Operation mode power flow coupling relation characterization method based on N / N-1 power flow embedded graph convolutional neural network

Abstract

The invention discloses an operation mode power flow coupling relation characterization method based on an N / N-1 power flow embedded graph convolutional neural network, and the method comprises the following steps: 1) proposing a graph convolution forward propagation strategy based on N / N-1 power flow embedding, and deducing and designing a graph convolution module forward propagation expression through an N / N-1 power flow physical model; 2) constructing a neural network architecture based on multi-layer graph convolution and convolution module cooperation; and 3) training to obtain an operation mode power flow coupling relation model, and representing the power flow coupling relation of the power system in the N / N-1 state. The operation mode power flow coupling relation model constructed by the method can represent the power flow coupling relation in the N / N-1 state with high precision, so that key problems such as fault diagnosis, stability evaluation and optimization control in a power system can be analyzed subsequently.

Description

Technical Field

[0001] The present invention relates to the field of power systems and automation thereof, and in particular to a method for characterizing the coupling relationship of operating modes and power flows based on an N / N-1 power flow embedded graph convolutional neural network. Background Art

[0002] With the continuous expansion of the scale of power systems and the increase in the proportion of renewable energy access, the complexity of power systems and the uncertainty of operation have increased significantly. In the operation of power systems, power flow calculation is a basic tool for analyzing and controlling the state of the power grid, especially under the N-1 verification condition, the system needs to be able to continue to operate normally when a single component fails. Therefore, how to efficiently and accurately characterize and analyze the power flow coupling relationship of the power system under the N / N-1 state has become an important topic in power system research.

[0003] Traditional power flow calculation methods rely on physical models and numerical calculations. These methods are usually based on the power system equations under steady-state conditions and obtain the power flow distribution of the system through iterative solutions. Although these methods can provide a certain degree of accuracy in theory, their calculation complexity is high and the speed is slow. Especially when facing large-scale power grids, it is difficult to meet the real-time requirements. Moreover, with the increase in the scale and complexity of power systems, traditional methods are unable to cope with various operating states and fault conditions. Especially in the case of N-1 verification, traditional methods need to characterize the power flow coupling relationship under each N-1 state one by one, and the amount of calculation is extremely large, which is difficult to complete in a short time.

[0004] In recent years, data-driven methods have shown great potential in power system analysis. These methods use big data and machine learning techniques to quickly and accurately model and predict power systems by training historical data of power system operation. Data-driven methods can extract complex nonlinear relationships from large amounts of data and provide power flow calculation results in a short time through trained models. It has a powerful ability to approximate high-dimensional complex nonlinear features, can capture the complex nonlinear relationships existing in the power system, directly map power flow inputs and outputs, and does not require iterative calculations, thereby significantly improving the calculation speed. However, the current data-driven methods still have some shortcomings in application: First, existing methods usually fail to fully consider the physical rules satisfied by the graph node features and the physical meaning of the graph nodes, resulting in the loss of physical coupling relationships; second, it is difficult to accurately consider the power flow coupling relationships under multiple discrete N-1 states at the same time. Summary of the invention

[0005] The object of the present invention is to provide a method for characterizing the coupling relationship of operating modes and power flows based on an N / N-1 power flow embedded graph convolutional neural network, comprising the following steps:

[0006] 1) Based on the graph convolution forward propagation strategy embedded in the N / N-1 power flow, the forward propagation expression of the graph convolution module is derived and designed using the N / N-1 power flow physical model;

[0007] 2) Construct an N / N-1 power flow coupling relationship model based on the collaborative operation of multi-layer graph convolution and convolution modules;

[0008] 3) Training operation mode N / N-1 power flow coupling relationship model;

[0009] 4) Use the N / N-1 power flow coupling relationship model of operation mode to obtain the power flow coupling relationship.

[0010] Further, the graph convolution module includes an N-node feature aggregation graph convolution submodule, an N / N-1 node feature mapping graph convolution submodule, and an N / N-1 node-branch feature conversion graph convolution submodule;

[0011] The output of the N-node feature aggregation graph convolution submodule is the input of the N / N-1-node feature mapping graph convolution submodule;

[0012] The outputs of the N-node feature aggregation graph convolution submodule and the N / N-1 node feature mapping graph convolution submodule are simultaneously used as the inputs of the N / N-1 node-branch feature conversion graph convolution submodule.

[0013] Furthermore, the input feature X of the N-node feature aggregation graph convolution submodule is as follows:

[0014] X={U in ,θ in ,P in ,Q in} (1)

[0015] In the formula, U i is the voltage amplitude of the i-th node; They represent the voltage amplitude eigenvector, voltage phase angle eigenvector, injected active power eigenvector and injected reactive power eigenvector of the power flow node under the N state respectively; n is the number of nodes in the power system;

[0016] Furthermore, the forward propagation expression of the graph convolution module is as follows:

[0017]

[0018] Where σ is the activation function; P i , Q i are the injected active power and injected reactive power of the ith node respectively; G ij , B ij Respectively represent the elements of the i-th row and j-th column in the conductance matrix G and the susceptance matrix B; θ ijrepresents the voltage phase angle difference from node i to node j; N(i) represents the adjacent set of node i; l represents the number of graph convolution layers, l = 1, 2, ..., m, m is the total number of graph convolution layers in graph convolution submodule 1; unit matrix F l The dimension of the output feature of the lth layer; are the voltage amplitude characteristics and phase angle characteristics of the N states output by the lth graph convolution layer. i , U j is the voltage amplitude of the i-th and j-th nodes. is the weight matrix of the lth layer.

[0019] Furthermore, the output feature Y1 of the N-node feature aggregation graph convolution submodule is as follows:

[0020] Y1={U Nm ,θ Nm} (5)

[0021] Where U Nm ,θ Nm It is the voltage amplitude characteristic and phase angle characteristic in N state.

[0022] Furthermore, the output feature Y2 of the N / N-1 node feature map graph convolution submodule is as follows:

[0023] Y2={U N-1,k ,θ N-1,k} (6)

[0024] Where k = 1, 2, ..., K, K is the number of N-1 states considered. N-1,k ,θ N-1,k It is the voltage amplitude characteristic and phase angle characteristic in N-1 state.

[0025] Furthermore, the output Y3 of the N / N-1 node-branch feature conversion graph convolution submodule is as follows:

[0026] Y3={P N ,Q N ,P N-1,k ,Q N-1,k} (7)

[0027] In the formula, They respectively represent the branch active power characteristic matrix and reactive power characteristic matrix in the N state, and the branch active power characteristic matrix and reactive power characteristic matrix in the N-1 state.

