Power grid model parameter estimation method and system based on physical information graph neural network
Through the grid model parameter estimation method based on physical infographic neural network, the problem of sensitive to measurement errors and manual threshold setting in the existing technology is solved. Combined with graph neural network technology, model topology errors and measurement errors are considered, and the model topology errors and measurement errors are realized. More accurate grid parameter estimation improves the data governance efficiency of the power scheduling and control system.
Patent Information
- Application Number
- CN202411034037.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-30
- Publication Date
- 2024-11-22
AI Technical Summary
The existing power system parameter identification methods are sensitive to measurement errors and manually set thresholds, and fail to effectively consider the impact of model topology errors and measurement errors on parameter estimation, resulting in large deviations in the estimates and low actual availability.
The grid model parameter estimation method based on physical infographic neural network is adopted. By obtaining the grid model data to be screened, the state estimation unqualified points are located, the sub-graph is constructed, the problem nodes are marked, the graph neural network model is trained, and the graph embedding layer and volume are used. Parameter estimation is performed by the accumulated layer and the fully connected layer, combining the Adam optimizer and the MSELoss function optimization model.
It improves the efficiency of model data governance, provides more accurate estimation of parameter problem nodes, enhances the simulation computing power of the power scheduling control system, and reduces the dependence on measurement errors and manually set thresholds.
Smart Images

Figure CN119004965A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of smart grids and relates to a method and system for estimating parameters of a grid model based on a physical information graph neural network. Background Art
[0002] The application of state estimation in power dispatching and control systems is the basic application of network analysis software. With the expansion of the scale of power grid computing, the difficulty of model data management has increased. At present, the power system parameter identification methods can be divided into two categories: the first category is based on sensitivity analysis. This type of method is based on the state estimation results. According to the measurement residual and the sensitivity relationship between the measurement and network parameters, possible network parameter errors are identified. The biggest advantage of this type of method is that there is no need to change the original state estimation program, which is easy to implement. Its disadvantage is that its identification results are very sensitive to measurement errors and some manually set thresholds; the second type of method expands the state variables of state estimation and uses generalized state estimation to calculate the estimated values of network parameters, thereby distinguishing the wrong network parameters in the power system. This type of method has numerical stability problems. The existing methods do not consider the impact of model topology errors and measurement errors on the accuracy of parameter estimation. In scenarios where other problems exist at the same time, the estimated values given have large deviations and low actual availability. Summary of the invention
[0003] The purpose of the present invention is to provide a method and system for estimating power grid model parameters based on physical information graph neural network, aiming to solve the problem that in the process of sensitivity analysis of power system parameters in the prior art, the identification results are very sensitive to measurement errors and some manually set thresholds, and the state variable method of extended state estimation does not consider the impact of model topology errors and measurement errors on parameter estimation accuracy.
[0004] In order to achieve the above object, the present invention adopts the following technical solutions:
[0005] The present invention proposes a method for estimating parameters of a power grid model based on a physical information graph neural network, comprising:
[0006] Acquire power grid model data to be screened;
[0007] Input the power grid model data to be screened into the pre-trained graph neural network parameter estimation model to obtain the parameter problem node;
[0008] Among them, the pre-trained graph neural network parameter estimation model is obtained through the following steps: obtaining state estimation unqualified points in the power grid model data; locating the problem nodes of the equipment near the measurement unqualified points in the state estimation unqualified points, and constructing a subgraph; labeling the types of problem nodes existing in the power grid model data, and screening out the parameter problem nodes in the subgraph as training samples for problem classification; constructing a graph neural network parameter estimation model; training the graph neural network parameter estimation model based on the problem classification training samples to obtain a pre-trained graph neural network parameter estimation model.
[0009] Preferably, the problem nodes of the equipment near the measurement unqualified points in the state estimation unqualified points are located to construct a subgraph, specifically: the unqualified points are collected, a topological search is performed with the unqualified point equipment as the starting point, the number of layers n_st of adjacent plants and stations is set, all the equipment of the adjacent n_st-layer plants and stations is retained to form a subgraph; and the subgraph and equipment parameter measurement information are saved in a file; and all the unqualified points are traversed at the same time until all the unqualified points are traversed to form graph structure data for neural network training.
[0010] Preferably, the problem nodes existing in the power grid model data are labeled by type, and the parameter problem nodes in the subgraph are screened out as training samples for problem classification, specifically: the problem node types include normal nodes, topology problem nodes, measurement problem nodes and parameter problem nodes; the problem nodes existing in the power grid model data are labeled by type including: manual labeling and automatic labeling; the manual labeling is to label the problem nodes existing in the power grid model based on the on-site operation and maintenance experience and generate training samples; the manual labeling information is saved in the label file; the automatic labeling is to search by plant station, if a plant station and the surrounding If there are no unqualified points in the surrounding plant station, the model data will be automatically modified at the plant station, the normal model data will be changed into abnormalities, and model data problems will be generated; the model data problems include deleting some equipment, changing equipment connection points, generating topological problems; modifying equipment measurement or measurement quality positions, generating measurement problems; modifying equipment parameters, generating parameter problems; recording the original parameters and the modified problem parameters, and the original parameters are used as the initial values for subsequent physical information neural network training; performing state estimation calculations on the modified section, and if deviation problems occur near the modified location, the section is used as a problem classification training sample, and the sub-graph file and label file are automatically saved.
