A multi-scale large-scale power system scenario simulation and generation method

Abstract

The application relates to the technical field of scene simulation generation, and discloses a multi-scale large-scale power system scene simulation and generation method, which comprises the following steps: calculating operation data sequences under different time scales to obtain operation characteristics, and constructing a time-varying operation characteristic map of a power system based on a topological structure and the operation characteristics; calculating activation scores of power nodes under different time scales, performing weighted clustering on the power nodes, and extracting group operation characteristics in a group node set; converting the group operation characteristics into a multi-behavior probability distribution map by using a conditional generation adversarial network; and performing behavior simulation on the power nodes in the group node set based on the multi-behavior probability distribution map limitation to obtain a power system simulation scene. By fusing multi-scale operation characteristics of the power nodes and topological information and by simulating multi-behavior distribution of the group nodes by using the conditional generation adversarial network, dynamic simulation of large-scale power nodes in a power system scene is realized.

Description

Technical Field

[0001] This invention relates to the field of scene simulation and generation, and in particular to a method for simulating and generating multi-scale large-scale power system scenes. Background Technology

[0002] With the continuous development of new power systems, the power grid is expanding in scale and becoming increasingly complex, exhibiting highly dynamic and uncertain operating modes. On the one hand, the large-scale integration of distributed energy resources, renewable energy sources, and flexible loads has led to power system operation characterized by multi-source inputs and multi-scenario switching. On the other hand, the operating environment of the power system is also affected by multiple factors such as climate change, external disturbances, and load fluctuations, resulting in significant complexity and diversity in operating states across both time and space. Against this backdrop, the safe and stable operation of the power system not only requires real-time monitoring and control but also relies heavily on high-precision, multi-dimensional scenario simulation and prediction to identify potential risks in advance and formulate effective response strategies.

[0003] In practical applications, the operating behavior of power systems often exhibits comprehensive characteristics across scales and levels. For example, transient oscillations may exist on millisecond timescales, power flow fluctuations on minute timescales, and load evolution trends on hourly or even longer timescales. If simulations only focus on a single scale or local characteristics, they cannot accurately reproduce the overall picture of the system under multi-source disturbances. Furthermore, the lack of modeling for collective behavior can cause simulation results to deviate from the systematic patterns observed in actual operation. These shortcomings significantly limit the application of existing scenario simulation methods in new power systems, making it difficult to meet the high requirements for simulation realism, versatility, and adaptability in complex environments.

[0004] In existing research, CN120124448B proposes a relay protection simulation system. This system includes a power system model, a protection device simulation module, and a fault simulation module, capable of simulating the response of protection devices under different fault conditions within a created simulation environment. This patent also incorporates machine learning algorithms to optimize the simulation results, achieving dynamic adjustment of protection action parameters and effectively improving the accuracy and response speed of the protection system. However, this method still has certain technical limitations: the simulation focuses on response testing at the protection device level, lacking multi-scale modeling of the overall power system operation scenario; the generated simulation scenario is relatively deterministic, making it difficult to cover random disturbances and diverse operating states in large-scale systems.

[0005] To address this problem, this invention proposes a multi-scale large-scale power system scenario simulation and generation method. By intelligently analyzing the behavioral characteristics of different power nodes, it outputs a multi-behavior probability distribution map, thereby realizing large-scale power system scenario simulation and providing scenario support and scientific basis for power system scheduling, planning and risk analysis. Summary of the Invention

[0006] This invention provides a method for simulating and generating large-scale power system scenarios at multiple scales. Addressing the limitations of traditional power system simulations, such as single time scale, lack of group behavior modeling, and rigid scenario generation methods, step S1 introduces multi-timescale operational feature extraction and time-varying operational feature map construction. This not only reflects the dynamic evolution of power nodes but also characterizes the global dependency structure of the power system by combining topological relationships, overcoming the limitations of traditional methods that rely solely on single-point or single-time-series data. Step S2 introduces a weighted clustering method based on activation scoring, enabling power nodes to aggregate differentially at multiple scales, forming a hierarchical set of group nodes, thus solving the technical problem of insufficient interpretability of power node clustering results. Step S3 utilizes conditional generative adversarial networks to map group operational features into multi-behavior probability distribution maps, achieving probabilistic expression and diversified generation of group behavior, overcoming the bottleneck of traditional deterministic modeling's inability to reflect the heterogeneity and randomness of group nodes. Step S4, by mapping multi-behavior probability distributions to node sets and driving behavior simulation, more realistic power node behavior simulation results under multiple disturbances and scenarios can be generated, resulting in a power system simulation scenario.

