Wind power and conventional thermal power combined operation power transmission system load flow calculation method, system, equipment and medium
By combining a graph hybrid neural network model with a hybrid driving method based on physical constraints, the complexity and accuracy issues of optimal power flow calculation in the joint operation of wind power and conventional thermal power were solved, achieving efficient and accurate power flow distribution prediction, and improving the grid operation efficiency and renewable energy absorption capacity.
Patent Information
- Application Number
- CN202511547769.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-28
- Publication Date
- 2026-02-13
AI Technical Summary
In power transmission systems where wind power and conventional thermal power are operating together, it is difficult to balance computational complexity and accuracy in the optimal power flow problem. Traditional methods suffer from high computational complexity and insufficient accuracy in the solution process.
By employing a graph hybrid neural network model combined with physical constraints, a hybrid solution method that integrates data-driven and model-driven approaches is achieved through data generation, network training, and model solving stages. The graph hybrid neural network model learns historical data and maps generator output and node voltage. The model training is optimized by combining a penalty loss function and a supervised loss term to ensure the feasibility and accuracy of the solution results.
It achieves fast and efficient near-optimal solution calculation in combined wind and fire systems, with a solution speed increased by 220 times and an accuracy of less than 0.41% for active power error rate and less than 0.01% for voltage amplitude error rate, which is significantly better than traditional methods.
Smart Images
Figure CN121529609A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of power system operation optimization technology, and in particular to a method, system, equipment and medium for calculating power flow in a transmission system that combines wind power and conventional thermal power. Background Technology
[0002] In regions rich in wind energy resources, the installed capacity of wind power in the system continues to increase, and wind power is gradually becoming an important part of the power system. The application of wind power has played a key role in promoting the use of clean energy and reducing costs. However, due to the significant intermittency and volatility of wind energy, its output is greatly affected by natural conditions such as climate and seasons, typically exhibiting characteristics of "high randomness and weak controllability." In wind-thermal combined power supply scenarios, this uncertainty makes the system operation more complex and seriously affects power flow distribution.
[0003] Optimal power flow is a mathematical optimization problem that optimizes the allocation strategy between power generation and load, as well as various controllable devices in the control system, to minimize system operating costs, fuel consumption, and network losses while meeting various operational constraints. Optimal power flow models typically include active and reactive power balance equations and various security constraints of the power grid. Traditional model-driven methods are based on modeling the physical characteristics of the power grid and have good interpretability; while data-driven methods rely on historical operating data, which can uncover potential patterns in complex scenarios and improve the model's adaptability to actual operating conditions. Combining the advantages of both approaches creates a complementary hybrid approach that integrates the strengths of model-driven and data-driven methods. This type of method retains the physical optimization framework of the power system while introducing data-driven techniques to improve the accuracy of power flow distribution prediction and achieve coordinated operation between wind and thermal power. Summary of the Invention
[0004] In view of the above-mentioned existing problems, the present invention provides a method, system, equipment and medium for calculating power flow in a transmission system that combines wind power and conventional thermal power.
[0005] Therefore, the technical problem addressed by this invention is that the Optimal Power Flow (OPF) problem is essentially an optimization problem with nonlinear and nonconvex characteristics, exhibiting significant complexity in its solution process and making efficient and reliable computation difficult. Traditional linearization methods simplify the nonlinear power flow equations to some extent, thereby reducing the computational difficulty, but these methods often come at the cost of computational accuracy. Fast solution of the optimal power flow problem remains one of the significant challenges in the field of power system dispatch optimization. To solve the above-mentioned technical problems, the present invention provides the following technical solution: a power flow calculation method for a transmission system that combines wind power and conventional thermal power, comprising: in the data generation stage, setting an optimal power flow optimization physical model, using a solver to solve for system parameters and optimization results, and generating a graph dataset containing node features and power flow distribution labels; During the network training phase, the graph dataset is divided into training data and test data. The graph hybrid neural network model is used to train the training data. The graph hybrid neural network model includes different MG layers and a Read layer, which are used to learn historical data and map generator output and node voltage. During the model solving phase, the data of the system to be solved is input, and the trained graph hybrid neural network model is used for data-driven solution. If the solution does not meet the feasibility requirements, the model-driven module is used for secondary solution until the optimal power flow solution is obtained.