[0011] Preferably, the construction of the graph neural network parameter estimation model is specifically as follows: the input dimension of the model is: node feature vector dimension × number of nodes; the output vector dimension is: node parameter dimension × number of nodes; the neural network model includes a graph embedding layer, three convolution and pooling layers, four fully connected layers, and an activation function; the graph embedding layer uses the feature representation of nodes and edges on the graph to capture the relationships and attributes in the graph structure data, converts the graph structure data into vectors of other dimensions, and the vectors are used as inputs to subsequent convolutional layers; the convolution layer is SAGEConv, and multiple SAGEConv graph convolution modules are stacked to construct a multi-layer graph neural network to generate vector representations of nodes; and the number of nodes in the last fully connected layer is reduced through the pooling layer; the fully connected layer connects each node to all nodes in the previous layer to integrate the features extracted previously; the activation function is a tanh function.
[0012] Preferably, the tanh function is:
[0013]
[0014] Here, e represents the base of natural logarithm; the output value of the tanh function is mapped between -1 and 1.
[0015] Preferably, the training samples based on the problem classification are used to train the graph neural network parameter estimation model to obtain a pre-trained graph neural network parameter estimation model, specifically:
[0016] The Adam optimizer is used to optimize the neural network parameters; the data deviation uses the MSELoss function, which is used to calculate the mean square error, specifically:
[0017]
[0018] Where N represents the number of samples, y i represents the true label of the i-th sample, represents the predicted probability of the i-th sample;
[0019] The output of the neural network is used as the initial estimate of the node with parameter problems, and is substituted into the state estimation calculation to obtain the average estimate deviation. The average estimate deviation is added to the loss output by MSELoss to obtain the total loss value. When the total loss value is divided by the total number of data, the average loss is obtained. When the average loss value is less than the preset threshold, training is stopped.
[0020] Preferably, the average estimated deviation is the weighted average of the average estimated deviation of active power, the average estimated deviation of reactive power and the average estimated deviation of voltage amplitude; the average estimated deviation of active power is the sum of the absolute values of the deviations between active power measurement and estimated value / the total number of active power measurements, the average estimated deviation of reactive power is the sum of the absolute values of the deviations between reactive power measurement and estimated value / the total number of reactive power measurements, and the average estimated deviation of voltage amplitude is the sum of the absolute values of the deviations between voltage amplitude measurement and estimated value / the total number of voltage amplitude measurements.
[0021] The present invention proposes a power grid model parameter estimation system based on physical information graph neural network, comprising:
[0022] A data acquisition module, wherein the data acquisition module is used to acquire power grid model data to be screened;
[0023] A data processing module, wherein the data processing module is used to input the power grid model data to be screened into a pre-trained graph neural network parameter estimation model to obtain a parameter problem node;
[0024] Among them, the pre-trained graph neural network parameter estimation model is obtained through the following steps: obtaining state estimation unqualified points in the power grid model data; locating the problem nodes of the equipment near the measurement unqualified points in the state estimation unqualified points, and constructing a subgraph; labeling the types of problem nodes existing in the power grid model data, and screening out the parameter problem nodes in the subgraph as training samples for problem classification; constructing a graph neural network parameter estimation model; training the graph neural network parameter estimation model based on the problem classification training samples to obtain a pre-trained graph neural network parameter estimation model.
[0025] A terminal device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of a power grid model parameter estimation method based on a physical information graph neural network are implemented.
[0026] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a power grid model parameter estimation method based on a physical information graph neural network.
[0027] Compared with the prior art, the present invention has the following beneficial effects:
[0028] The present invention proposes a method for estimating parameters of a power grid model based on a physical information graph neural network. The method locates the problem nodes of the equipment near the measurement failure points in the state estimation failure points, and constructs a subgraph for neural network training; the problem nodes existing in the power grid model data are marked by type, and the parameter problem nodes in the subgraph are screened out as training samples for problem classification; based on the problem classification training samples, the graph neural network parameter estimation model is trained to obtain a pre-trained graph neural network parameter estimation model; based on the pre-trained graph neural network parameter estimation model, the power grid model data containing the problem nodes is screened to obtain the parameter problem nodes. The present invention combines a large amount of state estimation operation data and operation and maintenance experience, and adds state estimation algorithm constraints to the neural network based on the physical information graph neural network technology, which can improve the efficiency of model data management, provide more accurate model data for the simulation calculation of the power dispatching control system, and realize intelligent estimation of parameter problem nodes.
