Multi-correlation combined driving water-wind-light scene generation method and system

Abstract

The invention belongs to the related technical field of electric power scene generation, and discloses a multi-correlation joint-driven water-wind-light scene generation method and system, and the method comprises the steps: collecting the historical data of water-wind-light three resources, determining the scene generation sequence, and extracting the existing variable information of each resource t-1 and t time period; constructing conditional distribution of a variable x2 when a variable x1 is known through a Gaussian mixture model, and optimizing parameters of the Gaussian mixture model according to existing variable information; for the first resource, optimizing the model parameters by using the existing variable information of the t-1 and t time periods; for the second resource and the third resource, combining the existing variable information of the t time period of the previous resource, the t-1 time period of the resource and the t time period of the resource to optimize the model parameters; and performing scene generation of corresponding resources by using the optimized Gaussian mixture model. The method not only considers the autocorrelation of a single resource, but also considers the cross-correlation between resources.

Description

Technical Field

[0001] The present invention belongs to the technical field related to power scene generation, and more specifically, relates to a method and system for generating a water-wind-solar scene with multiple related joint drives. Background Art

[0002] In recent years, with the rapid development of renewable energy, the proportion of hydropower, wind power, and photovoltaic power generation in the global energy structure has continued to increase. These resources are not only clean and renewable, but can also effectively reduce greenhouse gas emissions and help achieve carbon neutrality goals. However, hydropower, wind power, and photovoltaic power generation all have obvious randomness and volatility, and are greatly affected by climate and natural conditions, which makes their power generation on different time scales have greater uncertainty. How to accurately quantify these uncertainties is the key to ensuring the safe and stable operation of the power system.

[0003] In this context, scenario generation technology has become an effective means to characterize uncertainty. By constructing a reasonable set of scenarios, it is possible to simulate a variety of possible situations under different natural resource conditions, providing a scientific basis for scheduling decisions and planning analysis. However, current research focuses on the scenario generation of a single resource or only considers simple correlations between single resources, lacking systematic analysis and modeling of the complex relationships between hydropower, wind power, and photovoltaic power generation. Since these three resources are often affected by climate conditions at the same time in nature and have significant multiple correlations, ignoring this correlation will lead to deviations between the scenario generation results and the actual situation, thereby affecting the accuracy of subsequent analysis. Summary of the invention

[0004] In response to the above defects or improvement needs of the prior art, the present invention provides a water-wind-light scene generation method and system with multiple correlation joint drives, which is used to solve the problem that the existing scene generation technology mostly focuses on the scene generation of a single resource or only considers the simple correlation between a single resource, and lacks systematic analysis and modeling of the complex relationships among hydropower, wind power and photovoltaic power generation.

[0005] To achieve the above object, according to one aspect of the present invention, a method for generating a water-landscape scene driven by multiple correlations is provided, comprising: Offline training phase: S1, collect historical data on water resource runoff, wind resource wind power output, and light resource photovoltaic output, calculate the relationship between water, wind, and light resources based on the historical data, and determine the order of generating water, wind, and light scenes based on the relationship, which are the first resource, the second resource, and the third resource in order of priority; and extract the existing variable information of each resource in the t-1 period and the t period based on the historical data; S2, construct Gaussian mixture models corresponding to the three resources respectively: construct the two-dimensional joint distribution of variables x1 and variables x2 and the conditional distribution of variable x2 when variable x1 is known through the Gaussian mixture model, and then generate the simulated value of variable x2 according to variable x1, and establish a sample data set according to the existing variable information to optimize the parameters of the Gaussian mixture model corresponding to each resource, and obtain the Gaussian mixture model with optimized parameters corresponding to each resource; For the first resource, the existing variable information in the time period t-1 is made to correspond to the variable x1, and the existing variable information in the time period t is made to correspond to the variable x2, and the Gaussian mixture model parameters are optimized; For any resource among the second resource and the third resource, the existing variable information of the time period t of the resource ranked before the any resource and the existing variable information of the time period t-1 of the any resource are combined and corresponded to the variable x1, the existing variable information of the time period t of the any resource is corresponded to the variable x2, and the Gaussian mixture model parameters corresponding to the any resource are optimized; Online scene generation phase: S3, using the Gaussian mixture model with optimized parameters corresponding to each resource to generate scenes for the corresponding resources.

[0006] According to the method for generating water, scenery and light scenes driven by multiple correlation joint drive provided by the present invention, in S2, a sample data set is established according to the existing variable information to optimize the Gaussian mixture model parameters corresponding to each resource, specifically including: For the second resource, the existing variable information of the first resource in the t period and the existing variable information of the second resource in the t-1 period are arranged in sequence and correspond to the variable x1, and the existing variable information of the second resource in the t period is corresponded to the variable x2, and the Gaussian mixture model parameters are optimized; For the third resource, the existing variable information of the first resource in time period t, the existing variable information of the second resource in time period t, and the existing variable information of the third resource in time period t-1 are arranged in sequence and correspond to the variable x1, and the existing variable information of the third resource in time period t is corresponded to the variable x2, and the Gaussian mixture model parameters are optimized.