[0028] Further, in step 2), the step of constructing an N-node-branch characteristic transfer expression in the N state and an N-1-node-branch characteristic transfer expression in the N-1 state includes:

[0029] 2.1) Construct the power flow equation, namely:

[0030]

[0031] 2.2) P i , Q i , G ij , B ij , U i ,θ i They are respectively constructed into corresponding vector or matrix forms, thereby converting equation (8) into matrix form and obtaining:

[0032]

[0033] In the formula, θ=[θ1,θ1,…,θ n ] T ; P = [P1, P2, ... P n ] T ; Q=[Q1,Q2,…Q n ] T ; D(*) extracts the diagonal elements in * and constructs a diagonal matrix;

[0034] 2.3) Matrix merging is performed on equation (9) to obtain the relationship between the node injection power P, Q and the voltage U, θ in the N state, namely:

[0035]

[0036] Among them, the voltage-power conversion coefficient matrix X in the N state is as follows:

[0037]

[0038] 2.4) According to formula (10), the node injection power P, Q and voltage U in the N-1 state are obtained N-1,k ,θ N-1,k The relationship between them is:

[0039]

[0040] In the formula, * N-1,k Indicates the * corresponding to the N state when the kth branch is disconnected; X N-1,k is the voltage-power conversion coefficient matrix in the N-1 state;

[0041] 2.5) Based on formulas (10) and (12), construct the N / N-1 node feature mapping expression, namely:

[0042]

[0043] 2.6) Build The forward propagation formula of the convolution submodule of the N / N-1 node feature map of the convolution kernel is:

[0044]

[0045] 2.7) Establish the branch power equation, that is:

[0046]

[0047] In the formula, PF ij , QF ij are the active power and reactive power of branch ij respectively.

[0048] 2.8) Establish the branch-node association matrix A a =[a (ij)i ] b×n , where A a Each element of is defined as follows:

[0049]

[0050] 2.9) PF ij , QF ij , G ij , B ij , U i ,θ i They are constructed into vector form respectively, using the branch-node association matrix A a , convert equation (16) into matrix form, that is:

[0051]

[0052] Where, PF = [PF ij ] b×1 ; QF=[QF ij ] b×1 ; G br =[G ij ] b×1 ; B br =[B ij ] b×1 ; b is the total number of branches; diag(*) is a diagonal matrix constructed with * as the diagonal element.

[0053] 2.10) Matrix merge equation (18) to obtain the node-branch feature transfer expression of the N / N-1 node-branch feature conversion graph convolution submodule, that is:

[0054]

[0055] Among them, the coefficient matrix ω is as follows:

[0056]

[0057] 2.11) Construct the forward propagation equation of the N / N-1 node-branch feature conversion graph convolution submodule, that is:

[0058]

[0059] In the formula, ω N ,ω N-1,k are the corresponding ω in the N state and the kth N-1 state respectively.

[0060] Further, the operation mode N / N-1 power flow coupling relationship model includes an N-node feature aggregation graph convolution submodule, an N / N-1 node feature mapping graph convolution submodule, an N / N-1 node-branch feature conversion graph convolution submodule, an N / N-1 node feature prediction convolution submodule, and an N / N-1 branch feature prediction convolution submodule;

[0061] The N / N-1 node feature prediction convolution submodule is used to predict node voltage features in N states and K N-1 states;

[0062] The input features of the N / N-1 node feature prediction convolution submodule include the output of the N node feature aggregation graph convolution submodule Y1 = {U Nm ,θ Nm}, and the output Y2 of the N / N-1 node feature map convolution submodule = {U N-1,k ,θ N-1,k};

[0063] The convolution forward propagation formula of the N / N-1 node feature prediction convolution submodule is as follows:

[0064]

[0065] Where l = 1, 2, ..., c1, c1 is the total number of convolutional layers in the convolutional submodule 1; U l ,θ l They are the voltage amplitude feature and phase angle feature of the input of the lth convolutional layer respectively; are the convolution kernel and bias of the lth layer of the convolution layer respectively; when l = 1, When l>1, g l is the number of convolution kernel channels of the lth layer;

[0066] The output of the N / N-1 node feature prediction convolution submodule is as follows:

[0067] When l = 1, U 1 =[U Nm,U N-1,k ],θ 1 =[θ Nm ,θ N-1,k ], k = 1, 2, ..., K, and forward propagation is performed layer by layer from the 1st layer to the c1th layer through the forward propagation formula (23), and the output feature is finally obtained: the node voltage amplitude feature vector under the N state Phase angle eigenvector And K node voltage amplitude characteristic vectors under N-1 states Phase angle eigenvector

[0068] The output features are represented as follows:

[0069]

[0070] In the formula, is the node voltage amplitude characteristic vector and phase angle characteristic vector in the N state; is the node voltage amplitude characteristic vector and phase angle characteristic vector in the N-1 state; U o ,θ o is the set of output voltage amplitude characteristics and phase angle characteristics;

[0071] The N / N-1 branch feature prediction convolution submodule is used to predict the branch power features in N states and K N-1 states;

[0072] The input features of the N / N-1 branch feature prediction convolution submodule are the N / N-1 node-branch feature conversion graph and the output matrix of the convolution submodule is Y3 = {PF N ,QF N ,PF N-1,k ,QF N-1,k};

[0073] The convolution forward propagation formula of the N / N-1 branch feature prediction convolution submodule is as follows:

[0074]

[0075] Where l = 1, 2, ..., c2, c2 is the total number of convolutional layers in the convolutional submodule 2; PF l , QF l They are the branch active power features and branch reactive power features input to the lth convolutional layer respectively; are the convolution kernel and bias of the lth convolution layer, which need to be obtained through training. When l = 1, When l>1, r l is the number of convolution kernel channels of the lth layer;

[0076] The output of the N / N-1 branch feature prediction convolution submodule is as follows:

[0077] When l = 1, PF 1 =[PF N ,PF N-1,k ], QF 1 =[QF N ,QF N-1,k ], k = 1, 2, ..., K, according to the forward propagation formula (25), forward propagation is performed layer by layer from the 1st layer to the c2th layer, and the output feature is finally obtained: the branch active power feature vector under the N state Branch reactive power characteristic vector And the branch active power characteristic vector in the N-1 state Branch reactive power characteristic vector

[0078] The output features are represented as follows:

[0079]

[0080] In the formula, is the branch active power characteristic vector and branch reactive power characteristic vector in the N state; is the branch active power characteristic vector and branch reactive power characteristic vector in the N-1 state; PF o and QF o Represents the set of branch active power characteristics and branch reactive power characteristics.

[0081] Furthermore, the input of the N / N-1 power flow coupling relationship model in the N-state is the characteristic vector X={U in ,θ in ,P in ,Q in}, the output is the feature vector Y in the N / N-1 state o = {U o ,θ o ,θF o ,QF o}.

