Power system typical scene generation method

Abstract

The invention discloses a typical scene generation method for an electric power system. The method comprises the following steps: step 1, collecting power grid model data and multi-dimensional measurement data of the electric power system; 2, preprocessing the power grid model data and the multi-dimensional measurement data, and integrating the preprocessed data into a unified data matrix to form a modeling data set; 3, selecting input features for PCA dimension reduction in the modeling data set, calculating influence factors of the input features through a Pearson's correlation coefficient, removing redundant features, fusing the screened input feature set with load flow calculation constraint and voltage stability criterion features, and then performing PCA dimension reduction to form principal component data; and 4, inputting the principal component data into a clustering model for clustering, and outputting a typical scene of the power system. According to the method, deep combination of a power system physical mechanism and a data driving technology is realized, so that an identification result not only conforms to a data statistics rule, but also strictly follows a power grid operation essential rule.

Description

Technical Field

[0001] This invention relates to the field of power system operation monitoring and dispatch optimization technology, and in particular to a method for generating typical power system scenarios. Background Technology

[0002] As new power systems evolve towards higher penetration of renewable energy, random load fluctuations, and coordinated operation of power generation, grid, load, and storage, the complexity and uncertainty of power grid operation have significantly increased. Traditional scenario segmentation methods based on human experience are no longer able to accurately capture dynamic changes in the power grid. There is an urgent need to combine multi-dimensional data acquisition with machine learning modeling to achieve accurate identification of typical scenarios, providing support for core operations such as power grid dispatch optimization and reactive power and voltage control. However, existing technologies for identifying typical power system scenarios have significant shortcomings, resulting in scenario identification results lacking engineering practicality and failing to meet the requirements for safe and stable operation of new power systems. Specific shortcomings are as follows: Existing typical scenario identification methods mostly adopt a pure data-driven architecture, relying solely on multi-dimensional measurement data such as voltage amplitude, load power, and renewable energy output for feature extraction. Since the measurement data only reflects the superficial characteristics of power grid operation and lacks a characterization of the essential laws governing power grid operation, the feature system lacks physical correlation, and the analysis results are prone to invalid scenarios with similar data but contradictory physical structures.

[0003] Existing feature selection processes focus solely on data redundancy removal without prioritizing features based on the core requirements for safe power system operation. This results in the loss of critical physical information after dimensionality reduction: on the one hand, features such as voltage exceeding limits and sudden drops in renewable energy output in measurement data lack correlation with voltage stability criteria and power flow constraints, failing to effectively characterize safety-related scenarios; on the other hand, purely statistical screening tends to retain features that are highly relevant at the data level but lack engineering significance, crowding out the weight of core physical features. This makes it impossible to distinguish between voltage safety scenarios and voltage instability risk scenarios, hindering the support of critical operations such as reactive power voltage control and fault early warning. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing power system typical scenario identification technologies, which are mostly purely data-driven and do not integrate power system power flow calculation constraints and voltage stability criteria features, resulting in a lack of physical rationality and engineering practicality in scenario identification. This invention provides a method for generating typical power system scenarios. By collecting power grid models and multi-dimensional measurement data, preprocessing and integrating them, filtering features, and integrating power flow calculation constraints and voltage stability criteria features, and then performing PCA dimensionality reduction and clustering modeling, the invention achieves the effect of accurately outputting typical power system scenarios.

[0005] The objective of this invention is achieved through the following technical solution: The method for generating typical power system scenarios includes the following steps: Step 1: Collect power grid model data and multi-dimensional measurement data of the power system; Step 2: Preprocess the power grid model data and multi-dimensional measurement data, and convert the preprocessed dataset into a unified data matrix to form the modeling dataset; Step 3: Select input features for PCA dimensionality reduction from the modeling dataset, calculate the influence factor of the input features using the Pearson correlation coefficient and remove redundant features, fuse the filtered input feature set with power flow calculation constraints and voltage stability criterion features, and then perform PCA dimensionality reduction to form principal component data; Step 4: Input the principal component data into the clustering model for clustering and output typical power system scenarios.

[0006] Preferably, in step 1, the power grid model data includes the power grid topology, line parameters, transformer parameters, reactive power compensation equipment parameters, and node connection relationships. The multi-dimensional measurement data includes voltage measurement data, reactive power measurement data, load measurement data, generator operation measurement data, new energy power generation measurement data, and auxiliary measurement data. The multi-dimensional measurement data is updated at a set time frequency.

