Power system extreme scene generation method and related equipment

Abstract

The invention relates to the technical field of power system extreme scene generation, and discloses a power system extreme scene generation method and related equipment, and the method comprises the steps: extracting a meteorological feature factor representing extreme weather according to a historical meteorological sample; based on the correlation and causality of the meteorological characteristic factors and historical power scene samples, performing time sequence clustering on the meteorological characteristic factors to obtain a clustering result, and taking the clustering result as a classification label of a power scene; training the extreme scene generation model by improving the structure of the extreme scene generation model and the loss function; and inputting the power scene with the classification label into the trained extreme scene generation model, outputting to obtain a synthetic sample of the power system scene, and generating the extreme scene of the power system according to the synthetic sample of the power system scene. According to the method, the accuracy and credibility of scene generation are improved while the effective generation of the extreme scene of the power system is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power system extreme scenario generation, and specifically to a power system extreme scenario generation method and related devices. Background Technique

[0002] As an important infrastructure for ensuring the national economic lifeline and people's livelihood, the power system is extremely vulnerable to severe impacts of frequent extreme disasters such as typhoons, earthquakes, and ice disasters, resulting in large-scale power outages. Although the probability of extreme disasters occurring is low, due to their strong destructive power and wide influence range, it has seriously hindered the construction and development process of the power system to be green, clean, safe, controllable, intelligent, and friendly. Under the strong interference of extreme events, the failure probability of components will increase significantly, easily causing large-scale "N-K" fault situations such as power system line disconnection and tower collapse, posing a huge challenge to power safety. In order to enhance the extreme disaster resistance ability of the power system and minimize the losses caused by extreme disasters as much as possible, it is necessary to carry out medium- and long-term planning and operation research on the power system for extreme weather scenarios. In this process, the high-quality generation of power system scenarios under extreme weather directly affects the effectiveness and reliability of decision-making, which is of great significance.

[0003] However, extreme scenarios exhibit the characteristics of low occurrence probability and large impact. It is difficult to provide sufficient sample support only relying on historical data. Currently, scenario generation methods can be mainly divided into sampling methods and prediction methods. Sampling methods include Monte Carlo sampling method, Latin hypercube sampling method, Copula sampling method, etc. However, these methods all require certain prior assumptions and complex probability distribution modeling, which may lead to large parameter fitting errors and make it difficult to handle the complex probability distribution characteristics of actual new energy power generation and load. Prediction methods include autoregressive moving average model, autoregressive integrated moving average model, variational autoencoder, etc. This type of method focuses on the time series relationship of the scenario sequence while ignoring the statistical characteristics, which can avoid complex probability distribution modeling, but the accuracy of the generated scenarios is insufficient.

[0004] The generative adversarial network model is a machine learning-based scenario generation method that can effectively capture the time-domain characteristics of time-series scenarios, has low requirements for prior assumptions, and is highly adaptable to the training conditions of a small number of samples. The generative adversarial network model can effectively solve the small-sample learning problem of extreme scenario generation. However, how to maintain the generalization ability of the model and avoid mode collapse in the case of extremely few samples, and how to capture key features in high-dimensional and complex power data to generate practical power scenarios to effectively construct an extreme scenario generative adversarial network model under small-sample conditions still need to be solved. In addition, the limited scenario sample set is also difficult to cover various types of extreme scenarios, and there is a lack of comprehensive characterization of the impact of extreme weather events. Existing conventional data-driven methods rarely consider the feature correlations between multiple types of extreme scenarios (such as typhoons, cold snaps, and high temperatures), and there are still technical deficiencies in capturing the time-series characteristics of new energy and load curves in the power system under extreme scenarios and generating controllable and high-quality power curves. Summary of the Invention

[0005] To overcome the defects of the above-mentioned prior art, the purpose of the present invention is to provide a method and related equipment for generating extreme scenarios of a power system, so as to solve the technical problems in the prior art of how to improve the effective generation of extreme scenarios of a power system and how to improve the accuracy and credibility of scenario generation.

[0006] The present invention is realized through the following technical solutions:

[0007] In the first aspect, the present invention provides a method for generating extreme scenarios of a power system, including:

[0008] Obtain historical meteorological samples and historical power scenario samples under extreme weather, and extract meteorological feature factors representing extreme weather based on the historical meteorological samples;

[0009] Based on the correlation and causality between the meteorological feature factors and the historical power scenario samples, perform time-series clustering on the meteorological feature factors to obtain a clustering result, and use the clustering result as the classification label of the power scenario;

[0010] Construct an extreme scenario generation model, and train the extreme scenario generation model by improving the structure and loss function of the extreme scenario generation model;

[0011] Input the power scenario with the classification label into the trained extreme scenario generation model, and then output to obtain a synthetic sample of the power system scenario, and generate an extreme scenario of the power system according to the synthetic sample of the power system scenario.

[0012] Preferably, in the step of extracting meteorological feature factors representing extreme weather based on the historical meteorological samples, a random forest is used to extract the meteorological feature factors under extreme weather, including:

[0013] Evaluate the importance of extreme weather meteorological characteristics based on the contribution of different extreme weather historical meteorological samples to each tree in the random forest;

[0014] Rank the importance, and use the extreme weather meteorological characteristics with high rankings as the meteorological characteristic factors representing extreme weather.

[0015] Preferably, in the step based on the correlation and causality between the meteorological characteristic factors and historical power scenario samples, the meteorological characteristic factors under extreme weather and historical power scenario samples are respectively tested for the stationarity of the time series and the cointegration relationship between time series through the unit root test and the cointegration test method, and the correlation and causality between the meteorological characteristic factors under extreme weather and historical power scenario samples are obtained through the Granger causality test based on the stationarity of the time series and the cointegration relationship between time series;

[0016] Among them, in the unit root test, the stationarity test result is obtained. If the statistical result rejects the null hypothesis, then the time series does not have a unit root, and the stationarity test result is that the variable time series has stationarity. On the contrary, the time series has a unit root, and the stationarity test result is that the variable time series is a non-stationary time series;

[0017] According to the stationarity test result, when two variable time series are of the same order of integration, they are tested through the cointegration test method, the cointegration regression equation between variables is estimated, and a residual series is generated accordingly; the stationarity of the residual series is tested. If the residual series is stationary, there is a cointegration relationship between variables, otherwise there is no cointegration relationship;

[0018] Among them, when there is a cointegration relationship between variables, the correlation and causality between the meteorological characteristic factors under extreme weather and historical power scenario samples are obtained through the Granger causality test.

