A power system extreme scenario identification method and system

Abstract

The application belongs to the technical field of extreme scenario recognition, and particularly relates to a power system extreme scenario recognition method and system. The method comprises the following steps: obtaining net load data of a power system based on time series data of wind power, photovoltaic and load of the power system, and forming daily net load samples by daily slicing; adopting a dynamic time warping (DTW) algorithm to construct a difference matrix between the daily net load samples; performing principal component analysis on the difference matrix after standardization processing, obtaining a low-dimensional feature vector, and weighting the low-dimensional feature vector according to the inverse square root of the corresponding eigenvalue of the principal component in the principal component dimension to obtain a weighted low-dimensional feature vector; calculating the Euclidean distance from the weighted low-dimensional feature vector corresponding to each day to the space mean point in the feature space formed by the weighted low-dimensional feature vector, and constructing a threshold value by using a standard deviation method according to the mean and standard deviation of the Euclidean distance, and identifying the extreme scenario of the power system based on the threshold value.

Description

Technical Field

[0001] This invention belongs to the field of extreme scene recognition technology, specifically relating to a method and system for identifying extreme scenes in power systems. Background Technology

[0002] With the increasing penetration of renewable energy sources, primarily wind and solar power, into the power system, the uncertainty and volatility of power system operation have significantly increased. In particular, the frequent occurrence of extreme weather events has made balancing power supply and renewable energy consumption more difficult, leading to more complex supply-demand imbalances. Against this backdrop, accurately identifying and responding to extreme scenarios has become a core task for power system optimization dispatch and flexibility resource planning. Therefore, establishing scientific and efficient methods for acquiring extreme scenarios plays a crucial role in improving the dispatch flexibility of the power system, optimizing resource allocation, and ensuring the stable operation of the system.

[0003] Currently, methods for obtaining extreme scenarios mainly fall into the following categories: 1) Extreme Scene Generation Method These methods reconstruct extreme scenario sequences in known systems by using meteorological data and its mapping relationship with equipment operation, analyzing their distribution characteristics, and building models based on this to generate more extreme scenarios. Specifically, these include model-based methods such as the Copula function method, Markov chain method (MC), and extreme value theory method (EVT), as well as data-driven methods such as kernel density estimation (KDE), generative adversarial networks (GAN), and diffusion models (DM).

[0004] 2) Extreme Scene Extraction Method These methods extract extreme scenarios from historical data or actual operation to evaluate the performance of existing systems under these conditions. Common techniques include thresholding based on human expert experience, which uses empirical thresholds to extract scenarios of sustained low output for wind and solar power under extreme weather conditions, or to screen scenarios where renewable energy output is higher than the benchmark and load is lower than the benchmark.

[0005] Although these methods have been applied to the identification of extreme scenarios in power systems, they still have the following significant drawbacks and shortcomings: 1) Insufficient representativeness of extreme scenarios: Current methods for identifying extreme scenarios generally rely on thresholds set based on expert experience or statistical analysis of historical data, leading to insufficient representativeness in extreme scenario identification. For example, thresholding methods typically rely on pre-set empirical thresholds to extract specific scenarios, a process often highly subjective, and can only identify scenarios that meet the set criteria, failing to comprehensively capture extreme scenarios that may occur under different meteorological conditions. Furthermore, methods for reconstructing extreme scenarios based on historical or meteorological data also suffer from subjectivity. Human intervention in data acquisition and analysis results in a lack of comprehensiveness and representativeness in the generated extreme scenarios. Limited by the scarcity of extreme samples, model assumption biases, and extrapolation uncertainties, the generated scenarios may have insufficient coverage or be inconsistent with actual operational characteristics. These limitations make it difficult for existing methods to fully cover the various extreme scenarios that high-penetration renewable energy systems may encounter, restricting their application effectiveness and applicability in practical systems.

