A wind and light scene generation method based on LHS-GRU

Abstract

A wind and solar scene generation method based on LHS-GRU includes the following steps: Step 1: Establish a wind and solar power output probability model based on historical wind and solar power data, and perform Latin hypercube sampling based on the obtained wind and solar power output probability model to obtain an initial wind and solar scene set; Step 2: Learn the power output correlation between wind power and photovoltaic at various times through a neural network model, and establish correlation constraint models for wind power and photovoltaic respectively; Step 3: Import the initial scene set into the model to generate a corrected scene set, i.e., a corrected scene set. The purpose of this invention is to solve the technical problems in existing wind and solar scene generation technologies, where statistical methods are difficult to accurately represent the correlation between short-term data, resulting in the power points in the generated scene set not conforming to the actual power output, and the technical problems in using time series methods to obtain atypical power output scenes, resulting in the difficulty in obtaining atypical scenes. This method is provided to address these issues.

Description

Technical Field

[0001] This invention relates to the field of power system technology, and more particularly to renewable energy grid connection technology, specifically to a method for generating wind and solar scenarios based on LHS-GRU. Background Technology

[0002] Although a large number of research papers on the prediction of renewable energy output have emerged in recent years, very few people have studied a universal algorithm to improve prediction accuracy. The large-scale grid connection of renewable energy will bring a great impact to the power grid, and conventional deterministic optimization scheduling methods are no longer able to adapt to the large-scale grid connection of renewable energy.

[0003] Scenario analysis is a method suitable for handling the uncertainty of new energy sources. By analyzing historical data on new energy output, a probabilistic model of output is established, generating a large set of stochastic scenarios that conform to the historical data. This set is then applied to robust and stochastic optimization of power grid dispatch. Scenario analysis transforms the uncertainty of new energy output into a variety of deterministic scenario sets for calculation. For this reason, scenario analysis has been widely applied and developed in uncertainty analysis.

[0004] There are two main methods for generating scenario sets in scenario analysis: statistical methods and time series methods. Statistical methods involve probabilistically modeling historical power output data of new energy sources to obtain the model's probability density function, then randomly sampling the model and combining the sampling results to form a scenario set. Time series methods primarily rely on historical power output data, analyzing historical time series data to characterize the changing trends of the time series data, and generating future scenario sets through the analysis of these trends.

[0005] In existing technologies, the statistical method used in "Nonparametric Kernel Density Estimation Model of Photovoltaic Power Output" provided by Yan Wei, Ren Zhouyang, Zhao Xia, et al. has a good description of the distribution patterns of long-term and large-scale data, and also covers the output distribution description under extreme weather conditions, but lacks a description of the correlation between short-term data (between intraday data). The time series method used in "Two-way Optimization Technology for Generating Wind Power Time Series Scenarios" provided by Li Jinghua, Sun Haishun, Wen Jinyu, et al. can reflect the correlation between short-term data by analyzing the trends and relationships between historical data. However, since the amount of typical daily data in historical data is much larger than the amount of extreme daily data, the data generated by the time series method can fit the typical daily situation well, but it is difficult to accurately fit the extreme daily situation. Summary of the Invention

[0006] The object of the present invention is to provide a wind-solar scene generation method based on LHS-GRU to solve the technical problems in the existing wind-solar scene generation technology that it is difficult to accurately represent the correlation between short-term data by using statistical methods, resulting in the power points between different moments in the generated scene set not conforming to the actual output situation, and it is difficult to obtain atypical output scenarios by using time series methods, resulting in the difficulty of obtaining atypical scenarios.

[0007] To solve the above technical problems, the technical solution adopted by the present invention is as follows:

[0008] A wind-solar scene generation method based on LHS-GRU includes the following steps:

[0009] Step 1: Establish a wind-solar output probability model based on historical wind-solar data, and perform Latin hypercube sampling according to the obtained wind-solar output probability model to obtain an initial wind-solar scene set;

[0010] Step 2: Respectively learn the output correlation between each moment of wind power and photovoltaic power through a neural network model, and respectively establish correlation limit models for wind power and photovoltaic power;

[0011] Step 3: Import the initial scene set into the model to generate a corrected scene set, that is, the corrected scene set.

[0012] In Step 1, it specifically includes the following sub-steps:

[0013] Step 1-1: Build a wind-solar output probability model;

[0014] Step 1-2: Determine the bandwidth of the non-parametric kernel density estimation of wind-solar;

[0015] Step 1-3: Generate a wind-solar scene set based on Latin hypercube sampling.

[0016] In Step 1-1, specifically:

[0017] Suppose the output situation P of new energy has a probability density function f(p), then let F(p) be the distribution function of the overall new energy output P, and its samples P1, P2,..., P n The empirical distribution function is:

[0018]

[0019] where n refers to the number of samples; P i refers to the sample data; I(P i <p) refers to the indicator function of the set P;

[0020]

[0021] h refers to the window width, therefore, from equations (12) and (13), the natural estimate of the density function f(p) can be obtained as:

[0022]

[0023] Approximately:

[0024]

[0025] Let K(·) = 2 -1 I(·), and thus we have:

[0026]

[0027] Equation (5) is the probability model for wind and solar power output, and can also be called the kernel density estimate of f(p), where This refers to sample P. i The increase in the density function when it falls within the interval (ph, p+h), and K(·) is called its corresponding weight function, or kernel function. n This refers to the window width when there are n samples.

[0028] In steps 1-2, specifically:

[0029] First, using the integral squared error (ISE) as the loss function, the expression is:

[0030]

[0031] For equation (17), consider with With the relevant terms, equation (17) can be transformed into:

[0032]

[0033] in, This refers to the estimate after removing the i-th data point. Specifically:

[0034]

[0035] To simplify the calculation, approximate (n-1) in equation (19) as n, which yields a loss function equivalent to equation (17):

[0036]

[0037] Its optimal bandwidth is represented by h. opt =argminM1(h n ).

[0038] In steps 1-3, specifically:

[0039] (1) First, for each time t of the wind and solar power output, where t = 1, 2, ..., 96, kernel density estimation is performed to obtain the power output probability density estimation function f at time t. t (p);

[0040] (2) Calculate the cumulative probability density function F based on the kernel density estimation function obtained at each time step. t (P), and simultaneously the cumulative probability density function F t (P) is divided into N t There are N equal and non-overlapping intervals, where the length of each interval is 1 / N. t ;

[0041] (3) Based on the obtained probability sub-intervals, perform a random sampling once in each probability sub-interval, that is:

[0042]

[0043] Where i refers to the i-th subinterval, i = 1, 2, ..., N t ;f t,i This refers to the cumulative probability density of the i-th subinterval at time t; random refers to a random number between 0 and 1.

[0044] (4) Substitute the obtained value into its inverse function to obtain the Latin hypercube sampled value P of the scenery in the i-th subinterval at time t. t,i , can be represented as:

[0045]

[0046] in, It refers to the inverse function of the cumulative probability density function at time t;

[0047] One LHS sampling can be completed through steps (1)-(4);

[0048] The above methods are used to sample the probability density functions of wind power output and photovoltaic power output at each moment to generate data for the initial wind power scenario set and the initial photovoltaic scenario set.

[0049] In step 2, the established correction model includes:

[0050] Establish a GRU-corrected model for wind power output;

[0051] Establish a GRU modified model for photovoltaic power output.

[0052] When establishing the GRU modified model for wind power output, the following are included:

[0053] (1) Divide the wind power output sample data W horizontally from left to right along the timeline. Each sample data has k power points. Thus, it can be divided into wind power output sample data W(t,t+k / 16), wind power output sample data W(t,t+k / 8), wind power output sample data W(t,t+k / 4), wind power output sample data W(t,t+k / 2), and wind power output sample data W(t,t+k), which are the training sets of five correction models: wind power correction model GRU3-3, wind power correction model GRU6-6, wind power correction model GRU12-12, wind power correction model GRU24-24, and wind power correction model GRU48-48.

[0054] (2) Divide the wind power output sample data W(t,t+k / 16) into W(t,t+k / 32) and W(t+k / 32+1,t+k / 16) as the input and output of the wind power correction model GRU3-3 for training;

[0055] (3) The wind power output sample data W(t,t+k / 8) is divided into wind power output sample data W(t,t+k / 16) and wind power output sample data W(t+k / 16+1,t+k / 8) as the input and output of the wind power correction model GRU6-6, respectively, for training.

[0056] (4) The wind power output sample data W(t,t+k / 4) is divided into wind power output sample data W(t,t+k / 8) and wind power output sample data W(t+k / 8+1,t+k / 4) as the input and output of the wind power correction model GRU12-12, respectively, for training.

[0057] (5) The wind power output sample data W(t,t+k / 2) is divided into wind power output sample data W(t,t+k / 4) and wind power output sample data W(t+k / 4+1,t+k / 2) as the input and output of the wind power correction model GRU24-24, respectively, for training.

