Power system security and stability margin prediction method and device based on graph neural network

Abstract

The present invention provides a method and device for predicting the safety and stability margin of a power system based on a graph neural network. The method includes the following steps: constructing an analysis model based on the power system and obtaining key transmission lines based on transmission line impedance and transmission line voltage levels; initializing the analysis model; obtaining the current active transmission power of the key transmission lines under different operating conditions; presetting faults for each transmission line to obtain the limit transmission power according to stability criteria; and obtaining the safety and stability margin of the operating conditions based on the limit transmission power and the current active transmission power; forming a graph dataset based on the analysis model and the safety and stability margin value set, and training the graph neural network to obtain a graph neural network model, and then analyzing the power system to obtain a predicted safety and stability margin value. The present invention provides a fast and accurate method for real-time analysis of power system status and an intuitive safety and stability indicator for power system dispatching and control.

Description

Technical Field

[0001] The present invention relates to the field of power system safety technology, and in particular to a method and device for predicting power system safety and stability margin based on graph neural network. Background Art

[0002] The high proportion of power electronic equipment integrated into the power grid, the increasing penetration of renewable energy, and the random diversity of load behavior have led to the continuous expansion of the power grid and the complexity and variability of its operation. The safety and stability boundaries of the power system are constantly changing, and the safety and stability mechanisms of modern power systems are becoming increasingly complex. The integration of a high proportion of renewable energy and the increase in power electronic dynamic response equipment have led to problems such as power flow reversal, increased source-load uncertainty, and reduced system inertia. This has led to a more complex interweaving of frequency security, voltage stability, and power angle stability, posing serious challenges to the online assessment and operational control of power grid safety and stability. As the proportion of power electronic equipment, such as renewable energy, continues to increase, models based on time-domain simulation methods are becoming increasingly complex and computationally complex. This reduces the timeliness of time-domain simulation methods. Rapid changes in renewable energy or load power, coupled with discrepancies between the actual grid-related performance of computer groups and offline parameters, significantly impact the accuracy of time-domain simulation results.

[0003] Currently, power system transient stability assessment methods are categorized into time-domain simulation, direct methods, and data-driven approaches. Time-domain simulation, as the mainstream approach for power system transient stability assessment, is typically used for offline calculations. Its modeling accuracy is proportional to its computational accuracy and time consumption, but it faces accuracy and speed challenges in online applications. Direct methods determine system instability by establishing an energy function and comparing the energy value at fault removal with the critical energy value. However, it is difficult to establish a suitable energy function for a specific power grid, and the simplified physical model used in its application often produces conservative results. Summary of the Invention

[0004] In response to the problems existing in the prior art, the present invention provides a method for predicting the safety and stability margin of a power system based on a graph neural network, which can quickly provide intuitive safety and stability indicators and make accurate predictions.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a method for predicting power system security and stability margin based on graph neural network, comprising the following steps:

[0006] constructing an analysis model based on the power system, wherein the analysis model includes a generator, a busbar, and a transmission line;

[0007] obtaining a historical data set of the power system, transmission line impedance, and transmission line voltage level according to the analysis model;

[0008] Obtaining a key transmission line according to the transmission line impedance and the transmission line voltage level, wherein the key transmission line is the transmission line having the lowest transmission line impedance and the highest transmission line voltage level;

[0009] Obtain load variation range, preset variation step, generator output range, preset fault and stability criteria;

[0010] Initializing the analysis model according to the load variation range, the preset variation step and the generator output range;

[0011] Within the generator output range, changing the load of the analysis model within the load variation range with the preset variation step to obtain an operating condition;

[0012] Based on the operating conditions, obtaining the current active transmission power of the key transmission line according to the analysis model;

[0013] Under the operating conditions, respectively performing a preset fault on each of the transmission lines;

[0014] Based on the preset fault, a limit transmission power is obtained according to the stability criterion, where the limit transmission power is the current active transmission power of the critical transmission line corresponding to the extreme stable operating condition; the load corresponding to the extreme stable operating condition is increased by the preset change step size equal to the load corresponding to the first unstable operating condition;

[0015] Obtaining a safety and stability margin of the operating condition according to the limit transmission power and the current active transmission power;

[0016] Obtaining a safety and stability margin value set according to the safety and stability margin;

[0017] Forming a graph dataset and constructing an initial graph neural network model based on the analysis model, the historical dataset, the transmission line impedance, and the safety and stability margin value set;

[0018] Training the initial graph neural network model based on the graph dataset, and optimizing the parameters of the initial graph neural network model using the Adam algorithm to obtain a graph neural network model;

[0019] The power system is analyzed according to the graph neural network model to obtain a safety and stability margin prediction value.