[0019] Preferably, in the step of performing time series clustering on the meteorological characteristic factors to obtain a clustering result, K-MDTSC is used to calculate the distance of K-MDTSC based on the framework of the K-means clustering algorithm with N as the dimension of the clustering sample and T as the time scale of the clustering sample to obtain the clustering result. The expression of the distance of K-MDTSC is as follows:

[0020]

[0021] Among them, L is the metric distance, X N 、C N are multi-dimensional meteorological time series;

[0022] Evaluate the clustering result through the within-cluster sum of squared errors to obtain the optimal number of clusters, and its expression is as follows:

[0023]

[0024] Among them, C is a cluster of clustering, is the set of all clustering clusters.

[0025] Preferably, in the step of training the extreme scenario generation model by improving the structure and loss function of the extreme scenario generation model, by improving the Wasserstein distance and optimizing the gradient penalty weight, a WGAN-GP model is formed, and on the basis of the WGAN-GP model, the operation constraints of the power system are introduced to form a power system extreme scenario generation model, and the power system extreme scenario generation model is trained under different gradient penalty weights.

[0026] Preferably, in the step of inputting the power scenario with classification labels into the trained extreme scenario generation model and outputting the synthetic sample of the power system scenario, in the step of generating the extreme scenario of the power system according to the synthetic sample of the power system scenario, the synthetic sample of the power system scenario includes the time series of wind power, photovoltaic power and load. Among them, the wind power, photovoltaic power and load in the synthetic sample of the same power system scenario each form a power interval. When the generated scenario is within the power interval, the extreme scenario generation work of the power system is completed.

[0027] Furthermore, the generated scenarios are respectively determined by the point accuracy A p and the scenario accuracy A s Among them, the point accuracy A p is the ratio of the number of generated points within the effective range to the total number of generated points. When the time series lengths of the wind, light and load powers in the generated scenario are the same, the total number of generated points is fixed. The calculation formula of A p is as follows:

[0028]

[0029] In the formula, N pv,j is the number of generated points within the effective interval, j is the index of wind power, photovoltaic power and load, N s is the number of generated scenarios, and L is the length of each time series in the generated scenario;

[0030] Among them, the scenario accuracy A s is the ratio of the number of effective scenarios to the total number of generated scenarios. When the time series of wind power, photovoltaic power and load in the generated scenario are all within the effective interval, the generated scenario is effective. The calculation formula of A s is as follows:

[0031]

[0032] In the formula, Nsv is the number of generated points within the valid interval.

[0033] In a second aspect, the present invention further provides a power system extreme scenario generation system for implementing the power system extreme scenario generation method described above, including:

[0034] A meteorological feature factor extraction module for obtaining historical meteorological samples and historical power scenario samples under extreme weather, and extracting meteorological feature factors characterizing extreme weather based on the historical meteorological samples;

[0035] A clustering result classification module for performing time series clustering on the meteorological feature factors based on the correlation and causality between the meteorological feature factors and the historical power scenario samples to obtain a clustering result, and using the clustering result as a classification label for the power scenario;

[0036] A scenario generation model training module for constructing an extreme scenario generation model and training the extreme scenario generation model by improving the structure and loss function of the extreme scenario generation model;

[0037] An extreme scenario generation module for inputting the power scenario with a classification label into the trained extreme scenario generation model and outputting a synthetic sample of the power system scenario, and generating an extreme scenario of the power system based on the synthetic sample of the power system scenario.

[0038] In a third aspect, the present invention further provides a mobile terminal, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the power system extreme scenario generation method described above are implemented.

[0039] In a fourth aspect, the present invention further provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the steps of the power system extreme scenario generation method described above are implemented.

[0040] Compared with the prior art, the present invention has the following beneficial technical effects:

[0041] The present invention provides a method for generating extreme scenarios of a power system. By extracting meteorological characteristic factors representing extreme weather from historical meteorological samples under extreme weather, the quality of the basic data for subsequent analysis can be ensured, thereby improving the accuracy of prediction. Through the correlation and causality analysis of the meteorological characteristic factors under extreme weather and historical power scenario samples, the relationship between the two can be deeply understood, which helps to more accurately simulate and predict the impact of extreme weather on the power system, thus improving the accuracy of extreme scenario generation. Using time series clustering to cluster the meteorological characteristic factors under extreme weather and obtaining the clustering result as the classification label of the power scenario can capture the dynamic characteristics of time series data, improve the generalization ability of the model, and enable the model to handle more diverse extreme scenarios. By outputting synthetic samples of the power system scenario through the trained extreme scenario generation model, it can be used to simulate and test the operation of the power system under extreme weather, provide strong decision-making support for the operation and planning of the power system, improve the effective generation of extreme scenarios of the power system, and at the same time improve the accuracy and credibility of scenario generation.

[0042] Furthermore, the present invention proposes a method for constructing an extreme scenario sample set of a power system. Without manually identifying extreme weather and its characteristics, only through the importance ranking of quantified meteorological factors and the time series clustering of meteorological factors, the classification label of historical scenario samples can be automatically obtained, and a training sample set for the extreme scenario generation model can be constructed.

[0043] Furthermore, in order to enhance the adaptability of the generated extreme scenarios to the operation of the power system, the present invention captures the internal law of extreme weather scenario data by constructing a case generator and a discriminator, and introduces the operation constraints of the power system source and load into the training of the generation model, ensuring that the generated scenarios closely conform to the actual operation of the power system, and effectively improving the accuracy and credibility of the generated scenarios.