[0058] (6) Divide the wind power output sample data W(t,t+k) into wind power output sample data W(t,t+k / 2) and wind power output sample data W(t+k / 2+1,t+k), respectively, and use them as the input and output of the wind power correction model GRU48-48 for training.

[0059] When establishing the GRU correction model for photovoltaic output, the following are included:

[0060] (1) Divide the photovoltaic power output sample data V horizontally from the middle of the time line to both sides. Each sample data has k power points. Therefore, the sample data can be divided into photovoltaic power output sample data V(l-7*k / 96,l), photovoltaic power output sample data V(l-3*k / 96,l+4*k / 96), photovoltaic power output sample data V(l-19*k / 96,l+4*k / 96), and photovoltaic power output sample data V(l-7*k / 96,l+16*k / 96). The power output sample data V(l-33*k / 96,l+16*k / 96) and the photovoltaic power output sample data V(l-19*k / 96,l+30*k / 96) are the training sets of six modified photovoltaic models: GRU4-4L, GRU4-4R, GRU12-12L, GRU12-12R, GRU36-14L, and GRU36-14R.

[0061] (2) The photovoltaic power output sample data V(l-7*k / 96,l) can be divided into photovoltaic power output sample data V(l-3*k / 96,l) and photovoltaic power output sample data V(l-7*k / 96,l-4*k / 96), which are used as the input and output of the photovoltaic correction model GRU4-4L for training, respectively.

[0062] (3) The photovoltaic power output sample data V(l-3*k / 96,l+4*k / 96) can be divided into photovoltaic power output sample data V(l-3*k / 96,l) and photovoltaic power output sample data V(l+*k / 96,l+4*k / 96), which are used as the input and output of the photovoltaic correction model GRU4-4R respectively for training.

[0063] (4) The photovoltaic power output sample data V(l-19*k / 96,l+4*k / 96) can be divided into photovoltaic power output sample data V(l-7*k / 96,l+4*k / 96) and photovoltaic power output sample data V(l-19*k / 96,l-8*k / 96) as input and output of the photovoltaic correction model GRU12-12L for training;

[0064] (5) The photovoltaic power output sample data V(l-7*k / 96,l+16*k / 96) can be divided into photovoltaic power output sample data V(l-7*k / 96,l+4*k / 96) and photovoltaic power output sample data V(l+5*k / 96,l+16*k / 96) as input and output of the photovoltaic correction model GRU12-12R for training;

[0065] (6) The photovoltaic power output sample data V(l-33*k / 96,l+16*k / 96) can be divided into photovoltaic power output sample data V(l-19*k / 96,l+16*k / 96) and photovoltaic power output sample data V(l-33*k / 96,l-18*k / 96) as input and output of the photovoltaic correction model GRU36-14L for training;

[0066] (7) The photovoltaic power output sample data V(l-19*k / 96,l+30*k / 96) can be divided into photovoltaic power output sample data V(l-19*k / 96,l+16*k / 96) and photovoltaic power output sample data V(l+17*k / 96,l+30*k / 96) as input and output of the photovoltaic correction model GRU36-14R for training.

[0067] Step 3 specifically includes the following sub-steps:

[0068] Step 3-1: The trained wind power correction models GRU3-3, GRU6-6, GRU12-12, GRU24-24, and GRU48-48 from the established wind power output GRU correction model are spliced ​​together in chronological order to form a complete wind power correction model. The initial wind power scenario set obtained in Step 1 is then imported into the wind power correction model.

[0069] Step 3-2: The photovoltaic correction models GRU4-4L, GRU4-4R, GRU12-12L, GRU12-12R, GRU36-14L, and GRU36-14R trained in Step 2 are spliced ​​together in an expansion pattern from the middle to both sides to form a complete photovoltaic correction model. The initial photovoltaic scene set obtained in Step 1 is then imported into the photovoltaic correction model.

[0070] Step 3-1 includes the following steps:

[0071] Step 3-1-1) Assemble the wind power correction model in chronological order into a new wind power correction model in the form of GRU3-3-GRU6-6-GRU12-12-GRU24-24-GRU48-48;

[0072] Step 3-1-2) The wind power correction model GRU3-3 generates values ​​within ±m% of the values ​​of the first 3a time steps. It then selects appropriate values ​​from the corresponding time steps in the initial wind power scenario set without replacement as the wind power output value for the corresponding time step. If there are no values ​​within ±m% that meet the conditions, the range is expanded to ±2m%. If there are no values ​​within ±2m% that meet the conditions, the range is expanded to ±4m%. If there are no values ​​within ±4m% that meet the conditions, a random selection with replacement is performed from the initial wind power scenario set, where a is a positive integer and m is a real number.

[0073] Step 3-1-3) The wind power correction model GRU6-6 can generate values ​​within ±m% of the values ​​of the first 6a time points from the values ​​of the first 6a time points. It selects appropriate values ​​from the corresponding time points in the initial wind power scenario set without replacement as the wind power output values ​​for the corresponding time points. If there are no values ​​that meet the conditions within ±5%, the range is expanded to ±2m%. If there are no values ​​that meet the conditions within ±2m%, the range is expanded to ±4m%. If there are no values ​​that meet the conditions within ±4m%, random selection with replacement is performed from the initial wind power scenario set.

[0074] Step 3-1-4) The wind power correction model GRU12-12 can generate values ​​within ±m% of the values ​​of the previous 12a time points from the values ​​of the previous 12a time points. It selects appropriate values ​​from the corresponding time points in the initial wind power scenario set without replacement as the wind power output values ​​for the corresponding time points. If there are no values ​​that meet the conditions within ±m%, the range is expanded to ±2m%. If there are no values ​​that meet the conditions within ±2m%, the range is expanded to ±4m%. If there are no values ​​that meet the conditions within ±4m%, random selection with replacement is performed from the initial wind power scenario set.

[0075] 3-1-5) The wind power correction model GRU24-24 can generate values ​​within ±m% of the values ​​of the previous 24a time points from the values ​​of the previous 24a time points. It selects appropriate values ​​from the corresponding time points in the initial wind power scenario set without replacement as the wind power output values ​​for the corresponding time points. If there are no values ​​that meet the conditions within ±m%, the range is expanded to ±2m%. If there are no values ​​that meet the conditions within ±2m%, the range is expanded to ±4m%. If there are no values ​​that meet the conditions within ±4m%, random selection with replacement is performed from the initial wind power scenario set.

[0076] 3-1-6) The wind power correction model GRU48-48 can generate values ​​within ±m% of the values ​​of the first 48a time points from the values ​​of the first 48a time points. It selects appropriate values ​​from the corresponding time points in the initial wind power scenario set without replacement as the wind power output values ​​for the corresponding time points. If there are no values ​​that meet the conditions within ±m%, the range is expanded to ±2m%. If there are no values ​​that meet the conditions within ±2m%, the range is expanded to ±4m%. If there are no values ​​that meet the conditions within ±4m%, random selection with replacement is performed from the initial wind power scenario set.

[0077] The corrected wind power scenario set can be obtained by following the above steps.

[0078] Step 3-2 includes the following steps:

[0079] Step 3-2-1) Assemble the photovoltaic correction model in chronological order into a new photovoltaic correction model in the form of GRU36-14L-GRU12-12L-GRU4-4L-GRU4-4R-GRU12-12R-GRU36-14R;

[0080] Step 3-2-2) The photovoltaic correction model GRU4-4L can generate the values ​​of the left 4b times within ±m% from the values ​​of the middle 4b times, and select appropriate values ​​from the corresponding times in the photovoltaic initial scenario set in a non-replacement manner as the photovoltaic output value of the corresponding times. When there are no values ​​that meet the conditions within the ±m% range, the range is expanded to ±2m%. When there are no values ​​that meet the conditions within the ±2m% range, the range is expanded to ±4m%. When there are no values ​​that meet the conditions within the ±4m% range, random selection with replacement is performed from the photovoltaic initial scenario set.

[0081] Step 3-2-3) The photovoltaic correction model GRU4-4R can generate the values ​​of the right 4b times within ±m% from the values ​​of the middle 4b times, and select appropriate values ​​from the corresponding times in the photovoltaic initial scenario set in a non-replacement manner as the photovoltaic output value of the corresponding times. When there are no values ​​that meet the conditions within the ±m% range, the range is expanded to ±2m%. When there are no values ​​that meet the conditions within the ±2m% range, the range is expanded to ±4m%. When there are no values ​​that meet the conditions within the ±4m% range, random selection with replacement is performed from the photovoltaic initial scenario set.

