Method for building power grid transient voltage severity index calculation model

Abstract

The invention relates to a construction method of a power grid transient voltage severity index calculation model, which can construct a static scene set and a transient scene set of a power grid according to simulation random conditions, and further construct power grid heterogeneous graph structure data. Constructing a fault set according to the static scene set, performing time domain simulation on the fault set to calculate transient voltage severity indexes of all nodes, constructing a heterogeneous graph database according to the heterogeneous graph structure data of the power grid and the transient voltage severity indexes, and training the constructed heterogeneous graph attention network by using the heterogeneous graph database to obtain the heterogeneous graph attention network. And obtaining a power grid transient voltage severity index calculation model. According to the calculation model constructed by the method, the uncertainty of various static and transient parameters is considered in the construction of the database, and the node attention and semantic attention of the power grid are mined through the heterogeneous graph attention network, so that the rapid calculation accuracy of the transient voltage severity index of the power grid is improved; and the calculation requirement of the transient voltage stability of the alternating-current and direct-current hybrid power grid is met.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid index calculation, and in particular to a method for building a power grid transient voltage severity index calculation model. Background Art

[0002] In recent years, with the increase in installed capacity of new energy generators, the increase in the number of power electronic equipment, and the large-scale commissioning of high-voltage direct current transmission projects, the scale of the power grid has continued to expand and the structure has become increasingly complex, and its transient safe and stable operation faces huge challenges. Therefore, how to provide a fast calculation method for transient voltage stability for AC / DC hybrid power grids has become a key and difficult point in transient stability analysis of power systems.

[0003] At present, transient voltage assessment methods are divided into three categories: time domain simulation method, direct method and artificial intelligence method. The time domain simulation method obtains accurate results by constructing a complex system model, but it faces problems such as long time consumption and large amount of calculation. The direct method is difficult to accurately reflect the transient characteristics of the system when facing a complex AC / DC hybrid power grid. Artificial intelligence, with its ability to calculate quickly, provides new ideas for the assessment and calculation of transient voltage in power grids. However, some artificial intelligence methods lack consideration of the topological structure of the power grid, or lack consideration of transient parameters, or homogenize different types of nodes in the power grid, and cannot accurately distinguish the potential differences between them.

[0004] Therefore, existing artificial intelligence methods cannot accurately characterize AC / DC hybrid power grids and lack in-depth research on scenarios. Moreover, most fast calculation methods are limited to the binary classification of stability and lack further research on severity indicators. Summary of the invention

[0005] In view of this, the purpose of the present invention is to provide a method for building a grid transient voltage severity index calculation model to solve the problem that the technical solutions in the prior art cannot accurately characterize the AC / DC hybrid power grid and lack in-depth research on the scenarios.

[0006] According to a first aspect of an embodiment of the present invention, a method for building a grid transient voltage severity index calculation model is provided, comprising:

[0007] Step S1, constructing a static scenario set and a transient scenario set of the power grid according to simulated random conditions, wherein the simulated random conditions include random output conditions of new energy, random load change conditions, injection power of high-voltage direct current, proportion of induction motors, SVG configuration scheme and low voltage ride-through control logic;

[0008] Step S2, calculating the reactive power reserve of each scene in the static scene set; filtering the effective features in the static scene set and the transient scene set as node features according to the node type; adding directed edges and edge features to the static scene set according to the power flow direction and line admittance, and constructing the heterogeneous graph structure data of the power grid;

[0009] Step S3, constructing a fault set according to the static scenario set, and based on the power grid heterogeneous graph structure data, calculating the transient voltage severity index of all nodes by performing time domain simulation on the fault set, and constructing a heterogeneous graph database according to the transient voltage severity index;

[0010] Step S4, defining a meta-path according to the power grid current flow direction; constructing a heterogeneous graph attention network according to the meta-path, node characteristics and edge characteristics; and training the heterogeneous graph attention network using the heterogeneous graph database to obtain a power grid transient voltage severity index calculation model.