[0020] Furthermore, the step of obtaining the limit transmission power is:

[0021] Step S101: Initializing the analysis model so that the load level of the analysis model reaches the preset load level, and when the load level is changed with the preset change step size, the generator output matches the current load and is within the generator output range;

[0022] Step S102: setting the current load as the initial load, where the initial load is the lower limit of the load variation range;

[0023] Step S103: obtaining the operating condition according to the current load, and obtaining the current active transmission power of the key transmission line based on the operating condition;

[0024] Step S104: Under the operating condition, performing the preset fault detection on each of the transmission lines;

[0025] Step S105: Based on the preset fault, obtain a stability mark according to the stability criterion, wherein the stability mark indicates whether each of the transmission lines meets the stability criterion under the operating condition;

[0026] Step S106: Increase the load according to the preset change step to obtain the current load update value:

[0027] If the current load update value is not greater than the upper limit of the load variation range, the current load update value is set equal to the current load, and steps S103 to S106 are repeated;

[0028] If the current load update value is greater than the upper limit of the load variation range, obtaining a working condition stability table according to the stability mark;

[0029] Step S107: obtaining the extreme stable operating condition according to the operating condition stability table, and obtaining the extreme transmission power according to the extreme stable operating condition.

[0030] Furthermore, the preset fault is to cut off the corresponding transmission line after a three-phase short circuit fault occurs on each transmission line and lasts for a preset period of time; the stability criterion is:

[0031] During the three-phase short circuit fault, the maximum power angle difference of the generator cannot exceed 360°;

[0032] Also, starting from the time when the three-phase short circuit fault occurs, the minimum voltage of the generator is not lower than a first preset value and the duration does not exceed the first preset time, and the minimum frequency of the generator is not lower than a second preset value and the duration does not exceed the second preset time.

[0033] Further, the stability marker includes 1 or 0;

[0034] If, under the operating condition, each of the transmission lines satisfies the stability criterion, the stability mark of the operating condition is 1;

[0035] Otherwise, the stability of the operating condition is marked as 0.

[0036] Furthermore, the step of obtaining the limit transmission power is:

[0037] When the load is increased with the preset change step within the load change range, the first unstable operating condition is obtained according to the operating condition corresponding to the first appearance of the stability mark 0;

[0038] Obtaining a first instability load according to the first instability operating condition;

[0039] According to the first unstable load, the preset change step is reduced to obtain an ultimate stable operating load;

[0040] The limit transmission power is obtained according to the limit stable operation load.

[0041] Furthermore, the load variation range is 0.9 to 1.1, and the preset variation step is 0.01.

[0042] Furthermore, the safety and stability margin is:

[0043]

[0044] Where M i is the safety and stability margin of the i-th operating condition, P max is the maximum transmission power, P i is the current active transmission power under the i-th operating condition, and i is the operating condition number.

[0045] Furthermore, the historical data set includes a bus power characteristic value set; and the steps of forming the graph data set are:

[0046] constructing an adjacency matrix of a graph according to the analysis model, wherein the nodes of the graph are the buses and the edges of the graph are the transmission lines;

[0047] Obtaining a node characteristic matrix according to the busbar power characteristic value set, and obtaining an edge characteristic matrix according to the inverse of the transmission line impedance and the busbar power characteristic value set;

[0048] The graph dataset is obtained based on the adjacency matrix, the node feature matrix, and the edge feature matrix, with the safety and stability margin value set as a label.

[0049] Furthermore, the bus electrical quantity characteristic value set includes bus voltage amplitude, bus voltage phase angle, active power flowing into the bus, and reactive power flowing into the bus.

