Alternating current and direct current load flow calculation method based on switchable control mode graph learning
By building a physical guide map neural network in an AC-DC hybrid system, embedding physical information and using ALM optimization, the rapid accuracy problem of AC/DC current analysis in a large AC-DC hybrid system is solved, and the effectiveness of AC-DC current calculation is achieved and ability to follow Kirchoff's law.
Patent Information
- Application Number
- CN202411169082.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-23
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2044-08-23
AI Technical Summary
In large AC/DC hybrid systems, it is difficult for prior art to perform AC/DC current analysis quickly and accurately, especially due to the lack of effective methods to direct DC control mode switching and determine whether non-convergence originates from inappropriate control modes or violates the base Erhoff's law.
By constructing the current model and graph neural network model of AC/DC hybrid system, the physical information of AC/DC operation is embedded in the GNN learning process, and a physical guided graph neural network (PG-GNN) that supports control mode switching is built to violate the trigger/ Extinguishing the angle constraints as the minimization goal, and the enhanced Lagrangian method (ALM) is used to optimize the GNN model, ensuring that the model follows Kirchoff's law.
AC-DC current calculation is realized in custom graph modeling of AC and DC grids, and the control mode switching signal is provided to ensure that the system achieves the optimal operating state while meeting the constraints, improving computing efficiency and accuracy.
Smart Images

Figure CN119093378A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of AC and DC power transmission, and in particular to an AC and DC power flow calculation method based on switchable control mode graph learning. Background Art
[0002] AC / DC hybrid systems have widely promoted the consumption of renewable energy. Due to the significant uncertainties brought by renewable energy, large-scale interconnected grids, and nonlinearities of DC lines and converters, faster AC / DC power flow analysis is more necessary than ever. There are abundant model-based methods for AC / DC power flow, ranging from basic unified methods and sequential methods to enhanced parallel methods. However, slow convergence and diverse control methods still impose heavy computational burdens in large AC / DC hybrid systems. Emerging data-driven methods provide faster solutions, but their generalization may be limited by the lack of physical embedding and topological patterns. Physically guided graph neural networks (PG-GNNs) may be a forward-looking solution to the above problems. For pure AC power flow, two alternatives have emerged, one is to simulate the system process of power flow solvers, such as Newton-Raphson (NR) solvers, and the other is to directly treat the power flow equation as a loss function. However, they cannot be directly used for AC / DC power flow analysis because there is no effective way to guide the DC control mode switching. In addition, it is difficult for current GNN architectures to determine whether the non-convergence of AC / DC power flows stems from inappropriate control modes or violations of Kirchhoff's laws. Summary of the invention
[0003] The purpose of the present invention is to overcome the shortcomings of the prior art and provide an AC / DC power flow calculation method based on switchable control mode graph learning.
[0004] The object of the present invention is achieved by the following technical solution: A method for calculating AC and DC power flows based on learning of a switchable control mode diagram, comprising the following steps:
[0005] Step S1: construct a power flow model of an AC / DC hybrid system;
[0006] Step S2: construct a graph neural network model of AC / DC power grid; embed AC / DC operation physical information into the GNN learning process;
[0007] Step S3: Taking violation of the trigger / extinguishing angle constraint as the minimization target, and constructing a physical guidance graph neural network PG-GNN that supports control mode switching based on the power flow model of the AC / DC hybrid system, the DC control mode switching is mastered through multiple physical guidance graph neural networks PG-GNNs, and its specific PG-GNN is trained through control mode one and control mode two. Then, the PG-GNN that meets the trigger / extinguishing angle constraint outputs the AC / DC power flow calculation results.
[0008] Specifically, the power flow model of the AC / DC hybrid system in step S1 is:
[0009]
[0010] in, and Respectively represent PV busbar group, AC busbar group, DC busbar group and common axis point group, P i inj is the active power injected into the AC bus i, is the reactive power injected into the AC bus i, is the active power injected into the common axis point c, P cj (V c ,V j ,δ ij ) is the sum of the active powers transmitted through the lines connected to c, is the reactive power injected into the common axis point c, Q cj (V c ,V j ,δ ij ) is the sum of reactive powers transmitted through the lines connected to c, is the phase angle of the kth DC line, is the ratio k of the converter transformer on the rectifier side, is the ratio k of the converter transformer on the inverter side; represents the reactance of the converter connected to the inverter / rectifier DC bus k; α k / γ k is the triggering / extinguishing angle of the kth rectifier / inverter; represents the rectifier / inverter voltage of the kth DC line; R dc and I d,k represent the resistance and current of the kth DC line respectively; j∈i means that the jth bus is adjacent to the ith bus through the branch ij; Pij(V i ,V j ,δ ij ) and Q ij (V i ,V j ,δ ij is the function for calculating the active and reactive power of line ij; ij is the voltage phase difference between busbars i and j, V p is the voltage of the PV bus, is the PV bus voltage reference value, is the rectified current of the kth DC line, is the reference value of the rectified current of the kth DC line, is the reference value of the inverter voltage of the kth DC line, is the minimum value of the trigger angle of the kth rectifier / inverter, is the inverter current of the kth DC line; is the reference value of the inverter current of the kth DC line.