[0006] As a preferred embodiment of the power flow calculation method for a transmission system combining wind power and conventional thermal power as described in this invention, the data generation stage includes: setting wind farm and load conditions, using a solver to obtain system parameters and optimization results, and preprocessing the results into graphical data, wherein node features include load, voltage and phase angle, and labels include power flow distribution.
[0007] As a preferred embodiment of the power flow calculation method for a transmission system operating in conjunction with wind power and conventional thermal power as described in this invention, the data generation stage further includes obtaining graph data with node connection relationships by setting non-generator nodes as the first node type, generator nodes as the second node type, and wind turbine generator nodes as the third node type.
[0008] As a preferred embodiment of the power flow calculation method for a transmission system combining wind power and conventional thermal power as described in this invention, the network training stage includes training the training data using an MG layer formed by fusing graph convolutional networks, relational graph neural networks, and residual networks, and a Read layer formed by fusing graph convolutional networks and linear networks, to obtain a graph hybrid neural network model capable of outputting the active and reactive power output of generators and the voltage amplitude and phase angle of nodes.
[0009] As a preferred embodiment of the power flow calculation method for a transmission system combining wind power and conventional thermal power as described in this invention, the graph hybrid neural network model includes an approximation of the true value using MSEloss as the main component and applying a penalized loss function containing physical information, such as power balance, voltage upper and lower limits, and power upper and lower limits. Specifically, in, This is the total loss value. These are the weighting coefficients. Let y be the expected value, i and j be the variable indices, and y be the predicted output value. The actual label value. For physical penalty loss items, This is the deviation loss term. To monitor loss items; The loss function is divided into three parts: supervised feedback using MSEloss, penalized feedback using physical information constraints, and bias feedback for data approximation.
[0010] As a preferred embodiment of the power flow calculation method for a transmission system combining wind power and conventional thermal power as described in this invention, the data-driven solution includes: checking the feasibility of the output of the data-driven module; if all outputs are feasible, the result is taken as the final result; if only the active and reactive power of the generators are feasible, they are substituted into the power flow simplification model to solve for the node voltage magnitude and phase angle; if only the node voltage magnitude and phase angle are feasible, they are substituted into the power flow simplification model to solve for the generator active and reactive power; if all are infeasible, the corresponding variables are set as initial values for the optimal power flow model as a hot start, thus obtaining the final optimal power flow solution.
[0011] As a preferred embodiment of the power flow calculation method for a transmission system operating in conjunction with wind power and conventional thermal power as described in this invention, it also includes an evaluation step, which verifies the accuracy and efficiency of the optimal power flow solution by calculating the time saving rate, active power error rate and voltage amplitude error rate. The time saving rate is based on a comparison between the running time of the model solution steps and the running time of the solver: in, The time saving rate is N, where N is the total number of test samples. The runtime of a traditional solver in solving for optimal power flow on a test sample. The runtime of solving the proposed DMGOPF method on the test samples; The active power error rate is based on the comparison between the generator active power output output from the model solution steps and the generator active power output output from the solver: in, The active power error rate, This represents the total number of generators in the system. The active power output of the generator is calculated by a traditional solver. The active power output of the generator is predicted by the DMGOPF method; The voltage amplitude error rate is based on a comparison between the node voltage amplitude output by the model solution step and the node voltage amplitude output by the solver: in, For voltage amplitude error rate, This represents the total number of nodes in the system. The voltage magnitude at the node is calculated by a conventional solver. The voltage amplitude of the node is predicted by the DMGOPF method.
[0012] This invention provides a power flow calculation system for a power transmission system that combines wind power and conventional thermal power.
[0013] As a preferred embodiment of the power flow calculation system for a transmission system that combines wind power and conventional thermal power as described in this invention, it includes a data generation and processing module, a hybrid drive calculation module, and a verification and output module. The data generation and processing module is used to construct an optimal power flow physical model that includes wind farms and conventional thermal power plants, call the solver to generate a graph dataset containing node features and power flow distribution labels, and perform node type classification and data preprocessing. The hybrid drive computing module is connected to the data generation and processing module. It is used to load the trained graph hybrid neural network model, receive the node feature data of the system to be solved, obtain the preliminary generator output and node voltage parameters through the data-driven method, and selectively combine the power flow simplification model to solve the problem based on the feasibility of the parameters. The verification and output module is connected to the hybrid drive calculation module and is used to verify and evaluate the calculated optimal power flow solution, calculate the time saving rate, active power error rate and voltage amplitude error rate, and output the final reliable power flow result.