[0050] A device for implementing the power system security and stability margin prediction method based on graph neural network, comprising a first modeling unit, a calculation unit and a second modeling unit;

[0051] The first modeling unit:

[0052] for constructing an analysis model based on the power system;

[0053] for obtaining a historical data set of the power system, transmission line impedance and transmission line voltage level according to the analysis model;

[0054] Used to obtain key transmission lines based on the transmission line impedance and transmission line voltage level; used to obtain load change range, preset change step, generator output range, fault duration and stability criterion;

[0055] for initializing the analysis model according to the load variation range, the preset variation step and the generator output interval;

[0056] for changing the load of the analysis model within the load variation range with the preset variation step size within the generator output range to obtain an operating condition;

[0057] Used to obtain the current active transmission power of the transmission line according to the operating condition;

[0058] for disconnecting the corresponding transmission line after a three-phase short circuit fault occurs on each of the transmission lines and lasts for the fault duration under the operating condition;

[0059] for obtaining a limit transmission power according to the stability criterion based on the three-phase short circuit fault;

[0060] The computing unit:

[0061] Used to obtain the safety and stability margin of the transmission line according to the limit transmission power and the current active transmission power;

[0062] Used to obtain a safety and stability margin value set according to the safety and stability margin;

[0063] The second modeling unit:

[0064] Used to form a graph dataset and construct an initial graph neural network model based on the analysis model, the historical dataset, the transmission line impedance and the safety and stability margin value set;

[0065] Used to train the graph dataset based on a graph neural network, and use the Adam algorithm to optimize the parameters of the graph neural network to obtain a graph neural network model;

[0066] Used to analyze the power system according to the graph neural network model to obtain a safety and stability margin prediction value.

[0067] Compared with the prior art, the present invention has the following beneficial effects:

[0068] (1) The present invention adopts artificial intelligence to evaluate the safety and stability of the power system and quickly gives the safety and stability margin, providing a fast and accurate method for real-time analysis of the power system status and providing a new technical means for the dispatching and operation of the power grid.

[0069] (2) The present invention defines the safety and stability margin by the current active transmission power of the key transmission lines, providing an intuitive safety and stability indicator for power system dispatching and control.

[0070] (3) The present invention converts the power system operating parameter information and topology information into graph data, which enhances the interpretability of artificial intelligence algorithm applications in the power system field. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 Schematic diagram of the flow of the power system security and stability margin prediction method based on graph neural network of the present invention;

[0072] Figure 2 This is a topology diagram of a 39-node standard example system in an embodiment of the present invention;

[0073] Figure 3 This is an architecture diagram of the initial graph neural network model in an embodiment of the present invention;

[0074] Figure 4 This is a graph showing the prediction effect of the graph neural network model based on the training set in the implementation of the present invention;

[0075] Figure 5 This is a diagram showing the prediction effect of the graph neural network model based on the test set in the implementation of the present invention. DETAILED DESCRIPTION

[0076] Data-driven artificial intelligence technology offers a new approach to power system transient stability analysis and decision-making. Compared to traditional methods, this data-driven model is highly efficient and fast when applied online, enabling predictions within milliseconds and effectively meeting the requirements for speed and accuracy in online security assessments.

[0077] In order to clearly illustrate the technical features of this solution, this solution is described below through specific implementation methods.

[0078] A method for predicting power system security and stability margin based on graph neural network includes the following steps:

[0079] Constructing an analysis model based on the power system, the analysis model includes generators, busbars and transmission lines; preferably, using PSASP to construct the analysis model;

[0080] The historical data set of the power system, transmission line impedance, and transmission line voltage level are obtained based on the analysis model. The historical data set includes a busbar electrical quantity characteristic value set, which includes busbar voltage amplitude, busbar voltage phase angle, active power flowing into the bus, and reactive power flowing into the bus. The historical data set is obtained based on the PSASP example temp file.