[0082] Step 3-2-4) The photovoltaic correction model GRU12-12L can generate the values ​​of the left 12b times within ±m% from the values ​​of the middle 12b times, and select appropriate values ​​from the corresponding times in the photovoltaic initial scenario set in a non-replacement manner as the photovoltaic output value of the corresponding times. When there are no values ​​that meet the conditions within the ±m% range, the range is expanded to ±2m%. When there are no values ​​that meet the conditions within the ±2m% range, the range is expanded to ±4m%. When there are no values ​​that meet the conditions within the ±4m% range, random selection with replacement is performed from the photovoltaic initial scenario set.

[0083] Step 3-2-5) The photovoltaic correction model GRU12-12R can generate the values ​​of the right 12b times within ±m% from the values ​​of the middle 12b times. It selects appropriate values ​​from the corresponding times in the photovoltaic initial scenario set without replacement as the photovoltaic output value for the corresponding times. If there are no values ​​that meet the conditions within ±m%, the range is expanded to ±2m%. If there are no values ​​that meet the conditions within ±2m%, the range is expanded to ±4m%. If there are no values ​​that meet the conditions within ±4m%, random selection with replacement is performed from the photovoltaic initial scenario set.

[0084] Step 3-2-6) The photovoltaic correction model GRU36-14L can generate the values ​​of the left 14b times within ±m% from the values ​​of the middle 36b times. It selects appropriate values ​​from the corresponding times in the photovoltaic initial scenario set in a non-replacement manner as the photovoltaic output value of the corresponding times. If there are no values ​​that meet the conditions within the ±m% range, the range is expanded to ±2m%. If there are no values ​​that meet the conditions within the ±2m% range, the range is expanded to ±4m%. If there are no values ​​that meet the conditions within the ±4m% range, a random selection with replacement is performed from the photovoltaic initial scenario set.

[0085] Step 3-2-7) The photovoltaic correction model GRU36-14R can generate the values ​​of the right 14b times within ±m% from the values ​​of the middle 36b times. It selects appropriate values ​​from the corresponding times in the photovoltaic initial scenario set without replacement as the photovoltaic output value for the corresponding times. If there are no values ​​that meet the conditions within ±m%, the range is expanded to ±2m%. If there are no values ​​that meet the conditions within ±2m%, the range is expanded to ±4m%. If there are no values ​​that meet the conditions within ±4m%, random selection with replacement is performed from the photovoltaic initial scenario set.

[0086] The above steps will yield the corrected photovoltaic scenario set.

[0087] Compared with the prior art, the present invention has the following technical effects:

[0088] 1) In order to accurately reflect the power output of new energy on typical days and extreme days, this invention proposes a method that can generate typical day scenarios and extreme day scenarios at the same time. While ensuring the superiority of statistical methods and time series methods, it avoids the shortcomings of both, so that the generated scenario set can accurately reflect various situations in the dispatch of power systems containing wind and solar power, and better promote the power system's consumption of new energy.

[0089] 2) This invention models the cumulative probability density of wind and solar power output data at 96 moments (each moment is 15 minutes) over the past year using kernel density estimation, and forms 96 cumulative probability density functions. This method can reconstruct the power output probability density model at 96 moments over the year, ensuring that the model can simultaneously include data from typical days and extreme days, thereby improving the realism of the constructed model.

[0090] 3) This invention achieves the fitting of the trend relationship between 96 time points by constructing and training a gated recurrent unit (GRU) model, finding the best match between adjacent time points, enhancing the correlation between adjacent time points, and thus ensuring that the correlation between adjacent time points conforms to the real wind and solar power output curve. Attached Figure Description

[0091] The present invention will be further described below with reference to the accompanying drawings and embodiments:

[0092] Figure 1 This is a flowchart of the present invention;

[0093] Figure 2 Here is the GRU forward propagation graph at time t;

[0094] Figure 3 This is a schematic diagram of an extended form of the GRU neural network;

[0095] Figure 4 This is a schematic diagram illustrating the training direction of the GRU model for wind power.

[0096] Figure 5 A schematic diagram illustrating the training direction of the GRU model for photovoltaics;

[0097] Figure 6 This is a diagram showing the comparison of the BS index obtained by comparing the wind power scenario set generated under Event 1 with the real scenario set in 2022.

[0098] Figure 7 This is a diagram illustrating the comparison of the BS index between the wind power scenario set generated under Event 1 and the real scenario set in 2023.

[0099] Figure 8This is a diagram illustrating the comparison of the BS index between the wind power scenario set generated under Event 2 and the real scenario set in 2022.

[0100] Figure 9 This is a diagram illustrating the comparison of the BS index between the wind power scenario set generated under Event 2 and the real scenario set in 2023.

[0101] Figure 10 This is a diagram illustrating the comparison of the BS index between the photovoltaic scenario set generated under Event 1 and the real scenario set in 2022.

[0102] Figure 11 This is a diagram showing the comparison of the BS index between the photovoltaic scene set generated under Event 1 and the real scene set in 2023.

[0103] Figure 12 This is a diagram showing the comparison of the BS index between the photovoltaic scenario set generated under Event 2 and the real scenario set in 2022.

[0104] Figure 13 This is a diagram showing the comparison of the BS index between the photovoltaic scenario set generated under Event 2 and the real scenario set in 2023.

[0105] Figure 14 This is a diagram showing the comparison of the BS index obtained by comparing the wind power scenario set generated under Event 3 with the real scenario set in 2022.

[0106] Figure 15 This is a diagram showing the comparison of the BS index between the wind power scenario set generated under Event 3 and the real scenario set in 2023.

[0107] Figure 16 This is a diagram showing the comparison of the BS index obtained by comparing the wind power scenario set generated under Event 4 with the real scenario set in 2022.

[0108] Figure 17 This is a diagram showing the comparison of the BS index obtained by comparing the wind power scenario set generated under Event 4 with the real scenario set in 2023.

[0109] Figure 18 This is a diagram illustrating the comparison of the BS index between the wind power scenario set generated under Event 5 and the real scenario set in 2022.

[0110] Figure 19 This is a diagram illustrating the comparison of the BS index between the wind power scenario set generated under Event 5 and the real scenario set in 2023.

[0111] Figure 20 This is a diagram showing the comparison of the BS index between the photovoltaic scene set generated under Event 4 and the real scene set in 2022.

[0112] Figure 21 Schematic diagram of BS index comparison obtained by comparing the photovoltaic scenario set generated under Event 4 with the real scenario set in 2023;

[0113] Figure 22 Schematic diagram of BS index comparison obtained by comparing the photovoltaic scenario set generated under Event 5 with the real scenario set in 2022;

[0114] Figure 23 Schematic diagram of BS index comparison obtained by comparing the photovoltaic scenario set generated under Event 5 with the real scenario set in 2023. Detailed implementation method

[0115] A method for generating wind-solar scenarios based on LHS-GRU includes the following steps:

[0116] Step 1: Establish a wind-solar output probability model based on historical wind-solar data, and perform Latin hypercube sampling according to the obtained wind-solar output probability model to obtain an initial wind-solar scenario set;

[0117] Step 2: Use a neural network model to learn the output correlation between each moment of wind power and photovoltaic power respectively, and establish a correlation limitation model for wind power and photovoltaic power respectively;

[0118] Step 3: Import the initial scenario set into the model to generate a corrected scenario set, that is, the corrected scenario set.

[0119] In Step 1, it specifically includes the following sub-steps:

[0120] Step 1-1: Model the wind-solar output probability model;

[0121] Step 1-2: Determine the bandwidth of the non-parametric kernel density estimation of wind-solar;

[0122] Step 1-3: Generate a wind-solar scenario set based on Latin hypercube sampling.

[0123] In Step 1-1, specifically:

[0124] Assume that the output situation P of new energy has a probability density function f(p), then assume that F(p) is the distribution function of the overall new energy output P, and its samples P1, P2,..., P n The empirical distribution function of is:

[0125]

[0126] where n refers to the number of samples; P i refers to the sample data; I(P i <p) refers to the indicator function of the set P;

[0127]

[0128] h refers to the window width, therefore, from equations (12) and (13), the natural estimate of the density function f(p) can be obtained as:

[0129]

[0130] Approximately:

[0131]

[0132] Let K(·) = 2 -1 I(·), and thus we have:

[0133]

[0134] Equation (16) is the probability model for wind and solar power output, and can also be called the kernel density estimate of f(p), where This refers to sample P. i The increase in the density function when it falls within the interval (ph, p+h), and K(·) is called its corresponding weight function, or kernel function. n This refers to the window width for n samples.

[0135] In steps 1-2, specifically:

[0136] First, using the integral squared error (ISE) as the loss function, the expression is:

[0137]

[0138] For equation (17), consider with With the relevant terms, equation (17) can be transformed into:

[0139]

[0140] in, This refers to the estimate after removing the i-th data point. Specifically:

[0141]

[0142] To simplify the calculation, approximate (n-1) in equation (19) as n, which yields a loss function equivalent to equation (17):

[0143]

[0144] Its optimal bandwidth is represented by h. opt=argminM1(h n ).