[0014] The present invention provides a computer device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of a power flow calculation method for a transmission system that combines wind power and conventional thermal power.
[0015] The present invention provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the steps of a power flow calculation method for a transmission system that combines wind power and conventional thermal power are implemented.
[0016] The beneficial effects of this invention are as follows: By integrating data-driven and model-driven methods, this invention effectively solves the problem of balancing computational complexity and accuracy in traditional optimal power flow calculations for combined wind and thermal systems. The proposed DMGOPF model, combined with a graph hybrid neural network and physical constraints, achieves rapid calculation of near-optimal solutions. Actual tests show that this method maintains ultra-high computational accuracy (active power error rate less than 0.41%, voltage amplitude error rate less than 0.01%) while increasing the solution speed by more than 220 times, significantly outperforming traditional iterative solution methods. Attached Figure Description
[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0018] Figure 1 This is a schematic flowchart of a power flow calculation method for a transmission system that combines wind power and conventional thermal power, provided as an embodiment of the present invention.
[0019] Figure 2 This is a schematic diagram of a graph hybrid neural network model for a power flow calculation method for a transmission system that combines wind power and conventional thermal power, provided as an embodiment of the present invention.
[0020] Figure 3 This is a schematic diagram of the DMGOPF model solution framework for a power flow calculation method for a transmission system operating in conjunction with wind power and conventional thermal power, provided as an embodiment of the present invention.
[0021] Figure 4 This invention provides a modified IEEE 30 system for calculating the power flow of a transmission system that combines wind power and conventional thermal power, as an embodiment of the present invention.
[0022] Figure 5 This diagram illustrates a comparison between the active power output solution and the exact solution of the DMGOPF in a power flow calculation method for a transmission system operating in conjunction with wind power and conventional thermal power, as provided in an embodiment of the present invention.
[0023] Figure 6 This diagram illustrates a comparison between the calculated voltage amplitude of DMGOPF and the exact solution in a power flow calculation method for a transmission system operating in conjunction with wind power and conventional thermal power, provided in one embodiment of the present invention. Detailed Implementation
[0024] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.
[0025] Example 1, referring to Figure 1 This is the first embodiment of the present invention, which provides a power flow calculation method for a transmission system operating in conjunction with wind power and conventional thermal power, including: S1: In the data generation phase, set up the optimal power flow optimization physical model, use the solver to solve for system parameters and optimization results, and generate a graph dataset containing node features and power flow distribution labels.
[0026] S2: During the network training phase, the graph dataset is divided into training data and test data. The graph hybrid neural network model is used to train the training data. The graph hybrid neural network model includes different MG layers and a Read layer, which are used to learn historical data and map generator output and node voltage.
[0027] S3: In the model solving stage, input the data of the system to be solved, and use the trained graph hybrid neural network model to perform data-driven solution. If the solution does not meet the feasibility requirements, use the model-driven module to perform a second solution until the optimal power flow solution is obtained.
[0028] It should be noted that traditional optimal power flow calculation methods face severe challenges when dealing with high proportions of wind power integration: model-driven methods, while physically meaningful, are computationally complex and time-consuming; data-driven methods, while highly adaptable, struggle to guarantee the physical feasibility of the solution. This invention innovatively constructs a data-model hybrid driving framework, deeply integrating the optimal power flow physical model with a graph hybrid neural network, successfully overcoming the bottleneck of the trade-off between computational efficiency and accuracy. Practical applications demonstrate that this method achieves a perfect balance between ultra-high-speed computation (more than 220 times faster) and near-optimal accuracy (error rate less than 0.5%) in combined wind and thermal power systems, providing a fast and reliable optimized scheduling solution for power systems with high proportions of renewable energy, significantly improving grid operating efficiency and renewable energy absorption capacity.
[0029] Example 2, refer to Figures 2-6 As an embodiment of the present invention, based on the above embodiment, a power flow calculation method for a transmission system operating in conjunction with wind power and conventional thermal power is provided.