[0081] Obtain the key transmission line based on the transmission line impedance and transmission line voltage level. The key transmission line is the transmission line with the lowest transmission line impedance and the highest transmission line voltage level. Obtain the load variation range, preset variation step, generator output range, preset fault and stability criterion. The preset fault is an "N-1" fault, that is, the preset fault is to cut off the corresponding transmission line after a three-phase short circuit fault occurs on each transmission line and lasts for a preset period of time.

[0082] Preferably, the load variation range is 0.9 to 1.1, the preset variation step is 0.01; the preset duration is 100ms; the stability criterion is: during a three-phase short circuit fault, the maximum power angle difference of the generator cannot exceed 360°; and, starting from the start of the three-phase short circuit fault, the minimum voltage of the generator is not lower than a first preset value and the duration does not exceed the first preset time, and the minimum frequency of the generator is not lower than a second preset value and the duration does not exceed the second preset time; preferably, the first preset value is 0.7 pu, the first preset time is 1s, the second preset value is 47 Hz, and the second preset time is 1s, that is, starting from the start of the three-phase short circuit fault, the minimum voltage of the generator is not lower than 0.7 and the duration does not exceed 1s, and the minimum frequency of the generator is not lower than 47 Hz and the duration does not exceed 1s;

[0083] Initialize the analysis model according to the load variation range, preset variation step size and generator output range;

[0084] Before obtaining the safety and stability margin, the analysis model is initialized, that is, the load level of the analysis model is adjusted and the generator output is matched, that is, the load level of the analysis model is changed within the load variation range with a preset variation step size, and the generator output matches the load level within the generator output range; specifically:

[0085] The analysis model is divided according to the key transmission line to obtain the sending end and the receiving end. In the analysis model, the power is transmitted from the sending end to the receiving end after passing through the key transmission line;

[0086] Adjusting the sending end and the receiving end so that the load level of the sending end and the load level of the receiving end are respectively within the load variation range and change with a preset variation step size, and the generator output matches the load level within the generator output range. Preferably, for each load level, the sending end generator output is changed first, and when the sending end generator output cannot meet the load demand, the receiving end generator output is changed;

[0087] Within the generator output range, the load of the analysis model is changed within the load variation range with a preset change step size to obtain the operating condition;

[0088] Based on the operating conditions, the current active transmission power of the key transmission lines is obtained according to the analytical model;

[0089] Under operating conditions, a preset fault is set on each transmission line, that is, a three-phase short circuit fault is set on each transmission line and the corresponding transmission line is cut off after it lasts for a preset time.

[0090] Based on the preset fault, the limit transmission power is obtained according to the stability criterion. The limit transmission power is the current active transmission power of the key transmission line corresponding to the limit stable operating condition. The load corresponding to the limit stable operating condition is increased by a preset change step size equal to the load corresponding to the first unstable operating condition. The steps for obtaining the limit transmission power are as follows:

[0091] Step S101: Initializing the analysis model so that the load level of the analysis model reaches a preset load level and the generator output matches the current load and is within the generator output range when the load level is changed with a preset change step size;

[0092] Step S102: Let the current load be the initial load, and the initial load be the lower limit of the load variation range;

[0093] Step S103: obtaining an operating condition according to the current load, and obtaining the current active transmission power of the key transmission line based on the operating condition;

[0094] Step S104: Under operating conditions, a three-phase short circuit fault is generated on each transmission line and the corresponding transmission line is cut off after the fault lasts for a preset time period;

[0095] Step S105: Based on the three-phase short circuit fault, a stability flag is obtained according to the stability criterion. The stability flag indicates whether each transmission line meets the stability criterion under the operating condition. The stability flag includes 1 or 0.

[0096] If, under the operating condition, each transmission line satisfies the stability criterion, the stability mark of the operating condition is 1;

[0097] Otherwise, the stability of the operating condition is marked as 0;

[0098] Step S106: Increase the load according to the preset change step to obtain the current load update value:

[0099] If the current load update value is not greater than the upper limit of the load variation range, the current load update value is set equal to the current load, and steps S103 to S106 are repeated;

[0100] If the current load update value is greater than the upper limit of the load change range, the working condition stability table is obtained according to the stability mark;

[0101] Step S107: Obtain the extreme stable operating condition according to the operating condition stability table, and obtain the extreme transmission power according to the extreme stable operating condition; the steps for obtaining the extreme transmission power are:

[0102] When the load is increased with a preset change step within the load change range, the first unstable operating condition is obtained according to the operating condition corresponding to the first appearance of the stability mark 0;

[0103] Obtain the first instability load according to the first instability operating condition;

[0104] According to the first unstable load, the preset change step is reduced to obtain the ultimate stable operating load;

[0105] The ultimate transmission power is obtained according to the ultimate stable operating load.