[0007] Preferably, step 2 involves preprocessing the power grid model data and multi-dimensional measurement data, specifically including: For power grid model data, outliers are removed by the 3σ criterion, and missing data are filled by nearest-time interpolation or mean filling method of similar scenarios to ensure data integrity. For multi-dimensional measurement data, outliers are removed by using the 3σ criterion and sliding window anomaly detection; for missing data, linear interpolation or power flow estimation based on the power grid model is used to ensure data integrity. Z-score standardization is used to eliminate the dimensional effects of power grid model data and multi-dimensional measurement data; The process of transforming the preprocessed dataset into a unified data matrix to form a modeling dataset specifically includes: Using timestamps and node numbers as keywords, power grid model data and multi-dimensional measurement data are integrated into a unified data matrix. The row dimension is the combination of time and node, and the column dimension is various indicators, forming a modeling dataset.

[0008] As a preferred embodiment, the method for generating typical power system scenarios further expands the modeling dataset by generating small samples and incorporating them into the modeling dataset, specifically as follows: Generative adversarial network (GAN) models are used to generate targeted samples of abnormal business scenarios. After the business anomaly scenario samples pass the consistency check, they are merged with the original modeling dataset to form a complete modeling dataset; In the modeling dataset, abnormal scene samples are resampled to increase the proportion of abnormal scene samples.

[0009] Preferably, the selection of input features for PCA dimensionality reduction from the modeling dataset specifically involves: By leveraging expert experience, effective features including voltage over-limit duration, number of voltage over-limit occurrences, magnitude of sudden increases / decreases in renewable energy output, and load mutation values ​​by industry are selected. These effective features are then used as input features for PCA dimensionality reduction. Simultaneously, based on the fluctuation characteristics of renewable energy in the power system, the priority of the effective features is determined, and they are used to calculate the influence factors of the input features using the Pearson correlation coefficient, while redundant features are eliminated.

[0010] Preferably, in step 3, the filtered input feature set is fused with power flow calculation constraints and voltage stability criterion features, specifically as follows: The node power balance constraint characteristics, voltage amplitude constraint characteristics, and line transmission capacity constraint characteristics are acquired and integrated to form a power flow constraint feature set that includes the active / reactive power imbalance degree, voltage over-limit coefficient, and power flow margin of all nodes. The local voltage stability L-index and voltage margin features are acquired and integrated to form a voltage stability criterion feature set. The input feature set, power flow constraint feature set, and voltage stability criterion feature set are standardized. Then, the three types of standardized features are merged in a parallel fusion manner according to a unified dimension to form a comprehensive fusion feature matrix that includes measurement data, power flow constraints, and voltage stability criteria.

[0011] Preferably, the clustering model is a K-Means clustering or hierarchical clustering model, and the number of clusters is determined by the elbow rule, the profile coefficient method, and voltage scenario discrimination analysis.

[0012] Preferably, in step 4, after the principal component data is input into the clustering model for clustering, the physical consistency of the clustering results is also checked. The voltage amplitude in this scenario is verified by power flow calculation to see if it meets the set standard. If it does not meet the standard, the cluster is deleted. Each remaining cluster corresponds to a different typical power system scenario.

[0013] As a preferred approach, the method for generating typical power system scenarios also names and classifies typical scenarios, combining seasonality, time period, weather, and load characteristics to form standardized scenario names.

[0014] A typical power system scenario generation system, applicable to methods for generating typical power system scenarios, including: The data acquisition module is used to collect power grid model data and multi-dimensional measurement data of the power system; The data preprocessing and integration module is used to remove outliers, fill in missing values ​​and standardize the above two types of data, and integrate them to form a modeling dataset. It can also expand the dataset by generating business anomaly scenario samples through generative adversarial networks. The feature fusion and dimensionality reduction module is used to select input features and remove redundant features, fuse power flow calculation constraints and voltage stability criterion features to form a comprehensive fusion feature matrix, and then output principal component data through PCA dimensionality reduction. The clustering and scene output module is used to cluster principal component data using K-Means or hierarchical clustering models, output typical scenes after physical consistency verification, and complete standardized naming by combining seasonal, time period, weather and load characteristics.

