proposed in the present invention is robust and can still obtain valid topology in scenarios with certain errors.

[0249] Based on six scenarios, namely, C33, C33_0.1, C33_0.2, IC69, IC69_0.1, and IC69_0.2, the influence of the error on the parameter estimation model proposed in the present invention is analyzed. The candidate topology identification results of different scenarios are used as the graph input of the line parameter estimation model of the corresponding scenario.

[0250] IEEE-33 system: In the parameter estimation model, the Adam optimizer is used in the parameter estimation layer, with a learning rate of lr = 0.01 and an iteration number of epoch = 30. The outer iteration sets the iteration threshold ξ = 0.05. Table 5 shows the accuracy of the final line parameter estimation under different error levels of the IEEE-33 system.

[0251] Table 5 Average MAPE values ​​of IEEE-33 system parameter estimation

[0252]

[0253] The experimental results show that the error will harm the parameter identification effect in general. g and mape b There are different degrees of growth, indicating that the accuracy of line parameter estimation has decreased. g and mape b All remain within the 9% range.

[0254] Improved IEEE-69 system: In the parameter estimation model, the Adam optimizer is used in the parameter estimation layer, with a learning rate of lr = 0.01 and an iteration number of epoch = 30. The iteration threshold ξ = 0.05 is set for the outer iteration. Table 6 shows the accuracy of the final line parameter estimation under different error levels of the improved IEEE-69 system.

[0255] Table 6 Average MAPE values ​​of improved IEEE-69 system parameter estimation

[0256]

[0257]

[0258] Similar to the IEEE-33 system, as the error grows, mape g and mape b All of them show an upward trend, which indicates that the line parameter estimation accuracy shows a downward trend. The error will harm the parameter identification effect to a certain extent, but all mape g and mape b Both are less than 10%.

[0259] In summary, the error experiment proves that the joint identification method proposed in the present invention has good robustness, and still shows high identification accuracy and robustness when there are certain random errors in the measurement data.

[0260] Example 2

[0261] This embodiment 2 provides a topology parameter joint identification system based on a power flow graph convolutional network, including:

[0262] A node correlation module, used to obtain power and voltage amplitude measurement data, and construct a node correlation matrix using the power and voltage amplitude measurement data;

[0263] A candidate topology module is used to iteratively calculate the node correlation matrix using a minimum distance iteration algorithm to generate a candidate topology set;

[0264] The parameter estimation model module is used to establish a line parameter estimation model based on a power flow graph convolutional network, input the candidate topology set into the line parameter estimation model for screening, and obtain the final topology structure.

[0265] Example 3

[0266] Fig.17 A schematic diagram of the structure of an electronic device provided in Embodiment 3 of the present invention is shown in FIG. Fig.17 As shown, the electronic device includes a processor 21, a memory 22, an input device 23 and an output device 24; the number of processors 21 in the computer device can be one or more. Fig.17 A processor 21 is taken as an example; the processor 21, the memory 22, the input device 23 and the output device 24 in the electronic device can be connected by a bus or other means. Fig.17 The example of connecting through bus is taken in the following.

[0267] The memory 22 is a computer-readable storage medium that can be used to store software programs, computer executable programs, and modules. The processor 21 executes various functional applications and data processing of the electronic device by running the software programs, instructions, and modules stored in the memory 22, that is, the topological parameter joint identification method based on the power flow graph convolution network of Example 1 is implemented.

[0268] The memory 22 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system and at least one application required for a function; the data storage area may store data created according to the use of the terminal, etc. In addition, the memory 22 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some instances, the memory 22 may further include a memory remotely arranged relative to the processor 21, and these remote memories may be connected to the electronic device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0269] The input device 23 can be used to receive the ID and password input by the user, etc. The output device 24 is used to output the network configuration page.

[0270] Example 4

[0271] Embodiment 4 of the present invention further provides a computer-readable storage medium, wherein the computer-executable instructions, when executed by a computer processor, are used to implement the method for joint identification of topological parameters based on a power flow graph convolutional network as provided in Embodiment 1.

[0272] An embodiment of the present invention provides a storage medium containing computer executable instructions, and its computer executable instructions are not limited to the method operations provided in Example 1, and can also execute related operations in the topological parameter joint identification method based on the flow graph convolutional network provided in any embodiment of the present invention.

[0273] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A topological parameter joint identification method based on a power flow graph convolutional network, characterized in that: The steps include: Acquiring power and voltage amplitude measurement data, and constructing a node correlation matrix using the power and voltage amplitude measurement data; Iteratively calculating the node correlation matrix using a minimum distance iteration algorithm to generate a candidate topology set; A line parameter estimation model based on a power flow graph convolutional network is established, and the candidate topology set is input into the line parameter estimation model for screening to obtain a final topology structure.