[0044] Furthermore, in order to more accurately reflect the quality of the generated scenarios and the performance of the generation model, the present invention proposes a method for evaluating the generated scenarios considering the point accuracy and the scenario accuracy. Starting from the requirements of power system operation and planning, it discriminates the effectiveness of the generation model of the generated scenarios. Compared with traditional evaluation indicators such as probability density function and autocorrelation coefficient, it has more reference significance for the actual operation and planning of the power system. Description of the Drawings

[0045] Figure 1 It is a flowchart of the method for generating extreme scenarios of a power system in an embodiment of the present invention;

[0046] Figure 2 It is a flowchart of the extreme scenario generation framework of a power system in an embodiment of the present invention;

[0047] Figure 3This is the clustering graph of historical samples of extreme weather and power system scenarios in the embodiments of the present invention;

[0048] Figure 4 This is the schematic diagram of the improved GAN model in the embodiments of the present invention;

[0049] Figure 5 This is the classification comparison graph of generated scenarios and real scenarios in the embodiments of the present invention;

[0050] Figure 6 This is the accuracy comparison graph of the improved GAN, CGAN, and WGAN-GP in the embodiments of the present invention;

[0051] Figure 7 This is the schematic diagram of the structure of the power system extreme scenario generation system in the embodiments of the present invention;

[0052] In the figure: 1. Meteorological feature factor extraction module; 2. Clustering result classification module; 3. Scenario generation model training module; 4. Extreme scenario generation module; 11. Feature factor extraction sub-module; 21. First classification processing sub-module; 22. Second classification processing sub-module; 31. Model processing sub-module; 41. Extreme scenario generation sub-module; 411. Scenario determination sub-module. Detailed implementation manners

[0053] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0054] The purpose of the present invention is to provide a method and related device for generating extreme scenarios of a power system to solve the technical problems of how to improve the effective generation of extreme scenarios of a power system and how to improve the accuracy and credibility of scenario generation in the prior art.

[0055] The following further describes the present invention in detail with reference to the drawings:

[0056] See Figure 1 , the present invention provides a method and related device for generating extreme scenarios of a power system, including the following steps:

[0057] Step 1, obtain historical meteorological samples and historical power scenario samples under extreme weather, and extract meteorological feature factors characterizing extreme weather based on the historical meteorological samples;

[0058] Specifically, in the step of extracting meteorological characteristic factors representing extreme weather based on historical meteorological samples, the random forest is used to extract meteorological characteristic factors under extreme weather, including:

[0059] Evaluating the importance of extreme weather meteorological characteristics according to the contribution of historical meteorological samples of different extreme weather to each tree in the random forest;

[0060] Sorting the importance, and taking the extreme weather meteorological characteristics with high ranking as the meteorological characteristic factors representing extreme weather.

[0061] Among them, the meteorological characteristic importance score is denoted as VIM, and the Gini index is denoted as GI. Suppose there are c meteorological characteristics x1, x2, x3, …, x c , and k types of extreme weather scenarios. Calculate the Gini index score of each meteorological characteristic x j as VIM j , that is, the average change in the node splitting impurity of the jth meteorological characteristic in all decision trees. The calculation formula of the Gini index of node m in the ith tree is as follows:

[0062]

[0063] Among them, p mk represents the proportion of meteorological category k in node m, that is, the probability that the class labels of two samples randomly selected from node m are inconsistent. The importance of meteorological characteristic x j in the ith tree node m can be represented by the Gini gain before and after the branch of node m:

[0064]

[0065] Among them, and represent the Gini indices of the two new nodes after branching respectively. If the meteorological characteristic x j appears in the node set M in decision tree i, then the importance of x j in the ith tree is:

[0066]

[0067] Considering n decision trees in the random forest, the normalized importance score of meteorological characteristic x j is:

[0068]

[0069] Sort the importance scores of various meteorological characteristics in extreme weather, and take the top 5 meteorological characteristics with higher scores as the key meteorological factors of extreme weather.

[0070] Step 2: Based on the correlation and causality between the meteorological characteristic factors and the historical power scenario samples, perform time series clustering on the meteorological characteristic factors to obtain a clustering result, and use the clustering result as the classification label for the power scenario;

[0071] Specifically, considering 5 key meteorological factors, the Granger causality test is used to mine the dominant characteristics of the meteorological characteristic factors of extreme weather scenarios affecting the continuous operation of the power supply side and the load side. The main principle is that for two time series X and Y, if the prediction ability of Y is improved after adding the variable X with a specific lag period to the autoregressive model of Y, it can be considered that the variable X is the Granger cause of Y. However, the time series applied for the Granger causality test must be a stationary time series or a non-stationary time series that satisfies the cointegration relationship to avoid the problem of "spurious regression" in the Granger test results. Therefore, in order to ensure the effectiveness of the test results, unit root test and cointegration test need to be carried out before the causality test. In the present invention, the augmented Dickey-Fuller (ADF) unit root test and the Engle-Granger two-step cointegration test method are respectively used to test the stationarity of the time series and the cointegration relationship between time series.

[0072] The unit root test is mainly to ensure that the time series data participating in the Granger test is stationary. The ADF test used in the present invention is the most commonly used method, which can be expressed as:

[0073]

[0074] where α is a constant, β is a function of time t, and δ i is the coefficient of the i-th autoregressive model with p-order lag.

[0075] The null hypothesis of the ADF test is H0: γ = 0. If the statistical result rejects the null hypothesis, it indicates that there is no unit root in this time series, the variable time series is stationary, and it also means that x t does not change with the change of time t. If the statistical result does not reject the null hypothesis, it indicates that the sequence has a unit root and is a non-stationary time series. For non-stationary time series, differencing is required. If the original sequence becomes a stationary sequence after d times of differencing, the sequence is called integrated of order d, denoted as I(d).

[0076] In the results of the stationarity test, if the orders of integration of the above variable integrated sequences are different, there cannot be a cointegration relationship between the variables. Only when the orders of integration of the above variable integrated sequences are the same, there may be a cointegration relationship between the variables. Therefore, on the basis that two time series variables are integrated of the same order, cointegration analysis needs to be continued to determine whether there is a cointegration relationship between the two variables. If two non-stationary and same-order integrated time series pass the cointegration test, Granger causality test can be continued to further explore whether the relationship between the two constitutes a causal relationship.