[0015] The beneficial effects of this invention are as follows: This invention integrates power flow calculation constraints and voltage stability criterion features into the core of modeling through a complete process of data acquisition, preprocessing integration, feature screening and fusion, PCA dimensionality reduction, and clustering modeling. This achieves a deep integration of the physical mechanism of the power system and data-driven technology, fundamentally solving the core problems of existing pure data-driven methods that lack physical constraints and are prone to scene physical distortion. This ensures that the identification results not only conform to the statistical laws of data, but also strictly follow the essential rules of power grid operation.

[0016] Relying on a refined data preprocessing mechanism, expansion of business anomaly scenario samples, and optimization of scientific clustering parameters, the model not only covers normal operation scenarios but also effectively captures low-probability business anomaly scenarios, avoiding scenario omissions and misjudgments caused by sample imbalance and data quality defects. At the same time, through feature redundancy elimination and physical constraint enhancement, the model's adaptability to complex operating conditions such as new energy fluctuations and load mutations has been improved.

[0017] Typical scenarios, after physical consistency verification, combined with standardized naming based on season, time period, weather, and load characteristics, possess clear business identification and safety compliance. They can directly provide precise decision support for core businesses such as power grid dispatch optimization, reactive power and voltage control, and reactive power compensation configuration, achieving efficient transformation from data classification to engineering applications and contributing to the refined operation and management of new power systems. Attached Figure Description

[0018] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0019] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make this application more comprehensive and complete, and to fully convey the concept of the exemplary embodiments to those skilled in the art.

[0020] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a thorough understanding of embodiments of this application. However, those skilled in the art will recognize that the technical solutions of this application can be practiced without one or more of the specific details, or other methods, components, apparatuses, steps, etc., can be employed. In other instances, well-known methods, apparatuses, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of this application.

[0021] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0022] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0023] Example: Methods for generating typical power system scenarios, such as Figure 1 As shown, it includes the following steps: Step 1: Collect power grid model data and multi-dimensional measurement data of the power system; Step 2: Preprocess the power grid model data and multi-dimensional measurement data, and convert the preprocessed dataset into a unified data matrix to form the modeling dataset; Step 3: Select input features for PCA dimensionality reduction from the modeling dataset, calculate the influence factor of the input features using the Pearson correlation coefficient and remove redundant features, fuse the filtered input feature set with power flow calculation constraints and voltage stability criterion features, and then perform PCA dimensionality reduction to form principal component data; Step 4: Input the principal component data into the clustering model for clustering and output typical power system scenarios.

[0024] In step 1, the power grid model data includes the power grid topology, line parameters, transformer parameters, reactive power compensation equipment parameters, and node connection relationships. The multi-dimensional measurement data includes voltage measurement data, reactive power measurement data, load measurement data, generator operation measurement data, new energy power generation measurement data, and auxiliary measurement data. The multi-dimensional measurement data is updated at a set time frequency.

[0025] Step 2 involves preprocessing the power grid model data and multi-dimensional measurement data, specifically including: For power grid model data, outliers are removed by the 3σ criterion, and missing data are filled by nearest-time interpolation or mean filling method of similar scenarios to ensure data integrity. For multi-dimensional measurement data, outliers are removed by using the 3σ criterion and sliding window anomaly detection; for missing data, linear interpolation or power flow estimation based on the power grid model is used to ensure data integrity. Z-score standardization is used to eliminate the dimensional effects of power grid model data and multi-dimensional measurement data; In this embodiment, the power grid model data is mainly static data (updated only when the power grid topology and equipment parameters change). The preprocessing focuses on ensuring the physical compliance and integrity of the data. The specific steps are as follows: For core static data such as power grid topology parameters, line impedance, transformer capacity, and rated capacity of reactive power compensation equipment, historical data from the power grid model over the past three years are first extracted, and the mean μ and standard deviation σ of each indicator are calculated. Data exceeding the range of "μ±3σ" are identified as outliers. For example, the historical mean resistance parameter of a 110kV line is 0.15Ω / km, and the standard deviation is 0.02Ω / km. If a value of 0.5Ω / km is collected in a certain instance, it exceeds the reasonable range of "0.15±3×0.02" (0.09Ω / km~0.21Ω / km) and is identified as an outlier and removed. Outliers mainly originate from data entry errors and deviations in equipment parameter labeling. After removal, corrections need to be made by checking the power grid design drawings or equipment ledgers.