[0029] The invention proposes a power grid model parameter estimation system based on physical information graph neural network, which divides the system into a data acquisition module and a data processing module, and finally realizes the power grid model parameter estimation based on physical information graph neural network. The modularization concept is adopted to make each module independent of each other, which is convenient for unified management of each module. BRIEF DESCRIPTION OF THE DRAWINGS
[0030] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.
[0031] Figure 1 It is a flow chart of the method for estimating parameters of a power grid model based on a physical information graph neural network of the present invention.
[0032] Figure 2 It is a detailed flow chart of the power grid model parameter estimation method based on physical information graph neural network of the present invention.
[0033] Figure 3 It is a schematic diagram of the model data processing flow of the present invention;.
[0034] Figure 4 It is a single-line diagram of a 5-node system of the present invention.
[0035] Figure 5 It is a schematic diagram of the structure of a 5-node system diagram of the present invention.
[0036] Figure 6Schematic diagram of features and node labels for problem classification according to the present invention.
[0037] Figure 7 It is a schematic diagram of the loss function of the physical information graph neural network of the present invention.
[0038] Figure 8 It is a schematic diagram of the average estimation deviation of the present invention.
[0039] Fig. 9 It is a structural schematic diagram of the power grid model parameter estimation system based on physical information graph neural network of the present invention.
[0040] Fig.10 The figure is a schematic diagram of the structure of an electronic device of the present invention. DETAILED DESCRIPTION
[0041] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.
[0042] 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 ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0043] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.
[0044] In the description of the embodiments of the present invention, it should be noted that if the terms "upper", "lower", "horizontal", "inner", etc. indicate an orientation or positional relationship based on the orientation or positional relationship shown in the drawings, or the orientation or positional relationship in which the product of the invention is usually placed when in use, it is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. In addition, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.
[0045] The present invention is further described in detail below in conjunction with the accompanying drawings:
[0046] Example 1
[0047] The present invention proposes a method for estimating parameters of a power grid model based on a physical information graph neural network, such as Figure 1 As shown, including:
[0048] S1. Obtaining power grid model data to be screened;
[0049] S2. Input the power grid model data to be screened into a pre-trained graph neural network parameter estimation model to obtain parameter problem nodes;
[0050] Among them, the pre-trained graph neural network parameter estimation model is obtained through the following steps:
[0051] S201, obtaining state estimation unqualified points in power grid model data;
[0052] The power grid model data includes: steady-state parameters of power grid equipment, topological connection relationships and measurement data of equipment during operation; unqualified points in state estimation include: topological problems, measurement problems and parameter problems of power grid model data.
[0053] S202, locate the problem node of the equipment near the measurement failure point in the state estimation failure point, and construct a subgraph;
[0054] The problem nodes of the equipment near the measurement unqualified points in the state estimation unqualified points are located, and a subgraph is constructed. Specifically, the unqualified points are collected, and a topological search is performed with the unqualified point equipment as the starting point, and the number of layers n_st of adjacent plants and stations is set, and all the equipment of the adjacent n_st-layer plants and stations is retained to form a subgraph; and the subgraph and equipment parameter measurement information are saved in a file; and all unqualified points are traversed at the same time until all unqualified points are traversed, so as to form graph structure data for neural network training.
[0055] S203, marking the types of problem nodes in the power grid model data, and selecting the parameter problem nodes in the subgraph as training samples for problem classification;
[0056] The problem nodes in the power grid model data are labeled by type, and the parameter problem nodes in the subgraph are selected as training samples for problem classification, specifically:
[0057] The problem node types include normal nodes, topology problem nodes, measurement problem nodes and parameter problem nodes; type labeling of problem nodes in power grid model data includes: manual labeling and automatic labeling; the manual labeling is to label the problem nodes in the power grid model based on on-site operation and maintenance experience to generate training samples; the manual labeling information is saved in the label file; the automatic labeling is: search according to the plant station, if there are no unqualified points in a plant station and the surrounding plants and stations, the model data is automatically modified at the plant station, the normal model data is changed to abnormal, and the model data problem is generated; the model data problem includes deleting some equipment, changing the equipment connection point, generating topology problems; modifying equipment measurement or measurement quality bit, generating measurement problems; modifying equipment parameters, generating parameter problems; recording the original parameters and the modified problem parameters, and the original parameters are used as the initial values for subsequent physical information neural network training; performing state estimation calculation on the modified section, and a large deviation problem occurs near the modification point, the section is used as a problem classification training sample, and the subgraph file and label file are automatically saved.
[0058] S204, constructing a graph neural network parameter estimation model suitable for graph node classification tasks;
[0059] The graph neural network parameter estimation model suitable for graph node classification tasks is constructed as follows: the input dimension of the model is: node feature vector dimension × number of nodes; the output vector dimension is: node parameter dimension × number of nodes; the neural network model includes a graph embedding layer, three convolution and pooling layers, four fully connected layers, and an activation function; the graph embedding layer uses the feature representation of nodes and edges on the graph to capture the relationships and attributes in the graph structure data, converts the graph structure data into vectors of other dimensions, and the vectors are used as inputs of subsequent convolutional layers; the convolution layer is SAGEConv, and multiple SAGEConv graph convolution modules are stacked to construct a multi-layer graph neural network to generate vector representations of nodes; and the number of nodes in the last fully connected layer is reduced through the pooling layer; the fully connected layer connects each node to all nodes in the previous layer to integrate the features extracted from the previous edge; the activation function is a tanh function.
