Network-constructing type energy storage addressing and sizing method based on depth map neural network

Abstract

This invention relates to the field of power system technology and discloses a method for addressing and determining the capacity of grid-based energy storage based on a deep graph neural network. The method includes: constructing a Thevenin equivalent admittance matrix; defining a grid strength feature matrix by combining the diagonal matrix of the feed-in equipment; using the minimum eigenvalue of the grid strength feature matrix as the system's generalized operating short-circuit ratio; calculating the right characteristic vector and left eigenvector of the weakest mode corresponding to the minimum eigenvalue of the grid strength feature matrix, and defining the participation factor for each feed-in node; constructing a capacity determination model; inputting the node feature vectors and edge feature vectors into the deep graph neural network model to output the predicted value of the power system's generalized operating short-circuit ratio and the predicted value of the participation factor; and calculating the configured capacity of the grid-based energy storage by substituting the predicted value of the system's generalized operating short-circuit ratio and a preset stability threshold into the capacity determination model. The advantage of this invention is that it significantly reduces computational complexity while maintaining clear physical constraints and stability criteria.

Description

Technical Field

[0001] This invention relates to the field of power system technology, and in particular to a method for addressing and determining the capacity of grid-type energy storage based on deep graph neural networks. Background Technology

[0002] With the rapid development of new energy power generation technologies, large-scale wind power, photovoltaic power, and other new energy sources are being integrated into the power grid through power electronic interfaces. The operation mode of the power system is gradually shifting from being dominated by traditional synchronous generators to being dominated by a high proportion of power electronic equipment. Against the backdrop of the continuous increase in the penetration rate of new energy sources, the short-circuit capacity of the power grid has decreased significantly. The system exhibits obvious low short-circuit ratio and weak grid characteristics, and the voltage support capability and small disturbance stability margin are continuously decreasing, which can easily lead to a series of stability problems such as voltage instability and subsynchronous oscillations.

[0003] Grid-based converters, due to their external characteristics approaching those of an ideal voltage source, can actively establish voltage and frequency references under weak grid conditions, providing effective voltage support for the system. They are considered a key technology for improving the stability of new energy power systems. In recent years, the application of energy storage systems based on grid-based converters in new energy power plants and power system planning has received widespread attention. Related research mainly focuses on the site selection and capacity configuration methods for grid-based converters or grid-based energy storage.

[0004] Most existing methods for site selection and capacity determination of grid-based energy storage are based on small-disturbance stability analysis theory. These methods involve establishing a system model with multiple renewable energy feed-in nodes, calculating system eigenvalues ​​or stability indices such as the generalized short-circuit ratio, and combining this with participation factor analysis to determine the preferred locations for grid-based energy storage. Finally, the capacity of the grid-based energy storage is optimized while satisfying system stability constraints. These methods can accurately characterize the mechanism by which grid-based energy storage improves system stability and have achieved certain results in theoretical analysis and engineering examples.

[0005] However, with the continuous expansion of the power grid and the increase in the number of new energy power plants, existing methods have gradually revealed the following shortcomings in practical applications: On the one hand, the stability assessment process based on eigenvalue decomposition or sensitivity analysis has high computational complexity. In multi-scenario, long-term planning or large-scale operation simulation, it is necessary to repeatedly perform system modeling and numerical solutions, which is computationally intensive and time-consuming, making it difficult to meet the needs of rapid assessment and engineering planning. On the other hand, existing methods usually rely on accurate system parameters and complete network models. When the power grid topology or operation mode changes, it is necessary to recalculate, which has limited adaptability and generalization ability.

[0006] Furthermore, with the diversification of new energy grid connection scenarios, the site selection and capacity determination problems of grid-based energy storage exhibit a significant combinatorial explosion characteristic. Traditional solution methods based on optimization or iterative search face insufficient computational efficiency and practicality in large-scale systems. Therefore, there is an urgent need for a method that can fully consider the characteristics of the power grid topology and, while ensuring physical rationality, achieve rapid site selection and capacity determination of grid-based energy storage to support the planning and operation decisions of new energy power systems. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method for addressing and determining the capacity of network-based energy storage based on deep graph neural networks.