[0007] According to the method for generating water-landscape scenes driven by multiple correlation joint drives provided by the present invention, S3 specifically includes: For the first resource, the average value of the existing variable information in the maximum period is used as variable x1 to generate the first simulation value of variable x2, and the first simulation value of variable x2 is used as the first generated value of the scenario. Then, the previous generated value is used as the new variable x1 to generate the current variable x2; For the second resource, the first generated value of the first resource is combined with the average value of the existing variable information of the second resource's maximum period as variable x1 to generate the first simulated value of variable x2. The first simulated value of variable x2 is used as the first generated value of the scenario. Then, the current generated value of the first resource is combined with the previous generated value of the second resource as the new variable x1 to generate the current variable x2. For the third resource, the first generated value of the first resource, the first generated value of the second resource and the average value of the existing variable information of the third resource's maximum time period are combined as variable x1 to generate the first simulated value of variable x2. The first simulated value of variable x2 is used as the first generated value of the scene. Then, the current generated value of the first resource, the current generated value of the second resource and the previous generated value of the third resource are combined as the new variable x1 to generate the current variable x2.

[0008] According to the method for generating water, wind and light scenes driven by multiple correlation joint drive provided by the present invention, in S1, the mutual relationship between water, wind and light resources is calculated according to historical data, and the order of generating water, wind and light scenes is determined according to the mutual relationship, which specifically includes: Calculate the mutual information between the water, wind and light resources based on historical data, use the mutual information to determine the relationship between the water, wind and light resources, and obtain the mutual information value between the water, wind and light resources; Calculate the sum of the mutual information values ​​between any resource and other resources, obtain the comprehensive mutual information index of any resource, and determine the order of generating the three resource scenarios according to the comprehensive mutual information index.

[0009] According to the method for generating water, scenery and light scenes driven by multiple correlations provided by the present invention, in S1, the existing variable information of each resource in the t-1 period and the t period is extracted according to historical data, specifically: ; ; ; Among them, DH is the multi-year historical data of water resource runoff h, DW is the multi-year historical data of wind resource wind power output w, and DP is the multi-year historical data of light resource photovoltaic output p; the existing variable information of the three resources of water, wind and light in the t-1 period and the t period are h t-1 With h t 、w t-1 With w t 、p t-1 With p t ; , N is determined according to the target time scale.

[0010] According to the method for generating water, scenery and light scenes driven by multiple correlations provided by the present invention, in S2, a sample data set is established according to the existing variable information to optimize the Gaussian mixture model parameters corresponding to each resource, and the Gaussian mixture model after the parameter optimization corresponding to each resource is obtained, which specifically includes: For any resource, variables x1 and x2 change with the increase of time period t. The existing variable information of each time period t is used to optimize the parameters of the Gaussian mixture model, and N models with optimized parameters corresponding to any resource are obtained for scene generation of N time periods for any resource.

[0011] According to the method for generating water-landscape scenes driven by multiple correlation joint drive provided by the present invention, the Gaussian mixture model in S2 is specifically: ; ; Where: is the two-dimensional joint distribution of variables x1 and x2, is the conditional distribution of variable x2 when variable x1 is known, is the binary probability density function, is the conditional probability density function of variable x2 when variable x1 is known; The bivariate probability density function , the mathematical expression is: ; Where: represents the binary Gaussian probability density function; K is the number of components of the Gaussian mixture model, which is a hyperparameter; is the weight of the kth component; are the mean vector and covariance matrix of the kth component respectively.

[0012] According to the method for generating water, scenery and light scenes driven by multiple correlation joint drive provided by the present invention, in S2, the Gaussian mixture model corresponding to each resource is optimized for any resource, specifically, the hyperparameters K, weights in the Gaussian mixture model are optimized. , mean vector and the covariance matrix Optimize and determine, the specific optimization steps include: Pre-set several different hyperparameter K values; For any preset K value, the sample data set is divided into K clusters using clustering methods; the center of each cluster is used as the mean vector of each component in the Gaussian mixture model ; Calculate the difference between the data points in each cluster and the cluster center to obtain the covariance matrix ; Calculate the proportion of the number of samples in each cluster to the total number of samples and obtain the weight ; The obtained parameters are used as initial values, and the expectation maximization algorithm is used to optimize the model parameters to obtain the mean vector, covariance matrix and weight parameter value of the model after optimization corresponding to any preset K value; The comprehensive strategy of Akaike information criterion and Bayesian information criterion is used to evaluate the Gaussian mixture model under different preset K values, and the preset K value corresponding to the Gaussian mixture model with the smallest comprehensive strategy value is selected as the optimal K value; The Kolmogorov-Smilov test was performed on the theoretical distribution of the Gaussian mixture model corresponding to the optimal K value and the empirical distribution of the sample data set, with a significance level of Set to 0.2. If the test passes, the model training is completed. If the test fails, adjust the preset K value and re-optimize the model parameters.