[0060] The tanh function is:
[0061]
[0062] Here, e represents the base of natural logarithm; the output value of the tanh function is mapped between -1 and 1.
[0063] S205. Train the graph neural network parameter estimation model based on the training samples of the problem classification to obtain a pre-trained graph neural network parameter estimation model;
[0064] The training samples based on the problem classification are used to train the graph neural network parameter estimation model to obtain a pre-trained graph neural network parameter estimation model, specifically:
[0065] The Adam optimizer is used to optimize the neural network parameters; the data deviation uses the MSELoss function, which is used to calculate the mean square error, specifically:
[0066]
[0067] Where N represents the number of samples, y i represents the true label of the i-th sample, represents the predicted probability of the i-th sample;
[0068] The output of the neural network is used as the initial estimate of the node with parameter problems, and is substituted into the state estimation calculation to obtain the average estimate deviation. The average estimate deviation is added to the loss output by MSELoss to obtain the total loss value. When the total loss value is divided by the total number of data, the average loss is obtained. When the average loss value is less than the preset threshold, training is stopped.
[0069] The average estimated deviation is the weighted average of the average estimated deviation of active power, the average estimated deviation of reactive power and the average estimated deviation of voltage amplitude; the average estimated deviation of active power is the sum of the absolute values of the deviations between active power measurement and estimated value / the total number of active power measurements, the average estimated deviation of reactive power is the sum of the absolute values of the deviations between reactive power measurement and estimated value / the total number of reactive power measurements, and the average estimated deviation of voltage amplitude is the sum of the absolute values of the deviations between voltage amplitude measurement and estimated value / the total number of voltage amplitude measurements.
[0070] like Figure 2The present invention discloses a method for estimating parameters of a power grid model based on a physical information graph neural network, which includes: firstly locating the problem of the equipment near the state estimation measurement unqualified point (the point where the measured value and the estimated value have a large deviation), requiring the local network topology and parameter information near the unqualified point, taking the state estimation unqualified point as the center, searching outward according to a given depth, and forming a subgraph for neural network training; based on the field operation and maintenance experience and automatic problem generation, the type of the problem nodes in the power grid model is marked, and training samples are generated. A graph neural network parameter estimation model for intelligent classification of power grid model data problems is established, and the graph neural network parameter estimation model is trained by a supervised learning method, and the nodes are divided into four categories: normal nodes, topology problem nodes, measurement problem nodes and parameter problem nodes. After locating the node with parameter problems, other problems (topology problems, measurement problems) in the subgraph are excluded to avoid the influence of other problems (topology problems, measurement problems) on parameter estimation, and improve the accuracy of parameter estimation. A physical information graph neural network is used to add state estimation deviation constraints in network training, improve the accuracy of the training model, and realize intelligent estimation of parameter problem nodes.
[0071] The problem of the equipment near the measurement failure point in the state estimation is located, and the subgraph for neural network training is constructed.
[0072] See also Figure 3 , Figure 4 , Figure 5 and Figure 6 , model data classification problems and parameter estimation problems can share the same graph structure data. State estimation only diagnoses problems of equipment near measurement failure points (points where the measured value deviates greatly from the estimated value). Therefore, only the local network topology, parameters and measurement information near the failure point are needed. With the state estimation failure point as the center, search outward at a given depth to form graph structure data for neural network training.
[0073] Node characteristics include device ID, voltage level ID, device parameters, device measurements, and measurement estimates.
[0074] The equipment parameters include resistance, reactance, susceptance, tap type, rated voltage, rated capacity, etc., and the equipment measurements include active power, reactive power, voltage, gear, etc.
[0075] The node labels used to locate model data problems are divided into four categories: normal nodes, topology problem nodes, measurement problem nodes, and parameter problem nodes.
[0076] The problem nodes in the power grid model data are labeled by type, and the parameter problem nodes in the subgraph are screened out as training samples for problem classification.
[0077] Based on the normal section, model data problems are automatically generated, including deleting some equipment, changing equipment connection points, generating topology problems; modifying equipment measurement or measurement quality positions, generating measurement problems; modifying equipment parameters, generating parameter problems.
[0078] For parameter problems, record the original parameters and the modified problem parameters. The original parameters can be used as the initial values for subsequent physical information neural network training. Perform state estimation calculations on the modified section. If a large deviation problem occurs near the modified area, use the section as a problem classification training sample and automatically save the sub-graph file and label file. Solve the problem of insufficient actual samples by automatically generating samples. Generate training samples. The manually labeled information is saved in the label file. The format is shown in Table 1:
[0079] Table 1
[0080]
[0081] Construct a graph neural network parameter estimation model suitable for graph node classification tasks;
[0082] For parameter problems, determine whether there are other problems in the subgraph. If there are other problems, you need to notify the operation and maintenance personnel to solve the existing topological and measurement problems first. If there are no other problems, continue to use the following neural network model for parameter estimation.