[0011] Specifically, the method for constructing the graph neural network model of the AC / DC power grid in step S2 is:
[0012] By introducing converters in nodes and DC transformers and DC lines in edges, the DC grid is unified in the common graph of the AC grid. For an AC / DC system with N buses, b branches and 1 DC line, the corresponding GNN has N+2a nodes and b+3a edges:
[0013]
[0014] In the formula, H(·) (0) is the initial hidden layer representation of GNN, in particular, P i g and are the power generation and load power, V i is the voltage; for the inverter, or α min is the minimum value of the firing angle, is the rectifying reactance, is the reference value of the rectified current of the kth DC line, is the reference value of the inverter current of the kth DC line. For the rectifier, or γ min is the minimum value of the extinction angle, is the inverter reference voltage, represents the set of balanced nodes, P i g and are active generation power and active load power respectively, and are reactive power generation and reactive load power, respectively. It represents the DC bus group. For the PQ busbar group, is the common coupling point busbar group; the input edge feature is composed of branches, expressed as Edge feature matrix represents the eigenvector of each edge. For the AC branch, x ε=[G ε ,B ε ], where G ε and B ε is the ε line admittance parameter, for rectifier and inverter, For DC branch, or is a trainable transformation matrix that embeds edge features into nodes, G ε and B ε are the ε line conductance and susceptance parameters, and is the ratio of the minimum and maximum values of the converter transformer, R dc is the resistance of the DC line.
[0015] Specifically, the feedforward of the graph neural network model is:
[0016]
[0017] In the formula, σ is the activation function, F represents the order of adjacent nodes, For: The polynomial filter of the graph Laplacian operator, according to the graph structure To operate, H(·) (l) represents the lth layer response with the previous layer response as input, x is the input feature matrix of GNN, represents the initial input, θ j is the trainable weight of the j-th order filter, is the output of GNN, which is equal to the response of the last layer M, including the voltage and phase angle of the AC bus, the converter voltage and trigger angle of the DC bus, or [V, α (or γ)], φ θ (x) is the forward propagation function of the entire GNN, where λ is the trainable parameter of all layers, and [V,δ] is the output feature of the GNN, including the voltage V and phase angle δ of the AC bus.
[0018] Specifically, in step S3, a physical guided graph neural network PG-GNN supporting control mode switching is constructed with violation of the trigger / extinguishing angle constraint as the minimization target; and an equivalent optimization problem of the power flow model is formed:
[0019]
[0020] Among them, the state variables to be solved are represented by GNNφ θ (x) parameterization, f PQV , and They are respectively: formula (1)-
[0021] The equality constraints of (3), (4), and (5);
[0022] The process of tuning the trigger / extinguishing angle of the rectifier / inverter transformer is simulated by the following formula, using f DC Unified Representation and and omit the subscript:
[0023]
[0024] In the formula, To indicate the degree of violation of the trigger angle / extinguishing angle constraint, ReLU is the activation function, α min is the minimum trigger angle, α is the current trigger angle, K re is the ratio of the rectifier side converter, K re,max is the maximum allowable value of the converter ratio, K re,min is the minimum allowable value of the converter ratio.
[0025] Specifically, the following ALM-based learning strategy is used in step S3:
[0026] (i∈{PQV,DC})
[0027]
[0028] Where i is a variable, PQV is the PQ bus and V bus, DC is all DC buses, is the total loss function, which combines the constraint violation measure, linear combination term and quadratic penalty term, ρ>0, is the penalty parameter, which controls the penalty intensity of constraint violation; θ is the parameter of GNN, which is updated by minimizing the total loss function, λ i is the dual variable used to adjust the weight of the constraint term, φ θ (x) is the forward propagation function of NN, which is used to calculate the output of the current state. Formula (11a) updates θ by minimizing formula (10). By fixing θ, the dual variable is updated by formula (11b). Formula (11) is executed continuously until formula (10) remains stable.