[0106] The safety and stability margin of the operating condition is obtained based on the limit transmission power and the current active transmission power; the safety and stability margin is:

[0107]

[0108] Where M i is the safety and stability margin of the i-th operating condition, P max is the maximum transmission power, P i is the current active transmission power under the i-th operating condition, where i is the operating condition number;

[0109] obtaining a safety and stability margin value set according to the safety and stability margin;

[0110] A graph dataset is formed based on the analysis model, historical dataset, transmission line impedance, and safety and stability margin value set, and an initial graph neural network model is constructed. The steps for forming the graph dataset are as follows:

[0111] Construct the adjacency matrix of the graph based on the analytical model, where the nodes of the graph are buses and the edges of the graph are transmission lines;

[0112] The node characteristic matrix is ​​obtained according to the busbar power characteristic value set, and the edge characteristic matrix is ​​obtained according to the inverse of the transmission line impedance, the active power flowing into the busbar, and the reactive power flowing into the busbar;

[0113] Based on the adjacency matrix, node feature matrix and edge feature matrix, a graph dataset is obtained with the safety and stability margin value set as the label.

[0114] Train the initial graph neural network model based on the graph dataset, and use the Adam algorithm to optimize the parameters of the initial graph neural network model to obtain the graph neural network model;

[0115] The power system is analyzed based on the graph neural network model to obtain the predicted value of the safety and stability margin.

[0116] See also Figure 2 , the initial graph neural network model includes input layer, normalization layer, convolution layer, activation function, regularization layer and loss function, where:

[0117] The input layer is a graph dataset:

[0118] In the normalization layer, first calculate the degree of the first node in the original adjacency matrix A to form the degree matrix D, which is a diagonal matrix. Then calculate the inverse square root of the degree matrix D, and finally obtain the normalized adjacency matrix through the normalization formula The normalization formula is:

[0119]

[0120] There are three convolutional layers in total, the activation function is the RELU function, the regularization layer is Dropout, and the loss function is the root mean square function.

[0121] Calculation example:

[0122] Take the 39-node standard example system as an example. The topology diagram of the 39-node standard example system is as follows: Figure 2 As shown;

[0123] Since branch 4-14 has the lowest impedance and a voltage level of 220 kV (the highest voltage level in this example), branch 4-14 is selected as the key transmission line. Area 1, the sending end, includes generators 30, 37, and 39, while area 2, the receiving end, includes generators 31, 32, 33, 34, 35, 36, and 38.

[0124] The sending-end load level and the receiving-end load level k are each varied within a range of 0.9 to 1.1, with a preset step size of 0.01. For each load level, the sending-end generator output is prioritized. When the sending-end generator power cannot meet the load demand, the receiving-end generator output is then adjusted. The generator output changes are also matched to the load change. During the generator output matching process, the generator output is ensured to remain within the generator output range, i.e., the upper and lower limits of the generator output.

[0125] Perform power flow calculation to obtain the current active transmission power under each operating condition;

[0126] Under each operating condition, a safety and stability margin calculation was performed. A three-phase short-circuit fault was applied to all 33 transmission lines in the 39-node standard example system. After a duration of 100ms, the corresponding transmission line was disconnected. According to the stability criterion, if the system meets the transient stability criteria after the three-phase short-circuit fault was applied to all 33 transmission lines, the stability of this operating condition was marked as 1; otherwise, it was marked as 0. This generated a total of (21 * 21) different operating conditions. Table 1 shows the stability of nine of these operating conditions.