[0026] If the missing data is short-term data after the topology change, the nearest-time interpolation method is used: based on the most recent complete power grid model data before the topology change, combined with the design parameters of the new equipment, the missing parameters are interpolated to ensure the consistency between the topology structure and the parameters. If the missing data is historical model data that has not been updated for a long time, the average value filling method of similar scenarios is adopted: screen similar lines / equipment with the same grid voltage level, conductor type, and erection method, calculate their average parameter value, and use it as the filling value for the missing parameter. After filling, the rationality of the parameter needs to be verified by power flow calculation. The multi-dimensional measurement data consists of high-resolution dynamic data at the 5-minute level. Preprocessing needs to balance the removal of outliers and noise with the physical correlation of the data. The specific steps are as follows: For each type of measurement indicator, the mean μ and standard deviation σ are calculated based on the normal operation data of the past month, and data that exceed the range of μ±3σ are marked as suspected abnormalities; Set the sliding window size to 5 time points. Validate suspected abnormal data within the window. If only a single time point within the window is suspected of being abnormal, while the data at the other 4 time points is normal, it is determined to be isolated noise and is removed. If two or more consecutive time points within the window are suspected of being abnormal, and this is corroborated by auxiliary data, it is determined to be a real business anomaly, and the data is retained for subsequent anomaly scenario modeling.

[0027] If the missing data is short-term, linear interpolation is used. The effective measurement data adjacent to the missing time period are used as endpoints, and the missing value is calculated according to the linear relationship of time. For example, if the voltage at a certain node is 110.5kV at 10:00 and 110.3kV at 10:10, and the data at 10:05 is missing, then the voltage at 10:05 is calculated to be 110.4kV through interpolation. This method is suitable for measurement data with stable changes in load and voltage. If the missing data is to be filled in for a long period of time, the power flow estimation filling method based on the power grid model is adopted: based on the power grid admittance matrix, the measurement data of the known nodes are input, and the power flow equation is solved by the Newton-Raphson method to estimate the voltage and power measurement values ​​of the missing nodes. For example, if the reactive power data of a certain load node is missing, the reactive load value of the node can be estimated based on the voltage and line impedance parameters of its adjacent nodes, combined with the power balance constraints, to ensure that the filling data conforms to the physical laws of power grid operation.

[0028] The process of transforming the preprocessed dataset into a unified data matrix to form a modeling dataset specifically includes: Using timestamps and node numbers as keywords, power grid model data and multi-dimensional measurement data are integrated into a unified data matrix. The row dimension is the combination of time and node, and the column dimension is various indicators, forming a modeling dataset.

[0029] The method for generating typical power system scenarios further expands the modeling dataset by generating small samples and incorporating them into the modeling dataset, specifically: Generative adversarial network (GAN) models are used to generate targeted samples of abnormal business scenarios. After the business anomaly scenario samples pass the consistency check, they are merged with the original modeling dataset to form a complete modeling dataset; In the modeling dataset, abnormal scene samples are resampled to increase the proportion of abnormal scene samples.

[0030] The selection of input features for PCA dimensionality reduction from the modeling dataset specifically refers to: By leveraging expert experience, effective features including voltage over-limit duration, number of voltage over-limit occurrences, magnitude of sudden increases / decreases in renewable energy output, and load mutation values ​​by industry are selected. These effective features are then used as input features for PCA dimensionality reduction. Simultaneously, based on the fluctuation characteristics of renewable energy in the power system, the priority of the effective features is determined, and they are used to calculate the influence factors of the input features using the Pearson correlation coefficient, while redundant features are eliminated.

[0031] In step 3, the filtered input feature set is fused with power flow calculation constraints and voltage stability criterion features, specifically as follows: The node power balance constraint characteristics, voltage amplitude constraint characteristics, and line transmission capacity constraint characteristics are acquired and integrated to form a power flow constraint feature set that includes the active / reactive power imbalance degree, voltage over-limit coefficient, and power flow margin of all nodes. The local voltage stability L-index and voltage margin features are acquired and integrated to form a voltage stability criterion feature set. The input feature set, power flow constraint feature set, and voltage stability criterion feature set are standardized. Then, the three types of standardized features are merged in a parallel fusion manner according to a unified dimension to form a comprehensive fusion feature matrix that includes measurement data, power flow constraints, and voltage stability criteria.