[0013] According to the water-landscape scene generation method driven by multiple correlations provided by the present invention, the Akaike information criterion Specifically: ; Bayesian Information Criterion Specifically: ; Where: is the number of parameters of the Gaussian mixture model under K components, representing the complexity of the model; M is the amount of data in the sample data set; the comprehensive strategy value is the weighted sum of the Akaike information criterion and the Bayesian information criterion.

[0014] According to another aspect of the present invention, a system for generating water, landscape and light scenes by multiple related joint drives is provided, the system comprising a memory and a processor, the memory storing a computer program, and the processor executing any one of the above-mentioned methods for generating water, landscape and light scenes by multiple related joint drives when executing the computer program.

[0015] In general, compared with the prior art, the above technical solutions conceived by the present invention provide a method and system for generating water, scenery and light scenes with multiple related joint drives: 1. Not only the autocorrelation of a single resource but also the mutual correlation between resources are considered. The sample data set during Gaussian mixture model training associates related resources and uses the Gaussian mixture model to directly describe the joint distribution and conditional distribution between water, wind and solar resources. There is no need to pre-assume the specific distribution form of water, wind and solar. It has strong adaptability and can effectively characterize the autocorrelation of a single resource and the mutual correlation between resources. The scenarios generated by the optimized Gaussian mixture model can comprehensively characterize the multiple correlations between water, wind and solar, which helps to more realistically reflect the fluctuation characteristics of various renewable energy sources and further provide more accurate basic data for the dispatch and planning of power systems. 2. By embedding the joint relationship and conditional distribution between various resources into the scenario generation process, the time-series scenarios of water, wind and light resources are gradually generated, which can dynamically simulate the linkage changes between resources, making the generated scenarios more in line with the resource fluctuation characteristics in the actual operating environment, thereby significantly improving the adaptability and operation stability of the water, wind and light multi-energy complementary system; 3. The high-dimensional water, wind and light data generation problem is decomposed into multiple low-dimensional generation problems by using conditional distribution, which effectively avoids the difficulty of high-dimensional model parameter estimation. The nesting and step-by-step generation strategy of multi-model results is combined to truly reflect the mutual influence relationship between various resources, which not only reduces the computational complexity but also retains the multiple correlations between data, further improving the overall computational efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a flow chart of the method for generating water, scenery and light scenes by multiple correlation joint driving provided by the present invention. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0018] See also Figure 1 The first embodiment provides a method for generating a water-landscape scene with multiple related joint drives, and the method for generating a water-landscape scene includes: Offline training phase: S1, collect historical data on water resource runoff, wind resource wind power output, and light resource photovoltaic output, calculate the relationship between water, wind, and light resources based on the historical data, and determine the order of generating water, wind, and light scenes based on the relationship, which are the first resource, the second resource, and the third resource in order of priority; and extract the existing variable information of each resource in the t-1 period and the t period based on the historical data; S2, construct Gaussian mixture models corresponding to the three resources respectively: construct the two-dimensional joint distribution of variables x1 and variables x2 and the conditional distribution of variable x2 when variable x1 is known through the Gaussian mixture model, and then generate the simulated value of variable x2 according to variable x1, and establish a sample data set according to the existing variable information to optimize the parameters of the Gaussian mixture model corresponding to each resource, and obtain the Gaussian mixture model with optimized parameters corresponding to each resource; Among them, for the first resource, the existing variable information of the t-1 period is corresponded to the variable x1, and the existing variable information of the t period is corresponded to the variable x2, and the Gaussian mixture model parameters are optimized; that is, for the first resource, the existing variable information of the t-1 period is combined with the existing variable information of the t period as the sample data set input of the Gaussian mixture model corresponding to the t period; For any resource among the second resource and the third resource, the existing variable information of the time period t of the resource ranked before the any resource and the existing variable information of the time period t-1 of the any resource are combined and corresponded to the variable x1, the existing variable information of the time period t of the any resource is corresponded to the variable x2, and the Gaussian mixture model parameters corresponding to the any resource are optimized; that is, for any resource among the second resource and the third resource, the existing variable information of the time period t of the resource ranked before the any resource, the existing variable information of the time period t-1 of the any resource and the existing variable information of the time period t of the any resource are combined as the sample data set input of the Gaussian mixture model corresponding to the time period t of the any resource; Online scene generation phase: S3, using the Gaussian mixture model with optimized parameters corresponding to each resource to generate scenes for the corresponding resources.