[0127] According to Table 1, the active power of branches 4-14, i.e., the key transmission line, corresponding to the operating condition before the first unstable operating condition is used as the limit transmission power, i.e., 2.7595 is used as the limit transmission power. The safety and stability margin of each operating condition is further obtained according to the formula of the safety and stability margin, and a safety and stability margin value set is obtained. The safety and stability margin value set includes 441 safety and stability margins.

[0128] Table 1 Working condition stability table

[0129]

[0130] The historical dataset of the 39-node standard example system is converted into a graph dataset for the graph neural network. All buses are used as nodes of the graph and all transmission lines are used as edges of the graph to construct the adjacency matrix A. Since the system has a total of 33 transmission lines, A is a 33-row 33-column matrix. The adjacency matrix is ​​converted into a sparse form as shown in Table 2:

[0131] Table 2 Sparse form of adjacency matrix

[0132]

[0133]

[0134] The node characteristics in the node characteristic matrix are the bus voltage amplitude and bus voltage phase angle of the 28 buses, see Table 3;

[0135] Table 3 Node feature table

[0136]

[0137] The edge features in the edge feature matrix are the inverse of the transmission line impedance of the 33 transmission lines, the active power flowing into the bus, and the reactive power flowing into the bus, see Table 4;

[0138] Table 4 Edge feature table

[0139]

[0140] The historical data sets under each operating condition are converted into the data shown in Tables 2, 3, and 4, respectively. Finally, 441 such graph data are obtained. The 441 safety and stability margins are used as labels for each graph data in the graph data set.

[0141] The graph data in the graph dataset are trained on the graph neural network: first, each graph data is preprocessed and normalized;

[0142] There are a total of 441 data sets. The graph data sets are randomly divided into training sets and test sets in a ratio of 8:2. The number of training sets is 350 and the number of test sets is 91.

[0143] Build an initial graph neural network model with three convolutional layers, which are used for feature aggregation, information propagation, and graph structure learning:

[0144] Feature aggregation: Update the feature representation of a node by aggregating the features of its adjacent nodes, so that the node features contain local structural information;

[0145] Propagating information: The convolutional layers of GCN propagate information in the network. The features of the nodes in each convolutional layer are updated, and the updated features are used for calculations in the next layer, so that the representation of the node can include information about more distant neighbors.

[0146] Learning graph structure: The convolutional layer learns the structural information of the graph through weights. These weights represent the degree of influence of different adjacent nodes on the characteristics of the central node.

[0147] Through training, the initial graph neural network model can learn the importance of different relationships in the graph data; finally, through a layer of RELU activation function, the nonlinear expression ability of the initial graph neural network model is further improved.

[0148] The specific initial graph neural network model architecture is as follows Figure 3 As shown;

[0149] The node features and adjacency matrix in the graph data reflect the steady-state information and topological relationships of the 39-node standard example system. The safety and stability margin of the graph label already includes fault information. During the training process, the initial graph neural network model learns the information of the graph data through the convolutional layer, continuously performs forward propagation and backpropagation, and uses the Adam algorithm to optimize and update the parameters, ultimately obtaining a trained graph neural network model. The prediction results of the graph neural network model based on the training set and the prediction results of the graph neural network model based on the test set are respectively shown in Figure 4 and Figure 5 .

[0150] It can be seen that in the prediction results of the graph neural network model in the test set, the prediction error was less than 10% for 91% of the total prediction samples. Therefore, the prediction results of this graph neural network model can be used as the prediction results of the safety and stability margin.

[0151] A device for implementing a power system security and stability margin prediction method based on a graph neural network, comprising a first modeling unit, a calculation unit, and a second modeling unit;

[0152] The first modeling unit:

[0153] Used to build analysis models based on power systems;

[0154] Used to obtain historical data sets of power systems, transmission line impedances, and transmission line voltage levels based on analysis models;

[0155] Used to obtain key transmission lines based on transmission line impedance and transmission line voltage levels; used to obtain load change range, preset change step, generator output range, preset fault and stability criteria;

[0156] Used to initialize the analysis model according to the load change range, preset change step size and generator output range;

[0157] Used to change the load of the analysis model within the generator output range with a preset change step size within the load change range to obtain the operating condition;

[0158] Used to obtain the current active transmission power of key transmission lines based on operating conditions and analysis models;