[0006] 2) The temporal coupling relationship between the source and load was not considered. While existing methods can effectively identify or generate extreme scenarios, they suffer from a significant shortcoming: they fail to adequately consider the temporal coupling relationship among wind power, solar power, and load. This relationship refers to the synchronous changes and dynamic correlations exhibited by wind power, solar power, and load on the same time scale, reflecting the inherent temporal sequence of renewable energy output and load demand. However, when generating or extracting extreme scenarios, existing methods often sample the time series of wind power, solar power, and load independently or splice them across time periods, disrupting this inherent temporal correlation and resulting in "combined" extreme scenarios lacking physical plausibility. Although such scenarios possess statistical extremism, they fail to accurately reflect the temporal matching characteristics of wind, solar, and load under real-world operating conditions, making it difficult for the generated scenarios to accurately characterize the dynamic mismatch between load demand and renewable energy output under extreme scenarios. Summary of the Invention

[0007] To address the issues of insufficient representativeness and lack of temporal coupling in existing extreme scenario identification methods, especially in high-penetration renewable energy systems dominated by wind and solar power, this invention provides a power system extreme scenario identification method and system, aiming to achieve accurate identification of extreme scenarios through dynamic time warping (DTW), principal component analysis (PCA), and standard deviation method.

[0008] To achieve the above objectives, the present invention provides the following solution: A method for identifying extreme scenarios in a power system, the method comprising: Based on the time-series data of wind power, photovoltaic power and load in the power system, the net load data of the power system is obtained, and daily net load samples are formed by daily slices. The Dynamic Time Warping (DTW) algorithm is used to construct the pairwise difference matrix between daily net load samples; After standardizing the difference matrix, principal component analysis (PCA) is performed to obtain low-dimensional eigenvectors. The low-dimensional eigenvectors are then weighted according to the reciprocal square roots of the corresponding eigenvalues ​​of the principal components to obtain weighted low-dimensional eigenvectors. In the feature space formed by the weighted low-dimensional feature vectors, calculate the Euclidean distance from the weighted low-dimensional feature vectors corresponding to each day to the spatial mean point. Construct a threshold based on the mean and standard deviation of the Euclidean distance. Extract corresponding samples from the daily net load samples based on the threshold as extreme scenarios of the power system.

[0009] Preferably, the method for constructing the pairwise difference matrix between daily net load samples using the Dynamic Time Warping (DTW) algorithm includes: The DTW algorithm uses dynamic programming to find the minimum cumulative cost for any two-day net load sample. Based on the minimum cumulative cost, construct a difference matrix between each pair of daily net load samples.

[0010] Preferably, after standardizing the difference matrix, principal component analysis (PCA) is performed to obtain low-dimensional eigenvectors. The low-dimensional eigenvectors are then weighted according to the reciprocal square roots of the corresponding eigenvalues ​​of the principal components to obtain weighted low-dimensional eigenvectors. The method includes: After standardizing the difference matrix, principal component analysis is performed to obtain the principal component score vector for each day. The principal component score vector is then used as a low-dimensional feature vector obtained by reducing the dimensionality of the distance representation vector for each day, which is composed of the DTW distances between that day and the other days. Weights are constructed based on the reciprocal square roots of the eigenvalues ​​corresponding to the principal components, and the low-dimensional eigenvectors are then weighted dimension by dimension to obtain weighted low-dimensional eigenvectors.

[0011] Preferably, the method for calculating the Euclidean distance from the weighted low-dimensional feature vectors corresponding to each day to the spatial mean point in the feature space composed of the weighted low-dimensional feature vectors, constructing a threshold based on the mean and standard deviation of the Euclidean distances, and extracting corresponding samples from the daily net load samples as extreme scenarios of the power system based on the thresholds includes: First, calculate the Euclidean distance from the weighted low-dimensional feature vectors corresponding to each day to the spatial mean point: ; in, For the first The nth sample point in the weighted PCA low-dimensional space 3D coordinates; For all sample points at the th The mean of the dimensional coordinates; It is the dimension of a low-dimensional space; Then, the recognition threshold for extreme scenarios is set using the standard deviation method: ; In the formula, Euclidean distance The mean; Euclidean distance Standard deviation; This is the standard deviation factor, used to adjust the threshold level. Increasing the threshold will raise the number of samples identified as extreme scenarios. Decreasing the threshold will lower the threshold and increase the number of samples identified as extreme scenarios.

[0012] Extreme Scene Collection Defined as the index of all samples that exceed the recognition threshold: .