[0007] To achieve the above objectives, this invention provides a method for simulating and generating multi-scale large-scale power system scenarios, comprising the following steps:

[0008] S1: Collect the operating data sequence and topology of each power node in the power system, calculate the operating characteristics of the operating data sequence at different time scales, and construct the time-varying operating characteristic map of the power system based on the topology and operating characteristics;

[0009] S2: Based on the time-varying operation feature map of the power system, the activation scores of power nodes at different time scales are calculated, and the power nodes are weighted and clustered based on the activation scores to form a hierarchical group node set, and the group operation features in the group node set are extracted.

[0010] S3: Use conditional generative adversarial networks to convert group operation characteristics into multi-behavior probability distribution maps;

[0011] S4: Map the multi-behavior probability distribution map to the corresponding group node set, and perform behavior simulation on the power nodes in the group node set based on the multi-behavior probability distribution map constraints to obtain the power system simulation scenario.

[0012] As a further improvement of the present invention:

[0013] Furthermore, the operational data sequences and topology of each power node in the power system are collected, including:

[0014] The operating data sequence includes voltage, current, and power sequences;

[0015] The topology is in the form of a graph, with power nodes as nodes in the topology and the power connections between power nodes as edge relationships between nodes in the topology. The physical distance between power nodes is obtained, where the power connection relationship is determined by whether there are directly connected transmission lines between power nodes. If there are directly connected transmission lines between power nodes, the edge relationship between nodes in the topology is 1, indicating that the nodes corresponding to the two power nodes in the topology are neighbors. Otherwise, the edge relationship between nodes in the topology is 0, indicating that the nodes corresponding to the two power nodes in the topology are not neighbors.

[0016] Furthermore, the operational characteristics of the running data sequence at different time scales were calculated, including:

[0017] The running data sequence is divided into multiple sequence segments according to different time scales, and the mean of the sequence segments is calculated as the sequence segment mean sequence of the running data sequence at different time scales. The time scales include hourly, daily, and weekly scales.

[0018] The time-domain and frequency-domain features of the mean sequence of the sequence segment at different time scales are extracted respectively. The time-domain and frequency-domain features are used as the running features to obtain the running features of the running data sequence at different time scales. The time-domain features include the mean, variance, skewness and kurtosis of the mean sequence segment, and the frequency-domain features include the spectral energy distribution and dominant frequency components of the mean sequence segment.

[0019] Furthermore, a time-varying operating characteristic diagram of the power system is constructed based on the topology and operating characteristics, including:

[0020] Based on the topology and operational characteristics, the time-varying weight coefficients of any two power nodes in the power system at the same time scale are calculated.

[0021] The time-varying operation feature map of the power system is constructed based on the time-varying weight coefficient. The time-varying operation feature map consists of a time-varying operation feature matrix and a time-varying weight matrix. The time-varying operation feature matrix consists of the operation features of the power node's operation data sequence at different time scales, and the time-varying weight matrix consists of the time-varying weight coefficients of any two power nodes at the same time scale.

[0022] Furthermore, based on the time-varying operational characteristic map of the power system, the activation scores of power nodes at different time scales are calculated, including:

[0023] The activation score is calculated using the following formula:

[0024] ;

[0025] in, Indicates the first in the power system The activation score of each power node at time scale t. Indicates the first in the power system The operational characteristics of a power node at time scale t, where time scale t belongs to the set of time scales. , Describing the L2 norm, Represents the first power system determined based on the topology diagram. The set of neighboring nodes of a power node. Represents the set of neighboring nodes The number of neighboring nodes in the middle, Represents the set of neighboring nodes The Middle The neighboring nodes of a power node, Indicates the first in the power system The time-varying weight coefficients of each power node and its neighbor node e at time scale t. This represents the total number of power nodes in a power system. This represents the internal scale characteristic control coefficient. This represents the external time-varying weight control coefficient.