[0145] In kernel density estimation, it is necessary to select a kernel function and determine the optimal bandwidth, with the optimal bandwidth determining the fitting effect of the kernel density estimation. If the bandwidth is too large, the fitted curve will be too smooth and unable to reflect the detailed features of the data; if the bandwidth is too small, the fitted curve will contain too much noise and also fail to accurately reflect the features of the data. Methods for determining the optimal bandwidth mainly include the reference standard distribution method, the thumb rule, and cross-validation. Among these, cross-validation can effectively avoid the problems of overfitting or underfitting the probability density function in kernel density estimation. Therefore, this invention uses the cross-validation method to determine the optimal bandwidth.

[0146] In steps 1-3, specifically:

[0147] (1) First, for each time t of the wind and solar power output, where t = 1, 2, ..., 96, perform kernel density estimation to obtain f. t (p);

[0148] (2) Calculate the cumulative probability density function F based on the kernel density estimation function obtained at each time step. t (P), and simultaneously the cumulative probability density function F t (P) is divided into N t There are N equal and non-overlapping intervals, where the length of each interval is 1 / N. t ;

[0149] (3) Based on the obtained probability sub-intervals, perform a random sampling once in each probability sub-interval, that is:

[0150]

[0151] Where i refers to the i-th subinterval, i = 1, 2, ..., N t ;f t,i This refers to the cumulative probability density of the i-th subinterval at time t; random refers to a random number between 0 and 1.

[0152] (4) Substitute the obtained value into its inverse function to obtain the Latin hypercube sampled value P of the scenery in the i-th subinterval at time t. t,i , can be represented as:

[0153]

[0154] in, It refers to the inverse function of the cumulative probability density function at time t;

[0155] One LHS sampling can be completed through steps (1)-(4);

[0156] The above methods are used to sample the probability density functions of wind power output and photovoltaic power output at each moment to generate data for the initial wind power scenario set and the initial photovoltaic scenario set.

[0157] Latin Hypercube Sampling (LHS) evolved from Monte Carlo sampling. Monte Carlo sampling is a method of randomly sampling from a probability distribution. This method can generate a large number of random samples that conform to the sample distribution. However, its drawbacks include the potential for clustering of sampling results within a given distribution range, and the inaccurate representation of low-probability outcomes when the number of samples is small. In wind and solar power applications, however, some low-probability scenarios, i.e., extreme scenarios, are extremely important in power systems. LHS can reflect low-probability outcomes, thus ensuring that the sampling results simultaneously include both low-probability and high-probability results. LHS uses the concept of hierarchical fractal dimension to divide the cumulative probability density from 0 to 1.0 into multiple non-overlapping sub-intervals, and then uniformly and randomly samples from different sub-intervals.

[0158] In the above steps, kernel density estimation is used to probabilistically model the historical data of the wind and solar power samples at each moment. The probability density function at each moment is sampled using LHS to generate initial scene data for the wind and solar power. However, these generated data are not correlated at any given moment. Randomly combining these data leads to weak correlations, resulting in significant deviations between the generated scene data and the actual scene data, thus failing to reflect the true wind and solar power output. For this reason, this paper uses the Gate Recurrent Unit (GRU) model from recurrent neural networks (RNN) to correct the generated initial scene set, thereby enhancing the correlation between different moments and ensuring that the final generated scene set reflects the true wind and solar power output.

[0159] GRU is a type of RNN. GRU combines the forget gate and input gate into an update gate. The structure of GRU is simpler than that of Long Short-Term Memory networks, but its performance is almost the same as that of LSTM, and it is also relatively easy to train.

[0160] like Figure 2 As shown, r t This represents the reset gate. The function of the reset gate is to combine the input at the current time step with the input at the previous time step. Its expression is:

[0161] r t =σ(W r ·[ht-1 ,x t ])

[0162] z t This represents the update gate, which controls how much information from the current time step and previous time steps should be retained for the next time step. Its expression is:

[0163] z t =σ(W z ·[h t-1 ,x t ])

[0164] The candidate hidden state at time t is represented by the expression:

[0165]

[0166] Among them, h t-1 This represents the hidden state at time t-1, and also contains all the data information before time t, thus acting as the memory of the neural network; h t The hidden state that needs to be passed to time t+1 is expressed as follows:

[0167]

[0168] GRU can be scaled horizontally based on the length of the time series, and also vertically, forming deep GRU networks, such as... Figure 3 As shown, increasing the number of layers in the network can further enhance the GRU's ability to fit the data.

[0169] In scene generation, the scenes are usually randomly combined after generation to form a new scene set. However, this method has a large error, which makes the generated scene set unable to accurately reflect the real wind and solar power output scene, thus making the generated scene unsuitable for the optimized scheduling of the power grid.

[0170] This invention, based on the principle of GRU (Gross Root Relationship), learns the correlations between different moments in a landscape scene from historical data, thereby correcting the generated initial landscape scene set to a set that conforms to historical expectations. This invention generates a complete landscape power output scene set step-by-step by training multiple GRU neural network models. For wind power, firstly, the GRU model is trained using the wind power of the first six moments (00:00-1:15), that is, the possible values ​​of the wind power in the last three moments are limited based on the wind power of the first three moments. Secondly, the GRU model is trained using the wind power of the first twelve moments (00:00-3:00), that is, the values ​​of the wind power in the last six moments are limited based on the wind power of the first six moments, and so on, ultimately training five GRU models for the recombination of the initial wind power scene set. The training direction is as follows: Figure 4 As shown.

[0171] For photovoltaic (PV) power output scenarios, PV output has boundaries—specifically, it doesn't output power at night. Therefore, there are boundary conditions for PV output throughout the day. However, these boundaries vary over time. A common approach is to define a fixed, artificially imposed boundary, but this doesn't align with natural laws. Therefore, this paper expands the PV scenario generation process outwards from 11:30-12:15, using a GRU model to learn from historical data and analyze trends, thus obtaining a PV boundary condition that conforms to natural laws. The training direction is as follows... Figure 5 As shown.

[0172] In step 2, the established correction model includes:

[0173] Establish a GRU-corrected model for wind power output;

[0174] Establish a GRU modified model for photovoltaic power output.

[0175] When establishing the GRU modified model for wind power output, the following are included:

[0176] (1) Divide the wind power output sample data W horizontally from left to right along the timeline. Each sample data has k power points. Thus, it can be divided into wind power output sample data W(t,t+k / 16), wind power output sample data W(t,t+k / 8), wind power output sample data W(t,t+k / 4), wind power output sample data W(t,t+k / 2), and wind power output sample data W(t,t+k), which are the training sets of five correction models: wind power correction model GRU3-3, wind power correction model GRU6-6, wind power correction model GRU12-12, wind power correction model GRU24-24, and wind power correction model GRU48-48.

[0177] (2) Divide the wind power output sample data W(t,t+k / 16) into W(t,t+k / 32) and W(t+k / 32+1,t+k / 16) as the input and output of the wind power correction model GRU3-3 for training;

[0178] (3) The wind power output sample data W(t,t+k / 8) is divided into wind power output sample data W(t,t+k / 16) and wind power output sample data W(t+k / 16+1,t+k / 8) as the input and output of the wind power correction model GRU6-6, respectively, for training.

[0179] (4) The wind power output sample data W(t,t+k / 4) is divided into wind power output sample data W(t,t+k / 8) and wind power output sample data W(t+k / 8+1,t+k / 4) as the input and output of the wind power correction model GRU12-12, respectively, for training.

[0180] (5) The wind power output sample data W(t,t+k / 2) is divided into wind power output sample data W(t,t+k / 4) and wind power output sample data W(t+k / 4+1,t+k / 2) as the input and output of the wind power correction model GRU24-24, respectively, for training.

[0181] (6) Divide the wind power output sample data W(t,t+k) into wind power output sample data W(t,t+k / 2) and wind power output sample data W(t+k / 2+1,t+k), respectively, and use them as the input and output of the wind power correction model GRU48-48 for training.

[0182] When establishing the GRU correction model for photovoltaic output, the following are included:

[0183] (1) Divide the photovoltaic power output sample data V horizontally from the middle of the time line to both sides. Each sample data has k power points. Therefore, the sample data can be divided into photovoltaic power output sample data V(l-7*k / 96,l), photovoltaic power output sample data V(l-3*k / 96,l+4*k / 96), photovoltaic power output sample data V(l-19*k / 96,l+4*k / 96), and photovoltaic power output sample data V(l-7*k / 96,l+16*k / 96). The power output sample data V(l-33*k / 96,l+16*k / 96) and the photovoltaic power output sample data V(l-19*k / 96,l+30*k / 96) are the training sets of six modified photovoltaic models: GRU4-4L, GRU4-4R, GRU12-12L, GRU12-12R, GRU36-14L, and GRU36-14R.