[0008] The objective of this invention is achieved through the following technical solution: a method for addressing and determining the capacity of network-based energy storage based on deep graph neural networks, the method comprising,

[0009] Based on the network topology, line parameters, and operating point of the power system, the Thevenin equivalent admittance matrix is ​​constructed; combined with the diagonal matrix of the feeder equipment, the grid strength characteristic matrix is ​​defined; the minimum eigenvalue of the grid strength characteristic matrix is ​​used as the generalized operating short-circuit ratio of the system to characterize the grid strength of the system.

[0010] Calculate the right characteristic vector and left eigenvector of the weakest mode corresponding to the minimum eigenvalue of the power grid strength characteristic matrix, and define the participation factor of each feed-in node based on the right characteristic vector and left eigenvector; the node with the largest participation factor is taken as the priority access point for grid-type energy storage.

[0011] In the frequency band of small disturbance stability analysis, the grid-type energy storage is equivalent to the ground susceptance branch. An analytical relationship is established between the grid-type energy storage access capacity and the system generalized operating short-circuit ratio. Based on the analytical relationship, a fixed capacity model is constructed with the constraint that the system generalized operating short-circuit ratio is not lower than a preset stability threshold and the goal of minimizing the energy storage capacity.

[0012] A deep graph neural network is constructed to model the power system as a graph structure, and node feature vectors and edge feature vectors are constructed based on the graph structure.

[0013] The node feature vectors and edge feature vectors are input into a pre-trained deep graph neural network model, which outputs the predicted value of the generalized operation short-circuit ratio of the power system and the predicted value of the participation factor of each node.

[0014] The access nodes for grid-type energy storage are determined based on the predicted values ​​of the participation factors, and the configuration capacity of the grid-type energy storage is calculated by substituting the predicted values ​​of the generalized operating short-circuit ratio of the system and the preset stability threshold into the fixed-capacity model.

[0015] By introducing frequency change rate constraints under extreme disturbance scenarios, the obtained baseline access nodes and baseline capacity are optimized and corrected in a secondary manner to obtain the final configuration result of the grid-type energy storage that simultaneously satisfies the small disturbance stability constraint and the frequency security constraint.

[0016] Specifically, the participation factor is calculated as follows:

[0017] ;

[0018] In the formula, For nodes Participating factors; , These are the components of the left and right eigenvectors, respectively.

[0019] Specifically, the calculation of the priority access point for the grid-type energy storage is as follows:

[0020] ;

[0021] ;

[0022] ;

[0023] In the formula, Priority access nodes for grid-type energy storage; This is the normalized value of the participating factor; For nodes Participating factors; For all The sum of the participation factors of each candidate node.

[0024] Specifically, the parsing relationship is as follows:

[0025] At the access node, an equivalent susceptance increment proportional to the energy storage capacity is added to update the Thevenin equivalent admittance matrix, which in turn updates the grid strength characteristic matrix and its minimum eigenvalue, thus obtaining the generalized operating short-circuit ratio of the system after access.

[0026] Specifically, the volumetric model is as follows:

[0027] ;

[0028] ;

[0029] In the formula, For the node The added grid-type energy storage capacity; The feed-in device capacity is represented by a diagonal matrix. The Thevenin equivalent admittance matrix; It is a diagonal increment matrix; This is the stability threshold; It is an eigenvalue operator.

[0030] Specifically, the deep graph neural network includes a multi-layer feature graph neural network with edges. It aggregates information of neighboring nodes and edges through a message passing mechanism, updates the node embedding vector, and after fusing the embeddings of each layer, outputs the generalized short-circuit ratio of the system and the participation factor of each node through the global readout head and the node readout head, respectively.

[0031] Specifically, the frequency change rate constraint is as follows: for each extreme disturbance scenario, the absolute value of the system's frequency change rate at the initial moment of the disturbance does not exceed a preset threshold, expressed as:

[0032] ;

[0033] ;

[0034] In the formula, A collection of extreme disturbance scenarios; This represents an extreme perturbation scenario that needs to be verified. For the scene The frequency of the lower system over time The function of change; For the scene The net power loss of the lower system; The system reference frequency; The equivalent inertia of the system; For the scene The rate of change of the frequency of the system at the initial moment of the disturbance; The maximum allowable rate of change threshold.