[0077] Based on the above analysis, the present invention adopts the Engle-Granger two-step cointegration test method to test the cointegration relationship between variables and determine whether the linear combination between two variables is stationary. First, estimate the cointegration regression equation between variables to generate the residual term accordingly; then, conduct a stationarity test on the residual sequence. If the residual sequence is stationary, it indicates that there is a cointegration relationship between the variables, otherwise there is no cointegration relationship.

[0078] The present invention uses the least squares method (OLS) to estimate the cointegration regression equation of variables.

[0079] y t =α + βx t +e t (6)

[0080] Based on the cointegration regression equation, further calculate the residual estimate value of the regression model. By transposing, the residual term e t .

[0081] e t =y t -α - βx t (7)

[0082] In Equation (6), α and β are the estimated values of the regression coefficients. For Equation (6), the ADF unit root test method is used for the stationarity test. If the residual sequence e t passes the stationarity test, it indicates that the above residual sequence e t is stable, and it also shows that the non-stationary time series y t and x t have a cointegration relationship. On this basis, Granger causality test can be continued to test whether the relationship between variables constitutes a causal relationship.

[0083] Granger causality test is used to test the causal relationship between time series. For the meteorological time series X and the power time series Y, by increasing the historical values of one of the sequences to determine whether the prediction of the other sequence can be enhanced, so as to determine whether there is a Granger causal relationship. The detection model estimates the regression through the least squares method as follows:

[0084]

[0085] The null hypothesis of the Granger test is H0: the meteorological time series does not cause the Granger distribution of the power time series. The null hypothesis is expressed mathematically as follows:

[0086] H0: β1 = β2 =... = β k = 0 (9)

[0087] The Granger causality test is performed through a restricted F-test. First, assume that the null hypothesis holds, that is, assume that the parameters before each X lag term in the formula are all zero, and perform regressions with and without each X lag term respectively to obtain the residual sum of squares RSS of the restricted model R and the residual sum of squares RSS of the unrestricted model U . Then construct the F statistic as follows:

[0088]

[0089] In the formula, m is the number of lag terms of X, n is the number of samples, and k is the number of parameters to be estimated in the unrestricted regression model, including possible constant terms and other variables. If the calculated F value is greater than the critical value F of the F distribution at the given significance level α α (m, n - k), then the original hypothesis H0 can be rejected, and the conclusion can be drawn that the meteorological time series is the Granger cause of the power time series.

[0090] Specifically, meteorological factors have a higher dimension than power system scenarios, and their characteristics are more significant. Using the clustering results as labels can further clarify the characteristics of extreme scenarios and guide the generation of extreme scenarios.

[0091] Based on the framework of the K-means clustering algorithm, K-MDTSC improves the distance calculation of multi-dimensional time series. Taking N as the dimension of the clustering samples and T as the time scale of the clustering samples, the distance calculation formula of K-MDTSC is as follows:

[0092]

[0093] Among them, L is the metric distance (Euclidean distance when L = 2), X N , C N are multi-dimensional meteorological time series. Similar to traditional clustering algorithms, the sum of the squares of the distances between sample points and their cluster centers, that is, the within-cluster sum of squared errors (SSE), is used to evaluate the performance of the algorithm and determine the optimal number of clusters. The calculation formula of SSE is as follows:

[0094]

[0095] Where C is a cluster of clustering, is the set of all clustering clusters.

[0096] In the present invention, for power system scenarios such as wind power X w , photovoltaic X s , load X l etc., the clustering results of meteorological factors are marked with label c, which generally represents a certain specific extreme weather. The feature differences of different label scenarios are relatively large. Combined with the source-load time series scenario, learning samples {X w , X s , X l |c} with different labels are constructed to guide the directional generation of extreme scenarios.

[0097] Step 3, construct an extreme scenario generation model, and train the extreme scenario generation model by improving the structure and loss function of the extreme scenario generation model;

[0098] Specifically, the loss function of the generation model is generally constructed by the difference between the two probability distributions of the generated samples and the real samples. The Wasserstein distance, also known as the earth mover's distance, is used to construct the loss function of the generation model.

[0099]

[0100] Where Π(P r , P g ) is the set of all possible joint distributions combined by P1 and P2 distributions. For each possible joint distribution γ, a sample x and y can be sampled (x, y) ~ γ, and the distance ||x - y|| between this pair of samples can be calculated. Further, the expected value E (x,y)～γ [||x - y||] of the sample pair distance under this joint distribution γ can be obtained. The lower bound inf that can be taken for this expected value among all possible joint distributions is the Wasserstein distance. The advantage of the Wasserstein distance compared to the KL divergence and JS divergence is that even if the two distributions do not overlap, the Wasserstein distance can still reflect their proximity, while the JS divergence is a constant in this case and the KL divergence may be meaningless.

[0101] It can be seen from the above analysis that the Wasserstein distance has excellent smoothing characteristics and can theoretically solve the problem of gradient disappearance. However, the infimum in equation (13) is difficult to handle mathematically. According to the Kantorovich-Rubinstein duality, the following transformation can be made

[0102]

[0103] Where Pr is the real data distribution, P θ is the generated data distribution, and f is a K-Lipschitz function that satisfies the constraints in Equation (14). In this model, K = 1 is set to simplify the expression. The minimization problem of the original Wasserstein distance is transformed into a maximization problem in dual form, with the restriction that the function f maintains 1-Lipschitz continuity to ensure that the gradient remains smooth. Based on the Wasserstein distance, the loss function of the generative model is expressed as follows.

[0104]

[0105] where z is a random vector following the Gaussian distribution N. This loss function aims to evaluate the difference between the discriminator for the original data and the generated data to give a generation quality score. By minimizing the generator function G and maximizing the discriminator function D, the optimal synthetic data is output.

[0106] The gradients of the discriminator G and the generator D are defined as follows:

[0107]

[0108]

[0109] where and are the discriminator network parameters and the generator network parameters, respectively. However, the constraint imposed on the 1-Lipschitz function f is not only enforced through the gradient. Gradient clipping updates the weights through the optimizer to clip them between two hyperparameter constants (usually -0.01 to 0.01). Although gradient clipping ensures that the function f maintains 1-Lipschitz continuity, it actually brings instability problems to model training and does not solve the mode collapse problem. The main reason is that gradient clipping is highly sensitive to the values chosen for clipping and may lead to non-convergence or failure to reach the optimal solution. To address this issue, a penalty gradient is used to impose the 1-Lipschitz constraint on the function f. The loss function is redefined as a loss function with a regularization term as follows.