[0032] The clustering model is K-Means clustering or hierarchical clustering model, and the number of clusters is determined by elbow rule, profile coefficient method and voltage scene discrimination analysis.

[0033] In this embodiment, the specific implementation steps of PCA dimensionality reduction are as follows: Step 3.1 Input Feature Selection and Priority Determination: Select core input features from the modeling dataset, specifically including: Voltage-related characteristics: duration of voltage over-limit at each node (unit: min / 5-minute window), number of voltage over-limit occurrences (unit: times / 5-minute window), and voltage deviation rate (unit: %). New energy-related characteristics: wind power output surge / dip magnitude (unit: MW / 5 minutes, the sudden change threshold is defined as 20% of the rated output), photovoltaic output prediction error (unit: %). Load-related characteristics: Industrial / Residential / Commercial sector load abrupt change values ​​(unit: MW / 5 minutes), Regional total load change rate (unit: %); Reactive power related characteristics: output adjustment of reactive power compensation device (unit: Mvar), reactive power load deficit at nodes (unit: Mvar).

[0034] Based on the system characteristics of high penetration of new energy sources, priority weights are set (levels 1-5, with level 5 being the highest): Level 5 (Core Safety Features): Duration of voltage over-limit, Number of voltage over-limit occurrences, and magnitude of sudden drop in new energy output; Level 4 (Key Operating Characteristics): Voltage Deviation Rate, Industry-Specific Load Sudden Change Values, Reactive Power Deficit; Level 3 (Auxiliary Correlation Characteristics): Photovoltaic output prediction error, total load change rate, reactive power compensation adjustment.

[0035] Priority will serve as the basis for decision-making when removing redundant features. When the correlation between two types of features exceeds the standard, the feature with higher priority will be retained first.

[0036] Step 3.2 Redundant feature removal based on Pearson correlation coefficient: Using the combination of "time and node" as the sample dimension, the Pearson correlation coefficient r is calculated pairwise for each of the nine selected input features to construct a 9×9 correlation matrix; Set a redundancy threshold |r|≥0.8 and process feature pairs with excessive correlation. Step 3.3 Fusion of input feature set and physical constraint / stability criterion features: The active power imbalance, reactive power imbalance, and voltage limit exceedance coefficients of 432 nodes, as well as the power flow margin of 896 lines, have been pre-calculated, totaling 1760 feature items. Z-score standardization (mean → 0, variance → 1) is performed on the 6 types of input feature sets, the 1760 power flow constraint feature sets, and the 774 voltage stability criterion feature sets to eliminate dimensional differences. Using "time and node" as the row dimension, the three types of standardized features are concatenated in parallel columns to form a comprehensive fusion feature matrix with a dimension of 120,000 × 2540 (6 + 1760 + 774), ensuring that each row of samples simultaneously contains measurement statistical features, physical constraint features, and stability criterion features.

[0037] Step 3.4 PCA Dimensionality Reduction Core Calculation and Principal Component Selection: For the 2540-dimensional integrated feature matrix, its covariance matrix (dimension 2540×2540) is calculated. This matrix reflects the degree of linear correlation between various features, highlighting the intrinsic relationship between physical features and measurement features. The eigenvalues ​​λ and corresponding eigenvectors of the covariance matrix are solved using the Jacobi iteration method, resulting in 2540 eigenvalues ​​(sorted from largest to smallest: λ1 ≥ λ2 ≥ ... ≥ λ). 2540 ) and the corresponding unit eigenvectors; Calculate the variance contribution rate and cumulative variance contribution rate of each eigenvalue, and set a threshold of ≥85%; The integrated feature matrix and the 23-dimensional projection matrix are multiplied to obtain a 120,000×23 principal component data matrix, with each column being a principal component and each row corresponding to the dimensionality-reduced data of the original "time, node" samples. We analyzed the physical meaning of the first five core principal components.

[0038] Through the PCA dimensionality reduction steps described above, the high-dimensional fusion features of 2540 dimensions are compressed to 23 dimensions, with a dimensionality compression ratio of 99.1%. This not only significantly reduces the computational complexity of subsequent clustering modeling, but also, by incorporating physical constraints and stability criteria features, enables the principal component data to possess both statistical correlation and physical compliance. This lays the foundation for clustering practical engineering scenarios such as voltage stability critical scenarios and new energy fluctuation scenarios.