[0035] Specifically, the secondary optimization correction includes correcting the initial frequency change rate using the rapid active power support and equivalent inertia support provided by grid-type energy storage, as well as the scenario-node frequency support effectiveness coefficient, which characterizes the differences in frequency support effects of different access nodes under different scenarios. This correction is expressed as follows:

[0036] ;

[0037] ;

[0038] In the formula, Effective coefficients are provided to support scene-node frequency; Let i be the short-circuit capacity of node i; For node i to scene Equivalent electrical distance from the center of disturbance; Let be the short-circuit capacity of node j; For node j to scene Equivalent electrical distance from the center of disturbance; For the set of candidate access nodes; For nodes Grid-type energy storage in scenarios The amount of rapid active power support that can be provided below; A collection of grid-type energy storage access nodes; For nodes Grid-type energy storage in scenarios The equivalent inertia support that can be provided.

[0039] Specifically, if the baseline access node and baseline capacity already meet the frequency change rate constraints under all extreme disturbance scenarios, then they are directly output as the final configuration result; otherwise, a secondary optimization and correction process is triggered, which includes at least one of the following operations:

[0040] Increase grid-type energy storage capacity at the baseline access node;

[0041] Add grid-type energy storage to candidate nodes where the effective coefficient of scene-node frequency support is greater than a preset threshold;

[0042] The proportion of grid-type energy storage capacity is reallocated among multiple selected nodes;

[0043] Under the premise of satisfying the system's generalized operating short-circuit ratio threshold and the frequency change rate constraints for all scenarios, the optimal configuration scheme for grid-type energy storage with the minimum total additional capacity is solved, and the secondary optimization correction is expressed as:

[0044] ;

[0045] ;

[0046] ;

[0047] In the formula, The grid-type energy storage capacity configured at node i; A diagonal matrix of the rated capacities of each new energy power station in the system. The inverse matrix; This is the equivalent admittance matrix obtained from the receiving-end AC power grid via Thevenin equivalent; The equivalent admittance increment matrix caused by grid-connected energy storage; The stability threshold corresponding to the generalized operating short-circuit ratio of the system.

[0048] The present invention has the following advantages:

[0049] 1. This invention transforms the traditional model-driven approach of relying on eigenvalue analysis and numerical optimization for the location and capacity determination of grid-based energy storage into a learning-driven approach based on power grid topology and operating parameters. It uses a deep graph neural network to parametrically model the power system, fully characterizing the power grid topology and electrical coupling relationships between nodes. This rapidly establishes a mapping relationship between the power grid's operating state and system stability indicators, while simultaneously obtaining the overall system stability indicators and the degree of node participation in unstable modes. Based on this, the invention achieves rapid location determination of grid-based energy storage according to the participation factor results. Furthermore, based on the power grid strength analysis theory characterized by the Generalized Operating Short-Circuit Ratio (gOSCR), and combined with the equivalent impedance model of grid-based energy storage as a voltage source support device, it utilizes the analytical relationship between the grid-based energy storage access capacity and system stability indicators to complete capacity calculation. This achieves reasonable configuration of grid-based energy storage capacity under small disturbance stability constraints, significantly reducing computational complexity and avoiding repeated eigenvalue decomposition and iterative optimization.

[0050] 2. This invention, through a secondary optimization and correction process, can ensure acceptable initial frequency response capabilities of the system under various extreme disturbance scenarios while maintaining the required grid strength. Furthermore, it introduces frequency security constraints and configuration correction mechanisms for extreme disturbance scenarios, ensuring that the resulting grid-type energy storage configuration simultaneously satisfies small-disturbance stability constraints and frequency security constraints. This improves the engineering applicability and robustness of the method in the planning and configuration of high-proportion renewable energy integration into power systems. Attached Figure Description

[0051] Figure 1 This is a schematic diagram of the addressing and sizing method of the present invention. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only for explaining the invention and are not intended to limit the invention; that is, the described embodiments are merely some embodiments of the invention, and not all embodiments. The components of the embodiments of the invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0053] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0054] It should be noted that relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0055] The present invention will be further described below with reference to the accompanying drawings, but the scope of protection of the present invention is not limited to the following description.

[0056] like Figure 1 As shown, a method for addressing and determining the capacity of network-based energy storage based on deep graph neural networks is presented. This method includes:

[0057] Based on the network topology, line parameters, and operating point of the power system, the Thevenin equivalent admittance matrix is ​​constructed; combined with the diagonal matrix of the feeder equipment, the grid strength characteristic matrix is ​​defined; the minimum eigenvalue of the grid strength characteristic matrix is ​​used as the generalized operating short-circuit ratio of the system to characterize the grid strength of the system.