[0030] Furthermore, in this embodiment, step S1, during the data generation stage, involves setting an optimal power flow optimization physical model, using a solver to solve for system parameters and optimization results, and generating a graph dataset containing node features and power flow distribution labels. Specific steps include S101-S103: S101: Set up the optimal power flow optimization physical model, define the wind farm and load conditions, and use the solver to obtain system parameters and optimization results. Then, preprocess the stored results into graph data required by the graph hybrid neural network. Use the sample load, initial voltage, and phase angle as node features, and the solved power flow distribution as labels. Set non-generator nodes as node type 0, generator nodes as node type 1, and wind turbine nodes as type 2 as node relationships for node connection processing.
[0031] in, For the feature vector of the node, Let be the label vector of the node. For node type, For the active load on the node, For reactive loads on nodes, The initial voltage amplitude of the node. The initial voltage phase angle of the node is denoted as . To contribute one's efforts To exert effort for no reason The optimized voltage amplitude for the node. The optimized voltage phase angle for the node; In an optional embodiment, node partitioning can also be based on the partitioning of node electrical functions. Specifically, when constructing graph data, type identifier A is assigned to "pure consumption and transmission nodes", type identifier B is assigned to "controllable power generation nodes", and type identifier C is assigned to "intermittent power generation nodes".
[0032] In another alternative embodiment, node partitioning can also be based on energy injection characteristics. Specifically, when constructing the graph data structure, a relationship code X is assigned to "passive nodes," a relationship code Y is assigned to "traditional energy nodes," and a relationship code Z is assigned to "renewable energy nodes." The node relationship data encoded in this way, along with the node features, is input into the neural network, enabling the model to distinguish the impact of different energy injection forms on the system power flow.
[0033] S102: The optimal power flow model adopted is as follows: The objective function is set to minimize the sum of system cost and wind curtailment cost: In the formula, and These represent the outputs of conventional thermal power and wind power, respectively, both being generator outputs within the system. and These are the designations for thermal power and wind turbine generators, both belonging to... And indicates the generator set lead wire; The active power output provided by the wind turbine to the system. This represents all the active power actually generated by the wind turbine.
[0034] Node power balance equations: In the formula, n is the total number of nodes in the system, and V i V j Let i be the voltage magnitude at the node, and j be the variable indices. , The node voltage phase angle, For the active / reactive power output of the generator, For active / reactive load, For the system busbar and its lead wires; For the real / imaginary part of the admittance matrix of the corresponding node; and These represent the voltage magnitude and voltage phase angle of the node, respectively.
[0035] S103: Generator active power output constraint: In the formula, These represent the upper and lower limits of the generator's active power output.
[0036] Generator reactive power output constraints: In the formula, These represent the upper and lower limits of the generator's reactive power output.
[0037] Node voltage amplitude constraints: In the formula, These are the upper and lower limits of the node voltage.
[0038] Node voltage phase angle constraints: In the formula, These are the upper and lower limits of the node voltage phase angle.
[0039] Power flow constraints on the line: In the formula, This represents the upper limit of active power transmission for the line. For system branches and their leads.
[0040] Furthermore, in this embodiment, step S2, during the network training phase, divides the graph dataset into training data and test data, and uses a graph hybrid neural network model to train the training data. The graph hybrid neural network model includes different MG layers and a Read layer, used to learn historical data and map generator output and node voltage. Specific steps include S201-S202: S201: The graph hybrid neural network model is the core of DMGOPF. It learns historical data to map the required generator output and the voltage and phase angle of each bus. The core of the graph hybrid neural network is the MG layer, which is formed by fusing graph convolutional network (GCN), relational graph neural network (RGCN), and residual network (ResNet), and the Read layer, which is formed by fusing graph convolutional network and linear network to read the mapped data. It outputs two sets of outputs: generator active and reactive power output and node voltage amplitude and phase angle, respectively, as shown in Table 1.
[0041] Table 1. Structure and parameters of convolutional neural networks
[0042] S202: In order to approximate the optimal solution, MSEloss is used as the main body, and penalized loss functions containing physical information such as power balance, voltage upper and lower limits, and power upper and lower limits are applied to approximate the true value.
[0043] in, This is the total loss value. These are the weighting coefficients. Let y be the expected value, i and j be the variable indices, and y be the predicted output value. The actual label value. For physical penalty loss items, This is the deviation loss term. To monitor loss items; The loss function is divided into three parts: supervised feedback using MSEloss, penalized feedback using physical information constraints, and bias feedback for data approximation.