[0184] (2) The photovoltaic power output sample data V(l-7*k / 96,l) can be divided into photovoltaic power output sample data V(l-3*k / 96,l) and photovoltaic power output sample data V(l-7*k / 96,l-4*k / 96), which are used as the input and output of the photovoltaic correction model GRU4-4L for training, respectively.

[0185] (3) The photovoltaic power output sample data V(l-3*k / 96,l+4*k / 96) can be divided into photovoltaic power output sample data V(l-3*k / 96,l) and photovoltaic power output sample data V(l+*k / 96,l+4*k / 96), which are used as the input and output of the photovoltaic correction model GRU4-4R respectively for training.

[0186] (4) The photovoltaic power output sample data V(l-19*k / 96,l+4*k / 96) can be divided into photovoltaic power output sample data V(l-7*k / 96,l+4*k / 96) and photovoltaic power output sample data V(l-19*k / 96,l-8*k / 96) as input and output of the photovoltaic correction model GRU12-12L for training;

[0187] (5) The photovoltaic power output sample data V(l-7*k / 96,l+16*k / 96) can be divided into photovoltaic power output sample data V(l-7*k / 96,l+4*k / 96) and photovoltaic power output sample data V(l+5*k / 96,l+16*k / 96) as input and output of the photovoltaic correction model GRU12-12R for training;

[0188] (6) The photovoltaic power output sample data V(l-33*k / 96,l+16*k / 96) can be divided into photovoltaic power output sample data V(l-19*k / 96,l+16*k / 96) and photovoltaic power output sample data V(l-33*k / 96,l-18*k / 96) as input and output of the photovoltaic correction model GRU36-14L for training;

[0189] (7) The photovoltaic power output sample data V(l-19*k / 96,l+30*k / 96) can be divided into photovoltaic power output sample data V(l-19*k / 96,l+16*k / 96) and photovoltaic power output sample data V(l+17*k / 96,l+30*k / 96) as input and output of the photovoltaic correction model GRU36-14R for training.

[0190] Step 3 specifically includes the following sub-steps:

[0191] Step 3-1: The trained wind power correction models GRU3-3, GRU6-6, GRU12-12, GRU24-24, and GRU48-48 from the established wind power output GRU correction model are spliced ​​together in chronological order to form a complete wind power correction model. The initial wind power scenario set obtained in Step 1 is then imported into the wind power correction model.

[0192] Step 3-2: The photovoltaic correction models GRU4-4L, GRU4-4R, GRU12-12L, GRU12-12R, GRU36-14L, and GRU36-14R trained in Step 2 are spliced ​​together in an expansion pattern from the middle to both sides to form a complete photovoltaic correction model. The initial photovoltaic scene set obtained in Step 1 is then imported into the photovoltaic correction model.

[0193] Step 3-1 includes the following steps:

[0194] Step 3-1-1) Assemble the wind power correction model in chronological order into a new wind power correction model in the form of GRU3-3-GRU6-6-GRU12-12-GRU24-24-GRU48-48;

[0195] Step 3-1-2) The wind power correction model GRU3-3 generates values ​​within ±m% of the values ​​of the first 3a time steps. It then selects appropriate values ​​from the corresponding time steps in the initial wind power scenario set without replacement as the corresponding wind power output value. If no value meets the condition within ±m%, the range is expanded to ±2m%. If no value meets the condition within ±2m%, the range is expanded to ±4m%. If no value meets the condition within ±4m%, a random selection with replacement is performed from the initial wind power scenario set, where a is a positive integer and m is a real number.

[0196] Step 3-1-3) The wind power correction model GRU6-6 can generate values ​​within ±m% of the values ​​of the first 6a time points from the values ​​of the first 6a time points. It selects appropriate values ​​from the corresponding time points in the initial wind power scenario set without replacement as the wind power output values ​​for the corresponding time points. If there are no values ​​that meet the conditions within ±5%, the range is expanded to ±2m%. If there are no values ​​that meet the conditions within ±2m%, the range is expanded to ±4m%. If there are no values ​​that meet the conditions within ±4m%, random selection with replacement is performed from the initial wind power scenario set.

[0197] Step 3-1-4) The wind power correction model GRU12-12 can generate values ​​within ±m% of the values ​​of the previous 12a time points from the values ​​of the previous 12a time points. It selects appropriate values ​​from the corresponding time points in the initial wind power scenario set without replacement as the wind power output values ​​for the corresponding time points. If there are no values ​​that meet the conditions within ±m%, the range is expanded to ±2m%. If there are no values ​​that meet the conditions within ±2m%, the range is expanded to ±4m%. If there are no values ​​that meet the conditions within ±4m%, random selection with replacement is performed from the initial wind power scenario set.

[0198] 3-1-5) The wind power correction model GRU24-24 can generate values ​​within ±m% of the values ​​of the previous 24a time points from the values ​​of the previous 24a time points. It selects appropriate values ​​from the corresponding time points in the initial wind power scenario set without replacement as the wind power output values ​​for the corresponding time points. If there are no values ​​that meet the conditions within ±m%, the range is expanded to ±2m%. If there are no values ​​that meet the conditions within ±2m%, the range is expanded to ±4m%. If there are no values ​​that meet the conditions within ±4m%, random selection with replacement is performed from the initial wind power scenario set.

[0199] 3-1-6) The wind power correction model GRU48-48 can generate values ​​within ±m% of the values ​​of the first 48a time points from the values ​​of the first 48a time points. It selects appropriate values ​​from the corresponding time points in the initial wind power scenario set without replacement as the wind power output values ​​for the corresponding time points. If there are no values ​​that meet the conditions within ±m%, the range is expanded to ±2m%. If there are no values ​​that meet the conditions within ±2m%, the range is expanded to ±4m%. If there are no values ​​that meet the conditions within ±4m%, random selection with replacement is performed from the initial wind power scenario set.

[0200] The corrected wind power scenario set can be obtained by following the above steps.

[0201] Step 3-2 includes the following steps:

[0202] Step 3-2-1) Assemble the photovoltaic correction model in chronological order into a new photovoltaic correction model in the form of GRU36-14L-GRU12-12L-GRU4-4L-GRU4-4R-GRU12-12R-GRU36-14R;

[0203] Step 3-2-2) The photovoltaic correction model GRU4-4L can generate the values ​​of the left 4b times within ±m% from the values ​​of the middle 4b times, and select appropriate values ​​from the corresponding times in the photovoltaic initial scenario set in a non-replacement manner as the photovoltaic output value of the corresponding times. When there are no values ​​that meet the conditions within the ±m% range, the range is expanded to ±2m%. When there are no values ​​that meet the conditions within the ±2m% range, the range is expanded to ±4m%. When there are no values ​​that meet the conditions within the ±4m% range, random selection with replacement is performed from the photovoltaic initial scenario set.

[0204] Step 3-2-3) The photovoltaic correction model GRU4-4R can generate the values ​​of the right 4b times within ±m% from the values ​​of the middle 4b times, and select appropriate values ​​from the corresponding times in the photovoltaic initial scenario set in a non-replacement manner as the photovoltaic output value of the corresponding times. When there are no values ​​that meet the conditions within the ±m% range, the range is expanded to ±2m%. When there are no values ​​that meet the conditions within the ±2m% range, the range is expanded to ±4m%. When there are no values ​​that meet the conditions within the ±4m% range, random selection with replacement is performed from the photovoltaic initial scenario set.

[0205] Step 3-2-4) The photovoltaic correction model GRU12-12L can generate the values ​​of the left 12b times within ±m% from the values ​​of the middle 12b times, and select appropriate values ​​from the corresponding times in the photovoltaic initial scenario set in a non-replacement manner as the photovoltaic output value of the corresponding times. When there are no values ​​that meet the conditions within the ±m% range, the range is expanded to ±2m%. When there are no values ​​that meet the conditions within the ±2m% range, the range is expanded to ±4m%. When there are no values ​​that meet the conditions within the ±4m% range, random selection with replacement is performed from the photovoltaic initial scenario set.

[0206] Step 3-2-5) The photovoltaic correction model GRU12-12R can generate the values ​​of the right 12b times within ±m% from the values ​​of the middle 12b times. It selects appropriate values ​​from the corresponding times in the photovoltaic initial scenario set without replacement as the photovoltaic output value for the corresponding times. If there are no values ​​that meet the conditions within ±m%, the range is expanded to ±2m%. If there are no values ​​that meet the conditions within ±2m%, the range is expanded to ±4m%. If there are no values ​​that meet the conditions within ±4m%, random selection with replacement is performed from the photovoltaic initial scenario set.