[0011] Preferably, a static scene set of the power grid is constructed according to the simulated random conditions, specifically,

[0012] The random static parameters of the corresponding nodes are obtained according to the simulated random conditions, and the static scenario set of the power grid is constructed by modifying the initial static data of the power grid, including:

[0013] Acquire typical historical data of new energy plants and load nodes, fit the relationship between the typical historical data and time, obtain deterministic prediction results, add disturbance terms for simulating random fluctuations to the deterministic prediction results, and obtain random output conditions of new energy and random change conditions of load;

[0014] Generate a random floating factor within a first preset range as the injection power of the high voltage direct current;

[0015] According to the calculated active and reactive power shortage of the power grid, the balancing grid mode is set to adjust the output of synchronous generators and increase or decrease the local reactive power compensation;

[0016] According to the random output conditions of renewable energy, random load change conditions, the injected power of high-voltage direct current and the balanced grid mode, the random static parameters of the corresponding nodes are obtained. By modifying the initial static data of the grid, multiple static scenarios of the grid are constructed, and the scenarios that meet the flow convergence in multiple static scenarios are integrated into a static scenario set.

[0017] Preferably, a transient scenario set of a power grid is constructed according to simulated random conditions, specifically,

[0018] The random transient parameters of the corresponding nodes are obtained according to the simulated random conditions. By modifying the initial transient data of the power grid, a transient scenario set of the power grid is constructed, including:

[0019] generating a random proportion within a second preset range as the proportion of the induction motor at each load node;

[0020] Randomly generate SVGs of different capacities within a preset quantity range, and add the generated SVGs to any load node as an SVG configuration plan;

[0021] Generate a random number with a step size of 0.01 in the third preset range as the low voltage ride-through control logic of the new energy station;

[0022] According to the proportion of induction motors, SVG configuration scheme and low voltage ride-through control logic, the random transient parameters of the corresponding nodes are obtained. By modifying the initial transient data of the power grid, multiple transient scenarios of the power grid are constructed and integrated into a transient scenario set.

[0023] Preferably, in step S2, reactive power reserve calculation is performed, effective features are screened, and grid heterogeneous graph structure data is constructed, including:

[0024] Based on a static scenario set, a VQ curve of voltage versus reactive injection is obtained by controlling the node voltage, and the reactive reserve of each scenario is calculated according to the VQ curve;

[0025] Based on the static scenario set and transient scenario set, the voltage, phase angle, active output, reactive output, reactive reserve and low voltage ride-through control logic of new energy stations are selected as the six-dimensional point features of PV nodes;

[0026] Based on the static scenario set and transient scenario set, the voltage, phase angle, active load, reactive load, reactive reserve, local reactive compensation capacity, SVG capacity and induction motor proportion are selected as the eight-dimensional point features of the PQ node;

[0027] Based on the static scene set, directed edges are set according to the flow direction, and line admittance is selected as the two-dimensional edge feature to construct the heterogeneous graph structure data of the power grid.

[0028] Preferably, in step S3, time domain simulation is performed in batches according to the fault set and transient voltage severity index is calculated to construct a heterogeneous graph database, including:

[0029] Based on the static scenario set, N-1 faults with a breaking time of 0.1 seconds are set for lines with voltage levels of 220 kV and above, and the islanded faults are eliminated, and the remaining faults are integrated into a fault set;

[0030] Based on the static scenario set, transient scenario set and power grid heterogeneous graph structure data, the time domain simulation is performed by simulating the fault batches in the fault set, and the directed edges corresponding to the power grid heterogeneous graph structure data are removed to obtain the heterogeneous graph database input;

[0031] The transient voltage severity index of all nodes is calculated using the multi-binary table method based on the heterogeneous graph database input;

[0032] constructing an initial heterogeneous graph database according to the transient voltage severity index and the heterogeneous graph database input;

[0033] The data in the initial heterogeneous graph database is normalized to obtain a heterogeneous graph database.

[0034] Preferably, according to the power grid flow direction, a meta-path is defined, including:

[0035] According to the power flow direction of the power grid, three meta-paths, PV-PQ, PQ-PQ and PQ-PV, are defined to describe the power transmission relationship between nodes, and the three meta-paths are integrated into a meta-path set.

[0036] Preferably, the constructed heterogeneous graph attention network consists of two multi-layer perceptrons of different scales, three node attention layers, one semantic level aggregation layer and one multi-layer perceptron, and is used for:

[0037] Through two multi-layer perceptrons of different scales, nonlinear changes are performed on node features and edge features respectively to obtain node features and edge features of the same dimension;

[0038] According to the meta-path set and the node features and edge features of the same dimension, different attention values ​​are assigned to the adjacent nodes of the meta-path through the node-level attention mechanism of the three-layer node attention layer, and the node embedding representation set corresponding to the meta-path set is obtained;

[0039] According to the set of meta-paths and the set of node embedding representations, the semantic level aggregation layer is used to aggregate the different flow propagation semantic information extracted from different meta-paths to obtain the node features after semantic level aggregation.