[0044] In an optional embodiment, the loss function can also be implemented by simplifying the penalty term of the physical constraints. Specifically, when constructing the penalty loss term, only the node power balance equation constraints and the upper and lower limits of generator active power output constraints are retained. During the neural network training process, only the degree to which the model prediction value violates the above two types of constraints is calculated. This simplified penalty term is then weighted and summed with the original supervision loss term (such as MSE loss) and bias loss term to obtain the final total loss value used for model training.
[0045] Because the penalty term is simplified, the model may be easier to train and faster, but it is necessary to pay close attention to the feasibility of its output on constraints that are not included in the penalty (such as voltage constraints) and adjust the weight coefficients accordingly.
[0046] In an optional embodiment, the loss function can also be implemented using a penalty term that emphasizes safe operation. Specifically, when constructing the penalty loss term, node voltage magnitude constraints and line power flow transmission limit constraints are incorporated. During neural network training, the degree to which the model's predicted values violate voltage safety and line overload constraints is calculated. This penalty term, which emphasizes safe operation, is then weighted and summed with the original supervision loss term and bias loss term to jointly guide the model's training process.
[0047] The trained model will place particular emphasis on the feasibility of its output solutions in terms of voltage safety and line load, ensuring that the direct impact of the solutions on the safe operation of the power grid is minimized.
[0048] Furthermore, in this embodiment, step S3, during the model solving stage, involves inputting data from the system to be solved and using a trained graph hybrid neural network model for data-driven solving. If the solution does not meet the feasibility requirements, a second solution is performed using the model-driven module until the optimal power flow solution is obtained. Specific steps include S301-S302: S301: If all predicted quantities are feasible, they can be used as the final solution. If only the active and reactive power of generators are feasible, they are substituted into the simplified power flow (PF) model to obtain the node voltage magnitude and phase angle. If only the node voltage magnitude and phase angle are available, they are substituted into the simplified PF model to obtain the active and reactive power of generators. If all outputs are not feasible, they are set as the initial values of the corresponding variables as the hot start of the OPF model, reducing the solver's search time for the optimal value of the optimal power flow.
[0049] In an optional embodiment, processing partially feasible results can also be achieved through a unified simplified model. Specifically, when any part of the results output by the graph neural network (whether active or reactive power, or voltage phase angle) does not meet the requirement of complete feasibility, complex branching decisions are not performed. The generator active and reactive power output by the neural network is uniformly treated as known quantities. These power values are directly substituted into the power flow simplification model to resolve the node voltage magnitudes and phase angles of the entire system. The complete result obtained from the resolution is used as the final output.
[0050] In another alternative embodiment, processing partially feasible results can also be achieved through hierarchical iterative correction. Specifically, when partially feasible cases occur, all feasible output components are first retained. For infeasible output components, they are replaced with safe default values (such as nominal values or historical averages) of the corresponding variables. The complete result after the above correction is used as the initial solution of the optimal power flow model. A finite number of iterations of optimization are performed using a solver to quickly obtain a feasible optimal solution.
[0051] In this embodiment, when all outputs are infeasible, they are used as the initial warm-start of the OPF model. In an optional embodiment, the model can also be solved by setting initial values based on a safety threshold. Specifically, when all neural network outputs are infeasible, the degree of constraint violation in each variable is first identified. All out-of-limit variables are adjusted to their safe operating threshold range (e.g., adjusting the out-of-limit voltage to near the rated value). These safety-adjusted variable values are used as the initial warm-start values for the OPF model. The optimization process of the traditional solver is then initiated based on these initial values.
[0052] S302: Verify the accuracy and efficiency of the optimal power flow solution by calculating the time saving rate, active power error rate, and voltage amplitude error rate; Time savings, based on a comparison of the running time of the model solution steps with the running time of the solver: in, The time saving rate is N, where N is the total number of test samples. The runtime of a traditional solver in solving for optimal power flow on a test sample. The runtime of solving the proposed DMGOPF method on the test samples; Active power error rate, based on a comparison between the generator active power output output from the model solution steps and the generator active power output output from the solver: in, The active power error rate, This represents the total number of generators in the system. The active power output of the generator is calculated by a traditional solver. The active power output of the generator is predicted by the DMGOPF method; The voltage amplitude error rate is based on a comparison between the node voltage amplitudes output by the model solution steps and the node voltage amplitudes output by the solver: in, For voltage amplitude error rate, This represents the total number of nodes in the system. The voltage magnitude at the node is calculated by a conventional solver. The voltage amplitude of the node is predicted by the DMGOPF method.