[0013] The present invention also provides a power system extreme scenario identification system, the system being used to implement the aforementioned method, the system comprising: a data acquisition module, a difference processing module, a weighted dimensionality reduction module, and an identification module; The data acquisition module is used to obtain the net load data of the power system based on the time-series data of wind power, photovoltaic power and load of the power system, and to form a daily net load sample by daily slice; The difference processing module is used to construct a difference matrix between each pair of daily net load samples using the Dynamic Time Warping (DTW) algorithm. The weighted dimensionality reduction module is used to perform principal component analysis (PCA) on the difference matrix after standardization to obtain a low-dimensional feature vector, and to weight the low-dimensional feature vector according to the reciprocal square root of the corresponding eigenvalue of the principal component to obtain a weighted low-dimensional feature vector. The identification module is used to calculate the Euclidean distance from the weighted low-dimensional feature vector corresponding to each day to the spatial mean point in the feature space composed of the weighted low-dimensional feature vectors, construct a threshold based on the mean and standard deviation of the Euclidean distance, and extract corresponding samples from the daily net load samples as extreme scenarios of the power system based on the threshold.

[0014] Preferably, the difference processing module employs the Dynamic Time Warping (DTW) algorithm to construct a pairwise difference matrix between daily net load samples, comprising the following methods: The DTW algorithm uses dynamic programming to find the minimum cumulative cost for any two-day net load sample. Based on the minimum cumulative cost, construct a difference matrix between each pair of daily net load samples.

[0015] Preferably, after standardizing the difference matrix, principal component analysis (PCA) is performed to obtain low-dimensional eigenvectors. The low-dimensional eigenvectors are then weighted according to the reciprocal square roots of the corresponding eigenvalues ​​of the principal components to obtain weighted low-dimensional eigenvectors. The method includes: After standardizing the difference matrix, principal component analysis is performed to obtain the principal component score vector for each day. The principal component score vector is then used as a low-dimensional feature vector obtained by reducing the dimensionality of the distance representation vector for each day, which is composed of the DTW distances between that day and the other days. Weights are constructed based on the reciprocal square roots of the eigenvalues ​​corresponding to the principal components, and the low-dimensional eigenvectors are then weighted dimension by dimension to obtain weighted low-dimensional eigenvectors.

[0016] Preferably, the method for calculating the Euclidean distance from the weighted low-dimensional feature vectors corresponding to each day to the spatial mean point in the feature space composed of the weighted low-dimensional feature vectors, constructing a threshold based on the mean and standard deviation of the Euclidean distances, and extracting corresponding samples from the daily net load samples as extreme scenarios of the power system based on the thresholds includes: First, calculate the Euclidean distance from the weighted low-dimensional feature vectors corresponding to each day to the spatial mean point: ; in, For the first The nth sample point in the weighted PCA low-dimensional space 3D coordinates; For all sample points at the th The mean of the dimensional coordinates; It is the dimension of a low-dimensional space; Then, the recognition threshold for extreme scenarios is set using the standard deviation method: This is the standard deviation factor, used to adjust the threshold level. Increasing the threshold will raise the number of samples identified as extreme scenarios. Decreasing the threshold will lower the threshold and increase the number of samples identified as extreme scenarios.

[0017] Extreme Scene Collection Defined as the index of all samples exceeding the threshold, as shown in the following formula: Compared with the prior art, the beneficial effects of the present invention are as follows: The extreme scene identification method proposed in this invention, based on Dynamic Time Warping (DTW), Principal Component Analysis (PCA), and Standard Deviation, has significant technical advantages in addressing the problems of insufficient representativeness and lack of temporal coupling in existing extreme scene identification methods for power systems. It can achieve the following technical effects: 1) Improve the representativeness and accuracy of extreme scene recognition. Existing extreme scenario identification methods are highly subjective, relying primarily on statistical analysis of historical data or expert-set thresholds, which can easily lead to insufficient representativeness. This invention, by combining the DTW algorithm, PCA dimensionality reduction, and standard deviation method, can automatically extract the most representative extreme scenarios from the data without requiring separate empirical thresholds (such as sustained low output thresholds, peak-valley thresholds, etc.) for wind, solar, and load, thus avoiding the problem of relying on empirical parameters and lacking unified adaptive criteria. Especially in power systems with high penetration of renewable energy, this invention can effectively cover various extreme scenarios the system may encounter, ensuring that the identified extreme scenarios are more representative.