[0040] The node features after semantic-level aggregation are subjected to nonlinear changes to obtain the predicted node transient voltage severity index, and the loss function is set to compare the distance between the predicted value and the true value.

[0041] Preferably, two multi-layer perceptrons of different scales are used to perform nonlinear changes on the node features and the edge features respectively, so as to obtain the node features and the edge features of the same dimension, including:

[0042] The multi-layer perceptron and ReLU activation function are used to perform nonlinear changes on the six-dimensional point features to obtain the mapped PV feature h' pv ;

[0043] The multi-layer perceptron and ReLU activation function are used to perform nonlinear changes on the eight-dimensional point features to obtain the mapped PQ feature h' pq ;

[0044] The edge features are nonlinearly changed using a multi-layer perceptron and ReLU activation function to obtain the mapped edge features h e '.

[0045] Preferably, according to the meta-path set and the node features and edge features of the same dimension, different attention values ​​are assigned to adjacent nodes of the meta-path through the node-level attention mechanism of the three-layer node attention layer, and a node embedding representation set corresponding to the meta-path set is obtained, including:

[0046] According to the meta-path set and the mapped feature h' pv , characteristic h' pv and edge feature h' e , build a node attention layer with 3 layers;

[0047] For each node attention layer, the weight coefficient of each node on the meta-path to the neighboring nodes is calculated; the weight coefficient is used to aggregate the mapped features of the neighboring nodes to obtain an embedding representation based on the meta-path; the embedding representations obtained by all attention heads are concatenated to obtain the node embedding representation set corresponding to the meta-path set.

[0048] Preferably, according to the set of meta-paths and the set of node embedding representations, different flow propagation semantic information extracted from different meta-paths is aggregated through a semantic-level aggregation layer to obtain node features after semantic-level aggregation, including:

[0049] The meta-path weight coefficient of each meta-path is calculated through the node attention layer, and the meta-path weight coefficient is used to aggregate the embedded representation of the meta-path to obtain the node features after semantic-level aggregation.

[0050] The technical solution provided by the embodiments of the present invention may have the following beneficial effects:

[0051] It can be understood that the technical solution shown in the present invention can construct a static scene set and a transient scene set of the power grid according to simulated random conditions, and further construct the heterogeneous graph structure data of the power grid; construct a fault set according to the static scene set, calculate the transient voltage severity index of all nodes by performing time domain simulation on the fault set, construct a heterogeneous graph database according to the heterogeneous graph structure data of the power grid and the transient voltage severity index, and use the heterogeneous graph database to train the constructed heterogeneous graph attention network to obtain the calculation model of the transient voltage severity index of the power grid. The calculation model constructed by the present invention takes into account the uncertainty of various static and transient parameters in the construction of the database, and mines the attention and semantic attention of the power grid nodes through the heterogeneous graph attention network, which improves the accuracy of the rapid calculation of the transient voltage severity index of the power grid and meets the calculation requirements of the transient voltage stability of the AC / DC hybrid power grid.

[0052] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0054] Figure 1 It is a schematic diagram of the steps of a method for building a grid transient voltage severity index calculation model according to an exemplary embodiment;

[0055] Figure 2 It is a flowchart of a method for building a grid transient voltage severity index calculation model according to an exemplary embodiment. DETAILED DESCRIPTION

[0056] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0057] In one embodiment, Figure 1 is a schematic diagram of the steps of a method for building a grid transient voltage severity index calculation model according to an exemplary embodiment, see Figure 1 , a method for building a grid transient voltage severity index calculation model is provided, including:

[0058] Step S1, constructing a static scenario set and a transient scenario set of the power grid according to simulated random conditions, wherein the simulated random conditions include random output conditions of new energy, random change conditions of load, injection power of high-voltage direct current, proportion of induction motors, SVG configuration scheme and low voltage ride-through control logic.

[0059] The static scenario set and transient scenario set of the power grid constructed in this embodiment are constructed based on simulated random conditions, taking into account various uncertainties.

[0060] Step S2, calculate the reactive power reserve of each scene in the static scene set; according to the node type, filter the effective features in the static scene set and the transient scene set as node features; add directed edges and edge features to the static scene set according to the flow direction and line admittance, and construct the heterogeneous graph structure data of the power grid.

[0061] Step S3: construct a fault set according to the static scenario set, calculate the transient voltage severity index of all nodes by performing time domain simulation on the fault set based on the heterogeneous graph structure data of the power grid, and construct a heterogeneous graph database according to the transient voltage severity index.