[0029] Specifically, the formula (11a) is realized by the following formula:
[0030]
[0031] In the formula, θ is the trainable parameter of the graph neural network GNN, is the loss function, λ is the Lagrange multiplier, φ θ (x) is the output of GNN.
[0032] The present invention has the following advantages:
[0033] The present invention proposes an AC / DC power flow calculation method based on switchable control mode graph learning, in which the AC / DC operation physical information is embedded into the GNN learning process on the customized graph modeling of AC and DC grids. Then, in order to provide the control mode switching signal, the AC / DC power flow model is reconstructed as a parameterized optimization problem, in which the trigger / extinguishing angle constraint is relaxed to minimize the violation. In addition, the present invention proposes a gradient transformation algorithm based on the enhanced Lagrangian method (ALM), so that the PG-GNN strictly follows Kirchhoff's law and releases the converter constraint violation as the only signal to trigger the control mode switching. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] Figure 1 It is a schematic diagram of the method flow of the present invention. DETAILED DESCRIPTION
[0035] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention, that is, the embodiments described are only part of the embodiments of the present invention, rather than all of the embodiments. The components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations.
[0036] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present invention.
[0037] It should be noted that relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "including a..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.
[0038] The present invention is further described below in conjunction with the accompanying drawings, but the protection scope of the present invention is not limited to the following description. Figure 1As shown, a method for calculating AC and DC power flows based on switchable control mode graph learning includes the following steps:
[0039] Step S1: construct a power flow model of the AC / DC hybrid system; including defining the PV bus group, AC bus group, DC bus group and common axis point group, and establishing the corresponding AC / DC power flow model and control equations.
[0040] The power flow model is as follows:
[0041]
[0042] in, and Respectively represent PV busbar group, AC busbar group, DC busbar group and common axis point group, P i inj is the active power injected into the AC bus i, is the reactive power injected into the AC bus i, is the active power injected into the common axis point c, P cj (V c ,V j ,δ ij ) is the sum of the active powers transmitted through the lines connected to c, is the reactive power injected into the common axis point c, Q cj (V c ,V j ,δ ij ) is the sum of reactive powers transmitted through the lines connected to c, is the phase angle of the kth DC line, is the ratio k of the converter transformer on the rectifier side, is the ratio k of the converter transformer on the inverter side; represents the reactance of the converter connected to the inverter / rectifier DC bus k; α k / γ k is the triggering / extinguishing angle of the kth rectifier / inverter; represents the rectifier / inverter voltage of the kth DC line; R dc and I d,k represent the resistance and current of the kth DC line respectively; j∈i means that the jth bus is adjacent to the ith bus through the branch ij; Pij(V i ,V j ,δ ij ) and Q ij (V i ,V j ,δ ij ) is the function for calculating the active and reactive power of line ij; δ ij is the voltage phase difference between busbars i and j, Vp is the voltage of the PV bus, is the PV bus voltage reference value, is the rectified current of the kth DC line, is the reference value of the rectified current of the kth DC line, is the reference value of the inverter voltage of the kth DC line, is the minimum value of the trigger angle of the kth rectifier / inverter, is the inverter current of the kth DC line; is the reference value of the inverter current of the kth DC line. The sign of the third term on the right side in equations (2a) and (2b) is negative if it specifies the rectifier and positive otherwise. Equations (5a) and (5b) are the control equations of the DC system. The superscript ref indicates the control reference.