[0058] In this embodiment, the set of feed-in nodes is denoted as... , To determine the number of feed-in nodes involved in the analysis (typically grid connection points of renewable energy plants or power electronic interface nodes), and to characterize the supporting capability of these feed-in nodes "looking outward at the grid," the Thevenin equivalent admittance matrix is ​​obtained from the perspective of the feed-in nodes. Simultaneously, a diagonal matrix of the feed-in equipment capacity is introduced. ,in For nodes Capacity of feed-in equipment (such as the interface of a new energy converter); define the grid strength characteristic matrix. :

[0059] ;

[0060] ;

[0061] ;

[0062] In the formula, This is the power grid strength characteristic matrix; The feed-in device capacity is represented by a diagonal matrix. The Thevenin equivalent admittance matrix; For eigenvalue operators; For matrix The smallest eigenvalue corresponds to the weakest stable mode of the system; This refers to the system's power grid strength / stability margin index. The threshold is given based on stability requirements (such as damping ratio margin). If the generalized operating short-circuit ratio gOSCR of the system is not lower than this value, the grid strength of the system meets the stability requirements; otherwise, grid-type energy storage needs to be introduced to improve the support capacity.

[0063] Calculate the right characteristic vector and left eigenvector of the weakest mode corresponding to the minimum eigenvalue of the power grid strength characteristic matrix, and define the participation factor of each feed-in node based on the right characteristic vector and left eigenvector. The larger the participation factor, the greater the contribution of the node to the weakest mode. Therefore, the node with the largest participation factor is taken as the priority access point for grid-type energy storage.

[0064] Because the minimum characteristic value of the system's generalized operating short-circuit ratio gOSCR To identify which nodes have the greatest influence on the weakest stable mode of the system, the right eigenvector of the weakest mode is introduced. With left eigenvector ;in, Describe the response distribution of the weakest mode at each node. The directions that describe the sensitivity of this mode to parameter perturbations (and are thus used to construct the participation metric) satisfy the standard left-right characteristic relationship, specifically:

[0065] ;

[0066] ;

[0067] ;

[0068] ;

[0069] ;

[0070] ;

[0071] In the formula, For nodes Participating factors; , These are the components of the left and right eigenvectors, respectively; Priority access nodes for grid-type energy storage obtained through analytical addressing; This is the normalized value of the participating factor; For nodes Participating factors; For all The sum of the participation factors of each candidate node.

[0072] After determining the access node, it is necessary to establish an analytical relationship on "how the grid-type energy storage capacity changes the system's generalized operating short-circuit ratio (gOSCR)". The grid-type energy storage is equivalent to a voltage source support device in the small-disturbance stability analysis frequency band. Its grid connection effect can be described by "adding a ground susceptance branch at the access node". An analytical relationship between the grid-type energy storage access capacity and the system's generalized operating short-circuit ratio is established. Based on the analytical relationship, a fixed-capacity model is constructed with the constraint that the system's generalized operating short-circuit ratio is not lower than a preset stability threshold and the goal of minimizing the energy storage capacity.

[0073] Set at node Adding grid-type energy storage capacity is (Units can be MVA or equivalent capacity), and the equivalent susceptance parameter corresponding to a unit capacity grid-type energy storage is set as follows: (Calculated from the equivalent impedance of the grid-connected converter, the grid-connected transformer, and line parameters), the increase in equivalent susceptance caused by the connection is: It can be written as a diagonal increment matrix. Therefore, the system equivalent admittance matrix is ​​updated to... This leads to the generalized operating short-circuit ratio (gOSCR) of the system after access. The above relationship is expressed as:

[0074] ;

[0075] ;

[0076] ;

[0077] ;

[0078] In the formula, The equivalent admittance matrix after grid-connected energy storage is connected; The generalized operating short-circuit ratio (gOSCR) of the system after connection.

[0079] This embodiment describes the dimensionality problem as follows: In the nodes obtained by analytical addressing... At the location, adjust the grid-type energy storage capacity This ensures that the updated system's generalized operating short-circuit ratio (gOSCR) is not lower than the stability threshold. And under this constraint, find the minimum capacity to achieve the configuration goal of "meeting stability requirements and saving the most capacity";

[0080] The constant volume model is expressed as:

[0081] ;

[0082] ;

[0083] In the formula, For the node The added grid-type energy storage capacity; The feed-in device capacity is represented by a diagonal matrix. The Thevenin equivalent admittance matrix; It is a diagonal increment matrix; This is the stability threshold; It is an eigenvalue operator.