[0019] Furthermore, in S2, a sample data set is established based on the existing variable information to optimize the Gaussian mixture model parameters corresponding to each resource, specifically including: For the second resource, the existing variable information of the first resource in the t period and the existing variable information of the second resource in the t-1 period are arranged in sequence, that is, they are arranged in order and correspond to the variable x1, and the existing variable information of the second resource in the t period is corresponded to the variable x2, and the Gaussian mixture model parameters are optimized; For the third resource, the existing variable information of the first resource in time period t, the existing variable information of the second resource in time period t, and the existing variable information of the third resource in time period t-1 are arranged in sequence, that is, they are arranged together in order and correspond to the variable x1, and the existing variable information of the third resource in time period t is corresponded to the variable x2, and the Gaussian mixture model parameters are optimized.

[0020] Furthermore, S3 specifically includes: For the first resource, the average value of the existing variable information when the maximum time period, i.e., t, takes the maximum value, is used as variable x1 to generate the first simulated value of variable x2. The first simulated value of variable x2 is used as the first generated value of the scenario. After that, the previous generated value is used as the new variable x1 to generate the current variable x2. For the second resource, the first generated value of the first resource is combined with the average value of the existing variable information when the second resource takes the maximum value during the maximum period, i.e., t, as variable x1 to generate the first simulated value of variable x2. The first simulated value of variable x2 is used as the first generated value of the scenario. Then, the current generated value of the first resource is combined with the previous generated value of the second resource as the new variable x1 to generate the current variable x2. For the third resource, the first generated value of the first resource, the first generated value of the second resource and the average value of the existing variable information when the third resource takes the maximum value during the maximum period, i.e., t, are combined as variable x1 to generate the first simulated value of variable x2. The first simulated value of variable x2 is used as the first generated value of the scene. Then, the current generated value of the first resource, the current generated value of the second resource and the previous generated value of the third resource are combined as the new variable x1 to generate the current variable x2.

[0021] In some specific embodiments, in S1, the mutual relationship between the water, wind and light resources is calculated according to the historical data, and the order of generating the water, wind and light scenes is determined according to the mutual relationship, specifically including: Calculate the mutual information between the water, wind and light resources based on historical data, use the mutual information to determine the relationship between the water, wind and light resources, and obtain the mutual information value between the water, wind and light resources; Calculate the sum of the mutual information values ​​between any resource and other resources, obtain the comprehensive mutual information index of any resource, and determine the order of generating the three resource scenarios according to the comprehensive mutual information index.

[0022] The mutual information value between water, wind and light resources is as follows:

[0023] In the formula, is the mutual information value between random variables X and Y (such as water and wind, wind and light, light and water), x and y are the observed values ​​of random variables X and Y, which represent specific data points in random experiments or data sets. is the joint probability distribution of x and y, are the marginal probability distributions of x and y respectively.

[0024] The comprehensive mutual information index of resources indicates the overall correlation between a resource and the other two resources. The order of scene generation is determined according to the value of the comprehensive mutual information index of each resource. The larger the comprehensive mutual information index, the higher the priority. The formula of the comprehensive mutual information index is as follows:

[0025] In the formula, is the comprehensive mutual information index of the random variable X.

[0026] Furthermore, the historical measured water, wind and light data are all in the form of a matrix D with M rows and N columns:

[0027] In the formula, D represents the historical data of resources over many years, DH is the historical data of water resource runoff h, DW is the historical data of wind resource wind power output w, and DP is the historical data of light resource photovoltaic output p. M is the number of years; if it is daily scale data, N is 365; if it is ten-day scale data, N is 36; if it is monthly scale data, N is 12.

[0028] In S1, the existing variable information of each resource in the t-1 period and the t period is extracted based on historical data. Specifically, the relevant variables are extracted based on the historical measured water, wind and light data of the target time scale. The six existing variable information involving water, wind and light in the t-1 and t periods are as follows:

[0029] Among them, the existing variable information of the three resources of water, wind and light in period t-1 and period t are h t-1 With h t 、w t-1 With w t 、p t-1 With p t ; , N is determined according to the target time scale. That is, the existing variable information of any t period is a column of data corresponding to the t period in the historical data including many years.

[0030] Specifically, in S2, a sample data set is established according to the existing variable information to optimize the Gaussian mixture model parameters corresponding to each resource, and the Gaussian mixture model after the parameter optimization corresponding to each resource is obtained, which specifically includes: For any resource, the variables x1 and x2 in the Gaussian mixture model change with the increment of time period t. The existing variable information of each time period t is used to optimize the parameters of the Gaussian mixture model, and N models with optimized parameters corresponding to any resource are obtained for scene generation of any resource in N time periods. That is, for any resource, when t is 1-N, N Gaussian mixture models need to correspond, and the N Gaussian mixture models are respectively input into the corresponding sample data sets for optimization. When the scene is generated, the corresponding optimized Gaussian mixture models are used for scene generation in N time periods.

[0031] The Gaussian mixture model in S2 is specifically:

[0032] Where: is the two-dimensional joint distribution of variables x1 and x2, is the conditional distribution of variable x2 when variable x1 is known, is the binary probability density function, is the conditional probability density function of variable x2 when variable x1 is known; The bivariate probability density function , the mathematical expression is:

[0033] Where: represents the binary Gaussian probability density function; K is the number of components of the Gaussian mixture model, which is a hyperparameter; is the weight of the kth component; are the mean vector and covariance matrix of the kth component respectively.