[0207] Step 3-2-6) The photovoltaic correction model GRU36-14L can generate the values ​​of the left 14b times within ±m% from the values ​​of the middle 36b times. It selects appropriate values ​​from the corresponding times in the photovoltaic initial scenario set in a non-replacement manner as the photovoltaic output value of the corresponding times. If there are no values ​​that meet the conditions within the ±m% range, the range is expanded to ±2m%. If there are no values ​​that meet the conditions within the ±2m% range, the range is expanded to ±4m%. If there are no values ​​that meet the conditions within the ±4m% range, a random selection with replacement is performed from the photovoltaic initial scenario set.

[0208] Step 3-2-7) The photovoltaic correction model GRU36-14R can generate the values ​​of the right 14b times within ±m% from the values ​​of the middle 36b times. It selects appropriate values ​​from the corresponding times in the photovoltaic initial scenario set without replacement as the photovoltaic output value for the corresponding times. If there are no values ​​that meet the conditions within ±m%, the range is expanded to ±2m%. If there are no values ​​that meet the conditions within ±2m%, the range is expanded to ±4m%. If there are no values ​​that meet the conditions within ±4m%, random selection with replacement is performed from the photovoltaic initial scenario set.

[0209] The above steps will yield the corrected photovoltaic scenario set.

[0210] To verify the effectiveness of the proposed method, three metrics—Average Daily Output (ADO), Energy Score (ES), and Brier Score (BS)—were used to evaluate the scene set generated in this paper.

[0211] The calculation methods for the ADO and ES metrics are shown in the following formulas:

[0212]

[0213] Where J is the number of scene sets, t represents time, and P... i,t This represents the output at time t in scenario i, where z is a specific scenario. Let |j| represent the j-th scenario, ||·||2 be the L2 norm, and I represent the number of actual scenarios in a year.

[0214] The BS metric, also known as a scenario assessment method based on diagnostic events, focuses on the details of the scenario set.

[0215] The specific calculation process of the BS metric is as follows: First, diagnostic events are defined; second, the probability of occurrence of each diagnostic event is calculated for both the generated scene set and the actual scene; finally, the probability difference between the actual scene and the generated scene set under different diagnostic events is calculated. Therefore, the BS metric is also a negative indicator.

[0216] Based on this, the present invention defines 5 diagnostic events:

[0217] (1) Event 1: The power output fluctuates by more than 10% of the capacity within 1 hour.

[0218] (2) Event 2: The output power was greater than 30% of the capacity for a period of 4 hours.

[0219] (3) Event 3: The power output fluctuates by more than 30% of the capacity within 1 hour.

[0220] (4) Event 4: The output power is greater than 60% of the capacity for a period of 4 hours.

[0221] (5) Event 5: The output power was less than 10% of the capacity for a period of 4 hours.

[0222] Results analysis:

[0223] ADO indicator analysis:

[0224] Table 1 shows the ADO metrics for the four schemes in wind power and photovoltaic scenario generation. Compared with the real scenario set, it can be seen from the table that Scheme 1 and Scheme 3 are close to the real scenario ADO metrics in both wind power and photovoltaic scenario generation, Scheme 2 performs poorly, while Scheme 4 performs poorly in wind power scenario generation but shows excellent results in photovoltaic scenario generation.

[0225] Table 1. ADO metrics under different generation methods in different scenarios

[0226]

[0227] As shown in Table 1, the scene sets generated by Schemes 1, 3, and 4 are all superior to those generated by Scheme 2. This is because Scheme 2 only uses correlation coefficients and regression analysis to generate scene sets, which do not conform to the probability distribution of the real scene sets, resulting in its ADO index deviating significantly from the ADO index of the real scene sets. In contrast, the initial scene sets obtained by Schemes 1, 3, and 4 are all obtained by sampling the cumulative probability density function of the original data through LHS sampling, which conforms to the probability distribution of the real scene sets. Therefore, the ADO indices of Schemes 1, 3, and 4 are all closer to the ADO index of the real scene sets.

[0228] ES Metric Analysis:

[0229] Table 2 shows the ES (Effective Energy) indices obtained by comparing the generated wind power and solar power scenarios with the actual scenarios in 2022 for the four schemes. The relevant values ​​were normalized before calculation to obtain the ES indices. The table shows that Schemes 1, 3, and 4 generate wind power and solar power scenarios with minimal differences from the 2022 scenarios, accurately reflecting historical wind power and solar power output. However, the ES index of Scheme 2 indicates that it does not accurately reflect historical wind power and solar power output. Furthermore, the ES index of Scheme 3 shows that, compared to the other three schemes, both the generated wind power and solar power scenario sets are closer to the actual scenario sets in 2022.

[0230] Table 2 shows the ES metrics calculated in real-world scenarios in 2022.

[0231]

[0232] In addition to accurately reflecting historical power output scenarios, the generated scenarios also need to be able to predict future power output scenarios. Therefore, this paper compares the generated scenario set with the real scenario set from 2023. As shown in Table 3, the wind power scenario generated by Scheme 3 is closer to the real scenario, while the wind power scenarios generated by Schemes 1 and 4 are better than Scheme 2 but weaker than Scheme 3. In terms of photovoltaic scenario generation, the scenarios generated by all four schemes are relatively close to the real photovoltaic scenario, with Scheme 2 being better than the other three schemes, and Scheme 3 being better than Schemes 1 and 4.

[0233] Table 3 shows the ES metrics calculated in real-world scenarios in 2023.

[0234]

[0235] Based on the above analysis, it can be seen that, overall, among the four schemes, the scene set generated by scheme 3 is better than the other three schemes, while the scene sets generated by schemes 1 and 4 are weaker than those generated by scheme 3, but better than those generated by scheme 2.

[0236] As shown in Tables 2 and 3, Scheme 3 is superior. This is because the ES index calculates the difference between each generated scene and the real scene at the same moment. Since the scene sets obtained by Schemes 1 and 3 are obtained through LHS sampling, this can keep the output probability density of the scene set consistent with the real scene set. Scheme 2 determines the output value at each moment through the correlation coefficient, which makes the generated scenes have a strong linear relationship between different moments. However, the linear relationship of the real scene is not strong, so the output probability density of the scene set generated by Scheme 2 does not conform to the real scene. Schemes 3 and 4 are better than Scheme 2 (although their performance in photovoltaic scene generation is average, the difference between the four schemes is not significant). This is because Schemes 3 and 4 can fit the output trend of the real scene through the GRU machine learning model. Furthermore, the GRU machine learning model has the ability to fit nonlinear relationships, which allows the generated scene set to fit the real scene as closely as possible.

[0237] BS indicator analysis:

[0238] Figure 6 , Figure 7 , Figure 8 , Figure 9 , Figure 10 , Figure 11 , Figure 12 and Figure 13 The Black-Scholes (BS) index plots for the generated wind power and photovoltaic scenarios are compared with the real scenario sets of 2022 and 2023, respectively, to calculate typical events. The plots show that Scheme 3 outperforms the other three schemes in both fitting the 2022 and 2023 scenarios.

[0239] Specifically, in Event 1, for both the generated wind and solar power scene sets, Scheme 1 yielded a significantly higher Black-Scholes (BS) index than the other three schemes. This is because Scheme 3 uses a random combination method, resulting in no correlation between power values ​​at different times, leading to a large deviation between adjacent values ​​and thus an increase in the BS index. Furthermore, considering the overall index chart for Event 1, the wind power scene set generated by Scheme 3 is closer to the real scene set. In Event 2, for the generated wind power scene set, Scheme 2 is the least close to the real scene set, followed by Scheme 1, with Scheme 3 being the closest, and Scheme 4 falling between Schemes 1 and 2. In Event 2, for the generated solar power scene set, Schemes 1 and 2 are the least close to the real scene set, while Schemes 3 and 4 are the closest to the real scene set compared to Schemes 1 and 2. Overall, Schemes 1 and 2 yielded significantly higher BS values ​​than the other two schemes. For Scheme 1, the random combination of adjacent time points makes it difficult to maintain a constant power level, resulting in a probability of sustained power being much lower than the actual probability in real-world scenarios. For Scheme 2, the selection of power values ​​between adjacent time points strictly follows the linear relationship between regression equations, eliminating randomness between adjacent time points and increasing the probability of sustained power within a certain range, thus raising the BS value. Meanwhile, Schemes 3 and 4 employ the GRU machine learning model, which can effectively fit some nonlinearities in the scene set, making the generated scene set more consistent with real-world scenarios. Therefore, the BS values ​​obtained by Schemes 3 and 4 are lower than those of Schemes 1 and 2.