[0018] 2) Addressing the problem of insufficient source-load temporal coupling representation In extreme scenario identification driven by time-series data such as wind power, photovoltaics, and load, existing technologies often employ independent sampling or cross-period splicing, which easily disrupts the temporal matching relationship between source-side output and load demand, resulting in "combined" scenarios lacking operational consistency. This invention uses the net load sequence as a unified representation of the source-load relationship. By introducing the DTW algorithm and temporal weighting, the temporal correlation characteristics of wind, solar, and load are more fully preserved, thus giving the difference matrix more consistent statistical characteristics. This facilitates subsequent PCA dimensionality reduction and the use of the standard deviation method for extreme scenario discrimination. This invention not only overcomes the shortcomings of existing methods but also more realistically reflects the system's operating status under extreme scenarios. Attached Figure Description

[0019] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of the construction logic framework for the scenario set in an embodiment of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0023] Example 1 Based on the time-series data of wind power, photovoltaic power and load in the power system, the net load data of the power system is obtained, and daily net load samples are formed by daily slices. The Dynamic Time Warping (DTW) algorithm is used to construct the pairwise difference matrix between daily net load samples; After standardizing the difference matrix, principal component analysis (PCA) is performed to obtain low-dimensional eigenvectors. The low-dimensional eigenvectors are then weighted according to the reciprocal square roots of the corresponding eigenvalues ​​of the principal components to obtain weighted low-dimensional eigenvectors. In the feature space formed by the weighted low-dimensional feature vectors, calculate the Euclidean distance from the weighted low-dimensional feature vectors corresponding to each day to the spatial mean point. Construct a threshold based on the mean and standard deviation of the Euclidean distance. Extract corresponding samples from the daily net load samples based on the threshold as extreme scenarios of the power system.

[0024] like Figure 1 As shown, the specific implementation process of this invention includes the following steps: 1) Dataset preprocessing In a power system, net load, as the difference between the actual load and renewable energy generation, not only reflects the load's electricity demand but also effectively demonstrates the system's regulation capacity requirements in response to fluctuations in wind and solar power. Therefore, extracting time-series data on wind power, solar power, and load from historical data and calculating the net load at each point in time, as a direct representation of the power system's regulation capacity, is fundamental data for identifying extreme scenarios. Specifically, this includes: First, calculate the net load in the power system, which is the difference between the actual load of the power system and the power generation from renewable energy sources (wind and solar). Net load data reflects the power system's demand for regulation capacity and serves as the basis for identifying extreme scenarios. The net load calculation formula is as follows: In the formula, For at any time Total load power; and Wind power and solar power at different times The power generation capacity; Indexed by hour; Indexed by number of days.

[0025] 2) DTW Dissimilarity Matrix Calculation Methods for constructing pairwise difference matrices between daily net load samples using the Dynamic Time Warping (DTW) algorithm include: The DTW algorithm uses dynamic programming to find the minimum cumulative cost for any two-day net load sample. Based on the minimum cumulative cost, construct a difference matrix between each pair of daily net load samples.

[0026] Specifically, this invention uses the DTW algorithm to measure the temporal variability between daily net loads. The algorithm captures the variability in the time series by minimizing the distance between two daily net load sequences through alignment. Considering the significant temporal correlation between time points during power system operation, this invention introduces a temporal weighting matrix. Specifically, it includes: The DTW algorithm is used to compare the distance between two time series, and the temporal variability between the net load data is calculated by minimizing the matching difference between them. To construct the local cost matrix of DTW, the local cost is defined. Used to measure two daily net load sequences at time 1 With time The point-to-point difference in a one-dimensional time series is calculated using the absolute value of the difference between the net load values ​​at two time points, as shown in the following formula: In the formula, The first daily net load sequence is in the The value at each moment; For the second daily net load sequence in the 1st The value at each moment.