[0062] Step S4, defining a meta-path according to the power grid current flow direction; constructing a heterogeneous graph attention network according to the meta-path, node characteristics and edge characteristics; and training the heterogeneous graph attention network using the heterogeneous graph database to obtain a power grid transient voltage severity index calculation model.

[0063] It can be understood that the technical solution shown in this embodiment can construct a static scene set and a transient scene set of the power grid according to simulated random conditions, and further construct the heterogeneous graph structure data of the power grid; construct a fault set according to the static scene set, calculate the transient voltage severity index of all nodes by performing time domain simulation on the fault set, construct a heterogeneous graph database according to the heterogeneous graph structure data of the power grid and the transient voltage severity index, and use the heterogeneous graph database to train the constructed heterogeneous graph attention network to obtain the calculation model of the transient voltage severity index of the power grid. The calculation model constructed in this embodiment takes into account the uncertainty of various static and transient parameters in the construction of the database, and mines the attention and semantic attention of the power grid nodes through the heterogeneous graph attention network, which improves the accuracy of the rapid calculation of the transient voltage severity index of the power grid and meets the calculation requirements of the transient voltage stability of the AC / DC hybrid power grid.

[0064] Figure 2 is a flow chart of a method for building a grid transient voltage severity index calculation model according to an exemplary embodiment, see Figure 2 , and the following instructions are given.

[0065] In step S1, a static scenario set of the power grid is constructed according to the simulated random conditions, specifically,

[0066] The random static parameters of the corresponding nodes are obtained according to the simulated random conditions, and the static scenario set of the power grid is constructed by modifying the initial static data of the power grid, including:

[0067] S11. Obtain typical historical data of new energy plants and load nodes, fit the relationship between the typical historical data and time, obtain deterministic prediction results, add disturbance terms for simulating random fluctuations to the deterministic prediction results, and obtain random output conditions of new energy and random change conditions of load.

[0068] In specific practice, linear regression is used to fit the relationship between the typical historical data and time to obtain a deterministic prediction result. The added disturbance term conforms to the normal distribution to simulate random fluctuations.

[0069] In addition, to ensure that the simulation results are not negative, the results need to be constrained. The specific formula obtained is as follows:

[0070] P new (t) = max(P new,h (t)+ε(t),0)

[0071] P load (t) = max(P load,h (t)+ε(t),0)

[0072] Among them, P new (t) represents the active power of the new energy plant; P load (t) represents the active power of the load node; P new,h (t) represents the deterministic prediction result of the active power of the energy plant; P load,h (t) represents the deterministic prediction result of the active power of the load node; ε(t) represents the disturbance term that conforms to the normal distribution, with a standard deviation of 5 and a mean of 0.

[0073] S12: Generate a random floating factor within a first preset range as the injection power of the high voltage direct current.

[0074] The first preset range may be [0.8, 1.2], and a random floating factor in the range of [0.8, 1.2] is generated, so that the injection power of the high voltage direct current is randomly adjusted between 80% and 120%.

[0075] S13. According to the calculated active and reactive power shortage of the power grid, the grid balancing method is set to adjust the output of the synchronous generator and increase or decrease the local reactive power compensation.

[0076] S14. According to the random output conditions of new energy, random load change conditions, the injection power of high-voltage DC and the balanced grid mode, the random static parameters of the corresponding nodes are obtained, and the N of the grid is constructed by modifying the initial static data of the grid. dat Static scenes are integrated into a static scene set D. dat .

[0077] It should be noted that the transient scenario set of the power grid is constructed according to the simulated random conditions, specifically,

[0078] The random transient parameters of the corresponding nodes are obtained according to the simulated random conditions. By modifying the initial transient data of the power grid, a transient scenario set of the power grid is constructed, including:

[0079] S15. Generate a random proportion within a second preset range as the proportion of the induction motor at each load node.

[0080] The second preset range is [0.4, 0.6], and a random number (random proportion) in the range of [0.4, 0.6] is generated by uniform distribution to simulate the proportion of induction motors at each load node.

[0081] S16. Randomly generate SVGs of different capacities within a preset quantity range, and add the generated SVGs to any load node as an SVG configuration solution.

[0082] The preset quantity range is 1, 2 or 3, and the capacity range is 50, 100, 150, 200MVA. Randomly generate 1 to 3 SVGs with a capacity of 50, 100, 150, 200MVA and add them to any load node to simulate the grid SVG configuration plan.