[0082] The technical effect of the present invention is undoubted, and the beneficial effects of the present invention are as follows:

[0083] The present invention proposes a graph convolution forward propagation strategy based on N / N-1 power flow embedding, derives and designs the forward propagation expression of the graph convolution module based on the N / N-1 power flow physical model, and guides the forward propagation of the neural network through the forward propagation expression that implies the physical law of power flow, so that it can effectively extract the node characteristics and branch characteristics of the power system under the N / N-1 state.

[0084] The present invention proposes an N / N-1 power flow coupling relationship model based on the collaboration of multi-layer graph convolution and convolutional neural network modules. The power flow characteristics under the N / N-1 state of the power system are used as input / output characteristics, and a neural network architecture with collaborative multi-layer graph convolution and convolution modules is designed. The constructed operating mode power flow coupling relationship model can characterize the power flow coupling relationship under the N / N-1 state with high precision.

[0085] The present invention can be widely used in fault diagnosis, stability assessment and optimization control analysis in the power system industry.

[0086] The operation mode flow coupling relationship model constructed by the present invention can characterize the flow coupling relationship under the N / N-1 state with high precision, so as to facilitate the subsequent analysis of key issues such as fault diagnosis, stability assessment and optimization control in the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] Figure 1 It is a graph convolution forward propagation module framework based on N / N-1 power flow embedding;

[0088] Figure 2 It is a neural network architecture that combines multi-layer graph convolution and convolution modules. DETAILED DESCRIPTION

[0089] The present invention is further described below in conjunction with the embodiments, but it should not be understood that the above subject matter of the present invention is limited to the following embodiments. Without departing from the above technical ideas of the present invention, various substitutions and changes are made according to the common technical knowledge and customary means in the art, which should all be included in the protection scope of the present invention.

[0090] Embodiment 1:

[0091] See also Figure 1 to Figure 2 , a method for characterizing the coupling relationship of operation modes and power flows based on an N / N-1 power flow embedded graph convolutional neural network, comprising the following steps:

[0092] 1) Based on the graph convolution forward propagation strategy embedded in the N / N-1 power flow, the forward propagation expression of the graph convolution module is derived and designed using the N / N-1 power flow physical model;

[0093] 2) Construct an N / N-1 power flow coupling relationship model based on the collaborative operation of multi-layer graph convolution and convolution modules;

[0094] 3) Train the N / N-1 power flow coupling relationship model of operation mode to obtain the power flow coupling relationship.

[0095] Embodiment 2:

[0096] A method for characterizing the power flow coupling relationship of an operation mode based on an N / N-1 power flow embedded graph convolutional neural network, the technical content of which is the same as that of Example 1, and further, the graph convolution module includes an N-node feature aggregation graph convolution submodule, an N / N-1 node feature mapping graph convolution submodule, and an N / N-1 node-branch feature conversion graph convolution submodule;

[0097] The output of the N-node feature aggregation graph convolution submodule is the input of the N / N-1-node feature mapping graph convolution submodule;

[0098] The outputs of the N-node feature aggregation graph convolution submodule and the N / N-1 node feature mapping graph convolution submodule are simultaneously used as the inputs of the N / N-1 node-branch feature conversion graph convolution submodule.

[0099] Embodiment 3:

[0100] A method for characterizing the power flow coupling relationship of an operation mode based on an N / N-1 power flow embedded graph convolutional neural network, the technical content of which is the same as any one of Embodiments 1-2, and further, the input feature X of the N-node feature aggregation graph convolution submodule is as follows:

[0101] X={U in ,θ in ,P in ,Q in} (1)

[0102] In the formula, U i is the voltage amplitude of the i-th node; They represent the voltage amplitude eigenvector, voltage phase angle eigenvector, injected active power eigenvector and injected reactive power eigenvector of the power flow node under the N state respectively; n is the number of nodes in the power system;

[0103] Embodiment 4:

[0104] A method for characterizing the power flow coupling relationship of an operation mode based on an N / N-1 power flow embedded graph convolutional neural network, the technical content of which is the same as any one of Embodiments 1-3, and further, the forward propagation expression of the graph convolution module is as follows:

[0105]

[0106] Where σ is the activation function; P i , Q i are the injected active power and injected reactive power of the ith node respectively; G oj , B ij Respectively represent the elements of the i-th row and j-th column in the conductance matrix G and the susceptance matrix B; θ ijrepresents the voltage phase angle difference from node i to node j; N(i) represents the adjacent set of node i; l represents the number of graph convolution layers, l = 1, 2, ..., m, m is the total number of graph convolution layers in graph convolution submodule 1; unit matrix F l The dimension of the output feature of the lth layer; are the voltage amplitude characteristics and phase angle characteristics of the N states output by the lth graph convolution layer. i , U j is the voltage amplitude of the i-th and j-th nodes. is the weight matrix of the lth layer.

[0107] Embodiment 5:

[0108] A method for characterizing the power flow coupling relationship of an operation mode based on an N / N-1 power flow embedded graph convolutional neural network, the technical content of which is the same as any one of Embodiments 1-4, and further, the output feature Y1 of the N-node feature aggregation graph convolution submodule is as follows:

[0109] Y1={U Nm ,θ Nm} (5)

[0110] Where U Nm ,θ Nm It is the voltage amplitude characteristic and phase angle characteristic in N state.

[0111] Embodiment 6:

[0112] A method for characterizing the power flow coupling relationship of an operation mode based on an N / N-1 power flow embedded graph convolutional neural network, the technical content of which is the same as any one of Embodiments 1-5, and further, the output feature Y2 of the N / N-1 node feature map convolution submodule is as follows:

[0113] Y2={U N-1,k ,θ N-1,k} (6)

[0114] Where k = 1, 2, ..., K, K is the number of N-1 states considered. N-1,k ,θ N-1,k It is the voltage amplitude characteristic and phase angle characteristic in N-1 state.

[0115] Embodiment 7:

[0116] A method for characterizing the power flow coupling relationship of an operation mode based on an N / N-1 power flow embedded graph convolutional neural network, the technical content of which is the same as any one of Embodiments 1-6, and further, the output Y3 of the N / N-1 node-branch feature conversion graph convolution submodule is as follows:

[0117] Y3={P N ,QN ,P N-1,k ,Q N-1,k} (7)

[0118] In the formula, They respectively represent the branch active power characteristic matrix and reactive power characteristic matrix in the N state, and the branch active power characteristic matrix and reactive power characteristic matrix in the N-1 state.