[0026] Furthermore, based on activation scores, power nodes are weighted and clustered to form a hierarchical set of nodes, including:

[0027] The operational characteristics of power nodes at different time scales are weighted based on activation scores to obtain the weighted operational characteristics of power nodes.

[0028] A clustering algorithm that integrates topological relationships is used to perform weighted clustering of power nodes. The weighted clustering process is as follows:

[0029] Obtain the edge relationships between power nodes in the topology;

[0030] Calculate the similarity between weighted running features;

[0031] The distance between power nodes is calculated based on the similarity between edge relationships and weighted operational features.

[0032] Initially, select K power nodes as cluster centers, and assign power nodes that are not cluster centers to the cluster level of the nearest cluster center;

[0033] Repeat the update of cluster centers until the cluster centers no longer change;

[0034] All power nodes in the cluster level to which the cluster center belongs are considered as a group node set, resulting in K group node sets;

[0035] Extract the group operation characteristics from the group node set.

[0036] Furthermore, the group operation characteristics are the weighted average of the operation characteristics of all power nodes in the group node set.

[0037] Furthermore, step S3 includes:

[0038] The conditional generative adversarial network includes a generator and a discriminator. The generator's network structure consists of an input layer, a hidden layer, a conditional fusion layer, and a softmax output layer. The generator is used to receive group operation features and output the probability of each behavior corresponding to the group operation features. During the training process of the generator, the discriminator is used to check the output results of the generator and update the trainable model parameters of the generator.

[0039] Conditional generative adversarial networks are used to receive group operation features, output the probability of each behavior corresponding to the group operation features, and use the probability of each behavior as a multi-behavior probability distribution map transformed from the group operation features.

[0040] Furthermore, the multi-behavior probability distribution graph is mapped to the corresponding set of group nodes, and behavioral simulations of the power nodes in the group node set are performed based on the constraints of the multi-behavior probability distribution graph, including:

[0041] The power nodes in the group node set receive the multi-behavior probability distribution map, and based on the probability of different behaviors occurring, execute the behaviors in the multi-behavior probability distribution map to obtain the power system simulation scenario.

[0042] Compared with existing technologies, this invention proposes a method for simulating and generating multi-scale large-scale power system scenarios, which has the following advantages:

[0043] First, this invention utilizes activation scoring to evaluate the operational characteristics of power nodes at different time scales. The activation score consists of two parts: one part measures the overall contribution of the power node's operational characteristics at the current time scale relative to all time scales, specifically reflecting the dynamic importance of the power node at different time scales through the L2 norm ratio, which can highlight operational characteristics that are significant in an instantaneous or local manner; the other part reflects the coupling strength and centrality of the power node in the group topology by calculating the average of the comprehensive edge weights of all directly connected transmission lines of the power node, thereby capturing the potential collaboration or influence of the power node in the power system, and thus effectively integrating the dynamic characteristics of the power node itself with the neighborhood topology relationship, so that the activation score can reflect both the transient behavior of the node and its group correlation.

[0044] Meanwhile, by weighting the operational characteristics of power nodes at different time scales, this invention highlights the dynamic importance of power nodes at specific time scales, avoiding the information loss problem caused by traditional clustering methods relying solely on features from a single time scale. Furthermore, by incorporating the topological structure information of the power system into the clustering process and using edge relationships to correct the distances between power nodes, the physical coupling relationships between power nodes with edge relationships are preserved, improving the electrical-physical rationality and system interpretability of the clustering results. Attached Figure Description

[0045] Figure 1 This is a flowchart illustrating a method for simulating and generating multi-scale large-scale power system scenarios according to an embodiment of the present invention.

[0046] Figure 2 This is a flowchart of weighted clustering provided in an embodiment of the present invention. Detailed Implementation

[0047] The realization of the objectives, functional characteristics, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0048] This invention provides a method for simulating and generating multi-scale large-scale power system scenarios. The execution entity of this method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this invention: a server, a terminal, etc. In other words, the method can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster.

[0049] Reference Figure 1 as well as Figure 2 Embodiment 1 of the present invention is as follows:

[0050] S1: Collect the operating data sequence and topology of each power node in the power system, calculate the operating characteristics of the operating data sequence at different time scales, and construct the time-varying operating characteristic map of the power system based on the topology and operating characteristics.