[0034] In S2, the Gaussian mixture model parameters corresponding to each resource are optimized, specifically, the hyperparameters K, weights in the Gaussian mixture model are optimized. , mean vector and the covariance matrix Optimize and determine, the specific optimization steps of each parameter include: S21, pre-setting a number of different hyperparameter K values; for example, the K values ​​may be preset to be 1, 2, 3...20, that is, 20 K values ​​are pre-set, and the specific preset value of the K value may be selected based on experience, and is not specifically limited; S22, for any preset K value, use K-means clustering method to divide the sample data set into K clusters; the center of each cluster is used as the mean vector of each component in the Gaussian mixture model ; Calculate the difference between the data points in each cluster and the cluster center to obtain the covariance matrix ; Calculate the proportion of the number of samples in each cluster to the total number of samples and obtain the weight ; S23, taking the parameters obtained in S22 as initial values, optimizing the model parameters using the expectation maximization (EM) algorithm, and iteratively updating until the model converges to obtain the mean vector, covariance matrix, and weight parameter value of the model after optimization corresponding to any preset K value; S24, using the comprehensive strategy of Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) to evaluate the Gaussian mixture model under different preset K values, and selecting the preset K value corresponding to the Gaussian mixture model with the smallest comprehensive strategy value as the optimal K value; S25, the Kolmogorov-Smirnov (KS) test is performed on the theoretical distribution of the Gaussian mixture model corresponding to the optimal K value and the empirical distribution of the sample data set to evaluate their goodness of fit, with a significance level of Set to 0.2. If the test passes, it is considered that the model fits the data well, the parameter value is determined, and the model training is completed. If the test fails, adjust the preset K value and re-optimize the model parameters, that is, re-perform S21-S25.

[0035] Among them, Akaike information criterion Specifically:

[0036] Bayesian Information Criterion Specifically:

[0037] Where: is the number of parameters of the Gaussian mixture model under K components, representing the complexity of the model; M is the amount of data in the sample data set, i.e. the number of years; the comprehensive strategy value is the weighted sum of the Akaike information criterion and the Bayesian information standard.

[0038] A compromise strategy is adopted to select the model with the smallest score, taking into account the model's predictive performance, simplicity, and ability to prevent overfitting. :

[0039] After the parameters in the Gaussian mixture model are optimized and determined, the conditional probability density function of variable x2 when variable x1 is known is , the mathematical expression is:

[0040] Where: They are the conditional mean and conditional covariance of variable x2 when variable x1 is known, and the expressions are as follows:

[0041] Where: and are the mean vectors of variables x1 and x2 respectively; is the covariance matrix between variables x1 and x1; is the covariance matrix between variables x2 and x2; and Consistent, is the covariance matrix between variables x1 and x2; is the covariance matrix of variable x2 when variable x1 is known.

[0042] Substitute the result of formula (13) into formula (8) to calculate the conditional distribution of variable x2 when variable x1 is known; when each scene is generated, obtain a random number between 0 and 1 ,make and Equal, that is, according to the conditional distribution of variable x2 when variable x1 is known and the conditional distribution value of variable x2 when variable x1 is known, variable x2 when variable x1 is known can be obtained, that is, the simulation value of variable x2 is the generated scenario.

[0043] This embodiment provides a method for generating a water-landscape scene driven by multiple correlations, aiming to solve the problems that the existing scene generation method only considers the generation of a single resource or the generation of a single correlation. The method includes the following steps: Collect historical data of three resources: water resource runoff h, wind resource wind power output w, and light resource photovoltaic output p, calculate the relationship between water, wind and light, and determine the order of generating water, wind and light scenes according to the size of the calculation results. In this embodiment, it is assumed that w>p>h.

[0044] According to the sequence of scenario generation, the scenario of variable w is generated first. This step only considers the first-order autocorrelation of variable w, that is, in the sample data set of the Gaussian mixture model corresponding to time period t, x1= {w t-1}、x2={w t}, optimize the model parameters according to the sample data set according to steps S21-S25, obtain the optimized Gaussian mixture model, and then use the model to generate the scene.

[0045] Specifically: the sample data set is divided into t-1}、x2={w t} is substituted into the Gaussian mixture model corresponding to the time period t to train the optimization model, that is, to determine the model parameters; The above trained and optimized model is used for scene generation, and the scene data generated by wind resources is recorded as When the first data of the first year is generated, that is, when m= 1 and t= 1, let the known variable x1 be equal to the corresponding w at t= 1. t-1 The average value of the existing variable information in the maximum period generates the simulated value of the first variable x2, which is the first data of the first year scenario; then let t=t+1, and convert the calculated As new , at this time the known variable x1 is equal to ; When the simulation of the Nth data of the first year is completed, that is, t=N, the generation of the first year scenario is completed, and the simulation of the first data of the next year begins, that is, m=m+1; When the first data simulation of the new year is , t = 1, let That is, when simulating the first data of the next year, let the known variable x1 be equal to the last generated data of the previous year, and then continue to generate data for the new year until t is equal to the target time number N; Repeat the generation of scenario data for each year until m is equal to the target number of years, completing the scenario generation of the w variable.