[0043] Step S2: The power system is represented as an undirected graph G = (V, E, A), where V, E, and A represent the bus set, branch set, and adjacency matrix, respectively. By introducing converter nodes, DC transformers, and DC line edges, the DC grid is unified in the public graph of the AC grid, and a graph neural network model of the AC and DC grid is constructed; the AC / DC operation physical information is embedded into the GNN learning process; the AC / DC operation physical information includes the AC power balance equation, the DC power balance equation, and the voltage and current relationship equation;
[0044] The power system can be represented as an undirected graph G = (V, E, A), where V, E, and A are solved on the bus set, branch set, and adjacency matrix, respectively. For AC / DC hybrid systems, GNN is not easy to create because the necessary features characterizing the AC grid, DC grid, and their bridges do not always match each other. Therefore, the present invention proposes an alternative solution to unify these features, specifically:
[0045] By introducing converters in nodes and DC transformers and DC lines in edges, the DC grid is unified in the common graph of the AC grid. For an AC / DC system with N buses, b branches and 1 DC line, the corresponding GNN has N+2a nodes and b+3a edges:
[0046]
[0047] In the formula, H(·) (0) is the initial hidden layer representation of GNN, which is equal to the input feature matrix x. in particular, P i g and are the power generation and load power, V i is the voltage; for the inverter, or α min is the minimum value of the firing angle, is the rectifying reactance, is the reference value of the rectified current of the kth DC line, is the reference value of the inverter current of the kth DC line. For the rectifier, or γ min is the minimum value of the extinction angle, is the inverter reference voltage, represents the set of balanced nodes, P i g and are active generation power and active load power respectively, and are reactive power generation and reactive load power, respectively. It represents the DC bus group. For the PQ busbar group, is the common coupling point busbar group; the input edge feature is composed of branches, expressed as Edge feature matrix represents the eigenvector of each edge. For the AC branch, x ε =[G ε ,B ε ], where G ε and B ε is the ε line admittance parameter, for rectifier and inverter, For DC branch, or is a trainable transformation matrix that embeds edge features into nodes, G ε and B ε are the ε line conductance and susceptance parameters, and is the ratio of the minimum and maximum values of the converter transformer, R dc is the resistance of the DC line.
[0048] ChebNet is a graph filter because ChebNet can aggregate high-order neighbor information to achieve better performance. The feedforward of ChebNet is given by the following formula (L = DA):
[0049]
[0050] In the formula, F represents the order of adjacent nodes, H(·) (l) It represents the lth layer response with the previous layer response as input. The output is the response of the last layer m, including the voltage and phase angle of the AC bus, the converter voltage and trigger angle of the DC bus, Or [V, α (or γ)]; for simplicity, the present invention uses φ θ (x) to represent ChebNet;
[0051] Step S3: Reconstruct the AC / DC power flow model into a parameterized optimization problem to minimize the violation of the trigger / extinguishing angle constraint. Introduce the gradient transformation algorithm of the enhanced Lagrangian method (ALM) to make the PG-GNN strictly follow Kirchhoff's law and release the converter constraint violation as the only signal to trigger the control mode switching. Based on the power flow model of the AC / DC hybrid system, a physical guided graph neural network PG-GNN that supports control mode switching is constructed. Multiple physical guided graph neural networks PG-GNNs are used to master the DC control mode switching. Each control mode is experienced to train its specific PG-GNN. Then, the PG-GNN that meets the trigger / extinguishing angle constraint outputs the AC / DC power flow calculation results.
[0052] In order to realize the control mode switching, a physical guided graph neural network PG-GNN supporting the control mode switching is constructed with the violation of the trigger / extinguishing angle constraint as the minimization target; and the equivalent optimization problem of the power flow model is formed:
[0053]
[0054] Among them, the state variables to be solved are represented by GNNφ θ (x) parameterization, f PQV , and are the equality constraints of formulas (1)-(3), (4) and (5), respectively; formula (1) includes (1a) and (1b), formula (2) includes (2a) and (2b), formula (4) includes (4a) and (4b), and formula (5) includes (5a) and (5b).
[0055] In order to simulate the process of rectifier / inverter transformer tuning trigger / extinguishing angle, the following formula is given, using f DC Unified Representation and and omit the subscript:
[0056]
[0057] In the formula, To indicate the degree of violation of the trigger angle / extinguishing angle constraint, ReLU is the activation function, α min is the minimum trigger angle, α is the current trigger angle, K re is the ratio of the rectifier side converter, K re,maxis the maximum allowable value of the converter ratio, K re,min is the minimum allowable value of the converter ratio.
[0058] The solution of formula (10) can be approximated by the Lagrange dual method. Considering that formula (10) is non-convex, the present invention introduces ALM to alleviate the difficulty of solving. ALM can usually alleviate the strong convexity condition and obtain the tolerable performance of PG-GNN. The following ALM-based learning strategy is used:
[0059]
[0060] Where i is a variable, PQV is the PQ bus and V bus, DC is all DC buses, is the total loss function, which combines the constraint violation measure, linear combination term and quadratic penalty term, ρ>0, is the penalty parameter, which controls the penalty intensity of constraint violation; θ is the parameter of GNN, which is updated by minimizing the total loss function, λ i is the dual variable used to adjust the weight of the constraint term, φ θ (x) is the forward propagation function of NN, which is used to calculate the output of the current state. Formula (11a) updates θ by minimizing formula (10). By fixing θ, the dual variable is updated by formula (11b). Formula (11) is executed continuously until formula (10) remains stable. In formula (10), As a loss function, it guides the optimization of the GNN model in AC and DC power flow calculations. By combining the objective function and constraints, it achieves comprehensive optimization of the system state and ensures that the system reaches the best operating state while meeting various constraints.