[0051] Collect operational data sequences and topology of each power node in the power system, including:

[0052] As an embodiment of the present invention, the power node is a key component in the power system with observable operating status, including generators, transformers, transmission line nodes, circuit breakers and load nodes, etc. The operating data sequence consists of operating data at multiple operating times, wherein the operating data includes voltage, current and power.

[0053] The operating data sequence includes voltage, current, and power sequences;

[0054] As an embodiment of the present invention, an adaptive dynamic sampling method is used to collect the operating data sequence of power nodes. During the collection of the operating data sequence, the adaptive dynamic sampling method dynamically adjusts the sampling interval of the operating data sequence based on the instantaneous rate of change and standard deviation of the currently collected power sequence. The dynamic adjustment formula for the sampling interval is:

[0055] ;

[0056] in, This indicates the result of dynamic adjustment of the sampling interval. This indicates the preset minimum sampling interval (e.g., 1 minute). This indicates the preset maximum sampling interval (e.g., 10 minutes). This represents an exponential function with the natural constant as its base. This represents the standard deviation of the currently acquired power sequence. This represents the instantaneous rate of change of power in the currently acquired power sequence, where x represents the currently acquired power. This represents the power collected at the previous runtime. This represents the standard deviation control factor. Indicates the control coefficient for the instantaneous rate of change of power, set Set to 0.6 It is 0.4;

[0057] It should be noted that this invention dynamically adjusts the sampling interval by calculating the instantaneous rate of change and standard deviation of the collected power sequence of the power node in real time. This increases the sampling density when the power node is in a state of rapid fluctuation or abnormality, thereby accurately capturing key operating characteristics and sudden events. When the power node is in a stable state, the sampling interval is extended to reduce data redundancy and storage pressure, and optimize communication bandwidth and computing resource usage.

[0058] The topology is a graph structure, with power nodes as nodes and the power connections between them as edges. The physical distance between power nodes is determined by whether there are directly connected transmission lines between them. If there are directly connected transmission lines, the edge relationship is 1, indicating that the nodes corresponding to the two power nodes are neighbors; otherwise, the edge relationship is 0, indicating that the nodes corresponding to the two power nodes are not neighbors. Specifically, the physical distance between power nodes is the Euclidean distance between the coordinates of the power node's location.

[0059] The operational characteristics of the running data sequence at different time scales were calculated, including:

[0060] The running data sequence is divided into multiple sequence segments according to different time scales, and the mean of the sequence segments is calculated as the sequence segment mean sequence of the running data sequence at different time scales. The time scales include hourly, daily, and weekly scales. Specifically, the running data sequence includes voltage, current, and power sequences, and the sequence segment mean sequence of the running data sequence at different time scales includes the sequence segment mean sequence of the voltage part, the sequence segment mean sequence of the current part, and the sequence segment mean sequence of the power part.

[0061] Specifically, the running time range of the running data in the sequence segment at the hourly scale is 1 hour, the running time range of the running data in the sequence segment at the daily scale is 24 hours, and the running time range of the running data in the sequence segment at the weekly scale is 168 hours, wherein the running time range is the difference between the running times of the running data at both ends of the sequence segment;

[0062] The time-domain and frequency-domain features of the mean sequence of the sequence segment at different time scales are extracted respectively. The time-domain and frequency-domain features are used as the running features to obtain the running features of the running data sequence at different time scales. The time-domain features include the mean, variance, skewness and kurtosis of the mean sequence segment, and the frequency-domain features include the spectral energy distribution and dominant frequency components of the mean sequence segment.

[0063] Optionally, Fourier transform or short-time Fourier transform can be used to extract the spectral energy distribution and dominant frequency component of the mean sequence of the sequence segment. The dominant frequency component is the frequency component with the highest energy in the energy spectrum, and the spectral energy distribution is the energy spectrum of the mean sequence of the sequence segment. The energy spectrum is composed of the energy under different frequency components.