[0084] The calculation of the generalized operating short-circuit ratio (gOSCR) and participation factors of the system relies on the construction of the equivalent admittance matrix and eigenvalue decomposition. This process has high computational complexity in large-scale power grids or multi-operational scenarios, making it difficult to meet the needs of rapid planning and online analysis. Therefore, this embodiment introduces a deep graph neural network to parametrically model the mapping relationship between "power grid topology and operating parameters → gOSCR and participation factors." Specifically, a deep graph neural network is constructed to model the power system as a graph structure, and node feature vectors and edge feature vectors are constructed based on the graph structure; the power system is represented as a graph structure. , where the set of nodes Indicates busbar or new energy power station, side assembly This represents a power transmission line. For each node... Construct node feature vectors This describes the load level, power generation capacity, and operating status of the node; for each edge Construct edge features This is used to describe the electrical parameters of a power grid; it can simultaneously characterize the topological characteristics of the power grid and the electrical coupling relationships between nodes, providing a unified input representation for graph neural network learning, specifically as follows:

[0085] ;

[0086] ;

[0087] ;

[0088] In the formula, These represent the active / reactive power of the node load, respectively. These are the upper limits of active and reactive power output capacity of new energy sources, respectively. These are line conductance, susceptance, resistance, and reactance, respectively.

[0089] The node feature vectors and edge feature vectors are input into a pre-trained deep graph neural network model, which outputs the predicted value of the generalized operating short-circuit ratio of the power system and the predicted value of the participation factor of each node. The deep graph neural network includes a multi-layer feature graph neural network with edges. Through a message passing mechanism, each node can gradually aggregate information from neighboring nodes and edge lines to form a node embedding representation that simultaneously contains local electrical characteristics and global topology information. The node embedding vector is then updated. This process essentially achieves the learning of the implicit structural features of the equivalent admittance matrix without explicitly constructing the matrix or performing eigenvalue decomposition. Specifically, the initial node embedding is set as its node feature, and in the first... The layer completes the embedding update through message functions and update functions, specifically as follows:

[0090] ;

[0091] ;

[0092] ;

[0093] In the formula, For nodes In the Layer embedding vector; For node j at the th Embedding vectors of layer -1; For nodes The set of neighbors; and The first Message functions and update functions of the layer; For nodes The initial embedding vector is taken as the node. Original node feature vectors ; For nodes In the The message vector obtained by aggregating its neighboring nodes during the layer message passing process.

[0094] After fusing the embeddings from each layer, the generalized short-circuit ratio and the participation factor of each node are output through the global readout head and the node readout head, respectively. Since the node embeddings of different layers correspond to different scale topological receptive fields, in order to make full use of multi-scale information, this invention fuses the multi-layer node embeddings to obtain a fused node representation. A "single GNN encoder + dual-head readout structure" is constructed to simultaneously output system-level stability indicators and node-level stability features. The fusion and dual-head output process can be uniformly represented as:

[0095] ;

[0096] ;

[0097] ;

[0098] ;

[0099] ;

[0100] ;

[0101] In the formula, Represented as a merged node; For splicing or weighted merging operations; For graph-level aggregation operators; For global network readout; The system's generalized short-circuit ratio prediction value is obtained from the global readout network output; Read the network for the node; Read the original predicted values ​​of the participation factors from the network output for each node; This is the normalized value of the original predicted value of the participating factors; This is the system-level graph representation vector obtained by graph-level aggregation of the full graph node embeddings by the graph neural network encoder.

[0102] The access nodes for grid-type energy storage are determined based on the predicted participation factor values. The configured capacity of the grid-type energy storage is then calculated by substituting the predicted generalized operating short-circuit ratio of the system and a preset stability threshold into the capacity determination model. The deep graph neural network model is obtained through offline joint training. The supervision labels during training include the calculated generalized operating short-circuit ratio of the system and the participation factors of each node. After training, in the online application phase, only one forward inference is needed to obtain the addressing result of the grid-type energy storage, and the capacity is quickly calculated using the analytical capacity determination model. The training and inference process can be uniformly represented as follows:

[0103] ;

[0104] ;

[0105] ;

[0106] ;

[0107] In the formula, For parameters Graph neural network model; For network-based energy storage access nodes obtained based on network prediction participation factors; The normalized result of the node participation factor predicted by the deep graph neural network; A graph structure constructed based on the power system network topology; The node feature matrix; This invention transforms the traditional gOSCR process, which relies on admittance matrix construction and eigenvalue decomposition, into a parameterized prediction problem completed by a deep graph neural network. Based on the prediction results, an analytical capacity-determining model is then used to complete the configuration of grid-based energy storage. This method significantly reduces computational complexity while maintaining clear physical constraints and stability criteria, making it particularly suitable for rapid addressing and capacity determination analysis of grid-based energy storage in large-scale power grids and various operating scenarios.