[0083] S17. Generate a random number with a step size of 0.01 in a third preset range as the low voltage ride-through control logic of the new energy station.

[0084] The third preset range is [0.88, 0.92], and a random number with a range of [0.88, 0.92] and a step size of 0.01 is generated to simulate the low voltage ride-through control logic of the new energy station.

[0085] S18. According to the proportion of induction motors, SVG configuration scheme and low voltage ride-through control logic, the random transient parameters of the corresponding nodes are obtained, and the N of the power grid is constructed by modifying the initial transient data of the power grid. swi transient scenarios are integrated into the transient scenario set D swi .

[0086] It should be noted that in step S2, reactive power reserve calculation is performed, effective features are screened, and grid heterogeneous graph structure data is constructed, including:

[0087] Step S21: Based on a static scenario set, a VQ curve of voltage versus reactive injection is obtained by controlling the node voltage, and reactive reserve of each scenario is calculated according to the VQ curve.

[0088] In practice, based on the static scene set D dat , a virtual phase regulator is put into the node, and the VQ curve of voltage to reactive injection is obtained by controlling the node voltage, and the reactive reserve of all scenarios is obtained through multi-batch calculation.

[0089] Step S22: Based on the static scenario set and the transient scenario set, the voltage, phase angle, active output, reactive output, reactive reserve and low voltage ride-through control logic of the new energy station are selected as the six-dimensional point features of the PV node.

[0090] Step S23: Based on the static scenario set and the transient scenario set, the voltage, phase angle, active load, reactive load, reactive reserve, local reactive compensation capacity, SVG capacity and induction motor proportion are selected as eight-dimensional point features of the PQ node.

[0091] Step S24: Based on the static scene set, directed edges are set according to the power flow direction, line admittance is selected as a two-dimensional edge feature, and power grid heterogeneous graph structure data is constructed.

[0092] It should be noted that in step S3, time domain simulation is performed in batches according to the fault set and transient voltage severity index is calculated to construct a heterogeneous graph database, including:

[0093] Step S31: Based on the static scenario set, set N-1 faults with a breaking time of 0.1 seconds for lines with a voltage level of 220 kV and above, eliminate the island operation faults in the faults, and integrate the remaining faults into the fault set D lsd The number of faults in the fault set is set to N lsd .

[0094] Step S32: Based on the static scenario set, transient scenario set and the power grid heterogeneous graph structure data, a batch of faults in the simulated fault set are simulated in the time domain, and the directed edges corresponding to the power grid heterogeneous graph structure data are removed to obtain the heterogeneous graph database input. The number of heterogeneous graph database inputs is N. data =N dat *N swi *N lsd , that is, the number of static scenes N dat , Number of transient scenarios N sw i and the number of faults N lsd The product of .

[0095] Step S33: Calculate the transient voltage severity index of all nodes using a multi-binary table method based on the heterogeneous graph database input.

[0096] The calculation formula of transient voltage severity index is as follows:

[0097]

[0098] Among them, Λ n,i is the transient voltage severity index of node i under sample n; S n is the sum of the areas of the binary table subregions corresponding to node i under sample n; S0 is the critical area of ​​the corresponding binary table; t n is the moment when the corresponding voltage threshold value is not met; τ n is the moment when the corresponding voltage threshold is met; V e is the voltage rating; V n,i (t) is the actual voltage trajectory, V cr Voltage threshold value of binary meter.

[0099] Step S34: constructing an initial heterogeneous graph database according to the transient voltage severity index and the heterogeneous graph database input.

[0100] Step S35: normalize the data in the initial heterogeneous graph database to obtain a heterogeneous graph database. Use Min-Max normalization to scale all features of the data to the range of [0, 1] to obtain a heterogeneous graph database.

[0101] It should be noted that, in step S4, a meta-path is defined according to the power flow direction of the power grid, including:

[0102] According to the power flow direction of the power grid, three meta-paths, PV-PQ, PQ-PQ and PQ-PV, are defined to describe the power flow transmission relationship between nodes, and the three meta-paths are integrated into a meta-path set {Φ0, Φ1, Φ2}. The meta-path set {Φ0, Φ1, Φ2} is used to reflect the semantic information of power transmission in the power grid.

[0103] It should be noted that the constructed heterogeneous graph attention network consists of two multi-layer perceptrons of different scales, three node attention layers, one semantic level aggregation layer and one multi-layer perceptron, which are used for:

[0104] Step S41: Perform nonlinear changes on node features and edge features respectively through two multi-layer perceptrons of different scales to obtain node features and edge features of the same dimension.