[0046] According to the order of scenario generation, the scenario of variable p is generated next. This step considers the first-order autocorrelation of variable p and the cross-correlation between variable p and variable w, that is, in the sample data set of the Gaussian mixture model corresponding to time period t, x1= {w t ,p t-1 That is, the existing variable information of the first resource in period t and the existing variable information of the second resource in period t-1 are sequentially arranged together as x1, x2={p t According to the sample data set, the model parameters are optimized according to steps S21-S25 to obtain the optimized Gaussian mixture model, and then the model is used to generate the scene.

[0047] Specifically: the sample data set is divided into t ,p t-1}、x2={p t} is substituted into the Gaussian mixture model corresponding to the time period t to train the optimization model, that is, to determine the model parameters; The above trained and optimized model is used for scene generation, and the scene data generated by the light resource is recorded as When simulating the first data of the first year, that is, when m= 1, t= 1, let the known variable x1 be equal to ,in is the p corresponding to t = 1 t-1 The average value is the average value of the existing variable information in the maximum period. is the generated value of the corresponding scene of the first resource, and The variables x1 are arranged in sequence to generate the simulated value of the first variable x2, which is the first data of the first year scenario; then let t = t + 1, and the calculated As new , at this time the known variable x1 is equal to ; When the simulation of the Nth data of the first year is completed, that is, t=N, the generation of the first year scenario is completed, and the simulation of the first data of the next year begins, that is, m=m+1; When the first data simulation of the new year is , when t = 1, let That is, when simulating the first data of the next year, let the known variable x1 be equal to the generated value of the corresponding scenario of the first resource and the last generated data of the second resource in the previous year, and then continue to generate data for the new year until t is equal to the target time number N; Repeat the generation of scenario data for each year until m is equal to the target number of years, completing the scenario generation of the p variable.

[0048] According to the order of scenario generation, the scenario of h variable is finally generated. This step considers the first-order autocorrelation of h variable and the cross-correlation between h variable and w variable and p variable, that is, in the sample data set of Gaussian mixture model corresponding to time period t, x1= {w t ,p t ,h t-1 That is, the existing variable information of the first resource in period t, the existing variable information of the second resource in period t, and the existing variable information of the third resource in period t-1 are sequentially arranged together as x1, x2={h t According to the sample data set, the model parameters are optimized according to steps S21-S25 to obtain the optimized Gaussian mixture model, and then the model is used to generate the scene.

[0049] Specifically: the sample data set is divided into t ,p t ,h t-1}、x2={h t} is substituted into the Gaussian mixture model corresponding to the time period t to train the optimization model, that is, to determine the model parameters; The above trained and optimized model is used for scene generation, and the scene data generated by water resources is recorded as When simulating the first data of the first year, that is, when m= 1, t= 1, let the known variable x1 be equal to ,in is the h corresponding to t = 1 t-1 The average value is the average value of the existing variable information in the maximum period. is the generated value of the corresponding scenario of the first resource, is the generated value of the corresponding scene of the second resource, , and The variables x1 are arranged in sequence to generate the simulated value of the first variable x2, which is the first data of the first year scenario; then let t = t + 1, and the calculated As new , at this time the known variable x1 is equal to ; When the simulation of the Nth data of the first year is completed, that is, t=N, the generation of the first year scenario is completed, and the simulation of the first data of the next year begins, that is, m=m+1; When simulating the first data of a new year, that is, , when t = 1, let That is, when simulating the first data of the next year, let the known variable x1 be equal to the generated value of the first resource corresponding to the scenario, the generated value of the second resource corresponding to the scenario, and the last generated data of the third resource last year, and then continue to generate data for the new year until t is equal to the target time number N; Repeat the generation of scenario data for each year until m is equal to the target number of years, completing the scenario generation of the h variable.

[0050] Furthermore, the second embodiment provides a water-landscape scene generation system with multiple related joint drives, the system comprising a memory and a processor, the memory storing a computer program, and the processor executing the water-landscape scene generation method with multiple related joint drives as described in any of the above embodiments when executing the computer program.

[0051] The daily historical hydrological data of Xiluodu Hydropower Station and the daily historical wind speed and photovoltaic intensity of the planned wind power and photovoltaic installed areas in its vicinity are taken as the research objects, and the historical wind speed and photovoltaic intensity are converted into the corresponding wind power output and photovoltaic output according to the planned installed capacity. The time series of runoff, wind power output and photovoltaic output all cover the period from 1940 to 2021. According to the above time series, the method of the present invention is used to generate water, wind and light scenes, and each generates a scene sequence of 1000 years. During the implementation process, statistical indicators (mean, standard deviation, skewness coefficient, kurtosis coefficient), time correlation indicators (Pearson correlation coefficient) and spatial correlation indicators (average Kendall correlation coefficient absolute error) are used to evaluate the statistical characteristics and multiple correlations of the generated scenes.