[0119] Embodiment 8:

[0120] A method for characterizing the power flow coupling relationship of an operation mode based on an N / N-1 power flow embedded graph convolutional neural network, the technical content of which is the same as any one of embodiments 1-7, further, in step 2), the step of constructing an N node-branch feature transfer expression in the N state and an N-1 node-branch feature transfer expression in the N-1 state comprises:

[0121] 2.1) Construct the power flow equation, namely:

[0122]

[0123] 2.2) P i ,q i , G ij , B ij , U i ,θ i They are respectively constructed into corresponding vector or matrix forms, thereby converting equation (8) into matrix form and obtaining:

[0124]

[0125] In the formula, θ=[θ1,θ1,…,θ n ] T ; P = [P1, P2, ... P n ] T ; Q=[Q1,Q2,…Q n ] T ; D(*) extracts the diagonal elements in * and constructs a diagonal matrix;

[0126] 2.3) Matrix merging is performed on equation (9) to obtain the relationship between the node injection power P, Q and the voltage U, θ in the N state, namely:

[0127]

[0128] Among them, the voltage-power conversion coefficient matrix X in the N state is as follows:

[0129]

[0130] 2.4) According to formula (10), the node injection power P, Q and voltage U in the N-1 state are obtained N-1,k ,θ N-1,k The relationship between them is:

[0131]

[0132] In the formula, * N-1,k represents the * corresponding to the N state when the kth branch is disconnected; N-1,k is the voltage-power conversion coefficient matrix in the N-1 state;

[0133] 2.5) Based on formulas (10) and (12), construct the N / N-1 node feature mapping expression, namely:

[0134]

[0135] 2.6) Build The forward propagation formula of the convolution submodule of the N / N-1 node feature map of the convolution kernel is:

[0136]

[0137] 2.7) Establish the branch power equation, that is:

[0138]

[0139] In the formula, PF ij , QF ij are the active power and reactive power of branch ij respectively.

[0140] 2.8) Establish the branch-node association matrix A a =[a (ij)i ] b×n , where A a Each element of is defined as follows:

[0141]

[0142] 2.9) PF ij , QF ij , G ij , B ij , U i ,θ i They are constructed into vector form respectively, using the branch-node association matrix A a , convert equation (16) into matrix form, that is:

[0143]

[0144] Where, PF = [PF ij ]b×1 ; QF=[QF ij ] b×1 ; G br =[G ij ] b×1 ; B br =[B ij ] b×1 ; b is the total number of branches; diag(*) is a diagonal matrix constructed with * as the diagonal element.

[0145] 2.10) Matrix merge equation (18) to obtain the node-branch feature transfer expression of the N / N-1 node-branch feature conversion graph convolution submodule, that is:

[0146]

[0147] Among them, the coefficient matrix ω is as follows:

[0148]

[0149] 2.11) Construct the forward propagation equation of the N / N-1 node-branch feature conversion graph convolution submodule, that is:

[0150]

[0151] In the formula, ω N ,ω N-1,k are the corresponding ω in the N state and the kth N-1 state respectively.

[0152] Embodiment 9:

[0153] A method for characterizing the coupled relationship of power flow in an operation mode based on an N / N-1 power flow embedded graph convolutional neural network, the technical content of which is the same as any one of Embodiments 1-8, and further, the coupled relationship model of the N / N-1 power flow in an operation mode includes an N-node feature aggregation graph convolution submodule, an N / N-1 node feature mapping graph convolution submodule, an N / N-1 node-branch feature conversion graph convolution submodule, an N / N-1 node feature prediction convolution submodule, and an N / N-1 branch feature prediction convolution submodule;

[0154] The N / N-1 node feature prediction convolution submodule is used to predict node voltage features in N states and K N-1 states;

[0155] The input features of the N / N-1 node feature prediction convolution submodule include the output of the N node feature aggregation graph convolution submodule Y1 = {U Nm ,θ Nm}, and the output Y2 of the N / N-1 node feature map convolution submodule = {U N-1,k ,θ N-1,k};

[0156] The convolution forward propagation formula of the N / N-1 node feature prediction convolution submodule is as follows:

[0157]

[0158] Where l = 1, 2, ..., c1, c1 is the total number of convolutional layers in the convolutional submodule 1; U l ,θ l They are the voltage amplitude feature and phase angle feature of the input of the lth convolutional layer respectively; are the convolution kernel and bias of the lth layer of the convolution layer respectively; when l = 1, When l>1, g l is the number of convolution kernel channels of the lth layer;

[0159] The output of the N / N-1 node feature prediction convolution submodule is as follows:

[0160] When l = 1, U 1 =[U Nm ,U N-1,k ],θ 1 =[θ Nm ,θ N-1,k ], k = 1, 2, ..., K, and forward propagation is performed layer by layer from the 1st layer to the c1th layer through the forward propagation formula (23), and the output feature is finally obtained: the node voltage amplitude feature vector under the N state Phase angle eigenvector And K node voltage amplitude characteristic vectors under N-1 states Phase angle eigenvector Its available U o and θ o It is expressed as follows:

[0161]

[0162] In the formula, is the node voltage amplitude characteristic vector and phase angle characteristic vector in the N state; are the node voltage amplitude characteristic vector and phase angle characteristic vector in the N-1 state;

[0163] The N / N-1 branch feature prediction convolution submodule is used to predict the branch power features in N states and K N-1 states;

[0164] The input features of the N / N-1 branch feature prediction convolution submodule are the N / N-1 node-branch feature conversion graph and the output matrix of the convolution submodule is Y3 = {PF N ,QF N ,PFN-1,k ,QF N-1,k};

[0165] The convolution forward propagation formula of the N / N-1 branch feature prediction convolution submodule is as follows:

[0166]

[0167] Where l = 1, 2, ..., c2, c2 is the total number of convolutional layers in the convolutional submodule 2; PF l , QF l They are the branch active power features and branch reactive power features input to the lth convolutional layer respectively; are the convolution kernel and bias of the lth convolution layer, which need to be obtained through training. When l = 1, When l>1, r l is the number of convolution kernel channels of the lth layer;

[0168] The output of the N / N-1 branch feature prediction convolution submodule is as follows:

[0169] When l = 1, PF 1 =[PF N ,PF N-1,k ], QF 1 =[QF N ,QF N-1,k ], k = 1, 2, ..., K, according to the forward propagation formula (25), forward propagation is performed layer by layer from the 1st layer to the c2th layer, and the output feature is finally obtained: the branch active power feature vector under the N state Branch reactive power characteristic vector And the branch active power characteristic vector in the N-1 state Branch reactive power characteristic vector Its available PF o and QF o It is expressed as follows.

[0170]

[0171] In the formula, is the branch active power characteristic vector and branch reactive power characteristic vector in the N state; It is the branch active power characteristic vector and branch reactive power characteristic vector in the N-1 state.