[0105] Specifically, a multi-layer perceptron with a dimension of (6, 24) and a ReLU activation function are used to perform nonlinear changes on the six-dimensional point features (the feature dimension is 6) to obtain the mapped PV feature h' pv ; Use a multi-layer perceptron with dimension (8, 24) and ReLU activation function to perform nonlinear transformation on the eight-dimensional point feature (feature dimension is 8) to obtain the mapped PQ feature h' pq ; Use a multi-layer perceptron with dimension (2, 16) and ReLU activation function to perform nonlinear changes on the edge feature (feature dimension is 2) to obtain the mapped edge feature h e '.

[0106] The mapping formula is as follows:

[0107] h' pv =Relu(MLP(h pv ,θ mlp,pv ))

[0108] h' pq =Relu(MLP(h pq ,θ mlp,pq ))

[0109] h e '=Relu(MLP(h e ,θ mlp,e ))

[0110] Among them, h' pv represents the PV features after mapping, h' pq represents the PQ feature after mapping, h' erepresents the edge feature after mapping, MLP represents multi-layer perceptron, Relu represents activation function, h pv represents the initial characteristics of the PV node, h pq represents the initial characteristics of the PQ node, h e represents the initial feature of the edge, θ mlp,pv ,θ mlp,pq and θ mlp,e Represents a deep learning learnable parameter.

[0111] Step S42: According to the meta-path set and the node features (six-dimensional point features and eight-dimensional point features) and edge features of the same dimension, different attention values ​​are assigned to adjacent nodes of the meta-path through the node-level attention mechanism of the three-layer node attention layer to obtain the node embedding representation set corresponding to the meta-path set.

[0112] Specifically: According to the meta-path set {Φ0, Φ1, Φ2} and the mapped feature h' pv , characteristic h' pv and edge feature h' e , build a node attention layer with 3 layers; for each node attention layer, calculate the weight coefficient of each node i to the neighbor node j on the meta-path Φ; use the weight coefficient to aggregate the features of the neighbor nodes after mapping to obtain the embedded representation based on the meta-path; concatenate the embedded representations obtained by all attention heads, and pass through 3 layers of node attention layers to obtain the node embedded representation set corresponding to the meta-path set {Φ0, Φ1, Φ2}

[0113] The specific formula is as follows:

[0114]

[0115] in, represents the weight coefficient of node i to neighbor node j on meta-path Φ, represents the embedded features of node i after aggregation and concatenation on the meta-path Φ, Relu represents the activation function, σ represents the softmax activation function, and h' i 、h' j and h' k represents the features after mapping of nodes i, j and k, represents the node attention vector of the meta-path Φ, represents the set of all nodes d1 on the meta-path Φ of node i, and K represents the number of attention heads.

[0116] Step S43: According to the meta-path set and the node embedding representation set, different flow propagation semantic information extracted from different meta-paths is aggregated through a semantic-level aggregation layer to obtain node features after semantic-level aggregation.

[0117] Specifically: Calculate the meta-path weight coefficient of each meta-path through the node attention layer The meta-path weight coefficient is used to aggregate the embedded representation of the meta-path to obtain the node feature Z after semantic level aggregation. sem .

[0118] The formula is as follows:

[0119]

[0120] in, Represents the weight coefficient of each meta-path, atten sem represents the semantic level attention layer, represents the node embedding representation set, Z sem Represents node features after semantic-level aggregation.

[0121] Step S44: Perform nonlinear changes on the node features after semantic level aggregation to obtain a predicted node transient voltage severity index, and set a loss function to compare the distance between the predicted value and the true value.

[0122] Specifically: based on the node feature Z after semantic level aggregation sem , a multilayer perceptron with dimension (64, 1) and ReLU activation function are used for nonlinear changes to obtain the predicted node transient voltage severity index, and the loss function is set to compare the distance between the predicted value and the true value. The specific formula is as follows:

[0123]

[0124] in, represents the predicted transient voltage severity index of node i, L represents the mean square error loss function, MLP represents the multi-layer perceptron, and Z sem,i represents the node features after semantic level aggregation, Λ i represents the true value of the transient voltage severity index of node i, and N represents the total number of nodes in the heterogeneous graph data.

[0125] In step S4, the heterogeneous graph attention network is trained, including: inputting PV, PQ and edge features of the power grid, outputting the power grid transient voltage severity index, completing the training of the neural network, and obtaining the power grid transient voltage severity index calculation model.