[0052] Tables 1 to 3 respectively show the statistical characteristics and correlation coefficients of the daily-scale water, wind, and light scenes generated by the method of the present invention, the two-dimensional Copula method, and the seasonal autoregressive model method (SAR method), and the root mean square error (RMSE), mean absolute percentage error (MAPE), and mean absolute error (MAE) between the statistical characteristics and correlation coefficients of the historically measured daily-scale water, wind, and light scenes. Table 4 shows the spatial correlation indicators between the daily-scale wind, water, wind, and water-light scenes generated by these three methods. The smaller the root mean square error, mean absolute percentage error, and mean absolute error values, the higher the degree of fit between the generated scene and the historically measured scene in terms of statistical characteristics and autocorrelation; the smaller the spatial correlation index, the more accurately the simulation sequence describes the mutual correlation characteristics between the resources. In addition, the values ​​in Tables 1 to 4 are bolded to identify the best results in each method; ; ; ; .

[0053] Among the comparison methods, the two-dimensional Copula method only considers the mutual correlation between resources, while the SAR method only focuses on the first-order autocorrelation of a single resource. As can be seen from Tables 1 to 3, the method of the present invention can more comprehensively describe the statistical characteristics of the historical measured scenes, and the simulation effects in terms of statistical characteristics and first-order autocorrelation are better than the other two comparison methods. In addition, it can be seen from Table 4 that the method of the present invention also shows better results in simulating the mutual correlation between wind-solar, water-wind, and water-solar sequences, which is significantly better than the two-dimensional Copula method and the SAR method.

[0054] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for generating water-landscape scenes driven by multiple correlations, characterized in that: include: Offline training phase: S1, collect historical data on water resource runoff, wind resource wind power output, and light resource photovoltaic output, calculate the relationship between water, wind, and light resources based on the historical data, and determine the order of generating water, wind, and light scenes based on the relationship, which are the first resource, the second resource, and the third resource in order of priority; and extract the existing variable information of each resource in the t-1 period and the t period based on the historical data; S2, construct Gaussian mixture models corresponding to the three resources respectively: construct the two-dimensional joint distribution of variables x1 and variables x2 and the conditional distribution of variable x2 when variable x1 is known through the Gaussian mixture model, and then generate the simulated value of variable x2 according to variable x1, and establish a sample data set according to the existing variable information to optimize the parameters of the Gaussian mixture model corresponding to each resource, and obtain the Gaussian mixture model with optimized parameters corresponding to each resource; For the first resource, the existing variable information in the time period t-1 is made to correspond to the variable x1, and the existing variable information in the time period t is made to correspond to the variable x2, and the Gaussian mixture model parameters are optimized; For any resource among the second resource and the third resource, the existing variable information of the time period t of the resource ranked before the any resource and the existing variable information of the time period t-1 of the any resource are combined and corresponded to the variable x1, the existing variable information of the time period t of the any resource is corresponded to the variable x2, and the Gaussian mixture model parameters corresponding to the any resource are optimized; Online scene generation phase: S3, using the Gaussian mixture model with optimized parameters corresponding to each resource to generate scenes for the corresponding resources.

2. The method for generating water-landscape scenes driven by multiple correlation joint driving according to claim 1, characterized in that: In S2, a sample data set is established based on the existing variable information to optimize the Gaussian mixture model parameters corresponding to each resource, specifically including: For the second resource, the existing variable information of the first resource in the t period and the existing variable information of the second resource in the t-1 period are arranged in sequence and correspond to the variable x1, and the existing variable information of the second resource in the t period is corresponded to the variable x2, and the Gaussian mixture model parameters are optimized; For the third resource, the existing variable information of the first resource in time period t, the existing variable information of the second resource in time period t, and the existing variable information of the third resource in time period t-1 are arranged in sequence and correspond to the variable x1, and the existing variable information of the third resource in time period t is corresponded to the variable x2, and the Gaussian mixture model parameters are optimized.

3. The method for generating water-landscape scenes driven by multiple correlation joint driving according to claim 1, characterized in that: S3 specifically includes: For the first resource, the average value of the existing variable information in the maximum period is used as variable x1 to generate the first simulation value of variable x2, and the first simulation value of variable x2 is used as the first generated value of the scenario. Then, the previous generated value is used as the new variable x1 to generate the current variable x2; For the second resource, the first generated value of the first resource is combined with the average value of the existing variable information of the second resource's maximum period as variable x1 to generate the first simulated value of variable x2. The first simulated value of variable x2 is used as the first generated value of the scenario. Then, the current generated value of the first resource is combined with the previous generated value of the second resource as the new variable x1 to generate the current variable x2. For the third resource, the first generated value of the first resource, the first generated value of the second resource and the average value of the existing variable information of the third resource's maximum time period are combined as variable x1 to generate the first simulated value of variable x2. The first simulated value of variable x2 is used as the first generated value of the scene. Then, the current generated value of the first resource, the current generated value of the second resource and the previous generated value of the third resource are combined as the new variable x1 to generate the current variable x2.