[0053] Example 3 is the third embodiment of the present invention, which differs from the previous two embodiments in that: This embodiment also provides a power flow calculation system for a transmission system that combines wind power and conventional thermal power, including: a data generation and processing module, a hybrid drive calculation module, and a verification and output module; The data generation and processing module is used to construct an optimal power flow physical model that includes wind farms and conventional thermal power plants, call the solver to generate a graph dataset containing node features and power flow distribution labels, and perform node type classification and data preprocessing. The hybrid drive computing module, connected to the data generation and processing module, is used to load the trained graph hybrid neural network model, receive the node feature data of the system to be solved, obtain the preliminary generator output and node voltage parameters through the data-driven method, and selectively combine the power flow simplification model to solve the problem based on the feasibility of the parameters. The verification and output module, connected to the hybrid drive calculation module, is used to verify and evaluate the calculated optimal power flow solution, calculate the time saving rate, active power error rate and voltage amplitude error rate, and output the final reliable power flow result.
[0054] This embodiment also provides an electronic device, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the power flow calculation method for a transmission system that combines wind power and conventional thermal power as proposed in the above embodiment.
[0055] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a power flow calculation method for a transmission system operating in conjunction with wind power and conventional thermal power, as proposed in the above embodiments.
[0056] The storage medium proposed in this embodiment belongs to the same inventive concept as the power flow calculation method for a transmission system that enables the joint operation of wind power and conventional thermal power proposed in the above embodiments. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.
[0057] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.
[0058] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
[0059] Example 4 is an embodiment of the present invention, used to verify a power flow calculation method for a transmission system operating in conjunction with wind power and conventional thermal power.
[0060] This invention uses 30-node data provided by Matpower for case study verification, adding 3 wind turbines and setting a wind power penetration rate of 30% and a load fluctuation of 10%. The following metrics are used to evaluate the performance of the DMOPF model: (1) Solution rate: The ratio of the difference between the running time of the DMGOPF model and the solution time of the solver to the solution time of the solver is used as the test index, and its best performance is 100%.
[0061] (2) Prediction accuracy: The simulation accuracy is observed by the active power error rate and voltage amplitude error obtained from the DMGOPF model in the test index, and its best performance is 0%.
[0062] Table 1 compares the performance of the DMGOPF model and traditional optimal power flow solution methods.
[0063] Table 1 DMGOPF Test Results
[0064] The solution accuracy for active power output is 99.59%, and the solution accuracy for voltage amplitude is 99.99%, demonstrating that the DMGOPF solution is very close to the solver's solution for optimal power flow, and is a near-optimal solution. Figure 5 and Figure 6 As can be seen, the values solved by DMGOPF have high accuracy; compared with the solver, DMGOPF's running time is improved by at least 220 times. In summary, compared with traditional iterative optimal power flow solution methods, the DMGOPF method proposed in this patent can accelerate the solution speed of optimal power flow problems and has high accuracy.
Claims
1. A power flow calculation method for a transmission system combining wind power and conventional thermal power, characterized in that: include, In the data generation phase, an optimal power flow optimization physical model is set up, and the system parameters and optimization results are solved using a solver to generate a graph dataset containing node features and power flow distribution labels. During the network training phase, the graph dataset is divided into training data and test data. The graph hybrid neural network model is used to train the training data. The graph hybrid neural network model includes different MG layers and a Read layer, which are used to learn historical data and map generator output and node voltage. During the model solving phase, the data of the system to be solved is input, and the trained graph hybrid neural network model is used for data-driven solution. If the solution does not meet the feasibility requirements, the model-driven module is used for secondary solution until the optimal power flow solution is obtained.
2. The power flow calculation method for a transmission system combining wind power and conventional thermal power as described in claim 1, characterized in that: The data generation stage includes setting wind farm and load conditions, using a solver to obtain system parameters and optimization results, and preprocessing the results into graph data, where node features include load, voltage and phase angle, and labels include power flow distribution.