[0172] Embodiment 10:

[0173] A method for characterizing the power flow coupling relationship of an operation mode based on an N / N-1 power flow embedded graph convolutional neural network, the technical content of which is the same as any one of embodiments 1-9, and further, the input of the operation mode N / N-1 power flow coupling relationship model is the feature vector X={U in ,θ in ,P in ,Q in}, the output is the feature vector Y in the N / N-1 state o = {U o ,θ o ,PF o ,QF o}.

[0174] Embodiment 11:

[0175] A method for characterizing the power flow coupling relationship between operation modes based on N / N-1 power flow embedded graph convolutional neural network is as follows:

[0176] Firstly, a graph convolution forward propagation strategy based on N / N-1 power flow embedding is proposed, and the forward propagation expression of the graph convolution module is derived and designed based on the N / N-1 power flow physical model; then, a neural network architecture based on the collaboration of multi-layer graph convolution and convolutional neural network modules is constructed, and the operation mode power flow coupling relationship model obtained after training can characterize the power flow coupling relationship under the N / N-1 state of the power system; finally, through simulation verification on the IEEE 30-node system, the proposed method can characterize the power flow coupling relationship under the N / N-1 state of the power system with high precision, verifying the effectiveness of the method proposed in the present invention; the specific contents are shown as follows.

[0177] 1. Graph convolution forward propagation module framework based on N / N-1 power flow embedding

[0178] The graph convolution forward propagation module framework designed by the present invention based on N / N-1 power flow embedding is as follows: Figure 1 As shown, it includes N-node feature aggregation graph convolution submodule 1, N / N-1 node feature mapping graph convolution submodule 2, N / N-1 node-branch feature conversion graph convolution submodule 3, which consists of Figure 1 It can be seen that the relationship between the three is as follows: the flow node feature Y1 in the N state output by the graph convolution submodule 1 will be used as the input of the graph convolution submodule 2; the flow node feature Y1 output by the graph convolution submodule 1 and the flow node feature Y2 in the N-1 state output by the graph convolution submodule 2 will be used as the input of the graph convolution submodule 3 at the same time. The input and output features of each submodule and its forward propagation process are constructed as follows:

[0179] For graph convolution submodule 1, its input features include the voltage amplitude feature vector U of the power flow node under N states in , voltage phase angle characteristic vector θ in, injected active power characteristic vector P in and injected reactive power eigenvector Q in , we get the input feature X of the graph convolution submodule 1, which is specifically expressed as follows.

[0180] X={U in ,θ in ,P in ,Q in} (1)

[0181] In the formula, U i is the voltage amplitude of the i-th node; n is the number of nodes in the power system.

[0182] The forward propagation expression of the N-state power flow physical model embedded in the graph convolution submodule 1 can be derived from the power flow equation in the N-state as shown in formula (2). The domain aggregation feature of the l-th graph convolution layer in the N-state is derived by using the Gauss-Seidel iteration method.

[0183] U Nl and θ` Nl , as shown in formula (3).

[0184]

[0185]

[0186] Where P i , Q i are the injected active power and injected reactive power of the ith node respectively; G ij , B ij Respectively represent the elements of the i-th row and j-th column in the conductance matrix G and the susceptance matrix B; θ ij represents the voltage phase angle difference from node i to node j; N(i) represents the adjacent set of node i; l represents the number of graph convolution layers, l = 1, 2, ..., m, m is the total number of graph convolution layers in graph convolution submodule 1; unit matrix F l The dimension of the output feature of the lth layer; They are the voltage amplitude characteristics and phase angle characteristics of the N states output by the lth graph convolutional layer respectively.

[0187] Further according to the graph convolution calculation rules, U` Nl and θ` Nl Multiply by the convolution kernel of the lth graph convolution layer respectively (obtained through training), perform feature update, and obtain the forward propagation expression of the graph convolution submodule 1 as shown in formula (4).

[0188]

[0189] Where σ is the activation function.

[0190] When l = 1, U N1 ,θ N1 is the input feature of graph convolution module 1, namely U N1 =U in ,θ N1 =θ in , we can further forward propagate from layer 1 to layer m based on formula (4) to obtain the output features of graph convolution module 1 It can be expressed by Y1 as follows.

[0191] Y1={U Nm ,θ Nm} (5)

[0192] For graph convolution submodule 2, its input feature matrix is ​​the feature matrix Y1 output by graph convolution submodule 1. Through the graph convolution operation of submodule 2 (specifically derived in the graph convolution forward propagation derivation design based on N / N-1 power flow embedding), its output features include: K node voltage amplitude feature matrices U under N-1 state N-1,k and phase angle characteristic matrix θ N-1,k ,in It can be expressed by Y2 as follows.

[0193] Y2={U N-1,k ,θ N-1,k} (6)

[0194] In the formula, k = 1, 2, ..., K, K is the number of N-1 states considered.

[0195] For graph convolution submodule 3, its input features are the feature matrix Y1 output by graph convolution submodule 1 and the feature matrix Y2 output by graph convolution submodule 2. Through the graph convolution operation of submodule 3 (specifically derived in the graph convolution forward propagation derivation design based on N / N-1 power flow embedding), its output features include: the branch active power feature matrix P under N state N , reactive power characteristic matrix Q N And the branch active power characteristic matrix P under K N-1 states N-1,k , reactive power characteristic matrix Q N-1,k ,in, It can be expressed by Y3 as follows.

[0196] Y3={P N ,Q N ,P N-1,k ,Q N-1,k} (7)

[0197] 2. Design of graph convolution forward propagation based on N / N-1 power flow embedding

[0198] First, the forward propagation expression of the design graph convolution submodule 2 is derived based on the power flow equation in the N state shown in formula (2). ij Expand to θ i and θ j , the power flow equation can be converted into the following form.

[0199]

[0200] P i , Q i , G ij , B ij , U i ,θ i They are respectively constructed into their corresponding vector or matrix forms, and equation (8) is further converted into the following matrix form.

[0201]

[0202] In the formula, θ=[θ1,θ1,…,θ n ] T ; P = [P1, P2, ... P n ] T ; Q=[Q1,Q2,…Q n ] T ; D(*) extracts the diagonal elements in * and constructs a diagonal matrix.

[0203] Formula (9) is further matrix-merged to form formula (10), which gives the relationship between the node injected power P, Q and the voltage U, θ in the N state, where χ is the voltage-power conversion coefficient matrix in the N state.

[0204]

[0205] in,

[0206]

[0207] When the power system disconnects the kth branch, that is, in the kth N-1 state, the node injection power P and Q composed of the new energy output, traditional generator output and load remain unchanged. Therefore, according to formula (10), the node injection power P, Q and voltage U in the N-1 state can be obtained: N-1,k ,θ N-1,k The relationship between is shown in formula (12), where χ N-1,k is the voltage-power conversion coefficient matrix in the N-1 state, and its specific expression is shown in formula (13).