[0240] Figure 14 , Figure 15 , Figure 16 , Figure 17 , Figure 18 , Figure 19 , Figure 20 , Figure 21 , Figure 22 and Figure 23 This displays the BS (Browser-Side Scale) index under extreme events. Because photovoltaic power inherently exhibits significant volatility (lower output at night, early morning, and dusk, and higher output at midday), the BS calculation for event 3 is not performed for the photovoltaic scenario. The graph shows that the BS value of the wind power scenario in Scheme 1 under event 3 is significantly higher than that of the other schemes, while the BS values ​​of the other three schemes are at relatively low levels. In event 4, for the generated wind power scenario, from... Figure 8Looking at the data, Scheme 2 is the least similar to the real-world scenario set, followed by Schemes 1 and 4, with Scheme 3 being the closest. Schemes 1 and 4 overlap because event 4 does not occur in the scenarios generated by Schemes 1 and 4, resulting in a BS value that represents the probability of event 4 occurring in the real-world scenario set. For the generated photovoltaic scenarios, Schemes 1 and 2 have higher BS values ​​than Schemes 3 and 4, indicating that Schemes 3 and 4 are closer to the real-world scenario set. In event 5, the BS value of the scenario set generated by Scheme 3 is lower than other schemes, especially its wind power scenario set, which is closer to the real-world scenario. Furthermore, when evaluating the generated photovoltaic scenarios based on event 5, Schemes 1 and 2 exhibit a bimodal characteristic at both ends. This is because Scheme 1 uses a random combination and does not consider the natural process of photovoltaic power at the beginning and end, i.e., a process that gradually approaches zero power at the beginning and end. Scheme 2 exhibits a bimodal characteristic because its linear regression fails to consider its directionality; that is, photovoltaic power should gradually increase in the morning and gradually decrease in the afternoon. The scenario generation methods of Scheme 1 and Scheme 2 will result in a phenomenon that violates the laws of nature: the power of photovoltaic power generation continuously decreases in the morning and continuously increases in the afternoon.

[0241] Based on the analysis results, it can be seen that Scheme 3 outperforms the other three schemes in terms of the Black-Scholes ratio. The reason for this is that Scheme 3's GRU modified model provides redundancy for the transitions between different time points. This allows the transitions between adjacent time points to not be fixed values, but rather allow the values ​​between adjacent time points to fluctuate within a certain range. This fluctuation can be expected or unexpected. It is precisely this fluctuation that allows for the diversity of the generated scenario set, and also enables the generation of some extreme but realistically permissible scenarios.

Claims

1. A method for generating landscape scenes based on LHS-GRU, characterized in that, Includes the following steps: Step 1: Establish a wind and solar power output probability model based on historical wind and solar power output data, and obtain the initial wind and solar scene set by Latin hypercube sampling based on the obtained wind and solar power output probability model; Step 2: Learn the output correlation between wind power and photovoltaic power at different times using a neural network model, and establish correlation constraint models for wind power and photovoltaic power respectively; Step 3: Import the initial scene set into the model to generate the corrected scene set, also known as the corrected scene set; Step 1 specifically includes the following sub-steps: Step 1-1: Model the probability model of wind and solar power output; Step 1-2: Determine the bandwidth for wind and solar nonparametric kernel density estimation; Steps 1-3: Generate a landscape scene set based on Latin hypercube sampling; In steps 1-3, specifically: (1) First, for each time t of the wind and solar power output, where t = 1, 2, ..., 96, kernel density estimation is performed to obtain the power output probability density estimation function f at time t. t (p); (2) Calculate the cumulative probability density function F based on the kernel density estimation function obtained at each time step. t (P), and simultaneously the cumulative probability density function F t (P) is divided into N t There are N equal and non-overlapping intervals, where the length of each interval is 1 / N. t ; (3) Based on the obtained probability sub-intervals, perform a random sampling once in each probability sub-interval, that is: Where i refers to the i-th subinterval, i = 1, 2, ..., N t ;f t,i This refers to the cumulative probability density of the i-th subinterval at time t; random refers to a random number between 0 and 1. (4) Substitute the obtained value into its inverse function to obtain the Latin hypercube sampled value P of the scenery in the i-th subinterval at time t. t,i , can be represented as: in, It refers to the inverse function of the cumulative probability density function at time t; One LHS sampling can be completed through steps (1)-(4); The above methods are used to sample the probability density functions of wind power output and photovoltaic power output at each moment to generate data for the initial wind power scenario set and the initial photovoltaic scenario set.

2. The method according to claim 1, characterized in that, In step 1-1, specifically: Let the power output of new energy sources P have a power distribution density function f(p). Then let F(p) be the distribution function of the total power output of new energy sources P, with samples P1, P2, ..., P... n The empirical distribution function is: Among them, n refers to the number of samples; P i refers to the sample data; I(P i <p) refers to the indicator function of the set P; h refers to the window width, therefore, from equations (1) and (2), the natural estimate of the density function f(p) can be obtained as: Approximately: Let K(·) = 2 -1 I(·), and thus we have: Equation (5) is the probability model for wind and solar power output, and can also be called the kernel density estimate of f(p), where This refers to sample P. i The increase in the density function when it falls within the interval (ph, p+h), and K(·) is called its corresponding weight function, or kernel function. n This refers to the window width when there are n samples.

3. The method according to claim 1, characterized in that, In steps 1-2, specifically: First, using the integral squared error (ISE) as the loss function, the expression is: For equation (6), consider with With the relevant terms, equation (6) can be transformed into: in, This refers to the estimate after removing the i-th data point. Specifically: To simplify the calculation, approximating (n-1) in equation (8) as n yields a loss function equivalent to equation (6): Its optimal bandwidth is represented by h. opt =argminM1(h n ).

4. The method according to any one of claims 1 to 3, characterized in that, In step 2, the established correction model includes: Establish a GRU-corrected model for wind power output; Establish a GRU modified model for photovoltaic power output.

5. The method according to claim 4, characterized in that, When establishing the GRU modified model for wind power output, the following are included: (1) Divide the wind power output sample data W horizontally from left to right along the timeline. Each sample data has k power points. Thus, it can be divided into wind power output sample data W(t,t+k / 16), wind power output sample data W(t,t+k / 8), wind power output sample data W(t,t+k / 4), wind power output sample data W(t,t+k / 2), and wind power output sample data W(t,t+k), which are the training sets of five correction models: wind power correction model GRU3-3, wind power correction model GRU6-6, wind power correction model GRU12-12, wind power correction model GRU24-24, and wind power correction model GRU48-48. (2) Divide the wind power output sample data W(t,t+k / 16) into W(t,t+k / 32) and W(t+k / 32+1,t+k / 16) as the input and output of the wind power correction model GRU3-3 for training; (3) The wind power output sample data W(t,t+k / 8) is divided into wind power output sample data W(t,t+k / 16) and wind power output sample data W(t+k / 16+1,t+k / 8) as the input and output of the wind power correction model GRU6-6, respectively, for training. (4) The wind power output sample data W(t,t+k / 4) is divided into wind power output sample data W(t,t+k / 8) and wind power output sample data W(t+k / 8+1,t+k / 4) as the input and output of the wind power correction model GRU12-12, respectively, for training. (5) The wind power output sample data W(t,t+k / 2) is divided into wind power output sample data W(t,t+k / 4) and wind power output sample data W(t+k / 4+1,t+k / 2) as the input and output of the wind power correction model GRU24-24, respectively, for training. (6) Divide the wind power output sample data W(t,t+k) into wind power output sample data W(t,t+k / 2) and wind power output sample data W(t+k / 2+1,t+k), respectively, and use them as the input and output of the wind power correction model GRU48-48 for training. When establishing the GRU correction model for photovoltaic power output, the following are included: (1) Divide the photovoltaic power output sample data V horizontally from the middle of the time line to both sides. Each sample data has k power points. Therefore, the sample data can be divided into photovoltaic power output sample data V(l-7*k / 96,l), photovoltaic power output sample data V(l-3*k / 96,l+4*k / 96), photovoltaic power output sample data V(l-19*k / 96,l+4*k / 96), and photovoltaic power output sample data V(l-7*k / 96,l+16*k / 96). The power output sample data V(l-33*k / 96,l+16*k / 96) and the photovoltaic power output sample data V(l-19*k / 96,l+30*k / 96) are the training sets of six modified photovoltaic models: GRU4-4L, GRU4-4R, GRU12-12L, GRU12-12R, GRU36-14L, and GRU36-14R. (2) The photovoltaic power output sample data V(l-7*k / 96,l) can be divided into photovoltaic power output sample data V(l-3*k / 96,l) and photovoltaic power output sample data V(l-7*k / 96,l-4*k / 96), which are used as the input and output of the photovoltaic correction model GRU4-4L for training, respectively. (3) The photovoltaic power output sample data V(l-3*k / 96,l+4*k / 96) can be divided into photovoltaic power output sample data V(l-3*k / 96,l) and photovoltaic power output sample data V(l+*k / 96,l+4*k / 96), which are used as the input and output of the photovoltaic correction model GRU4-4R respectively for training. (4) The photovoltaic power output sample data V(l-19*k / 96,l+4*k / 96) can be divided into photovoltaic power output sample data V(l-7*k / 96,l+4*k / 96) and photovoltaic power output sample data V(l-19*k / 96,l-8*k / 96) as input and output of the photovoltaic correction model GRU12-12L for training; (5) The photovoltaic power output sample data V(l-7*k / 96,l+16*k / 96) can be divided into photovoltaic power output sample data V(l-7*k / 96,l+4*k / 96) and photovoltaic power output sample data V(l+5*k / 96,l+16*k / 96) as input and output of the photovoltaic correction model GRU12-12R for training; (6) The photovoltaic power output sample data V(l-33*k / 96,l+16*k / 96) can be divided into photovoltaic power output sample data V(l-19*k / 96,l+16*k / 96) and photovoltaic power output sample data V(l-33*k / 96,l-18*k / 96) as input and output of the photovoltaic correction model GRU36-14L for training; (7) The photovoltaic power output sample data V(l-19*k / 96,l+30*k / 96) can be divided into photovoltaic power output sample data V(l-19*k / 96,l+16*k / 96) and photovoltaic power output sample data V(l+17*k / 96,l+30*k / 96) as input and output of the photovoltaic correction model GRU36-14R for training.