[0159] Used to preset faults on each transmission line under operating conditions;

[0160] Used to obtain the limit transmission power based on the stability criterion based on the preset fault;

[0161] Computational Unit:

[0162] Used to obtain the safety and stability margin of the operating conditions based on the limit transmission power and the current active transmission power;

[0163] Used to obtain a safety and stability margin value set according to the safety and stability margin;

[0164] Second modeling unit:

[0165] Used to form a graph dataset and build an initial graph neural network model based on the analysis model, historical dataset, transmission line impedance and safety and stability margin value set;

[0166] It is used to train the initial graph neural network model based on the graph dataset and use the Adam algorithm to optimize the parameters of the initial graph neural network model to obtain the graph neural network model;

[0167] Used to analyze the power system based on the graph neural network model to obtain the safety and stability margin prediction value.

[0168] Finally, it should be noted that the above content is only used to illustrate the technical solution of the present invention, rather than to limit the scope of protection of the present invention. Simple modifications or equivalent substitutions of the technical solution of the present invention by ordinary technicians in this field do not deviate from the essence and scope of the technical solution of the present invention.

Claims

1. A method for predicting power system security and stability margin based on graph neural network, characterized by: The following steps are involved: constructing an analysis model based on the power system, wherein the analysis model includes a generator, a busbar, and a transmission line; obtaining a historical data set of the power system, transmission line impedance, and transmission line voltage level according to the analysis model; Obtaining a key transmission line according to the transmission line impedance and the transmission line voltage level, wherein the key transmission line is the transmission line having the lowest transmission line impedance and the highest transmission line voltage level; Obtain load variation range, preset variation step, generator output range, preset fault and stability criteria; Initializing the analysis model according to the load variation range, the preset variation step and the generator output range; Within the generator output range, changing the load of the analysis model within the load variation range with the preset variation step to obtain an operating condition; Based on the operating conditions, obtaining the current active transmission power of the key transmission line according to the analysis model; Under the operating conditions, respectively performing a preset fault on each of the transmission lines; Based on the preset fault, a limit transmission power is obtained according to the stability criterion, where the limit transmission power is the current active transmission power of the critical transmission line corresponding to the extreme stable operating condition; the load corresponding to the extreme stable operating condition is increased by the preset change step size equal to the load corresponding to the first unstable operating condition; Obtaining a safety and stability margin of the operating condition according to the limit transmission power and the current active transmission power; Obtaining a safety and stability margin value set according to the safety and stability margin; Forming a graph dataset and constructing an initial graph neural network model based on the analysis model, the historical dataset, the transmission line impedance, and the safety and stability margin value set; Training the initial graph neural network model based on the graph dataset, and optimizing the parameters of the initial graph neural network model using the Adam algorithm to obtain a graph neural network model; The power system is analyzed according to the graph neural network model to obtain a safety and stability margin prediction value.

2. The method for predicting power system security and stability margin based on graph neural network according to claim 1, characterized in that: The steps of obtaining the limit transmission power are: Step S101: Initializing the analysis model so that the load level of the analysis model reaches a preset load level, and when the load level is changed with the preset change step size, the generator output matches the current load and is within the generator output range; Step S102: setting the current load as the initial load, where the initial load is the lower limit of the load variation range; Step S103: obtaining the operating condition according to the current load, and obtaining the current active transmission power of the key transmission line based on the operating condition; Step S104: Under the operating condition, performing the preset fault detection on each of the transmission lines; Step S105: Based on the preset fault, obtain a stability mark according to the stability criterion, wherein the stability mark indicates whether each of the transmission lines meets the stability criterion under the operating condition; Step S106: Increase the load according to the preset change step to obtain the current load update value: If the current load update value is not greater than the upper limit of the load variation range, the current load update value is set equal to the current load, and steps S103 to S106 are repeated; If the current load update value is greater than the upper limit of the load variation range, obtaining a working condition stability table according to the stability mark; Step S107: obtaining the extreme stable operating condition according to the operating condition stability table, and obtaining the extreme transmission power according to the extreme stable operating condition.