[0126] It can be understood that the same or similar parts of the above embodiments can be referenced to each other, and the contents not described in detail in some embodiments can refer to the same or similar contents in other embodiments.

[0127] It should be noted that, in the description of the present invention, the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. In addition, in the description of the present invention, unless otherwise specified, the meaning of "plurality" refers to at least two.

[0128] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code that includes one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred embodiments of the present invention includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present invention belong.

[0129] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0130] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0131] In addition, each functional unit in each embodiment of the present invention may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0132] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc.

[0133] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0134] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A method for building a grid transient voltage severity index calculation model, characterized in that: include: Step S1, constructing a static scenario set and a transient scenario set of the power grid according to simulated random conditions, wherein the simulated random conditions include random output conditions of new energy, random load change conditions, injection power of high-voltage direct current, proportion of induction motors, SVG configuration scheme and low voltage ride-through control logic; Step S2, calculating the reactive power reserve of each scene in the static scene set; filtering the effective features in the static scene set and the transient scene set as node features according to the node type; adding directed edges and edge features to the static scene set according to the power flow direction and line admittance, and constructing the heterogeneous graph structure data of the power grid; Step S3, constructing a fault set according to the static scenario set, and based on the power grid heterogeneous graph structure data, calculating the transient voltage severity index of all nodes by performing time domain simulation on the fault set, and constructing a heterogeneous graph database according to the transient voltage severity index; Step S4, defining a meta-path according to the power grid current flow direction; constructing a heterogeneous graph attention network according to the meta-path, node characteristics and edge characteristics; and training the heterogeneous graph attention network using the heterogeneous graph database to obtain a power grid transient voltage severity index calculation model.

2. The method for constructing a grid transient voltage severity index calculation model according to claim 1, characterized in that: A static scenario set of the power grid is constructed according to the simulated random conditions, specifically, The random static parameters of the corresponding nodes are obtained according to the simulated random conditions, and the static scenario set of the power grid is constructed by modifying the initial static data of the power grid, including: Acquire typical historical data of new energy plants and load nodes, fit the relationship between the typical historical data and time, obtain deterministic prediction results, add disturbance terms for simulating random fluctuations to the deterministic prediction results, and obtain random output conditions of new energy and random change conditions of load; Generate a random floating factor within a first preset range as the injection power of the high voltage direct current; According to the calculated active and reactive power shortage of the power grid, the balancing grid mode is set to adjust the output of synchronous generators and increase or decrease the local reactive power compensation; According to the random output conditions of renewable energy, random load change conditions, the injected power of high-voltage direct current and the balanced grid mode, the random static parameters of the corresponding nodes are obtained. By modifying the initial static data of the grid, multiple static scenarios of the grid are constructed, and the scenarios that meet the flow convergence in multiple static scenarios are integrated into a static scenario set.

3. The method for constructing a grid transient voltage severity index calculation model according to claim 2, characterized in that: The transient scenario set of the power grid is constructed according to the simulated random conditions, specifically, The random transient parameters of the corresponding nodes are obtained according to the simulated random conditions. By modifying the initial transient data of the power grid, a transient scenario set of the power grid is constructed, including: generating a random proportion within a second preset range as the proportion of the induction motor at each load node; Randomly generate SVGs of different capacities within a preset quantity range, and add the generated SVGs to any load node as an SVG configuration plan; Generate a random number with a step size of 0.01 in the third preset range as the low voltage ride-through control logic of the new energy station; According to the proportion of induction motors, SVG configuration scheme and low voltage ride-through control logic, the random transient parameters of the corresponding nodes are obtained. By modifying the initial transient data of the power grid, multiple transient scenarios of the power grid are constructed and integrated into a transient scenario set.

4. The method for constructing a grid transient voltage severity index calculation model according to claim 1, characterized in that: In step S2, reactive power reserve calculation is performed, effective features are screened, and grid heterogeneous graph structure data is constructed, including: Based on a static scenario set, a VQ curve of voltage versus reactive injection is obtained by controlling the node voltage, and the reactive reserve of each scenario is calculated according to the VQ curve; Based on the static scenario set and transient scenario set, the voltage, phase angle, active output, reactive output, reactive reserve and low voltage ride-through control logic of new energy stations are selected as the six-dimensional point features of PV nodes; Based on the static scenario set and transient scenario set, the voltage, phase angle, active load, reactive load, reactive reserve, local reactive compensation capacity, SVG capacity and induction motor proportion are selected as the eight-dimensional point features of the PQ node; Based on the static scene set, directed edges are set according to the flow direction, and line admittance is selected as the two-dimensional edge feature to construct the heterogeneous graph structure data of the power grid.