[0108] The above-mentioned baseline access nodes and capacity for grid-connected energy storage meet the small-disturbance stability constraints. Considering that the aforementioned baseline scheme primarily focuses on configuring the system grid strength and small-disturbance stability margin, and does not explicitly account for frequency security issues under extreme disturbance scenarios such as high-power imbalance, low-inertia operation, and critical component failures, this invention further introduces a frequency change rate constraint based on extreme disturbance scenarios to perform secondary optimization and correction on the baseline scheme, thereby obtaining a grid-connected energy storage configuration that simultaneously meets both grid strength and frequency security constraints.

[0109] By introducing frequency change rate constraints under extreme disturbance scenarios, the obtained baseline access nodes and baseline capacity are optimized and corrected in a secondary manner to obtain the final configuration result of the grid-type energy storage that simultaneously satisfies small disturbance stability constraints and frequency security constraints.

[0110] The frequency change rate constraint is as follows: for each extreme disturbance scenario, the absolute value of the system's frequency change rate at the initial moment of the disturbance does not exceed a preset threshold, expressed as:

[0111] ;

[0112] ;

[0113] In the formula, This is a set of extreme disturbance scenarios, which include at least one or more of the following: large-capacity grid disconnection of new energy power plants, disconnection of important interconnection lines, sudden increase in load, low-inertia operation mode, and so on. Scenario where a fault results in a maximum net power deficit; This represents an extreme perturbation scenario that needs to be verified. For the scene The frequency of the lower system over time The function of change; For the scene The net power loss of the lower system; The system reference frequency; The equivalent inertia of the system; For the scene The rate of change of the frequency of the system at the initial moment of the disturbance; Define the maximum allowable rate of change threshold; define the scenario. Rate of frequency change of the lower system at the initial moment of the disturbance The absolute value of the frequency safety criterion is set at a preset threshold. If the aforementioned baseline scheme fails to meet the above constraints under any extreme disturbance scenario, it indicates that the grid-type energy storage configuration obtained solely based on small-disturbance stability constraints still has insufficient frequency safety and requires further secondary correction.

[0114] Considering that grid-based energy storage can improve system frequency response in the initial stage of a fault through virtual inertia control and fast active power support control, a corrected expression for the initial frequency change rate after grid-based energy storage is established based on the baseline scheme. This secondary optimization correction includes correcting the initial frequency change rate using the fast active power support and equivalent inertia support provided by grid-based energy storage, as well as the scenario-node frequency support effectiveness coefficient, which characterizes the difference in frequency support effects of different access nodes under different scenarios. The scenario-initial frequency change rate after grid-based energy storage is defined as follows:

[0115] ;

[0116] ;

[0117] In the formula, Effective coefficients are provided to support scene-node frequency; Let i be the short-circuit capacity of node i; For node i to scene Equivalent electrical distance from the center of disturbance; Let be the short-circuit capacity of node j; For node j to scene Equivalent electrical distance from the center of disturbance; For the set of candidate access nodes; For nodes Grid-type energy storage in scenarios The amount of rapid active power support that can be provided below; A collection of grid-type energy storage access nodes; For nodes Grid-type energy storage in scenarios The equivalent inertia support that can be provided.