[0064] Based on topology and operational characteristics, a time-varying operational characteristic diagram of the power system is constructed, including:

[0065] Based on the topology and operational characteristics, the time-varying weight coefficients of any two power nodes in the power system at the same time scale are calculated. The formula for calculating the time-varying weight coefficients is as follows:

[0066] ;

[0067] ;

[0068] ;

[0069] ;

[0070] in, Indicates the first in the power system The time-varying weight coefficients of the j-th power node and the j-th power node at time scale t, where time scale t belongs to the set of time scales. , This represents the first control factor (set to 0.4). This represents the second control factor (set to 0.6). , This represents the total number of power nodes in a power system;

[0071] Indicates the first in the power system The correlation coefficient between the mean sequence of the current component of the j-th power node and the j-th power node at time scale t. Indicates the first in the power system The correlation coefficient between the voltage segment mean sequence of the j-th power node and the voltage segment mean sequence of the j-th power node at time scale t;

[0072] Optionally, the correlation coefficient is calculated using either the Pearson correlation coefficient or the Spearman correlation coefficient.

[0073] Indicates the first in the power system The power coupling strength between a power node and the j-th power node at a time scale t, representing the mean sequence of the power components. Indicates the first in the power system The sequence mean of the power component of a power node at a time scale t. Let represent the sequence mean of the power component of the j-th power node in a power system at time scale t. This represents the maximum value of the sequence mean of the power component of all power nodes in a power system at a time scale t.

[0074] Indicates the first in the power system The distance factor between the j-th power node and the j-th power node Indicates the first in the power system The physical distance between the j-th power node and the j-th power node. This represents the distance attenuation factor (set to 0.2).

[0075] Specifically, when the voltage and current of power nodes are highly correlated and their power is close, the edge weights are increased to ensure that key coupling relationships are accurately captured; when the physical distance between power nodes is far or the power difference is large, the edge weights are decreased to effectively suppress false or weak couplings and improve the authenticity and robustness of the time-varying operation feature map representation. This method can enhance the ability of the time-varying operation feature map to perceive the dynamic interaction of power nodes and realize the accurate expression of multi-dimensional coupling information in the time-varying operation feature map modeling of power systems.

[0076] The time-varying operation feature map of the power system is constructed based on the time-varying weight coefficient. The time-varying operation feature map consists of a time-varying operation feature matrix and a time-varying weight matrix. The time-varying operation feature matrix consists of the operation features of the power node's operation data sequence at different time scales, and the time-varying weight matrix consists of the time-varying weight coefficients of any two power nodes at the same time scale.

[0077] S2: Based on the time-varying operation feature map of the power system, the activation scores of power nodes at different time scales are calculated, and the power nodes are weighted and clustered based on the activation scores to form a hierarchical group node set. The group operation features in the group node set are then extracted.

[0078] Based on the time-varying operation characteristic map of the power system, the activation scores of power nodes at different time scales are calculated, including:

[0079] The activation score is calculated using the following formula:

[0080] ;

[0081] in, Indicates the first in the power system The activation score of each power node at time scale t. Indicates the first in the power system The operational characteristics of a power node at time scale t, where time scale t belongs to the set of time scales. , Describing the L2 norm, Represents the first power system determined based on the topology diagram. The set of neighboring nodes of a power node. Represents the set of neighboring nodes The number of neighboring nodes in the middle, Represents the set of neighboring nodes The Middle The neighboring nodes of a power node, Indicates the first in the power system The time-varying weight coefficients of each power node and its neighbor node e at time scale t. This represents the total number of power nodes in a power system. This represents the internal scale feature control coefficient (set to 0.5). This represents the external time-varying weight control coefficient (set to 0.5).

[0082] Based on activation scores, power nodes are weighted and clustered to form a hierarchical set of nodes, including:

[0083] The operational characteristics of power nodes at different time scales are weighted based on activation scores to obtain the weighted operational characteristics of power nodes.