[0208]

[0209] In the formula, * N-1,k Indicates the * corresponding to the N state when the kth branch is disconnected.

[0210] It can be seen that equation (10) and equation (12) can be connected, removing the injection power term in the equation and then replacing χ N-1,k After inverting, multiply both sides of the equation on the left at the same time, and finally obtain the N / N-1 node feature mapping expression shown in equation (14).

[0211]

[0212] Relying on type (14), it is possible to design The forward propagation formula of the graph convolution submodule 2 of the convolution kernel is shown in formula (15), and its input feature is Y1 = {U Nm ,θ Nm}, the output feature is Y2 = {U N-1,k ,θ N-1,k}.

[0213]

[0214] Next, based on the branch power equation shown in formula (16), the forward propagation expression of the graph convolution submodule 3 is derived and designed.

[0215]

[0216] In the formula, PF ij , QF ij are the active power and reactive power of branch ij respectively.

[0217] First, establish the branch-node association matrix A a =[a (ij)i ] b×n , where A a Each element of is defined as follows.

[0218]

[0219] Secondly, PF ij , QF ij , G ij , B ij , U i ,θ i They are constructed into vector form respectively, and further used the branch-node association matrix A a , convert equation (16) into the following matrix form.

[0220]

[0221] Where, PF = [PF ij ] b×1 ; QF=[QF ij ] b×1 ; G br =[G ij ] b×1 ; B br =[B ij ] b×1 ; b is the total number of branches; diag(*) is a diagonal matrix constructed with * as the diagonal element.

[0222] By further performing matrix merging on equation (18), we can obtain the node-branch feature transfer expression, as shown in equation (19).

[0223]

[0224] in,

[0225]

[0226] Based on formula (19), we can obtain the N node-branch characteristic transfer expression in the N state and the N-1 node-branch characteristic transfer expression in the N-1 state, as shown in formulas (21) to (22) respectively.

[0227]

[0228] In the formula, ω N ,ω N-1,k are the corresponding ω in the N state and the kth N-1 state respectively.

[0229] Formulas (21) to (22) are used as the forward propagation formulas of graph convolution submodule 3, where ω N ,ω N-1,k is the convolution kernel, and the input feature is Y1={U Nm ,θ Nm} and Y2={U N-1,k ,θ N-1,k}, based on the forward propagation formula, the output feature is Y3 = {PF N ,QF N ,PF N-1,k ,QF N-1,k}.

[0230] 3. N / N-1 power flow coupling relationship model based on the collaborative operation mode of multi-layer graph convolution and convolution module

[0231] Based on the graph convolution forward propagation strategy derived and designed above, the neural network architecture designed by the present invention with multi-layer graph convolution and convolution modules is as follows: Figure 2As shown. The architecture includes: N-node feature aggregation graph convolution submodule 1, N / N-1 node feature mapping graph convolution submodule 2, N / N-1 node-branch feature conversion graph convolution submodule 3, N / N-1 node feature prediction convolution submodule 1 and N / N-1 branch feature prediction convolution submodule 2. The forward propagation of the graph convolution submodule has been proposed in the previous section. Next, the forward propagation expressions and neural network architectures of convolution submodule 1 and convolution submodule 2 will be designed.

[0232] For the convolution submodule 1, its function is to predict the node voltage characteristics in the N state and K N-1 states. The input feature is the feature matrix Y1={U Nm ,θ Nm} and the feature matrix output by the graph convolution submodule 2

[0233] Y2={U N-1,k ,θ N-1,k}, the convolution forward propagation formula is shown in formula (23).

[0234]

[0235] Where l = 1, 2, ..., c1, c1 is the total number of convolutional layers in the convolutional submodule 1; U l ,θ l They are the voltage amplitude feature and phase angle feature of the input of the lth convolutional layer respectively; are the convolution kernel and bias of the lth convolution layer, which need to be obtained through training. When l = 1, When l>1, g l is the number of convolution kernel channels in layer l.

[0236] When l = 1, U 1 =[U Nm ,U N-1,k ],θ 1 =[θ Nm ,θ N-1,k ], k = 1, 2, ..., K, and forward propagation is performed layer by layer from the 1st layer to the c1th layer through the forward propagation formula (23), and the output feature is finally obtained: the node voltage amplitude feature vector under the N state Phase angle eigenvector And K node voltage amplitude characteristic vectors under N-1 states Phase angle eigenvector Its available U o and θ o It is expressed as follows.

[0237]

[0238] Convolution submodule 2 is used to predict the branch power characteristics in N state and K N-1 states. Its input feature is the feature matrix Y3 output by graph convolution submodule 3 = {PF N ,QF N ,PF N-1,k ,QF N-1,k}, and its convolution forward propagation formula is shown in formula (25).

[0239]

[0240] Where l = 1, 2, ..., c2, c2 is the total number of convolutional layers in the convolutional submodule 2; PF l , QF l They are the branch active power features and branch reactive power features input to the lth convolutional layer respectively; are the convolution kernel and bias of the lth convolution layer, which need to be obtained through training. When l = 1, PF l , When l>1, r l is the number of convolution kernel channels in layer l.

[0241] When l = 1, PF 1 =[PF N ,PF N-1,k ], QFF 1 =[QF N ,QF N-1,k ], k = 1, 2, ..., K, according to the forward propagation formula (25), forward propagation is performed layer by layer from the 1st layer to the c2th layer, and the output feature is finally obtained: the branch active power feature vector under the N state Branch reactive power characteristic vector And the branch active power characteristic vector in the N-1 state Branch reactive power characteristic vector Its available PF o and QF o It is expressed as follows.

[0242]

[0243] based on Figure 2 The neural network architecture and the forward propagation expressions proposed in Chapters 2 and 3 can be obtained by using the flow training samples under the N / N-1 state corresponding to the operation mode and using training algorithms such as gradient descent to obtain the following operation mode N / N-1 flow coupling relationship model f, which can be obtained by the input feature vector X={U in ,θ in ,P in ,Q in} Directly map to obtain the output feature vector Y in the N / N-1 state o = {U o ,θ o ,PF o ,QF o}, as shown in formula (27).

[0244] Y o =f(X) (27)

[0245] Embodiment 12:

[0246] A verification of a method for characterizing the power flow coupling relationship of operation mode based on N / N-1 power flow embedded graph convolutional neural network is as follows:

[0247] (1) Introduction to the simulation process

[0248] In this embodiment, the IEEE-30 system is used for simulation. In this simulation, Rectified LinearUnit (ReLU) is selected as the activation function, 10,000 training samples and 1,000 test samples are set, and the output characteristics of each sample consider 11 states, that is, K = 10. In order to verify that the power flow coupling relationship model of the operating mode proposed in the present invention can accurately characterize the power flow coupling relationship under the N / N-1 state, in the IEEE 30-node system, the models constructed by the following four comparison methods M0-M3 are trained using the same training samples and gradient descent algorithm, and then the test samples are used to test their accuracy.