[0118] This invention describes the final configuration problem as follows: Based on the requirement that the system's generalized operating short-circuit ratio is not lower than a preset stability threshold, it further requires that the initial frequency change rate under all extreme disturbance scenarios does not exceed a preset threshold. The final configuration result of the grid-type energy storage is then obtained under these dual constraints. If the baseline access node and baseline capacity already meet the frequency change rate constraints under all extreme disturbance scenarios, they are directly output as the final configuration result; otherwise, a secondary optimization and correction process is triggered, which includes at least one of the following operations:

[0119] Increase grid-type energy storage capacity at the baseline access node;

[0120] Add grid-type energy storage to candidate nodes where the effective coefficient of scene-node frequency support is greater than a preset threshold;

[0121] The proportion of grid-type energy storage capacity is reallocated among multiple selected nodes;

[0122] Under the premise of satisfying the system's generalized operating short-circuit ratio threshold and the frequency change rate constraints for all scenarios, the optimal configuration scheme for grid-type energy storage with the minimum total additional capacity is solved, and the secondary optimization correction is expressed as:

[0123] ;

[0124] ;

[0125] ;

[0126] In the formula, The grid-type energy storage capacity configured at node i; A diagonal matrix of the rated capacities of each new energy power station in the system. The inverse matrix; This is the equivalent admittance matrix obtained from the receiving-end AC power grid via Thevenin equivalent; The equivalent admittance increment matrix caused by grid-connected energy storage; This represents the stability threshold corresponding to the system's generalized operating short-circuit ratio. Through a secondary optimization and correction process, while maintaining the required grid strength, the system can be guaranteed to have an acceptable initial frequency response capability under various extreme disturbance scenarios.

[0127] Through the aforementioned secondary optimization and correction process, while maintaining the required grid strength, the system can be guaranteed to have an acceptable initial frequency response capability under various extreme disturbance scenarios. Therefore, this invention further introduces a frequency security constraint and configuration correction mechanism for extreme disturbance scenarios, ensuring that the resulting grid-type energy storage configuration simultaneously satisfies small-disturbance stability constraints and frequency security constraints. This improves the engineering applicability and robustness of the method in the planning and configuration of power systems with a high proportion of new energy sources.

[0128] The above description is merely a preferred embodiment of the present invention and does not constitute any limitation on the present invention. Any person skilled in the art can make many possible variations and modifications to the technical solution of the present invention, or modify it into equivalent embodiments, without departing from the scope of the present invention. Therefore, any modifications, equivalent changes, and alterations made to the above embodiments based on the technology of the present invention without departing from the scope of the present invention are within the protection scope of the present invention.

Claims

1. A method for addressing and determining the capacity of network-based energy storage based on deep graph neural networks, characterized in that: The method includes, Based on the power system's network topology, line parameters, and operating point, the Thevenin equivalent admittance matrix is ​​constructed; combined with the feeder equipment diagonal matrix, the power grid strength characteristic matrix is ​​defined. The minimum eigenvalue of the power grid strength characteristic matrix is ​​used as the generalized operating short-circuit ratio of the system to characterize the power grid strength of the system. Calculate the right characteristic vector and left eigenvector of the weakest mode corresponding to the minimum eigenvalue of the power grid strength characteristic matrix, and define the participation factor of each feed-in node based on the right characteristic vector and left eigenvector; the node with the largest participation factor is taken as the priority access point for grid-type energy storage. In the frequency band of small disturbance stability analysis, the grid-type energy storage is equivalent to the ground susceptance branch. An analytical relationship is established between the grid-type energy storage access capacity and the system generalized operating short-circuit ratio. Based on the analytical relationship, a fixed capacity model is constructed with the constraint that the system generalized operating short-circuit ratio is not lower than a preset stability threshold and the goal of minimizing the energy storage capacity. A deep graph neural network is constructed to model the power system as a graph structure, and node feature vectors and edge feature vectors are constructed based on the graph structure. The node feature vectors and edge feature vectors are input into a pre-trained deep graph neural network model, which outputs the predicted value of the generalized operating short-circuit ratio of the power system and the predicted value of the participation factor of each node. The baseline access node for grid-type energy storage is determined based on the predicted value of the participation factor, and the baseline capacity is calculated by substituting the predicted value of the generalized operating short-circuit ratio of the system and the preset stability threshold into the fixed capacity model. By introducing frequency change rate constraints under extreme disturbance scenarios, the obtained baseline access nodes and baseline capacity are optimized and corrected in a secondary manner to obtain the final configuration result of the grid-type energy storage that simultaneously satisfies the small disturbance stability constraint and the frequency security constraint.

2. The addressing and capacity determination method for network-based energy storage based on deep graph neural networks according to claim 1, characterized in that: The participation factor is calculated as follows: ； In the formula, For nodes Participating factors; , These are the components of the left and right eigenvectors, respectively.