[0027] Based on this, a time-series weighting matrix is ​​introduced. Differentiated weighting of contributions at different times enhances the ability to model temporal dependencies and correlations between time points, resulting in a time-series weighting matrix. This is used to suppress long-distance time point alignment and strengthen nearest-neighbor time point alignment, thereby enhancing the constraint of the DTW matching process on timing consistency.

[0028] Finally, the minimum cumulative cost between the two days' net load samples is calculated as the inter-day DTW distance between the two days, as follows: First, construct the weighted local cost matrix of the two days' daily net load samples under the condition of pairwise alignment at intraday time points. The formula is as follows: In the formula, This represents the first day of the two-day net load sample. The time point and the second day of the sample Weighted local cost when aligning at each time point.

[0029] Based on this, dynamic programming is used to calculate the cumulative cost matrix, denoted as... .in, Indicates the starting alignment position To the alignment position The minimum cumulative cost is obtained using the following recursive formula: initialization ; Recursively apply to the first column ,in For matrix row index; Recursively apply to the first row ,in Matrix column index; Recursively apply to the remaining positions This ensures that the global minimum cumulative matching cost is obtained while satisfying the time alignment constraint.

[0030] Ultimately, two days of samples and The daytime DTW distance between them is This serves as the matrix element for the corresponding two-day sample in the difference matrix. The above intraday DTW distance calculation process is repeated for the daily net load samples of any two days throughout the year to obtain the intraday DTW distance values ​​between any two days throughout the year, and an annual difference matrix is ​​constructed. .

[0031] 3) PCA weighted dimensionality reduction After standardizing the difference matrix, principal component analysis (PCA) is performed to obtain low-dimensional eigenvectors. The low-dimensional eigenvectors are then weighted according to the reciprocal square roots of the corresponding eigenvalues ​​of the principal components to obtain weighted low-dimensional eigenvectors. The method for obtaining weighted low-dimensional eigenvectors includes: After standardizing the difference matrix, principal component analysis is performed to obtain the principal component score vector for each day. The principal component score vector is then used as a low-dimensional feature vector obtained by reducing the dimensionality of the distance representation vector for each day, which is composed of the DTW distances between that day and the other days. Weights are constructed based on the reciprocal square roots of the eigenvalues ​​corresponding to the principal components, and the low-dimensional eigenvectors are then weighted dimension by dimension to obtain weighted low-dimensional eigenvectors.

[0032] Specifically, due to the high dimensionality of the dissimilarity matrix, this invention utilizes PCA to reduce its dimensionality and selects the optimal number of dimensions based on the cumulative variance contribution rate. Dimensionality reduction using PCA preserves important variation information and removes redundant data, thereby reducing computational load and improving the efficiency of subsequent analysis. Specifically, this includes: To reduce the dimensionality of the data, this invention utilizes PCA to reduce the dimensionality of the difference matrix and determines the optimal number of dimensions using the cumulative variance contribution rate. Change information is retained while redundant data is removed. Finally, the reciprocal square root of the eigenvalues ​​is used for weighting, and the weighted PCA data is calculated as follows: In the formula, Principal component matrix, This is the eigenvalue vector.

[0033] 4) Extreme scene identification and extraction In the feature space formed by the weighted low-dimensional feature vectors, the Euclidean distance from the weighted low-dimensional feature vectors corresponding to each day to the spatial mean point is calculated. A threshold is constructed based on the mean and standard deviation of the Euclidean distances. Based on the threshold, corresponding samples are extracted from the daily net load samples as extreme scenarios of the power system. Specifically: On the dimensionality-reduced data, the standard deviation method is used to identify extreme scenarios. First, the Euclidean distance from the weighted low-dimensional feature vectors for each day to the spatial mean point is calculated. Then, a standard deviation threshold is set; all sample points with a distance greater than this threshold are considered extreme scenarios. This method can effectively identify extreme scenarios that are significantly different from the system's normal operating state and has strong adaptability. Specifically, it includes: First, calculate the Euclidean distance from the weighted low-dimensional feature vectors corresponding to each day to the spatial mean point: ; in, For the first The nth sample point in the weighted PCA low-dimensional space 3D coordinates; For all sample points at the th The mean of the dimensional coordinates; It is the dimension of a low-dimensional space; Then, the recognition threshold for extreme scenarios is set using the standard deviation method: ; In the formula, Euclidean distance The mean; Euclidean distance Standard deviation; This is the standard deviation factor, used to adjust the threshold level. Increasing the threshold will raise the number of samples identified as extreme scenarios. Decreasing the threshold will lower the threshold and increase the number of samples identified as extreme scenarios.