[0083] Construct a neural network model suitable for graph node classification tasks. The input dimension of the model is: node feature vector dimension × number of nodes; the output vector dimension is: node parameter dimension × number of nodes. The neural network model mainly includes a graph embedding layer (embedding), three convolution and pooling layers, four fully connected layers (Linear), and an activation function.
[0084] The graph embedding layer can use the feature representations of nodes and edges on the graph to capture the relationships and attributes in the graph structured data and convert the graph structured data into vectors of other dimensions, which can be used as input for subsequent convolutional layers.
[0085] The convolutional layer is SAGEConv, which is derived from an algorithm called GraphSAGE (Graph Sample and Aggregation), which is designed to effectively learn node features on graph data. The core idea of SAGEConv is to update the feature representation of each node by aggregating the information of neighboring nodes.
[0086] Multiple SAGEConv graph convolution modules are stacked to build a multi-layer graph neural network (more layers mean more neighbor information can be accessed) to generate vector representations of nodes. According to experience, a certain failure is closely related to model parameter problems of nearby devices within a range of less than three branches, so three graph convolutions are performed here.
[0087] The number of nodes in the final fully connected layer is reduced through the pooling layer. In a fully connected layer, each node is connected to all nodes in the previous layer to combine the features extracted previously. The parameter estimation problem has a high demand for fully connected layers. Four fully connected layers are used here, and the dimension of the fully connected layer is higher than that of the classification problem.
[0088] The activation function is the tanh function, and its mathematical definition is:
[0089]
[0090] Where e represents the base of natural logarithm. The output value of tanh function is mapped between -1 and 1. Tanh function has many advantages in deep learning:
[0091] (1) Solving nonlinear problems: Due to the S-shaped curve characteristics of the tanh function, it can produce nonlinear output within the input range, which helps the deep learning model better fit nonlinear data and provides powerful nonlinear expression capabilities.
[0092] (2) Gradient vanishing problem: Compared with the sigmoid function, the tanh function is centered at 0, which helps to reduce the gradient vanishing problem that occurs during backpropagation in deep neural networks. The gradient vanishing problem may cause the model to converge more slowly.
[0093] (3) Bilaterality: The tanh function is symmetric around the origin (x = 0), i.e., tanh(-x) = -tanh(x). This property is particularly useful in certain tasks, especially when both positive and negative weights need to be considered.
[0094] (4) Continuous and differentiable: The tanh function is continuously differentiable, which allows the use of derivative-based methods such as gradient descent during optimization. Therefore, when the tanh function is used as an activation function, these optimization algorithms can be used to optimize the model.
[0095] Based on the training samples of problem classification, the graph neural network parameter estimation model is trained to obtain a pre-trained graph neural network parameter estimation model.
[0096] join Figure 7In the early application of state estimation, the measurement accuracy of power plants was not high and the delay was obvious. The telemetry pass rate was mainly used to evaluate the measurement quality and the level of state estimation maintenance and application. In recent years, the measurement accuracy and real-time performance of power plants have generally improved. The traditional pass rate involves subjective factors such as benchmark values and error percentage definitions, and the indicators have gradually become blunt.
[0097] Starting from the actual data needs of power grid calculation and analysis, the focus is on the deviation of measured values and calculated values, and the assessment indicators also focus on the measurement deviation evaluation indicators. The present invention mainly considers the average value of power deviation and the average value of voltage deviation. The advantages of the measurement deviation indicator are that it is simple and direct, has physical meaning, is easily accepted by primary power grid operators, and can guide various units to continuously improve the consistency and accuracy of measured values and calculated values.
[0098] See also Figure 8 , average active power estimation deviation = sum of absolute values of deviations between active power measurement and estimated value / total number of active power measurements, average reactive power estimation deviation = sum of absolute values of deviations between reactive power measurement and estimated value / total number of reactive power measurements, average voltage amplitude estimation deviation = sum of absolute values of deviations between voltage amplitude measurement and estimated value / total number of voltage amplitude measurements. The average estimation deviation is equal to the weighted average of the three estimation deviations.
[0099] The Adam optimizer is used to optimize the neural network parameters.
[0100] Data deviation uses the MSELoss function, which is used to calculate the mean square error, a loss function commonly used in regression problems, which is used to measure the difference between the target value and the predicted value.
[0101]
[0102] Where N represents the number of samples, y i represents the true label of the i-th sample, represents the predicted probability of the i-th sample.
[0103] The output of the neural network is used as the initial estimate of the node with parameter problems, and is substituted into the state estimation calculation to find the average estimate deviation. The average estimate deviation is added to the loss output by MSELoss to obtain the total loss value.