[0061] Furthermore, the formula (11a) is realized by the following formula:
[0062]
[0063] In the formula, θ is the trainable parameter of the graph neural network GNN, is the loss function, λ is the Lagrange multiplier, φ θ (x) is the output of GNN.
[0064] Finally, multiple PG-GNNs are used to master the DC control mode switching, and each control mode is experienced to train its specific PG-GNN. Then, the PG-GNN that meets the trigger / extinguishing angle constraints outputs the AC and DC power flow calculation results.
[0065] The above is only a preferred embodiment of the present invention and does not limit the present invention in any form. Any technician familiar with the art can make many possible changes and modifications to the technical solution of the present invention by using the above-mentioned technical content without departing from the scope of the technical solution of the present invention, or modify it into an equivalent embodiment of equivalent changes. Therefore, any changes, modifications, equivalent changes and modifications made to the above embodiments based on the technology of the present invention without departing from the content of the technical solution of the present invention belong to the protection scope of the present technical solution.
Claims
1. A method for calculating AC and DC power flows based on switchable control mode graph learning, characterized in that: The following steps are involved: Step S1: construct a power flow model of an AC / DC hybrid system; Step S2: construct a graph neural network model of AC / DC power grid; embed AC / DC operation physical information into the GNN learning process; Step S3: Taking violation of the trigger / extinguishing angle constraint as the minimization target, and constructing a physical guidance graph neural network PG-GNN that supports control mode switching based on the power flow model of the AC / DC hybrid system, the DC control mode switching is mastered through multiple physical guidance graph neural networks PG-GNNs, and each control mode is experienced to train its specific PG-GNN. Then, the PG-GNN that meets the trigger / extinguishing angle constraint outputs the AC / DC power flow calculation results.
2. The method for calculating AC and DC power flows based on switchable control mode graph learning according to claim 1, characterized in that: The power flow model of the AC / DC hybrid system in step S1 is: P i inj -∑ j∈i Pij(V i ,V j ,δ ij )=0 (1a) Control model 1: Control model 2: in, and Respectively represent PV busbar group, AC busbar group, DC busbar group and common axis point group, P i inj is the active power injected into the AC bus i, is the reactive power injected into the AC bus i, is the active power injected into the common axis point c, P cj (V c ,V j ,δ ij ) is the sum of the active powers transmitted through the lines connected to c, is the reactive power injected into the common axis point c, Q cj (V c ,V j ,δ ij ) is the sum of reactive powers transmitted through the lines connected to c, is the phase angle of the kth DC line, is the ratio k of the converter transformer on the rectifier side, is the ratio k of the converter transformer on the inverter side; represents the reactance of the converter connected to the inverter / rectifier DC bus k; α k / γ k is the triggering / extinguishing angle of the kth rectifier / inverter; represents the rectifier / inverter voltage of the kth DC line; R dc and I d,k represent the resistance and current of the kth DC line respectively; j∈i means that the jth bus is adjacent to the ith bus through the branch ij; Pij(V i ,V j ,δ ij ) and Q ij (V i ,V j ,δ ij ) is the function for calculating the active and reactive power of line ij; δ ij is the voltage phase difference between busbars i and j, V p is the voltage of the PV bus, is the PV bus voltage reference value, is the rectified current of the kth DC line, is the reference value of the rectified current of the kth DC line, is the reference value of the inverter voltage of the kth DC line, is the minimum value of the trigger angle of the kth rectifier / inverter, is the inverter current of the kth DC line; is the reference value of the inverter current of the kth DC line.