[0034] Extreme Scene Collection Defined as the index of all samples that exceed the recognition threshold: .

[0035] By following the steps above, the extreme scenario set is completed. The construction enables the identification of extreme scenarios in power systems.

[0036] In summary, this invention provides a method for identifying extreme scenarios in power systems. Based on time-series data of wind power, photovoltaic power, and load in a power system, net load data is obtained and sliced ​​daily to form daily net load samples. A Dynamic Time Warping (DTW) algorithm is used to calculate the pairwise DTW distance between each daily net load sample and construct a difference matrix. After standardizing the difference matrix, principal component analysis is performed to obtain the principal component score vectors corresponding to each day. These principal component score vectors are then used as low-dimensional feature vectors obtained by dimensionality reduction of the distance representation vectors formed by the DTW distances between that day and all other days. The low-dimensional feature vectors are weighted according to the principal component dimensions based on the reciprocal square root of the corresponding eigenvalues ​​of the principal components to obtain weighted low-dimensional feature vectors. In the feature space formed by the weighted low-dimensional feature vectors, the Euclidean distance from the weighted low-dimensional feature vectors corresponding to each day to the spatial mean point is calculated. A threshold is constructed using the standard deviation method based on the mean and standard deviation of the Euclidean distances to determine the index of the day whose Euclidean distance exceeds the threshold. Based on this, corresponding samples are extracted from the daily net load samples as extreme scenarios of the power system.

[0037] The extreme scenario identification method proposed in this invention, based on dynamic time warping, principal component analysis, and standard deviation, has significant technical advantages in addressing the problems of insufficient representativeness and lack of temporal coupling relationships in existing power system extreme scenario identification. By combining the DTW algorithm, PCA dimensionality reduction, and standard deviation, it can automatically extract the most representative extreme scenarios from the data without setting empirical thresholds (such as sustained low output thresholds, peak-valley thresholds, etc.) for wind, solar, and load respectively. Especially in power systems with high penetration of renewable energy, this invention can effectively cover various extreme scenarios that the system may encounter, ensuring that the identified extreme scenarios have stronger representativeness. By introducing the DTW algorithm and temporal weighting, the inherent temporal coupling relationship between wind, solar, and load is fully preserved, thereby identifying extreme scenarios that conform to physical rationality. This invention not only makes up for the shortcomings of existing methods but also more realistically reflects the operating state of the system under extreme scenarios.

[0038] Example 2 Based on the same inventive concept, the present invention also provides a power system extreme scene identification system for implementing the methods described in the foregoing embodiments. The system includes: a data acquisition module, a difference processing module, a weighted dimensionality reduction module, and an identification module. The data acquisition module is used to obtain the net load data of the power system based on the time-series data of wind power, photovoltaic power and load of the power system, and to form a daily net load sample by daily slice; The difference processing module is used to construct a difference matrix between each pair of daily net load samples using the Dynamic Time Warping (DTW) algorithm. The weighted dimensionality reduction module is used to perform principal component analysis (PCA) on the difference matrix after standardization to obtain a low-dimensional feature vector, and to weight the low-dimensional feature vector according to the reciprocal square root of the corresponding eigenvalue of the principal component to obtain a weighted low-dimensional feature vector. The identification module is used to calculate the Euclidean distance from the weighted low-dimensional feature vector corresponding to each day to the spatial mean point in the feature space composed of the weighted low-dimensional feature vectors, construct a threshold based on the mean and standard deviation of the Euclidean distance, and extract corresponding samples from the daily net load samples as extreme scenarios of the power system based on the threshold.

[0039] Furthermore, the difference processing module employs the Dynamic Time Warping (DTW) algorithm to construct a pairwise difference matrix between daily net load samples, including: The DTW algorithm uses dynamic programming to find the minimum cumulative cost for any two-day net load sample. Based on the minimum cumulative cost, construct a difference matrix between each pair of daily net load samples.