4. The method for generating water-landscape scenes driven by multiple correlation joint driving according to claim 1, characterized in that: In S1, the relationship between water, wind and light resources is calculated based on historical data, and the order of generating water, wind and light scenes is determined based on the relationship, including: Calculate the mutual information between the water, wind and light resources based on historical data, use the mutual information to determine the relationship between the water, wind and light resources, and obtain the mutual information value between the water, wind and light resources; Calculate the sum of the mutual information values ​​between any resource and other resources, obtain the comprehensive mutual information index of any resource, and determine the order of generating the three resource scenarios according to the comprehensive mutual information index.

5. The method for generating water-landscape scenes driven by multiple correlation joint driving according to claim 1, characterized in that: In S1, according to historical data, the existing variable information of each resource in period t-1 and period t is extracted as follows: ； ； ； Among them, DH is the multi-year historical data of water resource runoff h, DW is the multi-year historical data of wind resource wind power output w, and DP is the multi-year historical data of light resource photovoltaic output p; the existing variable information of the three resources of water, wind and light in the t-1 period and the t period are h t-1 With h t 、w t-1 With w t 、p t-1 With p t ; , N is determined according to the target time scale.

6. The method for generating water-landscape-light scenes by multiple correlation joint driving according to claim 5, characterized in that: In S2, a sample data set is established based on the existing variable information to optimize the Gaussian mixture model parameters corresponding to each resource, and the Gaussian mixture model after the parameter optimization corresponding to each resource is obtained, which specifically includes: For any resource, variables x1 and x2 change with the increase of time period t. The existing variable information of each time period t is used to optimize the parameters of the Gaussian mixture model, and N models with optimized parameters corresponding to any resource are obtained for scene generation of N time periods for any resource.

7. The method for generating water-landscape scene with multiple correlation joint drive according to any one of claims 1 to 6, characterized in that: The Gaussian mixture model in S2 is specifically: ； ； Where: is the two-dimensional joint distribution of variables x1 and x2, is the conditional distribution of variable x2 when variable x1 is known, is the binary probability density function, is the conditional probability density function of variable x2 when variable x1 is known; The bivariate probability density function , the mathematical expression is: ； Where: represents the binary Gaussian probability density function; K is the number of components of the Gaussian mixture model, which is a hyperparameter; is the weight of the kth component; are the mean vector and covariance matrix of the kth component respectively.

8. The method for generating water-landscape scenes driven by multiple correlation joint driving according to claim 7, characterized in that: In S2, the Gaussian mixture model parameters corresponding to each resource are optimized, specifically, the hyperparameters K, weights in the Gaussian mixture model are optimized. , mean vector and the covariance matrix Optimize and determine, the specific optimization steps include: Pre-set several different hyperparameter K values; For any preset K value, the sample data set is divided into K clusters using clustering methods; the center of each cluster is used as the mean vector of each component in the Gaussian mixture model ; Calculate the difference between the data points in each cluster and the cluster center to obtain the covariance matrix ; Calculate the proportion of the number of samples in each cluster to the total number of samples and obtain the weight ; The obtained parameters are used as initial values, and the expectation maximization algorithm is used to optimize the model parameters to obtain the mean vector, covariance matrix and weight parameter value of the model after optimization corresponding to any preset K value; The comprehensive strategy of Akaike information criterion and Bayesian information criterion is used to evaluate the Gaussian mixture model under different preset K values, and the preset K value corresponding to the Gaussian mixture model with the smallest comprehensive strategy value is selected as the optimal K value; The Kolmogorov-Smilov test was performed on the theoretical distribution of the Gaussian mixture model corresponding to the optimal K value and the empirical distribution of the sample data set, with a significance level of Set to 0.

2. If the test passes, the model training is completed. If the test fails, adjust the preset K value and re-optimize the model parameters.

9. The method for generating water-landscape-light scenes by multiple correlation joint driving according to claim 8, characterized in that: Akaike Information Criterion Specifically: ； Bayesian Information Criterion Specifically: ； Where: is the number of parameters of the Gaussian mixture model under K components, representing the complexity of the model; M is the amount of data in the sample data set; the comprehensive strategy value is the weighted sum of the Akaike information criterion and the Bayesian information criterion.

10. A water-landscape scene generation system driven by multiple correlations, characterized in that: The system includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method for generating water, landscape and light scenes driven by multiple correlation joint drives described in any one of claims 1 to 9 is executed.

Images