[0039] The clustering process in this embodiment is as follows: First, the clustering model and parameters were determined: considering the large sample size and low dimensionality of the principal component data, the K-Means clustering model was selected first, while hierarchical clustering was used as a comparative verification model. The number of clusters was determined through triple verification. First, the elbow curve from K=8 to 12 was calculated. It was found that when K=10, the slope of the curve changed abruptly (elbow point), and the corresponding silhouette coefficient reached 0.72, which is higher than the reasonable threshold of 0.6. Then, combined with the voltage scenario discrimination analysis, when K=10, the voltage index difference of core scenarios such as "voltage stability critical scenario" and "new energy large-scale scenario" was ≥35%. Finally, the number of clusters was determined to be 10.

[0040] When performing K-Means clustering, the number of iterations was set to 100 and the random seed was fixed at 42. The principal component data matrix was used as input, and the cluster division was completed by calculating the Euclidean distance between the samples and the cluster centers. At the same time, a hierarchical clustering model was run for cross-validation. The clustering overlap between the two models reached 91%, which verified the stability of the clustering results.

[0041] After clustering, physical consistency verification was carried out: core feature parameters of the scenario were extracted from each of the 10 clusters and substituted into the power grid power flow calculation model to verify whether the voltage amplitude of all nodes in each scenario met the GB50057-2010 standard (110kV node voltage allowable range 10kV~117kV, 220kV node 198kV~235kV, 500kV node 485kV~550kV). One cluster was judged as a physically invalid scenario and removed because the voltage amplitude of 3 nodes exceeded the safe range. Finally, 9 valid clusters were obtained. Each cluster corresponds to a typical power system scenario with distinct characteristics, and the voltage, power flow and other indicators of the samples in the cluster all conform to the physical laws of power grid operation.

[0042] The method for generating typical power system scenarios also names and classifies typical scenarios, combining seasonality, time period, weather and load characteristics to form standardized scenario names, such as the summer midday sunny photovoltaic power generation scenario.

[0043] A typical power system scenario generation system, applicable to methods for generating typical power system scenarios, including: The data acquisition module is used to collect power grid model data and multi-dimensional measurement data of the power system; The data preprocessing and integration module is used to remove outliers, fill in missing values ​​and standardize the above two types of data, and integrate them to form a modeling dataset. It can also expand the dataset by generating business anomaly scenario samples through generative adversarial networks. The feature fusion and dimensionality reduction module is used to select input features and remove redundant features, fuse power flow calculation constraints and voltage stability criterion features to form a comprehensive fusion feature matrix, and then output principal component data through PCA dimensionality reduction. The clustering and scene output module is used to cluster principal component data using K-Means or hierarchical clustering models, output typical scenes after physical consistency verification, and complete standardized naming by combining seasonal, time period, weather and load characteristics.

[0044] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the embodiments disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein.

[0045] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A method for generating typical power system scenarios, characterized by: Includes the following steps: Step 1: Collect power grid model data and multi-dimensional measurement data of the power system; Step 2: Preprocess the power grid model data and multi-dimensional measurement data, and convert the preprocessed dataset into a unified data matrix to form the modeling dataset; Step 3: Select input features for PCA dimensionality reduction from the modeling dataset, calculate the influence factor of the input features using the Pearson correlation coefficient and remove redundant features, fuse the filtered input feature set with power flow calculation constraints and voltage stability criterion features, and then perform PCA dimensionality reduction to form principal component data; Step 4: Input the principal component data into the clustering model for clustering and output typical power system scenarios.

2. The method for generating typical power system scenarios according to claim 1, characterized in that, In step 1, the power grid model data includes the power grid topology, line parameters, transformer parameters, reactive power compensation equipment parameters, and node connection relationships. The multi-dimensional measurement data includes voltage measurement data, reactive power measurement data, load measurement data, generator operation measurement data, new energy power generation measurement data, and auxiliary measurement data. The multi-dimensional measurement data is updated at a set time frequency.