3. The method for calculating AC and DC power flows based on switchable control mode graph learning according to claim 2, characterized in that: The method for constructing the graph neural network model of the AC / DC power grid in step S2 is: By introducing converters in nodes and DC transformers and DC lines in edges, the DC grid is unified in the common graph of the AC grid. For an AC / DC system with N buses, b branches and 1 DC line, the corresponding GNN has N+2a nodes and b+3a edges: In the formula, H(·) (0) is the initial hidden layer representation of GNN, in particular, P i g and are the power generation and load power, V i is the voltage; for the inverter, or α min is the minimum value of the firing angle, is the rectifying reactance, is the reference value of the rectified current of the kth DC line, is the reference value of the inverter current of the kth DC line. For the rectifier, or γ min is the minimum value of the extinction angle, is the inverter reference voltage, represents the set of balanced nodes, P i g and are active generation power and active load power respectively, and are reactive power generation and reactive load power, respectively. It represents the DC bus group. For the PQ busbar group, is the common coupling point busbar group; the input edge feature is composed of branches, expressed as Edge feature matrix represents the eigenvector of each edge. For the AC branch, x ε =[G ε ,B ε ], where G ε and B ε is the ε line admittance parameter, for rectifier and inverter, For DC branch, or is a trainable transformation matrix that embeds edge features into nodes, G ε and B ε are the ε line conductance and susceptance parameters, and is the ratio of the minimum and maximum values of the converter transformer, R dc is the resistance of the DC line.
4. The method for calculating AC and DC power flows based on switchable control mode graph learning according to claim 3 is characterized in that: The feedforward of the graph neural network model is: In the formula, σ is the activation function, F represents the order of adjacent nodes, For: The polynomial filter of the graph Laplacian operator, according to the graph structure To operate, H(·) (l) represents the lth layer response with the previous layer response as input, x is the input feature matrix of GNN, represents the initial input, θ j is the trainable weight of the j-th order filter, is the output of GNN, which is equal to the response of the last layer M, including the voltage and phase angle of the AC bus, the converter voltage and trigger angle of the DC bus, or [V, α (or γ)], φ θ (x) is the forward propagation function of the entire GNN, where θ is the trainable parameter of all layers, and [V,δ] is the output feature of the GNN, including the voltage V and phase angle δ of the AC bus.
5. The method for calculating AC and DC power flows based on switchable control mode graph learning according to claim 4, characterized in that: In step S3, a physical guided graph neural network PG-GNN supporting control mode switching is constructed with violation of the trigger / extinguishing angle constraint as the minimization target; and an equivalent optimization problem of the power flow model is formed: Among them, the state variables to be solved are given by GNN φ θ (x) parameterization, f PQV , and are the equality constraints of formulas (1)-(3), (4) and (5) respectively; The process of tuning the trigger / extinguishing angle of the rectifier / inverter transformer is simulated by the following formula, using f DC Unified Representation and and omit the subscript: In the formula, To indicate the degree of violation of the trigger angle / extinguishing angle constraint, ReLU is the activation function, α min is the minimum trigger angle, α is the current trigger angle, K re is the ratio of the rectifier side converter, K re,max is the maximum allowable value of the converter ratio, K re,min is the minimum allowable value of the converter ratio.
6. The method for calculating AC and DC power flows based on switchable control mode graph learning according to claim 5, characterized in that: In step S3, the physical guided graph neural network PG-GNN is optimized using the following ALM-based learning strategy: (i∈{PQV,DC}) Where i is a variable, PQV is the PQ bus and V bus, DC is all DC buses, is the total loss function, · combines the constraint violation measure, linear combination term and quadratic penalty term, ρ>0, is the penalty parameter, which controls the penalty intensity of constraint violation; θ is the parameter of GNN, which is updated by minimizing the total loss function, λ i is the dual variable used to adjust the weight of the constraint term, φ θ (x) is the forward propagation function of GNN, which is used to calculate the output of the current state. Formula (11a) updates θ by minimizing formula (10). By fixing θ, the dual variable is updated by formula (11b). Formula (11) is executed continuously until formula (10) remains stable.
7. The method for calculating AC and DC power flows based on switchable control mode graph learning according to claim 6, characterized in that: The formula (11a) is realized by the following formula: In the formula, θ is the trainable parameter of the graph neural network GNN, is the loss function, λ is the Lagrange multiplier, φ θ (x) is the output of GNN.
Citation Information
Patent Citations
Method for calculating load flow of power grid alternating current and direct current power transmission system based on state switching
CN103138279A
Load transfer control method and device for AC / DC hybrid power distribution network, and electronic equipment
CN115149533A
Unified power flow model of alternating-current and direct-current hybrid power distribution network
CN115395588A
Direct current maximum power index and influence identification method considering optimal power flow
CN115764898A
Power system load flow calculation method of physical equation driven graph neural network
CN116345462A