[0084] Specifically, in the power system, the first The weighted operating characteristics of the power nodes are as follows: :

[0085] ;

[0086] For reference Figure 2 The weighted clustering flowchart shown illustrates the weighted clustering of power nodes using a clustering algorithm that integrates topological relationships. The weighted clustering process is as follows:

[0087] S201: Obtain the edge relationships between power nodes in the topology;

[0088] S202: Calculate the similarity between the weighted running features; optionally, the similarity is calculated using a cosine similarity algorithm;

[0089] S203: Based on the similarity between edge relationships and weighted operational features, the distance between power nodes is calculated; specifically, in the power system, the... The distance between the j-th power node and the j-th power node is:

[0090] ;

[0091] in, Indicates the first in the power system The distance between the j-th power node and the j-th power node. Indicates the first in the power system The similarity between the weighted operational characteristics of the j-th power node and the j-th power node. This represents the similarity control coefficient (set to 1.2). Indicates the first The edge relationship between the j-th power node and the j-th power node. Represents the topology penalty coefficient, set It is 0.4;

[0092] S204: Initialize and select K power nodes as cluster centers, and assign power nodes that are not cluster centers to the cluster level of the nearest cluster center; optionally, set K to 10;

[0093] S205: Repeatedly update the cluster centers until the cluster centers no longer change; optionally, the update method of the cluster centers is the update method in the K-means algorithm;

[0094] All power nodes in the cluster level to which the cluster center belongs are considered as a group node set, resulting in K group node sets;

[0095] Extract the group operation characteristics from the group node set.

[0096] Specifically, by defining a distance function based on a joint metric of similarity and edge relationships, the coupling strength and feature similarity between power nodes are considered simultaneously. This effectively avoids the problem of traditional Euclidean distance metrics being insensitive to high-dimensional features, significantly improving the precision and robustness of clustering. The iterative update mechanism of cluster centers ensures the stability and global consistency of clustering results, reducing biases caused by initialization or noise. The extracted group operation features not only retain the dynamics of local nodes but also reflect the overall synergy of the group, facilitating subsequent group behavior modeling and power system simulation.

[0097] The group operation characteristics are the weighted average of the operation characteristics of all power nodes in the group node set.

[0098] S3: Use conditional generative adversarial networks to convert group operation characteristics into a multi-behavior probability distribution map.

[0099] The conditional generative adversarial network includes a generator and a discriminator. The generator's network structure consists of an input layer, a hidden layer, a conditional fusion layer, and a softmax output layer. The generator is used to receive group operation features and output the probability of each behavior corresponding to the group operation features. During the training process of the generator, the discriminator is used to check the output results of the generator and update the trainable model parameters of the generator.

[0100] Specifically, for the generator, the input layer is used to receive the group's running features and random noise, the hidden layer is in the form of a fully connected layer + ReLU activation function, which is used to extract the hidden state of the group's running features, the conditional fusion layer uses an attention mechanism or a gated fusion mechanism to fuse the hidden state and random noise to obtain fused features, and the Softmax output layer uses the Softmax function to receive the fused features and output the probability of each behavior occurring.

[0101] The discriminator receives the probability of each behavior output by the generator and the probability of the actual behavior, and compares the distribution difference between the two probabilities. It constructs a training loss function to minimize the distribution difference and updates the trainable model parameters of the generator. Optionally, the discriminator adopts a multi-index distribution difference measurement method, including KL divergence, JS divergence and Wasserstein distance, and selects the optimal index according to the behavior model scenario to be simulated, thereby improving the stability and convergence accuracy of behavior distribution adversarial learning.

[0102] Optionally, the training loss function can be trained and optimized using the stochastic gradient descent algorithm or the Adam optimizer to obtain stable trainable model parameters of the generator and construct the generator in the conditional generative adversarial network.

[0103] Conditional generative adversarial networks are used to receive group operation features, output the probability of each behavior corresponding to the group operation features, and use the probability of each behavior as a multi-behavior probability distribution map transformed from the group operation features.

[0104] It should be noted that the hidden layer adopts the form of a fully connected layer + ReLU activation function to extract the hidden state of the group operation features, ensuring the model's expressive ability in capturing the nonlinear relationship of group features. The conditional fusion layer effectively realizes the adaptive fusion of features and noise, improving the generator's ability to characterize complex behavioral patterns. The Softmax output layer outputs the behavior probability distribution, enabling power nodes to exhibit randomness and diversity in multiple behavior choices, breaking through the limitations of traditional deterministic simulation methods. Combined with the discriminator's distribution difference measurement, it can achieve diversified group behavior and realistic reproduction of multiple power nodes in power system operation while ensuring simulation accuracy, providing innovative support for the simulation research of power systems under multiple disturbances and multiple scenarios.