[0249] M0: The operation mode power flow coupling relationship model based on the collaboration of multi-layer graph convolution and convolutional neural network modules proposed in the present invention;

[0250] M1: Power flow coupling relationship model based on fully connected neural network (replace the neural network in M0 with a fully connected neural network);

[0251] M2: Power flow coupling relationship model based on convolutional neural network (replace the neural network in M0 with a convolutional neural network);

[0252] M3: Power flow coupling relationship model based on graph convolutional neural network (replace the neural network in M0 with a graph convolutional neural network).

[0253] (2) Simulation results and analysis

[0254] The generated flow samples that can represent the N / N-1 state are divided into a training set and a test set. The training set is used to train the neural network. Subsequently, the four trained flow coupling relationship models (M0-M3) are used to calculate the flow on the test set, and finally the accuracy comparison table of the flow coupling relationship model of the operation mode is obtained as shown in Table 1.

[0255] Table 1 Comparison of the accuracy of the power flow coupling relationship model in the operation mode

[0256]

[0257] As shown in Table 1, the node voltage amplitude U predicted by M0 is O , phase angle θ O The accuracies of branch active power PF and branch reactive power QF are all higher than 99%, while the highest prediction accuracies of M1, M2 and M3 are 85.63%, 85.52% and 94.15% respectively. This shows that the power flow coupling relationship model of the operating mode proposed in the present invention can accurately characterize the power flow coupling relationship under the N / N-1 state, so as to facilitate the subsequent analysis of key issues such as fault diagnosis, stability assessment and optimization control in the power system.

[0258] In summary, the present invention proposes a method for characterizing the coupling relationship of operating mode and power flow based on N / N-1 power flow embedded graph convolutional neural network. First, a graph convolution forward propagation strategy based on N / N-1 power flow embedding is proposed, and the forward propagation expression of the graph convolution module is derived and designed based on the N / N-1 power flow physical model; then, a neural network architecture based on the collaboration of multi-layer graph convolution and convolutional neural network modules is constructed, and the operating mode power flow coupling relationship model obtained after training can characterize the power flow coupling relationship under the N / N-1 state of the power system; the effectiveness of the method proposed in the present invention is verified through example simulation analysis. Therefore, the present invention can provide technical support for key issues such as fault diagnosis, stability assessment and optimization control in power systems.

Claims

1. A method for characterizing the coupling relationship between operation modes and power flows based on N / N-1 power flow embedded graph convolutional neural network, characterized in that: The following steps are involved: 1) Based on the graph convolution forward propagation strategy embedded in the N / N-1 power flow, the forward propagation expression of the graph convolution module is derived using the N / N-1 power flow physical model; 2) Construct an N / N-1 power flow coupling relationship model based on the collaborative operation of multi-layer graph convolution and convolution modules; 3) Training operation mode N / N-1 power flow coupling relationship model; 4) Use the N / N-1 power flow coupling relationship model of operation mode to obtain the power flow coupling relationship.

2. According to claim 1, a method for characterizing the coupling relationship between operation modes and power flows based on an N / N-1 power flow embedded graph convolutional neural network is characterized in that: The graph convolution module includes an N-node feature aggregation graph convolution submodule, an N / N-1 node feature mapping graph convolution submodule, and an N / N-1 node-branch feature conversion graph convolution submodule; The output of the N-node feature aggregation graph convolution submodule is the input of the N / N-1-node feature mapping graph convolution submodule; The outputs of the N-node feature aggregation graph convolution submodule and the N / N-1 node feature mapping graph convolution submodule are simultaneously used as inputs of the N / N-1 node-branch feature conversion graph convolution submodule.

3. According to claim 2, a method for characterizing the coupling relationship of operation mode and power flow based on N / N-1 power flow embedded graph convolutional neural network is characterized in that: The input features X of the N-node feature aggregation graph convolution submodule are as follows: X={U in ,θ in ,P in ,Q in } (1) In the formula, U i is the voltage amplitude of the i-th node; They represent the voltage amplitude eigenvector, voltage phase angle eigenvector, injected active power eigenvector and injected reactive power eigenvector of the power flow node under N states respectively; n is the number of nodes in the power system.

4. According to claim 1, a method for characterizing the coupling relationship between operation modes and power flows based on an N / N-1 power flow embedded graph convolutional neural network is characterized in that: The forward propagation expression of the graph convolution module is as follows: Where σ is the activation function; P i , Q i are the injected active power and injected reactive power of the ith node respectively; G ij , B ij denote the elements of the i-th row and j-th column in the conductance matrix G and the susceptance matrix B respectively; θ ij represents the voltage phase angle difference from node i to node j; N(i) represents the adjacent set of node i; l represents the number of graph convolution layers, l = 1, 2, ..., m, m is the total number of graph convolution layers in graph convolution submodule 1; unit matrix F l The dimension of the output feature of the lth layer; They are the voltage amplitude characteristics and phase angle characteristics under N states output by the lth graph convolution layer respectively; U i , U j is the voltage amplitude of the i-th and j-th nodes; is the weight matrix of the lth layer.

5. According to claim 2, a method for characterizing the coupling relationship of operation mode and power flow based on N / N-1 power flow embedded graph convolutional neural network is characterized in that: The output feature Y1 of the N-node feature aggregation graph convolution submodule is as follows: Y1={U Nm ,θ Nm } (5) Where U Nm ,θ Nm It is the voltage amplitude characteristic and phase angle characteristic in N state.

6. According to claim 2, a method for characterizing the coupling relationship of operation mode and power flow based on N / N-1 power flow embedded graph convolutional neural network, characterized in that: The output feature Y2 of the N / N-1 node feature map graph convolution submodule is as follows: Y2={U N-1,k ,θ N-1,k } (6) Where k = 1, 2, ..., K, K is the number of N-1 states considered. N-1,k ,θ N-1,k It is the voltage amplitude characteristic and phase angle characteristic in N-1 state.

7. The method for characterizing the coupling relationship of operation modes and power flows based on N / N-1 power flow embedded graph convolutional neural network according to claim 2 is characterized in that: The output Y3 of the N / N-1 node-branch feature transformation graph convolution submodule is as follows: Y3={P N ,Q N ,P N-1,k ,Q N-1,k } (7) Where P N , Q N They represent the branch active power characteristic matrix and reactive power characteristic matrix under N state respectively; P N-1,k , They respectively represent the branch active power characteristic matrix and reactive power characteristic matrix under the state.