5. The method for constructing a grid transient voltage severity index calculation model according to claim 1, characterized in that: In step S3, time domain simulation is performed in batches according to the fault set and transient voltage severity index is calculated to construct a heterogeneous graph database, including: Based on the static scenario set, N-1 faults with a breaking time of 0.1 seconds are set for lines with voltage levels of 220 kV and above, and the islanded faults are eliminated, and the remaining faults are integrated into a fault set; Based on the static scenario set, transient scenario set and power grid heterogeneous graph structure data, the time domain simulation is performed by simulating the fault batches in the fault set, and the directed edges corresponding to the power grid heterogeneous graph structure data are removed to obtain the heterogeneous graph database input; The transient voltage severity index of all nodes is calculated using the multi-binary table method based on the heterogeneous graph database input; constructing an initial heterogeneous graph database according to the transient voltage severity index and the heterogeneous graph database input; The data in the initial heterogeneous graph database is normalized to obtain a heterogeneous graph database.

6. The method for constructing a grid transient voltage severity index calculation model according to claim 4, characterized in that: According to the power flow direction of the power grid, the meta-path is defined, including: According to the power flow direction of the power grid, three meta-paths, PV-PQ, PQ-PQ and PQ-PV, are defined to describe the power transmission relationship between nodes, and the three meta-paths are integrated into a meta-path set.

7. The method for constructing a grid transient voltage severity index calculation model according to claim 6, characterized in that: The constructed heterogeneous graph attention network consists of two multi-layer perceptrons of different scales, three node attention layers, one semantic level aggregation layer and one multi-layer perceptron layer, which are used to: Through two multi-layer perceptrons of different scales, nonlinear changes are performed on node features and edge features respectively to obtain node features and edge features of the same dimension; According to the meta-path set and the node features and edge features of the same dimension, different attention values ​​are assigned to the adjacent nodes of the meta-path through the node-level attention mechanism of the three-layer node attention layer, and the node embedding representation set corresponding to the meta-path set is obtained; According to the set of meta-paths and the set of node embedding representations, the semantic level aggregation layer is used to aggregate the different flow propagation semantic information extracted from different meta-paths to obtain the node features after semantic level aggregation. The node features after semantic-level aggregation are subjected to nonlinear changes to obtain the predicted node transient voltage severity index, and the loss function is set to compare the distance between the predicted value and the true value.

8. The method for constructing a grid transient voltage severity index calculation model according to claim 7, characterized in that: Through two multi-layer perceptrons of different scales, nonlinear changes are performed on node features and edge features respectively to obtain node features and edge features of the same dimension, including: The multi-layer perceptron and ReLU activation function are used to perform nonlinear changes on the six-dimensional point features to obtain the mapped PV feature h' pv ; The multi-layer perceptron and ReLU activation function are used to perform nonlinear changes on the eight-dimensional point features to obtain the mapped PQ feature h' pq ; The edge features are nonlinearly changed using a multi-layer perceptron and ReLU activation function to obtain the mapped edge features h e '.

9. The method for constructing a grid transient voltage severity index calculation model according to claim 8, characterized in that: According to the meta-path set and the node features and edge features of the same dimension, different attention values ​​are assigned to the adjacent nodes of the meta-path through the node-level attention mechanism of the three-layer node attention layer, and the node embedding representation set corresponding to the meta-path set is obtained, including: According to the meta-path set and the mapped feature h' pv , characteristic h' pv and edge feature h′ e , build a node attention layer with 3 layers; For each node attention layer, the weight coefficient of each node on the meta-path to the neighboring nodes is calculated; the weight coefficient is used to aggregate the mapped features of the neighboring nodes to obtain an embedding representation based on the meta-path; the embedding representations obtained by all attention heads are concatenated to obtain the node embedding representation set corresponding to the meta-path set.

10. The method for constructing a grid transient voltage severity index calculation model according to claim 9, characterized in that: According to the set of meta-paths and the set of node embedding representations, the semantic level aggregation layer aggregates the different flow propagation semantic information extracted from different meta-paths to obtain the node features after semantic level aggregation, including: The meta-path weight coefficient of each meta-path is calculated through the node attention layer, and the meta-path weight coefficient is used to aggregate the embedded representation of the meta-path to obtain the node features after semantic-level aggregation.

Images