[0104] In order to improve the performance, the training set is processed in batches, each batch contains 64 sets of test data, each set of data is input into the neural network for parallel calculation at the same time, and the output results and labels are used for MSELoss loss calculation; at the same time, the corresponding model parameters are replaced by the output results, and the state estimation calculation is performed on the model data in the subgraph to calculate the average estimation deviation, and the average estimation deviation is added to the loss output by MSELoss to obtain the total loss value of the set of data; finally, the average loss of 64 sets of data is calculated; when the average loss value is less than the threshold value (according to experience, the threshold value can be equal to 10), stop training.
[0105] Based on the pre-trained graph neural network parameter estimation model, the power grid mode data containing problem nodes are screened to obtain parameter problem nodes.
[0106] Example 2
[0107] The present invention proposes a power grid model parameter estimation system based on physical information graph neural network, such as Fig. 9 As shown, it includes a data acquisition module and a data processing module;
[0108] The data acquisition module is used to acquire the power grid model data to be screened;
[0109] The data processing module is used to input the power grid model data to be screened into the pre-trained graph neural network parameter estimation model to obtain the parameter problem node;
[0110] Among them, the pre-trained graph neural network parameter estimation model is obtained through the following steps:
[0111] Obtaining state estimation failure points in power grid model data;
[0112] The power grid model data includes: steady-state parameters of power grid equipment, topological connection relationships and measurement data of equipment during operation; unqualified points in state estimation include: topological problems, measurement problems and parameter problems of power grid model data.
[0113] Locate the problem nodes of the equipment near the measurement failure points in the state estimation failure points and construct a subgraph;
[0114] The problem nodes of the equipment near the measurement unqualified points in the state estimation unqualified points are located, and a subgraph is constructed. Specifically, the unqualified points are collected, and a topological search is performed with the unqualified point equipment as the starting point, and the number of layers n_st of adjacent plants and stations is set, and all the equipment of the adjacent n_st-layer plants and stations is retained to form a subgraph; and the subgraph and equipment parameter measurement information are saved in a file; and all unqualified points are traversed at the same time until all unqualified points are traversed, so as to form graph structure data for neural network training.
[0115] Label the types of problem nodes in the power grid model data, and select the parameter problem nodes in the subgraph as training samples for problem classification;
[0116] The problem nodes in the power grid model data are labeled by type, and the parameter problem nodes in the subgraph are selected as training samples for problem classification, specifically:
[0117] The problem node types include normal nodes, topology problem nodes, measurement problem nodes and parameter problem nodes; type labeling of problem nodes in the power grid model includes: manual labeling and automatic labeling; the manual labeling is to label the problem nodes in the power grid model based on on-site operation and maintenance experience to generate training samples; the manual labeling information is saved in the label file; the automatic labeling is: search according to the plant station, if a plant station and the surrounding plants and stations have no unqualified points, then the model data is automatically modified at the plant station, the normal model data is changed to abnormal, and the model data problem is generated; the model data problem includes deleting some equipment, changing the equipment connection point, generating topology problems; modifying equipment measurement or measurement quality bit, generating measurement problems; modifying equipment parameters, generating parameter problems; recording the original parameters and the modified problem parameters, and the original parameters are used as the initial values for subsequent physical information neural network training; performing state estimation calculation on the modified section, and a large deviation problem occurs near the modification point, the section is used as a problem classification training sample, and the subgraph file and label file are automatically saved.
[0118] Construct a graph neural network parameter estimation model suitable for graph node classification tasks;
[0119] The graph neural network parameter estimation model suitable for graph node classification tasks is constructed as follows: the input dimension of the model is: node feature vector dimension × number of nodes; the output vector dimension is: node parameter dimension × number of nodes; the neural network model includes a graph embedding layer, three convolution and pooling layers, four fully connected layers, and an activation function; the graph embedding layer uses the feature representation of nodes and edges on the graph to capture the relationships and attributes in the graph structure data, converts the graph structure data into vectors of other dimensions, and the vectors are used as inputs of subsequent convolutional layers; the convolution layer is SAGEConv, and multiple SAGEConv graph convolution modules are stacked to construct a multi-layer graph neural network to generate vector representations of nodes; and the number of nodes in the last fully connected layer is reduced through the pooling layer; the fully connected layer connects each node to all nodes in the previous layer to integrate the features extracted from the previous edge; the activation function is a tanh function.
[0120] The tanh function is:
[0121]
[0122] Here, e represents the base of natural logarithm; the output value of the tanh function is mapped between -1 and 1.