8. The method for characterizing the coupling relationship of operation modes and power flows based on N / N-1 power flow embedded graph convolutional neural network according to claim 1 is characterized in that: In step 2), the steps of constructing N node-branch characteristic transfer expressions in the N state and N-1 node-branch characteristic transfer expressions in the N-1 state include: 2.1) Construct the power flow equation, namely: 2.2) P i , Q i , G ij , B ij , U i ,θ i They are respectively constructed into corresponding vector or matrix forms, thereby converting equation (8) into matrix form and obtaining: In the formula, θ=[θ1,θ1,…,θ n ] T ; P = [P1, P2, ... P n ] T ; Q=[Q1,Q2,…Q n ] T ; D(*) extracts the diagonal elements in * and constructs a diagonal matrix; 2.3) Matrix merging is performed on equation (9) to obtain the relationship between the node injection power P, Q and the voltage U, θ in the N state, namely: Among them, the voltage-power conversion coefficient matrix χ under N states is as follows: 2.4) According to formula (10), the node injection power P, Q and voltage U in the N-1 state are obtained N-1,k ,θ N-1,k The relationship between them is: In the formula, * N-1,k represents the * corresponding to the N state when the kth branch is disconnected; N-1,k is the voltage-power conversion coefficient matrix in the N-1 state; 2.5) Based on formulas (10) and (12), construct the N / N-1 node feature mapping expression, namely: 2.6) Build The forward propagation formula of the convolution submodule of the N / N-1 node feature map of the convolution kernel is: 2.7) Establish the branch power equation, that is: In the formula, PF ij , QF ij are the active power and reactive power of branch ij respectively. 2.8) Establish the branch-node association matrix A a =[a (ij)i ] b×n , where A a Each element a (ij)i The definition is as follows: 2.9) PF ij , QF ij , G ij , B ij , U i ,θ o They are constructed into vector form respectively, using the branch-node association matrix A a , convert equation (16) into matrix form, that is: In the formula, PF = [PF ij ] b×1 ; QF=[QF ij ] b×1 ; G br =[G ij ] b×1 ; B br =[B ij ] b×1 ; b is the total number of branches; diag(*) is a diagonal matrix constructed with * as the diagonal element. 2.10) Matrix merge equation (18) to obtain the node-branch feature transfer expression of the N / N-1 node-branch feature conversion graph convolution submodule, that is: Among them, the coefficient matrix ω is as follows: 2.11) Construct the forward propagation equation of the N / N-1 node-branch feature conversion graph convolution submodule, that is: In the formula, ω N ,ω N-1,k are the coefficients ω corresponding to the N state and the kth N-1 state respectively.

9. The method for characterizing the coupling relationship of operation modes and power flows based on N / N-1 power flow embedded graph convolutional neural network according to claim 1 is characterized in that: The operation mode N / N-1 power flow coupling relationship model includes an N-node feature aggregation graph convolution submodule, an N / N-1 node feature mapping graph convolution submodule, an N / N-1 node-branch feature conversion graph convolution submodule, an N / N-1 node feature prediction convolution submodule, and an N / N-1 branch feature prediction convolution submodule; The N / N-1 node feature prediction convolution submodule is used to predict node voltage features in N states and K N-1 states; The input features of the N / N-1 node feature prediction convolution submodule include the output of the N node feature aggregation graph convolution submodule Y1 = {U Nm ,θ Nm }, and the output Y2 of the N / N-1 node feature map convolution submodule = {U N-1,k ,θ N-1,k }; The convolution forward propagation formula of the N / N-1 node feature prediction convolution submodule is as follows: Where l = 1, 2, ..., c1, c1 is the total number of convolutional layers in the convolutional submodule 1; U l ,θ l They are the voltage amplitude feature and phase angle feature of the input of the lth convolutional layer respectively; are the convolution kernel and bias of the lth layer of the convolution layer respectively; when l = 1, When l>1, g l is the number of convolution kernel channels of the lth layer; The output of the N / N-1 node feature prediction convolution submodule is as follows: When l = 1, U 1 =[U Nm ,U N-1,k ],θ 1 =[θ Nm ,θ N-1,k ], k = 1, 2, ..., K, and forward propagation is performed layer by layer from the 1st layer to the c1th layer through the forward propagation formula (23), and finally the output feature is obtained: the node voltage amplitude feature vector under the N state Phase angle eigenvector And K node voltage amplitude characteristic vectors under N-1 states Phase angle eigenvector The output features are represented as follows: In the formula, is the node voltage amplitude characteristic vector and phase angle characteristic vector in the N state; is the node voltage amplitude characteristic vector and phase angle characteristic vector in the N-1 state; U o ,θ o is the set of output voltage amplitude characteristics and phase angle characteristics; The N / N-1 branch feature prediction convolution submodule is used to predict the branch power features in N states and K N-1 states; The input features of the N / N-1 branch feature prediction convolution submodule are the N / N-1 node-branch feature conversion graph and the output matrix of the convolution submodule is Y3 = {PF N ,QF N ,PF N-1,k ,QF N-1,k }; The convolution forward propagation formula of the N / N-1 branch feature prediction convolution submodule is as follows: Where l = 1, 2, ..., c2, c2 is the total number of convolutional layers in the convolutional submodule 2; PF l , QF l They are the branch active power features and branch reactive power features input to the lth convolutional layer respectively; are the convolution kernel and bias of the lth convolution layer, which need to be obtained through training. When l = 1, When l>1, r l is the number of convolution kernel channels of the lth layer; The output of the N / N-1 branch feature prediction convolution submodule is as follows: When l = 1, PF 1 =[PF N ,PF N-1,k ], QF 1 =[QF N ,QF N-1,k ], k = 1, 2, ..., K, according to the forward propagation formula (25), forward propagation is performed layer by layer from the 1st layer to the c2th layer, and the output feature is finally obtained: the branch active power feature vector under the N state Branch reactive power characteristic vector And the branch active power characteristic vector in the N-1 state Branch reactive power characteristic vector The output features are represented as follows: In the formula, is the branch active power characteristic vector and branch reactive power characteristic vector in the N state; is the branch active power characteristic vector and branch reactive power characteristic vector in the N-1 state; PF o and QF o Represents the set of branch active power characteristics and branch reactive power characteristics.

10. The method for characterizing the coupling relationship of operation modes and power flows based on N / N-1 power flow embedded graph convolutional neural network according to claim 1, characterized in that: The input of the N / N-1 power flow coupling model is the characteristic vector X={U in ,θ in ,P in ,Q in }, the output is the feature vector Y in the N / N-1 state o = {U o ,θ o ,PF o ,QF o }.

Images