6. The method according to claim 5, characterized in that, Step 3 specifically includes the following sub-steps: Step 3-1: The trained wind power correction models GRU3-3, GRU6-6, GRU12-12, GRU24-24, and GRU48-48 from the established wind power output GRU correction model are spliced ​​together in chronological order to form a complete wind power correction model. The initial wind power scenario set obtained in Step 1 is then imported into the wind power correction model. Step 3-2: The photovoltaic correction models GRU4-4L, GRU4-4R, GRU12-12L, GRU12-12R, GRU36-14L, and GRU36-14R trained in Step 2 are spliced ​​together in an expansion pattern from the middle to both sides to form a complete photovoltaic correction model. The initial photovoltaic scene set obtained in Step 1 is then imported into the photovoltaic correction model.

7. The method according to claim 6, characterized in that, Step 3-1 includes the following steps: Step 3-1-1) Assemble the wind power correction model in chronological order into a new wind power correction model in the form of GRU3-3-GRU6-6-GRU12-12-GRU24-24-GRU48-48; Step 3-1-2) The wind power correction model GRU3-3 generates values ​​within ±m% of the values ​​of the first 3a time steps. It then selects appropriate values ​​from the corresponding time steps in the initial wind power scenario set without replacement as the wind power output value for the corresponding time step. If there are no values ​​within ±m% that meet the conditions, the range is expanded to ±2m%. If there are no values ​​within ±2m% that meet the conditions, the range is expanded to ±4m%. If there are no values ​​within ±4m% that meet the conditions, a random selection with replacement is performed from the initial wind power scenario set, where a is a positive integer and m is a real number. Step 3-1-3) The wind power correction model GRU6-6 can generate values ​​within ±m% of the values ​​of the first 6a time points from the values ​​of the first 6a time points. It selects appropriate values ​​from the corresponding time points in the initial wind power scenario set without replacement as the wind power output values ​​for the corresponding time points. If there are no values ​​that meet the conditions within ±5%, the range is expanded to ±2m%. If there are no values ​​that meet the conditions within ±2m%, the range is expanded to ±4m%. If there are no values ​​that meet the conditions within ±4m%, random selection with replacement is performed from the initial wind power scenario set. Step 3-1-4) The wind power correction model GRU12-12 can generate values ​​within ±m% of the values ​​of the previous 12a time points from the values ​​of the previous 12a time points. It selects appropriate values ​​from the corresponding time points in the initial wind power scenario set without replacement as the wind power output values ​​for the corresponding time points. If there are no values ​​that meet the conditions within ±m%, the range is expanded to ±2m%. If there are no values ​​that meet the conditions within ±2m%, the range is expanded to ±4m%. If there are no values ​​that meet the conditions within ±4m%, random selection with replacement is performed from the initial wind power scenario set. 3-1-5) The wind power correction model GRU24-24 can generate values ​​within ±m% of the values ​​of the previous 24a time points from the values ​​of the previous 24a time points. It selects appropriate values ​​from the corresponding time points in the initial wind power scenario set without replacement as the wind power output values ​​for the corresponding time points. If there are no values ​​that meet the conditions within ±m%, the range is expanded to ±2m%. If there are no values ​​that meet the conditions within ±2m%, the range is expanded to ±4m%. If there are no values ​​that meet the conditions within ±4m%, random selection with replacement is performed from the initial wind power scenario set. 3-1-6) The wind power correction model GRU48-48 can generate values ​​within ±m% of the values ​​of the first 48a time points from the values ​​of the first 48a time points. It selects appropriate values ​​from the corresponding time points in the initial wind power scenario set without replacement as the wind power output values ​​for the corresponding time points. If there are no values ​​that meet the conditions within ±m%, the range is expanded to ±2m%. If there are no values ​​that meet the conditions within ±2m%, the range is expanded to ±4m%. If there are no values ​​that meet the conditions within ±4m%, random selection with replacement is performed from the initial wind power scenario set. The corrected wind power scenario set can be obtained by following the above steps.

8. The method according to claim 6 or 7, characterized in that, Step 3-2 includes the following steps: Step 3-2-1) Assemble the photovoltaic correction model in chronological order into a new photovoltaic correction model in the form of GRU36-14L-GRU12-12L-GRU4-4L-GRU4-4R-GRU12-12R-GRU36-14R; Step 3-2-2) The photovoltaic correction model GRU4-4L can generate the values ​​of the left 4b times within ±m% from the values ​​of the middle 4b times, and select appropriate values ​​from the corresponding times in the photovoltaic initial scenario set in a non-replacement manner as the photovoltaic output value of the corresponding times. When there are no values ​​that meet the conditions within the ±m% range, the range is expanded to ±2m%. When there are no values ​​that meet the conditions within the ±2m% range, the range is expanded to ±4m%. When there are no values ​​that meet the conditions within the ±4m% range, random selection with replacement is performed from the photovoltaic initial scenario set. Step 3-2-3) The photovoltaic correction model GRU4-4R can generate the values ​​of the right 4b times within ±m% from the values ​​of the middle 4b times, and select appropriate values ​​from the corresponding times in the photovoltaic initial scenario set in a non-replacement manner as the photovoltaic output value of the corresponding times. When there are no values ​​that meet the conditions within the ±m% range, the range is expanded to ±2m%. When there are no values ​​that meet the conditions within the ±2m% range, the range is expanded to ±4m%. When there are no values ​​that meet the conditions within the ±4m% range, random selection with replacement is performed from the photovoltaic initial scenario set. Step 3-2-4) The photovoltaic correction model GRU12-12L can generate the values ​​of the left 12b times within ±m% from the values ​​of the middle 12b times, and select appropriate values ​​from the corresponding times in the photovoltaic initial scenario set in a non-replacement manner as the photovoltaic output value of the corresponding times. When there are no values ​​that meet the conditions within the ±m% range, the range is expanded to ±2m%. When there are no values ​​that meet the conditions within the ±2m% range, the range is expanded to ±4m%. When there are no values ​​that meet the conditions within the ±4m% range, random selection with replacement is performed from the photovoltaic initial scenario set. Step 3-2-5) The photovoltaic correction model GRU12-12R can generate the values ​​of the right 12b times within ±m% from the values ​​of the middle 12b times. It selects appropriate values ​​from the corresponding times in the photovoltaic initial scenario set without replacement as the photovoltaic output value for the corresponding times. If there are no values ​​that meet the conditions within ±m%, the range is expanded to ±2m%. If there are no values ​​that meet the conditions within ±2m%, the range is expanded to ±4m%. If there are no values ​​that meet the conditions within ±4m%, random selection with replacement is performed from the photovoltaic initial scenario set. Step 3-2-6) The photovoltaic correction model GRU36-14L can generate the values ​​of the left 14b times within ±m% from the values ​​of the middle 36b times. It selects appropriate values ​​from the corresponding times in the photovoltaic initial scenario set in a non-replacement manner as the photovoltaic output value of the corresponding times. If there are no values ​​that meet the conditions within the ±m% range, the range is expanded to ±2m%. If there are no values ​​that meet the conditions within the ±2m% range, the range is expanded to ±4m%. If there are no values ​​that meet the conditions within the ±4m% range, a random selection with replacement is performed from the photovoltaic initial scenario set. Step 3-2-7) The photovoltaic correction model GRU36-14R can generate the values ​​of the right 14b times within ±m% from the values ​​of the middle 36b times. It selects appropriate values ​​from the corresponding times in the photovoltaic initial scenario set without replacement as the photovoltaic output value for the corresponding times. If there are no values ​​that meet the conditions within ±m%, the range is expanded to ±2m%. If there are no values ​​that meet the conditions within ±2m%, the range is expanded to ±4m%. If there are no values ​​that meet the conditions within ±4m%, random selection with replacement is performed from the photovoltaic initial scenario set. The above steps will yield the corrected photovoltaic scenario set.

Images