3. The method for predicting power system security and stability margin based on graph neural network according to claim 2, characterized in that: The preset fault is to disconnect the corresponding transmission line after a three-phase short circuit fault occurs on each transmission line and lasts for a preset period of time; the stability criterion is: During the three-phase short circuit fault, the maximum power angle difference of the generator cannot exceed 360°; Also, starting from the time when the three-phase short circuit fault occurs, the minimum voltage of the generator is not lower than a first preset value and the duration does not exceed the first preset time, and the minimum frequency of the generator is not lower than a second preset value and the duration does not exceed the second preset time.

4. The method for predicting power system security and stability margin based on graph neural network according to claim 3 is characterized by: The stability marker includes 1 or 0; If, under the operating condition, each of the transmission lines satisfies the stability criterion, the stability mark of the operating condition is 1; Otherwise, the stability of the operating condition is marked as 0.

5. The method for predicting power system security and stability margin based on graph neural network according to claim 4 is characterized in that: The steps of obtaining the limit transmission power are: When the load is increased with the preset change step within the load change range, the first unstable operating condition is obtained according to the operating condition corresponding to the first appearance of the stability mark 0; Obtaining a first instability load according to the first instability operating condition; According to the first unstable load, the preset change step is reduced to obtain an ultimate stable operating load; The limit transmission power is obtained according to the limit stable operation load.

6. The method for predicting power system security and stability margin based on graph neural network according to claim 5, characterized in that: The load variation range is 0.9 to 1.1, and the preset variation step is 0.

01.

7. The method for predicting power system security and stability margin based on graph neural network according to any one of claims 1 to 6, characterized in that: The safety stability margin is: Where M i is the safety and stability margin of the i-th operating condition, P max is the maximum transmission power, P i is the current active transmission power under the i-th operating condition, and i is the operating condition number.

8. The method for predicting power system security and stability margin based on graph neural network according to claim 1, characterized in that: The historical data set includes a bus power characteristic value set; the steps of forming the graph data set are: constructing an adjacency matrix of a graph according to the analysis model, wherein the nodes of the graph are the buses and the edges of the graph are the transmission lines; Obtaining a node characteristic matrix according to the busbar power characteristic value set, and obtaining an edge characteristic matrix according to the inverse of the transmission line impedance and the busbar power characteristic value set; The graph dataset is obtained based on the adjacency matrix, the node feature matrix, and the edge feature matrix, with the safety and stability margin value set as a label.

9. The method for predicting power system security and stability margin based on graph neural network according to claim 8, characterized in that: The bus electrical quantity characteristic value set includes bus voltage amplitude, bus voltage phase angle, active power flowing into the bus, and reactive power flowing into the bus.

10. A device for implementing the power system security and stability margin prediction method based on graph neural network according to any one of claims 1 to 9, characterized in that: comprising a first modeling unit, a calculation unit and a second modeling unit; The first modeling unit: for constructing an analysis model based on the power system; for obtaining a historical data set of the power system, transmission line impedance and transmission line voltage level according to the analysis model; for obtaining a key transmission line according to the transmission line impedance and the transmission line voltage level; Used to obtain load variation range, preset variation step, generator output range, preset fault and stability criteria; for initializing the analysis model according to the load variation range, the preset variation step and the generator output interval; for changing the load of the analysis model within the load variation range with the preset variation step size within the generator output range to obtain an operating condition; for obtaining the current active transmission power of the key transmission line based on the operating condition and the analysis model; Used to perform the preset fault on each of the transmission lines under the operating conditions; for obtaining a limit transmission power according to the stability criterion based on the preset fault; The computing unit: A safety and stability margin for the operating condition is obtained using the limit transmission power and the current active transmission power; Used to obtain a safety and stability margin value set according to the safety and stability margin; The second modeling unit: Used to form a graph dataset and construct an initial graph neural network model based on the analysis model, the historical dataset, the transmission line impedance and the safety and stability margin value set; Used to train the initial graph neural network model based on the graph dataset, and use the Adam algorithm to optimize the parameters of the initial graph neural network model to obtain a graph neural network model; Used to analyze the power system according to the graph neural network model to obtain a safety and stability margin prediction value.

Images