[0105] As an embodiment of the present invention, the categories of the behaviors include, but are not limited to, current regulation behavior, voltage stabilization behavior, power distribution behavior, load transfer behavior, and backup power activation behavior; when a power node selects current regulation behavior, it performs fine adjustment or amplitude increase of the current amplitude; when a power node selects voltage stabilization behavior, it performs voltage offset correction, such as increasing the bus voltage or decreasing the line voltage; when a power node selects power distribution behavior, it transfers part of the load to neighboring nodes; when a power node selects backup power activation behavior, it activates backup equipment or switches to redundant lines.

[0106] S4: Map the multi-behavior probability distribution map to the corresponding group node set, and perform behavior simulation on the power nodes in the group node set based on the multi-behavior probability distribution map constraints to obtain the power system simulation scenario.

[0107] The multi-behavior probability distribution graph is mapped to the corresponding set of group nodes, and the behavior of the power nodes in the group node set is simulated based on the multi-behavior probability distribution graph constraints, including:

[0108] The power nodes in the group node set receive the multi-behavior probability distribution map, and based on the probability of different behaviors occurring, execute the behaviors in the multi-behavior probability distribution map to obtain the power system simulation scenario.

[0109] Specifically, during power system scenario simulation, each power node randomly selects and executes the chosen behavior according to a multi-behavior probability distribution diagram. For example, in the multi-behavior probability distribution diagram of a certain group of nodes, the probability of current regulation behavior is 0.30, the probability of voltage stabilization behavior is 0.25, the probability of power distribution behavior is 0.20, the probability of load transfer behavior is 0.15, and the probability of reserve activation behavior is 0.10. Then, the power node will select and execute different behaviors according to the corresponding probability proportions during operation.

[0110] This allows some power nodes to adjust current, others to regulate voltage, and still others to optimize power allocation or perform load transfer. Through this probability distribution-based simulation of collective node behavior, the power system can present an overall evolution process that more closely resembles its actual operating state, thereby improving the simulation realism and robustness of the power system under multiple scenarios and disturbance conditions.

[0111] It should be understood that the embodiments described are for illustrative purposes only and are not limited to this structure in the scope of the patent application.

[0112] It should be noted that the sequence numbers of the above embodiments of the present invention are merely for descriptive purposes and do not represent the superiority or inferiority of the embodiments. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method 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, apparatus, article, or method. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0113] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0114] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A method for simulating and generating multi-scale large-scale power system scenarios, characterized in that, The method includes: S1: Collect the operating data sequence and topology of each power node in the power system, calculate the operating characteristics of the operating data sequence at different time scales, and construct the time-varying operating characteristic map of the power system based on the topology and operating characteristics; S2: Based on the time-varying operation feature map of the power system, the activation scores of power nodes at different time scales are calculated, and the power nodes are weighted and clustered based on the activation scores to form a hierarchical group node set, and the group operation features in the group node set are extracted. The formula for calculating the activation score is: ； in, Indicates the first in the power system The activation score of each power node at time scale t. Indicates the first in the power system The operational characteristics of a power node at time scale t, where time scale t belongs to the set of time scales. , Describing the L2 norm, Represents the first power system determined based on the topology diagram. The set of neighboring nodes of a power node. Represents the set of neighboring nodes The number of neighboring nodes in the middle, Represents the set of neighboring nodes The Middle The neighboring nodes of a power node, Indicates the first in the power system The time-varying weight coefficients of each power node and its neighbor node e at time scale t. This represents the total number of power nodes in a power system. This represents the internal scale characteristic control coefficient. Indicates the external time-varying weight control coefficient; S3: Use conditional generative adversarial networks to convert group operation characteristics into multi-behavior probability distribution maps; S4: Map the multi-behavior probability distribution map to the corresponding group node set, and perform behavior simulation on the power nodes in the group node set based on the multi-behavior probability distribution map constraints to obtain the power system simulation scenario.