[0123] Train the graph neural network parameter estimation model based on the training samples of the problem classification to obtain a pre-trained graph neural network parameter estimation model;
[0124] The training samples based on the problem classification are used to train the graph neural network parameter estimation model to obtain a pre-trained graph neural network parameter estimation model, specifically:
[0125] The Adam optimizer is used to optimize the neural network parameters; the data deviation uses the MSELoss function, which is used to calculate the mean square error, specifically:
[0126]
[0127] Where N represents the number of samples, y i represents the true label of the i-th sample, represents the predicted probability of the i-th sample;
[0128] The output of the neural network is used as the initial estimate of the node with parameter problems, and is substituted into the state estimation calculation to obtain the average estimate deviation. The average estimate deviation is added to the loss output by MSELoss to obtain the total loss value. When the total loss value is divided by the total number of data, the average loss is obtained. When the average loss value is less than the preset threshold, training is stopped.
[0129] The average estimated deviation is the weighted average of the average estimated deviation of active power, the average estimated deviation of reactive power and the average estimated deviation of voltage amplitude; the average estimated deviation of active power is the sum of the absolute values of the deviations between active power measurement and estimated value / the total number of active power measurements, the average estimated deviation of reactive power is the sum of the absolute values of the deviations between reactive power measurement and estimated value / the total number of reactive power measurements, and the average estimated deviation of voltage amplitude is the sum of the absolute values of the deviations between voltage amplitude measurement and estimated value / the total number of voltage amplitude measurements.
[0130] Example 3
[0131] See also Fig.10 As shown, the present invention also provides an electronic device 100 for a power grid model parameter estimation method based on a physical information graph neural network; the electronic device 100 includes a memory 101, at least one processor 102, a computer program 103 stored in the memory 101 and executable on the at least one processor 102, and at least one communication bus 104.
[0132] The memory 101 can be used to store the computer program 103, and the processor 102 implements the steps of the power grid model parameter estimation method based on physical information graph neural network described in Example 1 by running or executing the computer program stored in the memory 101 and calling the data stored in the memory 101. The memory 101 can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area can store data (such as audio data) created according to the use of the electronic device 100, etc. In addition, the memory 101 can include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other non-volatile solid-state storage devices.
[0133] The at least one processor 102 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The processor 102 may be a microprocessor or any conventional processor, etc. The processor 102 is the control center of the electronic device 100, and uses various interfaces and lines to connect various parts of the entire electronic device 100.
[0134] The memory 101 in the electronic device 100 stores a plurality of instructions to implement a method for estimating parameters of a power grid model based on a physical information graph neural network, and the processor 102 can execute the plurality of instructions to implement:
[0135] Acquire power grid model data to be screened;
[0136] Input the power grid model data to be screened into the pre-trained graph neural network parameter estimation model to obtain the parameter problem node;
[0137] Among them, the pre-trained graph neural network parameter estimation model is obtained through the following steps: obtaining state estimation unqualified points in the power grid model data; locating the problem nodes of the equipment near the measurement unqualified points in the state estimation unqualified points, and constructing a subgraph; labeling the types of problem nodes existing in the power grid model data, and screening out the parameter problem nodes in the subgraph as training samples for problem classification; constructing a graph neural network parameter estimation model; training the graph neural network parameter estimation model based on the problem classification training samples to obtain a pre-trained graph neural network parameter estimation model.
[0138] Example 4
[0139] If the module / unit integrated in the electronic device 100 is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the process in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device that can carry the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory and read-only memory (ROM, Read-Only Memory).
[0140] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0141] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0142] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0143] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0144] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.
Claims
1. A method for estimating parameters of a power grid model based on a physical information graph neural network, characterized in that: include: Acquire power grid model data to be screened; Input the power grid model data to be screened into the pre-trained graph neural network parameter estimation model to obtain the parameter problem node; Among them, the pre-trained graph neural network parameter estimation model is obtained through the following steps: obtaining state estimation unqualified points in the power grid model data; locating the problem nodes of the equipment near the measurement unqualified points in the state estimation unqualified points, and constructing a subgraph; labeling the types of problem nodes existing in the power grid model data, and screening out the parameter problem nodes in the subgraph as training samples for problem classification; constructing a graph neural network parameter estimation model; training the graph neural network parameter estimation model based on the problem classification training samples to obtain a pre-trained graph neural network parameter estimation model.
2. The method for estimating power grid model parameters based on physical information graph neural network according to claim 1 is characterized in that: The problem nodes of the equipment near the measurement unqualified points in the state estimation unqualified points are located, and a subgraph is constructed. Specifically, the unqualified points are collected, and a topological search is performed with the unqualified point equipment as the starting point, and the number of layers n_st of adjacent plants and stations is set, and all the equipment of the adjacent n_st-layer plants and stations is retained to form a subgraph; and the subgraph and equipment parameter measurement information are saved in a file; and all unqualified points are traversed at the same time until all unqualified points are traversed, so as to form graph structure data for neural network training.