[0110]

[0111] where is the uniform sampling distribution on the straight line between P r and P θ , is a sample in the sampling distribution, and λ is the weight of the gradient penalty.

[0112] The goal of the generator is to obtain synthetic samples that are as close as possible to the true sample distribution. However, the operating scenarios of the power system not only have requirements for the probability distribution but also have strict constraints on the physical meaning. For example, the installed capacity of the power generation system has a constraint on the maximum power, and the device characteristics also have a constraint on the power change rate. In order to ensure the actual operating requirements of the generated sample load, it is necessary to constrain and correct the generated samples. The physical constraints of the power range and the ramp rate are as follows.

[0113]

[0114]

[0115] Where x lim is the numerical upper limit of the generated sample, Δt is the time interval for calculating the ramp rate, and Δx max is the maximum power change within the time interval of Δt.

[0116] Step 4: Input the power scenarios with classification labels into the trained extreme scenario generation model, and then output the synthetic samples of the power system scenarios. Generate the extreme scenarios of the power system according to the synthetic samples of the power system scenarios.

[0117] Specifically, there is a set of historical extreme scenarios under a certain type of extreme weather. The set of extreme scenario samples consists of the time series of wind power, photovoltaic power, and load. The wind power, photovoltaic power, and load within the same sample set each form a power interval. When the generated scenario is not or partially not within the corresponding power interval, the generated scenario is not effectively generated according to the type requirements. The accuracy of the generated scenario directly reflects the feasibility and effectiveness of the generation model.

[0118] Considering the data characteristics of the generated scenario, two accuracy quantification methods are proposed. The first accuracy is the point accuracy A p . The point accuracy considers the effectiveness of each point in the generated time series. The point accuracy is the ratio of the number of generated points within the valid range to the total number of generated points.

[0119] In the generated scenario, the time series lengths of each wind power, photovoltaic power, and load power are the same, so the total number of generated points is fixed. The calculation formula of A p is as follows.

[0120]

[0121] Where N pv,j is the number of generated points within the valid interval, j is the index of wind power, photovoltaic power, and load, N s is the number of generated scenarios, and L is the length of each time series in the generated scenario.

[0122] The second accuracy is the scenario accuracy A s。An effective scenario requires that the time series of wind power, photovoltaic power, and load are all within the effective range. Scenario accuracy reflects the ability of the model to generate completely effective scenarios.

[0123] Scenario accuracy is the ratio of the number of effective scenarios to the total number of generated scenarios. A s The calculation formula is as follows.

[0124]

[0125] In the formula, N sv is the number of generation points within the effective range. It can be seen from the definition of accuracy that A s is a more stringent metric than A p . Even if some points of the time series are not within the power range, or some types of time series are invalid, the entire scenario is considered invalid.

[0126] In the present invention, the extreme scenario generation method can improve the operation scenarios of the power system under various types of extreme weather, expand the extreme scenario set, provide data support and decision-making basis for power system dispatching and planning, and enhance the reliability of the power system under extreme events.

[0127] Embodiment 1

[0128] This embodiment provides a method for generating extreme scenarios of a power system, as Figure 2 shown, and the specific process is as follows:

[0129] First, to construct a training sample set for extreme scenarios, meteorological factor data such as temperature, precipitation, snowfall, light intensity, and wind speed during extreme weather occurrences on the main islands of Japan were collected, as well as wind power, photovoltaic power, and load power data. The time resolution of the data was adjusted to 5 minutes through a time resampling algorithm. Since various meteorological data and wind-solar-load data have different dimensions and significantly different value ranges, in order to reduce the computational complexity and enable the model to converge quickly, it is necessary to normalize the data:

[0130]

[0131] In the formula, x is each type of meteorological and source-load data, and x i is the i-th sample value in this type of data, and x i * is the value after normalizing x i . max{x} and min{x} are the maximum and minimum values in this type of data, respectively.

[0132] The random forest is used to screen the important features from the original meteorological feature sequences of the above extreme weather. Each decision tree is constructed based on the impurity of nodes, and the average impurity reduction of each feature in the decision tree node splitting is calculated to measure the importance of meteorological factors, and the key meteorological factors are extracted.

[0133] The Granger causality test is used to analyze the causality between meteorological factors such as temperature, precipitation, wind speed, light, and snowfall and the source-load operation scenarios of the power system. To test the effectiveness of the results, the unit root test and the cointegration test are carried out before the causality test. In this embodiment, the ADF unit root test and the Engle-Granger two-step cointegration test method are respectively used to test the stationarity of the time series and the cointegration relationship between time series. On this basis, the Granger causality test is carried out to test whether the relationship between variables constitutes a causal relationship. The Granger causality test results are shown in Table 1.

[0134] Table 1 Granger causality test results

[0135]

[0136] In the Granger test results, the null hypothesis is rejected at the significance level with a p-value less than 0.05. It can be seen that both temperature and light are Granger causes of load and photovoltaic, while the main characteristic factors affecting wind power are mainly wind speed and snowfall.

[0137] The K-MDTSC algorithm is used to cluster the meteorological time series of extreme scenarios. The elbow method is used to determine that the optimal number of clusters is 15. Based on the correlation and causality between meteorological conditions and extreme scenarios, 5 clusters that can represent extreme scenarios are selected from 15 scenarios. These five categories represent typhoon, high temperature, cold wave, heavy snow, and heavy rain respectively, and the number of samples is 41, 86, 37, 36, and 28 respectively. The classified extreme scenario training samples are as Figure 3 shown.

[0138] Then, a CGAN model is built, and on this basis, the improved Wasserstein distance and gradient penalty weight are introduced to form a WGAN-GP model. Further, the operation constraints of the power system are constructed to form an improved GAN model applicable to the generation of extreme scenarios of the power system. The constructed improved GAN model is as Figure 4 shown. To obtain the optimal weight, the improved GAN model is trained under different gradient penalty weights, and when the weight of the gradient penalty is 15, there is good convergence.