2. The method for simulating and generating multi-scale large-scale power system scenarios as described in claim 1, characterized in that, Collect operational data sequences and topology of each power node in the power system, including: The operating data sequence includes voltage, current, and power sequences; The topology is in the form of a graph, with power nodes as nodes in the topology and the power connections between power nodes as edge relationships between nodes in the topology. The physical distance between power nodes is obtained, where the power connection relationship is determined by whether there are directly connected transmission lines between power nodes. If there are directly connected transmission lines between power nodes, the edge relationship between nodes in the topology is 1, indicating that the nodes corresponding to the two power nodes in the topology are neighbors. Otherwise, the edge relationship between nodes in the topology is 0, indicating that the nodes corresponding to the two power nodes in the topology are not neighbors.

3. The method for simulating and generating multi-scale large-scale power system scenarios as described in claim 2, characterized in that, The operational characteristics of the running data sequence at different time scales were calculated, including: The running data sequence is divided into multiple sequence segments according to different time scales, and the mean of the sequence segments is calculated as the sequence segment mean sequence of the running data sequence at different time scales. The time scales include hourly, daily, and weekly scales. The time-domain and frequency-domain features of the mean sequence of the sequence segment at different time scales are extracted respectively. The time-domain and frequency-domain features are used as the running features to obtain the running features of the running data sequence at different time scales. The time-domain features include the mean, variance, skewness and kurtosis of the mean sequence segment, and the frequency-domain features include the spectral energy distribution and dominant frequency components of the mean sequence segment.

4. The method for simulating and generating multi-scale large-scale power system scenarios as described in claim 2, characterized in that, Based on topology and operational characteristics, a time-varying operational characteristic diagram of the power system is constructed, including: Based on the topology and operational characteristics, the time-varying weight coefficients of any two power nodes in the power system at the same time scale are calculated. The time-varying operation feature map of the power system is constructed based on the time-varying weight coefficient. The time-varying operation feature map consists of a time-varying operation feature matrix and a time-varying weight matrix. The time-varying operation feature matrix consists of the operation features of the power node's operation data sequence at different time scales, and the time-varying weight matrix consists of the time-varying weight coefficients of any two power nodes at the same time scale.

5. The method for simulating and generating multi-scale large-scale power system scenarios as described in claim 1, characterized in that, Based on activation scores, power nodes are weighted and clustered to form a hierarchical set of nodes, including: The operational characteristics of power nodes at different time scales are weighted based on activation scores to obtain the weighted operational characteristics of power nodes. A clustering algorithm that integrates topological relationships is used to perform weighted clustering of power nodes. The weighted clustering process is as follows: Obtain the edge relationships between power nodes in the topology; Calculate the similarity between weighted running features; The distance between power nodes is calculated based on the similarity between edge relationships and weighted operational features. Initially, select K power nodes as cluster centers, and assign power nodes that are not cluster centers to the cluster level of the nearest cluster center; Repeat the update of cluster centers until the cluster centers no longer change; All power nodes in the cluster level to which the cluster center belongs are considered as a group node set, resulting in K group node sets; Extract the group operation characteristics from the group node set.

6. The method for simulating and generating multi-scale large-scale power system scenarios as described in claim 5, characterized in that, The group operation characteristics are the weighted average of the operation characteristics of all power nodes in the group node set.

7. The method for simulating and generating multi-scale large-scale power system scenarios as described in claim 1, characterized in that, Step S3 includes: The conditional generative adversarial network includes a generator and a discriminator. The generator's network structure consists of an input layer, a hidden layer, a conditional fusion layer, and a softmax output layer. The generator is used to receive group operation features and output the probability of each behavior corresponding to the group operation features. During the training process of the generator, the discriminator is used to check the output results of the generator and update the trainable model parameters of the generator. The conditional generative adversarial network is used to receive the group operation characteristics, output the probability of each behavior corresponding to the group operation characteristics, and use the probability of each behavior as the multi-behavior probability distribution map obtained by transforming the group operation characteristics.

8. The method for simulating and generating multi-scale large-scale power system scenarios as described in claim 7, characterized in that, The multi-behavior probability distribution graph is mapped to the corresponding set of group nodes, and the behavior of the power nodes in the group node set is simulated based on the multi-behavior probability distribution graph constraints, including: The power nodes in the group node set receive the multi-behavior probability distribution map, and based on the probability of different behaviors occurring, execute the behaviors in the multi-behavior probability distribution map to obtain the power system simulation scenario.

Images