[0040] Furthermore, after standardizing the difference matrix, principal component analysis (PCA) is performed to obtain low-dimensional eigenvectors. The low-dimensional eigenvectors are then weighted according to the reciprocal square roots of the corresponding eigenvalues ​​of the principal components to obtain weighted low-dimensional eigenvectors. The method for obtaining weighted low-dimensional eigenvectors includes: After standardizing the difference matrix, principal component analysis is performed to obtain the principal component score vector for each day. The principal component score vector is then used as a low-dimensional feature vector obtained by reducing the dimensionality of the distance representation vector for each day, which is composed of the DTW distances between that day and the other days. Weights are constructed based on the reciprocal square roots of the eigenvalues ​​corresponding to the principal components, and the low-dimensional eigenvectors are then weighted dimension by dimension to obtain weighted low-dimensional eigenvectors.

[0041] Furthermore, the method of calculating the Euclidean distance from the weighted low-dimensional feature vectors corresponding to each day to the spatial mean point in the feature space composed of the weighted low-dimensional feature vectors, constructing a threshold based on the mean and standard deviation of the Euclidean distances, and extracting corresponding samples from the daily net load samples as extreme scenarios of the power system based on the thresholds includes: First, calculate the Euclidean distance from the weighted low-dimensional feature vectors corresponding to each day to the spatial mean point: ; in, For the first The nth sample point in the weighted PCA low-dimensional space 3D coordinates; For all sample points at the th The mean of the dimensional coordinates; It is the dimension of a low-dimensional space; Then, the recognition threshold for extreme scenarios is set using the standard deviation method: ; In the formula, Euclidean distance The mean; Euclidean distance Standard deviation; This is the standard deviation factor, used to adjust the threshold level. Increasing the threshold will raise the number of samples identified as extreme scenarios. Decreasing the threshold will lower the threshold and increase the number of samples identified as extreme scenarios.

[0042] Extreme Scene Collection Defined as the index of all samples that exceed the recognition threshold: .

[0043] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for identifying extreme scenarios in a power system, characterized in that, The method includes: Based on the time-series data of wind power, photovoltaic power and load in the power system, the net load data of the power system is obtained, and daily net load samples are formed by daily slices. The Dynamic Time Warping (DTW) algorithm is used to construct the pairwise difference matrix between daily net load samples; After standardizing the difference matrix, principal component analysis (PCA) is performed to obtain low-dimensional eigenvectors. The low-dimensional eigenvectors are then weighted according to the reciprocal square roots of the corresponding eigenvalues ​​of the principal components to obtain weighted low-dimensional eigenvectors. In the feature space formed by the weighted low-dimensional feature vectors, the Euclidean distance from the weighted low-dimensional feature vectors corresponding to each day to the spatial mean point is calculated. A threshold is constructed based on the mean and standard deviation of the Euclidean distance. Based on the threshold, corresponding samples are extracted from the daily net load samples as extreme scenarios of the power system. The specific method includes: First, calculate the Euclidean distance from the weighted low-dimensional feature vectors corresponding to each day to the spatial mean point: ； in, For the first The nth sample point in the weighted PCA low-dimensional space 3D coordinates; For all sample points at the th The mean of the dimensional coordinates; It is the dimension of a low-dimensional space; Then, the recognition threshold for extreme scenarios is set using the standard deviation method: ； In the formula, Euclidean distance The mean; Euclidean distance Standard deviation; This is the standard deviation factor, used to adjust the threshold level. Increasing the threshold will raise the number of samples identified as extreme scenarios. Decreasing the threshold will lower the number of samples identified as extreme scenarios; Extreme Scene Collection Defined as the index of all samples that exceed the recognition threshold: 。 2. The method according to claim 1, characterized in that, Methods for constructing pairwise difference matrices between daily net load samples using the Dynamic Time Warping (DTW) algorithm include: The DTW algorithm uses dynamic programming to find the minimum cumulative cost for any two-day net load sample. Based on the minimum cumulative cost, construct a difference matrix between each pair of daily net load samples.