[0139] According to the type of input sample data, the generation model can generate 5 types of scenarios, and 200 samples are set for each type of scenario, with a total of 1000 scenarios. The time series comparison between the generated scenarios and the real scenarios is as Figure 5As shown in the figure. The comparison results show that there are significant features among various generated scenarios, and it has good interpretability for extreme weather. For example, the load level is the highest in high-temperature weather represented by Scenario 2, and the power generation of wind power and solar energy is the lowest in heavy-rain weather represented by Scenario 5. In the same type of scenario, the generated scenarios can basically track the patterns and trends of real samples and have good consistency.

[0140] Finally, using the accuracy evaluation index of the generated scenarios, the improved GAN applicable to extreme scenario generation is compared with CGAN and WGAN-GP. The proposed improved GAN has certain advantages in accuracy, and the comparison results are as Figure 6 shown. A p ratio of A s evaluates the scenarios more precisely, resulting in little difference between various methods. By evaluating the complete scenarios, the proposed improved GAN has more significant advantages in effective scenario generation.

[0141] In summary, a method for generating extreme scenarios of a power system provided in this embodiment extracts meteorological characteristic factors representing extreme weather from historical meteorological samples under extreme weather, which can ensure the quality of the basic data for subsequent analysis, thereby improving the accuracy of prediction. Through the correlation and causality analysis of the meteorological characteristic factors under extreme weather and historical power scenario samples, the relationship between the two can be deeply understood, which helps to more accurately simulate and predict the impact of extreme weather on the power system, thereby improving the accuracy of extreme scenario generation. Using time series clustering to cluster the meteorological characteristic factors under extreme weather, and obtaining the clustering results as the classification labels of the power scenarios, can capture the dynamic characteristics of time series data, improve the generalization ability of the model, and enable the model to handle more diverse extreme scenarios. By outputting synthetic samples of the power system scenarios through the trained extreme scenario generation model, it can be used to simulate and test the operation of the power system under extreme weather, provide strong decision-making support for the operation and planning of the power system, improve the effective generation of extreme scenarios of the power system, and at the same time improve the accuracy and credibility of scenario generation.

[0142] Embodiment 2

[0143] According to Figure 7 as shown, the present invention also provides a power system extreme scenario generation system, including:

[0144] A meteorological characteristic factor extraction module 1, configured to obtain historical meteorological samples and historical power scenario samples under extreme weather, and extract meteorological characteristic factors representing extreme weather based on the historical meteorological samples;

[0145] In the step of the feature factor extraction sub-module 11 for extracting meteorological feature factors characterizing extreme weather based on historical meteorological samples, the random forest is used to extract meteorological feature factors under extreme weather, including:

[0146] Evaluating the importance of extreme weather meteorological features according to the contribution of historical meteorological samples of different extreme weather to each tree in the random forest;

[0147] Sorting the importance, and taking the extreme weather meteorological features with high sorting as the meteorological feature factors characterizing extreme weather.

[0148] The clustering result classification module 2 is used to perform time series clustering on the meteorological feature factors under extreme weather based on the correlation and causality between the meteorological feature factors characterizing extreme weather and historical power scenario samples to obtain a clustering result, and use the clustering result as the classification label of the power scenario;

[0149] The first classification processing sub-module 21 is used to, in the step of based on the correlation and causality between the meteorological feature factors and historical power scenario samples, respectively test the stationarity of the time series and the cointegration relationship between time series of the meteorological feature factors under extreme weather and historical power scenario samples through the unit root test and the cointegration test method, and obtain the correlation and causality between the meteorological feature factors under extreme weather and historical power scenario samples through the Granger causality test based on the stationarity of the time series and the cointegration relationship between time series;

[0150] Among them, in the unit root test, a stationarity test result is obtained. If the statistical result rejects the null hypothesis, then this time series has no unit root, and the stationarity test result is that the variable time series has stationarity. On the contrary, this time series has a unit root, and the stationarity test result is that the variable time series is a non-stationary time series;

[0151] According to the stationarity test result, when two variable time series are of the same order of integration, they are tested by the cointegration test method, the cointegration regression equation between the variables is estimated, and a residual sequence is generated accordingly; the stationarity of the residual sequence is tested. If the residual sequence is stationary, then there is a cointegration relationship between the variables, otherwise there is no cointegration relationship;

[0152] Among them, when there is a cointegration relationship between variables, the correlation and causality between the meteorological feature factors under extreme weather and historical power scenario samples are obtained through the Granger causality test.

[0153] The classification processing second sub-module 22 is used in the step of performing time series clustering on the meteorological characteristic factors to obtain a clustering result. Based on the framework of the K-means clustering algorithm, with N as the dimension of the clustering samples and T as the time scale of the clustering samples, the distance of K-MDTSC is calculated to obtain the clustering result. The expression of the distance of K-MDTSC is as follows:

[0154]

[0155] Among them, L is the metric distance, X N , C N are multi-dimensional meteorological time series;

[0156] The clustering result is evaluated by the sum of squared errors within the cluster to obtain the optimal number of clusters. The expression is as follows:

[0157]

[0158] Among them, C is a cluster of the clustering, is the set of all clustering clusters.

[0159] The scenario generation model training module 3 is used to construct an extreme scenario generation model and train the extreme scenario generation model by improving the structure and loss function of the extreme scenario generation model;

[0160] The model processing sub-module 31 is used in the step of constructing an extreme scenario generation model and training the extreme scenario generation model by improving the structure and loss function of the extreme scenario generation model. By improving the Wasserstein distance and optimizing the gradient penalty weight, a WGAN-GP model is formed. Based on the WGAN-GP model, the operating constraints of the power system are introduced to form a power system extreme scenario generation model, and the power system extreme scenario generation model is trained under different gradient penalty weights.

[0161] The extreme scenario generation module 4 is used to input the power scenario with classification labels into the trained extreme scenario generation model and output the synthetic sample of the power system scenario, and generate the extreme scenario of the power system according to the synthetic sample of the power system scenario.