3. The method for generating typical power system scenarios according to claim 1, characterized in that, Step 2 involves preprocessing the power grid model data and multi-dimensional measurement data, specifically including: For power grid model data, outliers are removed by the 3σ criterion, and missing data are filled by nearest-time interpolation or mean filling method of similar scenarios to ensure data integrity. For multi-dimensional measurement data, outliers are removed by using the 3σ criterion and sliding window anomaly detection; for missing data, linear interpolation or power flow estimation based on the power grid model is used to ensure data integrity. Z-score standardization is used to eliminate the dimensional effects of power grid model data and multi-dimensional measurement data; The process of transforming the preprocessed dataset into a unified data matrix to form a modeling dataset specifically includes: Using timestamps and node numbers as keywords, power grid model data and multi-dimensional measurement data are integrated into a unified data matrix. The row dimension is the combination of time and node, and the column dimension is various indicators, forming a modeling dataset.

4. The method for generating typical power system scenarios according to claim 1, characterized in that, Furthermore, the modeling dataset is expanded by generating smaller samples and incorporating them into the modeling dataset, specifically as follows: Generative adversarial network (GAN) models are used to generate targeted samples of abnormal business scenarios. After the business anomaly scenario samples pass the consistency check, they are merged with the original modeling dataset to form a complete modeling dataset; In the modeling dataset, abnormal scene samples are resampled to increase the proportion of abnormal scene samples.

5. The method for generating typical power system scenarios according to claim 1, characterized in that, The selection of input features for PCA dimensionality reduction from the modeling dataset specifically refers to: By leveraging expert experience, effective features including voltage over-limit duration, number of voltage over-limit occurrences, magnitude of sudden increases / decreases in renewable energy output, and load mutation values ​​by industry are selected. These effective features are then used as input features for PCA dimensionality reduction. Simultaneously, based on the fluctuation characteristics of renewable energy in the power system, the priority of the effective features is determined, and they are used to calculate the influence factors of the input features using the Pearson correlation coefficient, while redundant features are eliminated.

6. The method for generating typical power system scenarios according to claim 1, characterized in that, In step 3, the filtered input feature set is fused with power flow calculation constraints and voltage stability criterion features, specifically as follows: The node power balance constraint characteristics, voltage amplitude constraint characteristics, and line transmission capacity constraint characteristics are acquired and integrated to form a power flow constraint feature set that includes the active / reactive power imbalance degree of all nodes, voltage over-limit coefficient, and power flow margin of all lines. The local voltage stability L-index and voltage margin features are acquired and integrated to form a voltage stability criterion feature set. The input feature set, power flow constraint feature set, and voltage stability criterion feature set are standardized. Then, the three types of standardized features are merged in parallel according to a unified dimension to form a comprehensive fusion feature matrix that includes measurement data, power flow constraints, and voltage stability criteria.

7. The method for generating typical power system scenarios according to claim 1, characterized in that, The clustering model is K-Means clustering or hierarchical clustering model, and the number of clusters is determined by elbow rule, profile coefficient method and voltage scene discrimination analysis.

8. The method for generating typical power system scenarios according to claim 1, characterized in that, In step 4, after the principal component data is input into the clustering model for clustering, the physical consistency of the clustering results is also checked. The voltage amplitude in this scenario is verified by power flow calculation to see if it meets the set standard. If it does not meet the standard, the cluster is deleted. Each remaining cluster corresponds to a different typical power system scenario.

9. The method for generating typical power system scenarios according to claim 1, characterized in that, Typical scenarios are also named and categorized, and standardized scenario naming is formed by combining season, time period, weather and load characteristics.

10. A power system typical scenario generation system, applicable to the power system typical scenario generation method as described in any one of claims 1-9, characterized in that, include: The data acquisition module is used to collect power grid model data and multi-dimensional measurement data of the power system; The data preprocessing and integration module is used to remove outliers, fill in missing values ​​and standardize the above two types of data, and integrate them to form a modeling dataset. It can also expand the dataset by generating business anomaly scenario samples through generative adversarial networks. The feature fusion and dimensionality reduction module is used to select input features and remove redundant features, fuse power flow calculation constraints and voltage stability criterion features to form a comprehensive fusion feature matrix, and then output principal component data through PCA dimensionality reduction; The clustering and scene output module is used to cluster principal component data using K-Means or hierarchical clustering models, output typical scenes after physical consistency verification, and complete standardized naming by combining seasonal, time period, weather and load characteristics.

Images