3. The addressing and capacity determination method for network-based energy storage based on deep graph neural networks according to claim 2, characterized in that: The calculation of the priority access point for the grid-type energy storage is as follows: ； ； ； In the formula, Priority access nodes for grid-type energy storage; This is the normalized value of the participating factor; For nodes Participating factors; For all The sum of the participation factors of each candidate node.

4. The method for addressing and determining the capacity of network-based energy storage based on deep graph neural networks according to claim 3, characterized in that: The analytical relationship is as follows: At the access node, an equivalent susceptance increment proportional to the energy storage capacity is added to update the Thevenin equivalent admittance matrix, which in turn updates the grid strength characteristic matrix and its minimum eigenvalue, thus obtaining the generalized operating short-circuit ratio of the system after access.

5. The method for addressing and determining the capacity of network-based energy storage based on deep graph neural networks according to claim 4, characterized in that: The volumetric model is as follows: ； ； In the formula, For the node The added grid-type energy storage capacity; The feed-in device capacity is represented by a diagonal matrix. The Thevenin equivalent admittance matrix; It is a diagonal increment matrix; This is the stability threshold; It is an eigenvalue operator.

6. The addressing and capacity determination method for network-based energy storage based on deep graph neural networks according to claim 4, characterized in that: The deep graph neural network includes a multi-layer feature graph neural network with edges. It aggregates information of neighboring nodes and edges through a message passing mechanism, updates the node embedding vector, and after fusing the embeddings of each layer, outputs the generalized short-circuit ratio of the system and the participation factor of each node through the global readout head and the node readout head, respectively.

7. The addressing and capacity determination method for network-based energy storage based on deep graph neural networks according to claim 4, characterized in that: The frequency change rate constraint is as follows: for each extreme disturbance scenario, the absolute value of the system's frequency change rate at the initial moment of the disturbance does not exceed a preset threshold, expressed as: ； ； In the formula, A collection of extreme disturbance scenarios; This represents an extreme perturbation scenario that needs to be verified. For the scene The frequency of the lower system over time The function of change; For the scene The net power loss of the lower system; The system reference frequency; The equivalent inertia of the system; For the scene The rate of change of the frequency of the system at the initial moment of the disturbance; The maximum allowable rate of change threshold.

8. The addressing and capacity determination method for network-based energy storage based on deep graph neural networks according to claim 7, characterized in that: The secondary optimization correction includes correcting the initial frequency change rate using the rapid active power support and equivalent inertia support provided by the grid-type energy storage, as well as the scenario-node frequency support effectiveness coefficient, which characterizes the difference in frequency support effect of different access nodes in different scenarios. This correction is expressed as: ； ； In the formula, Effective coefficients are provided to support scene-node frequency; Let i be the short-circuit capacity of node i; For node i to scene Equivalent electrical distance from the center of disturbance; Let be the short-circuit capacity of node j; For node j to scene Equivalent electrical distance from the center of disturbance; For the set of candidate access nodes; For nodes Grid-type energy storage in scenarios The amount of rapid active power support that can be provided below; It is a collection of grid-type energy storage access nodes; For nodes Grid-type energy storage in scenarios The equivalent inertia support that can be provided.

9. The addressing and capacity determination method for network-based energy storage based on deep graph neural networks according to claim 7, characterized in that: If the baseline access node and baseline capacity already meet the frequency change rate constraints under all extreme disturbance scenarios, then they are directly output as the final configuration result; otherwise, a secondary optimization and correction process is triggered, which includes at least one of the following operations: Increase grid-type energy storage capacity at the baseline access node; Add grid-type energy storage to candidate nodes where the effective coefficient of scene-node frequency support is greater than a preset threshold; The proportion of grid-type energy storage capacity is reallocated among multiple selected nodes; Under the premise of satisfying the system's generalized operating short-circuit ratio threshold and the frequency change rate constraints for all scenarios, the optimal configuration scheme for grid-type energy storage with the minimum total additional capacity is solved, and the secondary optimization correction is expressed as: ； ； ； In the formula, The grid-type energy storage capacity configured at node i; A diagonal matrix of the rated capacities of each new energy power station in the system. The inverse matrix; This is the equivalent admittance matrix obtained from the receiving-end AC power grid via Thevenin equivalent; The equivalent admittance increment matrix caused by grid-connected energy storage; This is the stability threshold corresponding to the generalized operating short-circuit ratio of the system.

Images