3. The method according to claim 1, characterized in that, After standardizing the difference matrix, principal component analysis (PCA) is performed to obtain low-dimensional eigenvectors. The low-dimensional eigenvectors are then weighted according to the reciprocal square roots of the corresponding eigenvalues ​​of the principal components to obtain weighted low-dimensional eigenvectors. The method for obtaining weighted low-dimensional eigenvectors includes: After standardizing the difference matrix, principal component analysis is performed to obtain the principal component score vector for each day. The principal component score vector is then used as a low-dimensional feature vector obtained by reducing the dimensionality of the distance representation vector for each day, which is composed of the DTW distances between that day and the other days. Weights are constructed based on the reciprocal square roots of the eigenvalues ​​corresponding to the principal components, and the low-dimensional eigenvectors are then weighted dimension by dimension to obtain weighted low-dimensional eigenvectors.

4. A power system extreme scenario recognition system, the system being used to implement the method according to any one of claims 1-3, characterized in that, The system includes: a data acquisition module, a difference processing module, a weighted dimensionality reduction module, and a recognition module; The data acquisition module is used to obtain the net load data of the power system based on the time-series data of wind power, photovoltaic power and load of the power system, and to form a daily net load sample by daily slice; The difference processing module is used to construct a difference matrix between each pair of daily net load samples using the Dynamic Time Warping (DTW) algorithm. The weighted dimensionality reduction module is used to perform principal component analysis (PCA) on the difference matrix after standardization to obtain a low-dimensional feature vector, and to weight the low-dimensional feature vector according to the reciprocal square root of the corresponding eigenvalue of the principal component to obtain a weighted low-dimensional feature vector. The identification module is used to calculate the Euclidean distance from the weighted low-dimensional feature vector corresponding to each day to the spatial mean point in the feature space composed of the weighted low-dimensional feature vectors, construct a threshold based on the mean and standard deviation of the Euclidean distance, and extract corresponding samples from the daily net load samples as extreme scenarios of the power system based on the threshold.

5. The system according to claim 4, characterized in that, The difference processing module employs the Dynamic Time Warping (DTW) algorithm, and the method for constructing the pairwise difference matrix between daily net load samples includes: The DTW algorithm uses dynamic programming to find the minimum cumulative cost for any two-day net load sample. Based on the minimum cumulative cost, construct a difference matrix between each pair of daily net load samples.

6. The system according to claim 4, characterized in that, After standardizing the difference matrix, principal component analysis (PCA) is performed to obtain low-dimensional eigenvectors. The low-dimensional eigenvectors are then weighted according to the reciprocal square roots of the corresponding eigenvalues ​​of the principal components to obtain weighted low-dimensional eigenvectors. The method for obtaining weighted low-dimensional eigenvectors includes: After standardizing the difference matrix, principal component analysis is performed to obtain the principal component score vector for each day. The principal component score vector is then used as a low-dimensional feature vector obtained by reducing the dimensionality of the distance representation vector for each day, which is composed of the DTW distances between that day and the other days. Weights are constructed based on the reciprocal square roots of the eigenvalues ​​corresponding to the principal components, and the low-dimensional eigenvectors are then weighted dimension by dimension to obtain weighted low-dimensional eigenvectors.

7. The system according to claim 4, characterized in that, The method for calculating the Euclidean distance from the weighted low-dimensional feature vectors corresponding to each day to the spatial mean point in the feature space composed of the weighted low-dimensional feature vectors, constructing a threshold based on the mean and standard deviation of the Euclidean distances, and extracting corresponding samples from the daily net load samples as extreme scenarios of the power system based on the thresholds includes: First, calculate the Euclidean distance from the weighted low-dimensional feature vectors corresponding to each day to the spatial mean point: ； in, For the first The nth sample point in the weighted PCA low-dimensional space 3D coordinates; For all sample points at the th The mean of the dimensional coordinates; It is the dimension of a low-dimensional space; Then, the recognition threshold for extreme scenarios is set using the standard deviation method: ； In the formula, Euclidean distance The mean; Euclidean distance Standard deviation; This is the standard deviation factor, used to adjust the threshold level. Increasing the threshold will raise the number of samples identified as extreme scenarios. Decreasing the threshold will lower the number of samples identified as extreme scenarios; Extreme Scene Collection Defined as the index of all samples that exceed the recognition threshold: 。

Images