3. The power flow calculation method for a transmission system combining wind power and conventional thermal power as described in claim 2, characterized in that: The data generation stage also includes obtaining graph data with node connection relationships by setting non-generator nodes as the first node type, generator nodes as the second node type, and wind turbine nodes as the third node type.
4. The power flow calculation method for a transmission system combining wind power and conventional thermal power as described in claim 3, characterized in that: The network training phase includes training the training data using an MG layer formed by fusing graph convolutional networks, relational graph neural networks, and residual networks, and a Read layer formed by fusing graph convolutional networks and linear networks, to obtain a graph hybrid neural network model that can output the active and reactive power output of the generator and the voltage amplitude and phase angle of the nodes.
5. The power flow calculation method for a transmission system combining wind power and conventional thermal power as described in claim 4, characterized in that: The graph hybrid neural network model includes a method that uses MSEloss as the main loss function and applies a penalized loss function containing physical information, such as power balance, voltage upper and lower limits, and power upper and lower limits, to approximate the true value. Specifically, in, This is the total loss value. These are the weighting coefficients. Let y be the expected value, i and j be the variable indices, and y be the predicted output value. The actual label value. For physical penalty loss items, This is the deviation loss term. To monitor loss items; The loss function is divided into three parts: supervised feedback using MSEloss, penalized feedback using physical information constraints, and bias feedback for data approximation.
6. The power flow calculation method for a transmission system combining wind power and conventional thermal power as described in claim 5, characterized in that: The data-driven solution process includes checking the feasibility of the outputs of the data-driven module. If all outputs are feasible, the result is taken as the final result. If only the active and reactive power of the generators are feasible, they are substituted into the simplified power flow model to solve for the node voltage magnitude and phase angle. If only the node voltage magnitude and phase angle are feasible, they are substituted into the simplified power flow model to solve for the generator active and reactive power. If all outputs are infeasible, the initial values of the corresponding variables are set as the hot start of the optimal power flow model to obtain the final optimal power flow solution.
7. The power flow calculation method for a transmission system combining wind power and conventional thermal power as described in claim 6, characterized in that, It also includes an evaluation step, which verifies the accuracy and efficiency of the optimal power flow solution by calculating the time saving rate, active power error rate and voltage amplitude error rate; The time saving rate is based on a comparison between the running time of the model solution steps and the running time of the solver: in, The time saving rate is N, where N is the total number of test samples. The runtime of a traditional solver in solving for optimal power flow on a test sample. The runtime of solving the proposed DMGOPF method on the test samples; The active power error rate is based on the comparison between the generator active power output output from the model solution steps and the generator active power output output from the solver: in, The active power error rate, This represents the total number of generators in the system. The active power output of the generator is calculated by a traditional solver. The active power output of the generator is predicted by the DMGOPF method; The voltage amplitude error rate is based on a comparison between the node voltage amplitude output by the model solution step and the node voltage amplitude output by the solver: in, For voltage amplitude error rate, This represents the total number of nodes in the system. The voltage magnitude at the node is calculated by a conventional solver. The voltage amplitude of the node is predicted by the DMGOPF method.
8. A power flow calculation system for a transmission system operating in conjunction with wind power and conventional thermal power, comprising applying the power flow calculation method for a transmission system operating in conjunction with wind power and conventional thermal power as described in any one of claims 1 to 7, characterized in that, include: Data generation and processing module, hybrid-driven computing module, and verification and output module; The data generation and processing module is used to construct an optimal power flow physical model that includes wind farms and conventional thermal power plants, call the solver to generate a graph dataset containing node features and power flow distribution labels, and perform node type classification and data preprocessing. The hybrid drive computing module is connected to the data generation and processing module. It is used to load the trained graph hybrid neural network model, receive the node feature data of the system to be solved, obtain the preliminary generator output and node voltage parameters through the data-driven method, and selectively combine the power flow simplification model to solve the problem based on the feasibility of the parameters. The verification and output module is connected to the hybrid drive calculation module and is used to verify and evaluate the calculated optimal power flow solution, calculate the time saving rate, active power error rate and voltage amplitude error rate, and output the final reliable power flow result.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the power flow calculation method for a transmission system operating in conjunction with wind power and conventional thermal power, as described in any one of claims 1 to 7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the power flow calculation method for a transmission system operating in conjunction with wind power and conventional thermal power, as described in any one of claims 1 to 7.