2. The method for joint identification of topological parameters based on power flow graph convolutional network according to claim 1 is characterized in that: Constructing a node correlation matrix using the power and voltage amplitude measurement data includes the following steps: Extracting any two sets of node voltage data from the power and voltage amplitude measurement data; Calculating the Euclidean distance and the DDTW distance between the two groups of node voltage data, and establishing a Euclidean distance matrix and a DDTW distance matrix according to the Euclidean distance and the DDTW distance; The Euclidean distance matrix and the DDTW distance matrix are normalized, and a node correlation matrix is ​​constructed using the normalized Euclidean distance matrix and the DDTW distance matrix.

3. The method for joint identification of topological parameters based on power flow graph convolutional network according to claim 2 is characterized in that: The node correlation matrix is ​​iteratively calculated using a minimum distance iteration algorithm, comprising the following steps: Initialize the node pool, line pool and node adjacency matrix; Arrange the node correlation matrix in ascending order to obtain an index matrix; Traversing the index matrix, checking the connection conditions between nodes, if node i and node j meet the connection conditions, establishing a connection between the node i and the node j, and updating the line pool; Checking a minimum distance condition, establishing a connection between corresponding nodes that satisfy the minimum distance condition, and updating the line pool; Removing nodes with established connections from the initialization node pool based on the updated line pool; When the initialization node pool after removing the nodes with established connections is not empty, traverse the initialization node pool, establish connections between the remaining nodes in the initialization node pool and the corresponding nodes in the index matrix, and update the line pool; The node adjacency matrix is ​​updated using the updated line pool to obtain a candidate topology set.

4. The method for joint identification of topological parameters based on power flow graph convolutional network according to claim 1 is characterized in that: Establishing a line parameter estimation model based on a power flow graph convolutional network, and inputting the candidate topology set into the line parameter estimation model for screening, including: Constructing an AC power flow equation, mapping the AC power flow equation into a complex-valued graph convolution framework, and generating an AC power flow equation in a complex-valued convolution form; A phase angle estimation layer for pseudo power flow estimation is established, and a parameter estimation layer is established based on the AC power flow equation in complex valued convolution form; Using the phase angle estimation layer to perform pseudo power flow estimation on the candidate topology set to obtain phase angle data; Acquire topological power voltage amplitude data, perform regression estimation on the topological power voltage amplitude data, and obtain a regression estimation initial value; Inputting the regression estimation initial value and the phase angle data into the parameter estimation layer for iterative calculation to obtain estimated values ​​of conductance and susceptance matrices, and performing admittance matrix property correction on the estimated values ​​of conductance and susceptance matrices; The estimated values ​​of the conductance and susceptance matrices after the properties of the admittance matrix are corrected are input into the phase angle estimation layer for the next iteration until a stop iteration condition is met.

5. The method for joint identification of topological parameters based on power flow graph convolutional network according to claim 4 is characterized in that: The parameter estimation layer is composed of a power flow graph convolution network, wherein the neurons of the power flow graph convolution network and the connection relationship of the neurons correspond one-to-one to the calculation process of the AC power flow equation based on the complex-valued convolution form.

6. The method for joint identification of topological parameters based on power flow graph convolutional network according to claim 4 is characterized in that: Mapping the AC power flow equation into a complex-valued graph convolution framework includes: Establishing a distribution network topology model and preprocessing the distribution network topology model; defining an intermediate variable function, bringing the intermediate variable function and the preprocessed distribution network topology model into the AC power flow equation to obtain an AC power flow equation in a complex-valued convolution form; Among them, establishing the distribution network topology model includes: encoding the distribution network topology into a node adjacency matrix, expressing the voltage amplitude and phase angle with a diagonal matrix, and expressing the line parameters in complex form of an admittance matrix.

7. The method for joint identification of topological parameters based on power flow graph convolutional network according to claim 6 is characterized in that: Regression estimation is performed on the topology power voltage amplitude data, including: The topological power voltage amplitude data is calculated using an AC power flow equation to obtain an active power injection-voltage amplitude ratio and a reactive power injection-voltage amplitude ratio; The active power injection-voltage amplitude ratio and the reactive power injection-voltage amplitude ratio are estimated and calculated to obtain conductance estimation and susceptance matrix estimation.

8. A topological parameter joint identification system based on a power flow graph convolutional network, characterized in that: include: A node correlation module, used to obtain power and voltage amplitude measurement data, and construct a node correlation matrix using the power and voltage amplitude measurement data; A candidate topology module is used to iteratively calculate the node correlation matrix using a minimum distance iteration algorithm to generate a candidate topology set; The parameter estimation model module is used to establish a line parameter estimation model based on a power flow graph convolutional network, input the candidate topology set into the line parameter estimation model for screening, and obtain the final topology structure.

9. An electronic device, characterized in that: It includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the method for joint identification of topological parameters based on a tidal graph convolutional network as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for joint identification of topological parameters based on a power flow graph convolutional network as described in any one of claims 1 to 7 is implemented.

Images