[0162] The extreme scenario generation sub-module 41 is used in the step of inputting the power scenario with classification labels into the trained extreme scenario generation model and outputting the synthetic sample of the power system scenario. In the step of generating the extreme scenario of the power system according to the synthetic sample of the power system scenario, the synthetic sample of the power system scenario includes the time series of wind power, photovoltaic power, and load. Among them, the wind power, photovoltaic power, and load in the synthetic sample of the same power system scenario each form a power interval. When the generated scenario is within the power interval, the extreme scenario generation work of the power system is completed.

[0163] The scenario determination sub-module 411 is used to determine the generated scenario respectively through the point accuracy A p and the scenario accuracy A s where the point accuracy A p is the ratio of the number of generated points within the effective range to the total number of generated points. When the time series lengths of each wind, light, and load power in the generated scenario are the same, the total number of generated points is fixed. The calculation formula of A p is as follows:

[0164]

[0165] In the formula, N pv,j is the number of generated points within the effective interval, j is the index of wind power, photovoltaic power, and load, N s is the number of generated scenarios, and L is the length of each time series in the generated scenario;

[0166] Among them, the scenario accuracy A s is the ratio of the number of effective scenarios to the total number of generated scenarios. When the time series of wind power, photovoltaic power, and load in the generated scenario are all within the effective interval, the generated scenario is effective. The calculation formula of A s is as follows:

[0167]

[0168] In the formula, N sv is the number of generated points within the effective interval.

[0169] Embodiment 3

[0170] The present invention also provides a mobile terminal, including a memory, a processor, and a computer program stored in the memory and executable on the processor, such as the power system extreme scenario generation program.

[0171] When the processor executes the computer program, it realizes the steps of the above-mentioned power system extreme scenario generation method, such as:

[0172] Obtain historical meteorological samples and historical power scenario samples under extreme weather, and extract meteorological characteristic factors representing extreme weather based on the historical meteorological samples;

[0173] Based on the correlation and causality between the meteorological characteristic factors and the historical power scenario samples, perform time series clustering on the meteorological characteristic factors to obtain a clustering result, and use the clustering result as the classification label for the power scenario;

[0174] Construct an extreme scenario generation model, and train the extreme scenario generation model by improving the structure and loss function of the extreme scenario generation model;

[0175] Input the power scenario with the classification label into the trained extreme scenario generation model, and output to obtain a synthetic sample of the power system scenario, and generate the extreme scenario of the power system according to the synthetic sample of the power system scenario.

[0176] Alternatively, when the processor executes the computer program, it realizes the functions of each module in the above system. For example:

[0177] The meteorological characteristic factor extraction module 1 is used to obtain historical meteorological samples and historical power scenario samples under extreme weather, and extract meteorological characteristic factors representing extreme weather based on the historical meteorological samples;

[0178] The clustering result classification module 2 is used to perform time series clustering on the meteorological characteristic factors based on the correlation and causality between the meteorological characteristic factors and the historical power scenario samples to obtain a clustering result, and use the clustering result as the classification label for the power scenario;

[0179] The scenario generation model training module 3 is used to construct an extreme scenario generation model, and train the extreme scenario generation model by improving the structure and loss function of the extreme scenario generation model;

[0180] The extreme scenario generation module 4 is used to input the power scenario with the classification label into the trained extreme scenario generation model, and output to obtain a synthetic sample of the power system scenario, and generate the extreme scenario of the power system according to the synthetic sample of the power system scenario.

[0181] Exemplarily, the computer program can be divided into one or more modules / units. The one or more modules / units are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units can be a series of computer program instruction segments capable of performing specific functions, and these instruction segments are used to describe the execution process of the computer program in the mobile terminal.

[0182] For example, the computer program can be divided into a meteorological feature factor extraction module 1, a clustering result classification module 2, a scenario generation model training module 3, and an extreme scenario generation module 4;

[0183] The specific functions of each module are as follows:

[0184] The meteorological feature factor extraction module 1 is used to obtain historical meteorological samples and historical power scenario samples under extreme weather, and extract meteorological feature factors characterizing extreme weather based on the historical meteorological samples;

[0185] The clustering result classification module 2 is used to perform time series clustering on the meteorological feature factors based on the correlation and causality between the meteorological feature factors and the historical power scenario samples to obtain a clustering result, and use the clustering result as the classification label of the power scenario;

[0186] The scenario generation model training module 3 is used to construct an extreme scenario generation model and train the extreme scenario generation model by improving the structure and loss function of the extreme scenario generation model;

[0187] The extreme scenario generation module 4 is used to input the power scenario with a classification label into the trained extreme scenario generation model and output a synthetic sample of the power system scenario, and generate an extreme scenario of the power system according to the synthetic sample of the power system scenario.

[0188] The mobile terminal can be a computing device such as a desktop computer, a notebook, a palm computer, and a cloud server. The mobile terminal may include, but is not limited to, a processor and a memory.

[0189] The processor can be a central processing unit (CPU), or can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The processor is the control center of the mobile terminal, and connects various parts of the entire mobile terminal through various interfaces and lines.

[0190] The memory can be used to store the computer program and / or module. By running or executing the computer program and / or module stored in the memory, and invoking the data stored in the memory, the processor realizes various functions of the mobile terminal.

[0191] The memory may mainly include a program storage area and a data storage area. Among them, the program storage area can store an operating system, application programs required for at least one function (such as a sound playback function, an image playback function, etc.); the data storage area can store data created according to the use of the mobile phone (such as audio data, phone book, etc.). In addition, the memory can include high-speed random access memory, and can also include non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, at least one magnetic disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0192] Embodiment 4

[0193] The present invention also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for generating an extreme scenario of a power system are realized.

[0194] If the modules / units integrated in the mobile terminal are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium.

[0195] Based on such an understanding, to implement all or part of the processes in the above method, the present invention can also be completed by instructing relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps of the above aggregation reinforcement learning resource scheduling method can be realized. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, an executable file, or some intermediate form, etc.

[0196] The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard disk, a magnetic disk, an optical disc, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0197] It should be noted that the content included in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice within the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0198] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: modifications or equivalent substitutions can still be made to the specific implementation manners of the present invention, and any modification or equivalent substitution that does not depart from the spirit and scope of the present invention should be covered by the protection scope of the claims of the present invention.