3. The method for estimating power grid model parameters based on physical information graph neural network according to claim 1, characterized in that: The problem nodes in the power grid model data are labeled by type, and the parameter problem nodes in the subgraph are screened out as training samples for problem classification. Specifically, the problem node types include normal nodes, topology problem nodes, measurement problem nodes and parameter problem nodes. The problem nodes in the power grid model data are labeled by type, including manual labeling and automatic labeling. The manual labeling is based on the on-site operation and maintenance experience to label the problem nodes in the power grid model and generate training samples. The manual labeling information is saved in the label file. The automatic labeling is: search by plant station, if a plant station and surrounding plants If there is no unqualified point at the station, the model data will be automatically modified at the station, the normal model data will be changed into abnormality, and model data problems will be generated; the model data problems include deleting some equipment, changing equipment connection points, generating topological problems; modifying equipment measurement or measurement quality bits, generating measurement problems; modifying equipment parameters, generating parameter problems; recording the original parameters and the modified problem parameters, and the original parameters are used as the initial values for subsequent physical information neural network training; performing state estimation calculations on the modified section, and if a deviation problem occurs near the modified location, the section is used as a problem classification training sample, and the sub-graph file and label file are automatically saved.
4. The method for estimating power grid model parameters based on physical information graph neural network according to claim 1, characterized in that: The construction of the graph neural network parameter estimation model is specifically as follows: the input dimension of the model is: node feature vector dimension × number of nodes; the output vector dimension is: node parameter dimension × number of nodes; the neural network model includes a graph embedding layer, three convolution and pooling layers, four fully connected layers, and an activation function; the graph embedding layer uses the feature representation of nodes and edges on the graph to capture the relationship and attributes in the graph structure data, converts the graph structure data into vectors of other dimensions, and the vectors are used as inputs of subsequent convolutional layers; the convolutional layer is SAGEConv, and multiple SAGEConv graph convolution modules are stacked to construct a multi-layer graph neural network to generate vector representations of nodes; and the number of nodes in the last fully connected layer is reduced through the pooling layer; the fully connected layer connects each node to all nodes in the previous layer to integrate the features extracted from the previous edge; The activation function is the tanh function.
5. The method for estimating power grid model parameters based on physical information graph neural network according to claim 4 is characterized in that: The tanh function is: Here, e represents the base of natural logarithm; the output value of the tanh function is mapped between -1 and 1.
6. The method for estimating power grid model parameters based on physical information graph neural network according to claim 1, characterized in that: The training samples based on the problem classification are used to train the graph neural network parameter estimation model to obtain a pre-trained graph neural network parameter estimation model, specifically: The Adam optimizer is used to optimize the neural network parameters; the data deviation uses the MSELoss function, which is used to calculate the mean square error, specifically: Where N represents the number of samples, yi represents the true label of the i-th sample, represents the predicted probability of the i-th sample; The output of the neural network is used as the initial estimate of the node with parameter problems, and is substituted into the state estimation calculation to obtain the average estimate deviation. The average estimate deviation is added to the loss output by MSELoss to obtain the total loss value. When the total loss value is divided by the total number of data, the average loss is obtained. When the average loss value is less than the preset threshold, training is stopped.
7. The method for estimating power grid model parameters based on physical information graph neural network according to claim 6, characterized in that: The average estimated deviation is the weighted average of the average estimated deviation of active power, the average estimated deviation of reactive power and the average estimated deviation of voltage amplitude; the average estimated deviation of active power is the sum of the absolute values of the deviations between active power measurement and estimated value / the total number of active power measurements, the average estimated deviation of reactive power is the sum of the absolute values of the deviations between reactive power measurement and estimated value / the total number of reactive power measurements, and the average estimated deviation of voltage amplitude is the sum of the absolute values of the deviations between voltage amplitude measurement and estimated value / the total number of voltage amplitude measurements.
8. A power grid model parameter estimation system based on physical information graph neural network, characterized in that: include: A data acquisition module, wherein the data acquisition module is used to acquire power grid model data to be screened; A data processing module, wherein the data processing module is used to input the power grid model data to be screened into a pre-trained graph neural network parameter estimation model to obtain a parameter problem node; Among them, the pre-trained graph neural network parameter estimation model is obtained through the following steps: obtaining state estimation unqualified points in the power grid model data; locating the problem nodes of the equipment near the measurement unqualified points in the state estimation unqualified points, and constructing a subgraph; labeling the types of problem nodes existing in the power grid model data, and screening out the parameter problem nodes in the subgraph as training samples for problem classification; constructing a graph neural network parameter estimation model; training the graph neural network parameter estimation model based on the problem classification training samples to obtain a pre-trained graph neural network parameter estimation model.
9. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the power grid model parameter estimation method based on physical information graph neural network as claimed in any one of claims 1 to 7 are implemented.
10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the power grid model parameter estimation method based on physical information graph neural network as claimed in any one of claims 1 to 7 are implemented.
Citation Information
Cited By
Risk monitoring and early warning platform for power grid engineering
CN119602495A
Shear wall design method based on graph node classification graph neural network
CN119622886A
Gas molecule insulation strength prediction method and device based on graph attention neural network, terminal equipment and storage medium
CN119626349A
Collaborative analysis method, system and equipment for main and distribution networks and storage medium
CN121614813A