Claims

1. A method for generating extreme scenarios of a power system, characterized in that, Including: Obtain historical meteorological samples and historical power scenario samples under extreme weather, and extract meteorological characteristic factors representing extreme weather based on the historical meteorological samples; Based on the correlation and causality between the meteorological characteristic factors and the historical power scenario samples, perform time series clustering on the meteorological characteristic factors to obtain a clustering result, and use the clustering result as the classification label of the power scenario; Construct an extreme scenario generation model, and train the extreme scenario generation model by improving the structure and loss function of the extreme scenario generation model; Input the power scenario with the classification label into the trained extreme scenario generation model, and output to obtain a synthetic sample of the power system scenario, and generate the extreme scenario of the power system according to the synthetic sample of the power system scenario.

2. The method for generating an extreme scenario of a power system according to claim 1, wherein In the step of extracting meteorological characteristic factors representing extreme weather based on historical meteorological samples, use random forest to extract meteorological characteristic factors under extreme weather, including: Evaluate the importance of extreme weather meteorological characteristics according to the contribution of different extreme weather historical meteorological samples to each tree in the random forest; Rank the importance, and use the extreme weather meteorological characteristics with high ranking as the meteorological characteristic factors representing extreme weather.

3. A method for generating an extreme scenario of a power system according to claim 1, characterized in that, In the step of based on the correlation and causality between the meteorological characteristic factors and the historical power scenario samples, respectively test the stationarity of the time series and the cointegration relationship between time series for the meteorological characteristic factors under extreme weather and the historical power scenario samples through the unit root test and the cointegration test method, and obtain the correlation and causality between the meteorological characteristic factors under extreme weather and the historical power scenario samples through the Granger causality test based on the stationarity of the time series and the cointegration relationship between time series; Among them, in the unit root test, obtain the stationarity test result. If the statistical result rejects the null hypothesis, then the time series does not have a unit root, and the stationarity test result is that the variable time series has stationarity. On the contrary, the time series has a unit root, and the stationarity test result is that the variable time series is a non-stationary time series; According to the stationarity test result, when two variable time series are of the same order of integration, conduct a test through the cointegration test method, estimate the cointegration regression equation between the variables, and generate a residual series accordingly; conduct a stationarity test on the residual series. If the residual series is stationary, then there is a cointegration relationship between the variables, otherwise there is no cointegration relationship; Among them, when there is a cointegration relationship between variables, obtain the correlation and causality between the meteorological characteristic factors under extreme weather and the historical power scenario samples through the Granger causality test.

4. A method for generating an extreme scenario of a power system according to claim 1, characterized in that, In the step of performing time series clustering on the meteorological characteristic factors to obtain a clustering result, use K-MDTSC to calculate the distance of K-MDTSC with N as the dimension of the clustering sample and T as the time scale of the clustering sample based on the framework of the K-means clustering algorithm to obtain the clustering result, where the expression of the distance of K-MDTSC is as follows: Where L is the metric distance, X N , C N It is a multidimensional meteorological time series; Evaluate the clustering result through the within-cluster sum of squared errors to obtain the optimal number of clusters, and its expression is as follows: Among them, C is a cluster of clustering, which is the set of all clustering clusters.

5. A method for generating an extreme scenario of a power system according to claim 1, characterized in that In the step of training the extreme scenario generation model by improving the structure and loss function of the extreme scenario generation model, a WGAN-GP model is formed by improving the Wasserstein distance and optimizing the gradient penalty weight. Based on the WGAN-GP model, the operation constraints of the power system are introduced to form an extreme scenario generation model for the power system, and the extreme scenario generation model for the power system is trained under different gradient penalty weights.

6. A method for generating an extreme scenario of a power system according to claim 1, wherein In the step of inputting the power scenario with classification labels into the trained extreme scenario generation model and outputting the synthetic sample of the power system scenario, and in the step of generating the extreme scenario of the power system according to the synthetic sample of the power system scenario, the synthetic sample of the power system scenario includes the time series of wind power, photovoltaic power and load. Among them, the wind power, photovoltaic power and load in the synthetic sample of the same power system scenario form power intervals respectively. When the generated scenario is within the power interval, the extreme scenario generation work of the power system is completed.

7. A method for generating an extreme scenario of a power system according to claim 6, characterized in that, The generated scenarios are determined respectively by point accuracy A p and scenario accuracy A s wherein, the point accuracy A p is the ratio of the number of generated points within the effective range to the total number of generated points. When the time series lengths of each wind, light, and load power in the generated scenario are the same, the total number of generated points is fixed. The calculation formula of A p is as follows: where N pv,j is the number of generation points in the effective interval, j is the index of wind power, photovoltaic power, and load, and N s is the number of generation scenarios, and L is the length of each time series in the generation scenarios; Among them, the scenario accuracy A s is the ratio of the number of valid scenarios to the total number of generated scenarios. When the time series of wind power, photovoltaic, and load in the generated scenarios are all within the valid interval, the generated scenarios are valid. The calculation formula of A s is as follows: where N sv is the number of generated points in the effective interval.

8. A power system extreme scenario generation system for implementing the power system extreme scenario generation method according to any one of claims 1-7, characterized in that, It includes: A meteorological feature factor extraction module, which is used to obtain historical meteorological samples and historical power scenario samples under extreme weather, and extract meteorological feature factors representing extreme weather based on the historical meteorological samples; A clustering result classification module, which is used to perform time series clustering on the meteorological feature factors based on the correlation and causality between the meteorological feature factors and the historical power scenario samples to obtain clustering results, and use the clustering results as the classification labels of the power scenarios; A scenario generation model training module, which is used to construct an extreme scenario generation model and train the extreme scenario generation model by improving the structure and loss function of the extreme scenario generation model; An extreme scenario generation module, which is used to input the power scenario with classification labels into the trained extreme scenario generation model and output the synthetic sample of the power system scenario, and generate the extreme scenario of the power system according to the synthetic sample of the power system scenario.

9. A mobile terminal, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it realizes the steps of the power system extreme scenario generation method according to any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it realizes the steps of the power system extreme scenario generation method according to any one of claims 1-7.

Images