Scene generation method, system and device for power prediction error distribution and medium

Abstract

The invention discloses a scene generation method, system and device for power prediction error distribution and a medium, and the method comprises the steps: obtaining a power prediction value and an actual value, carrying out the calculation according to the power prediction value and the actual value, obtaining sample data, carrying out the non-parameter estimation of the power prediction error distribution through a first estimation method based on the sample data, and obtaining the power prediction error distribution. Obtaining a prediction error distribution result; sampling a power error scene by using a first sampling method based on the prediction error distribution result; and acquiring historical power data, estimating a power scene in combination with the power error scene, generating a future power scene, and better estimating the power.

Description

Technical Field

[0001] This invention relates to the field of power system technology, and in particular to a method, system, device and medium for generating power prediction error distribution scenarios. Background Technology

[0002] In recent years, new energy sources have developed rapidly, with the proportion of wind power installed capacity in the power grid continuously increasing. In regions rich in wind energy resources, wind power accounts for a high proportion of the power grid, and in some areas, the installed capacity of wind power even exceeds the local load level. Unlike conventional thermal power generating units, wind is affected by factors such as wind speed and direction, resulting in random, fluctuating, and intermittent power output. Large-scale grid connection of non-dispatchable new energy sources can adversely affect the safe, stable, and economical operation of the power system. When the proportion of new energy installed capacity in the power grid exceeds a certain value, it will seriously affect the safe and stable operation of the power system, even impacting conventional power generation methods and causing significant fluctuations in grid voltage and frequency. Therefore, it is necessary to conduct power prediction research to improve the dispatchability of wind power. Due to the randomness of wind farm output power, power prediction inevitably contains errors compared to the actual wind power output. Different wind power prediction technologies result in different magnitudes of prediction errors, and currently, there is no universally accepted power prediction technology with good prediction performance. The selection of wind power prediction error indicators is closely related to the accuracy of wind power prediction. Commonly used and effective wind power prediction error evaluation indicators include absolute error, average error, average relative error, and root mean square error.

[0003] The probability distribution of renewable energy power forecasting errors is a crucial basis for describing the accuracy of renewable energy resource forecasting, and many reliability assessment methods for power systems containing renewable energy rely on renewable energy probability distribution models. These models can be broadly classified into parametric and non-parametric models. Parametric models are simple to model, but require assumptions about the distribution of renewable energy output, making it difficult to guarantee universality. Non-parametric models, on the other hand, do not employ any prior knowledge or make any assumptions about the probability distribution of the samples; they purely fit the distribution based on the samples. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] This invention provides a method, system, device, and medium for generating power prediction error distribution scenarios to solve the problems of large power prediction errors and poor schedulability.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a method for generating a scene of power prediction error distribution, comprising:

[0008] Obtain the predicted power value and the actual power value, calculate sample data based on the predicted power value and the actual power value, and perform nonparametric estimation of the power prediction error distribution based on the sample data using the first estimation method to obtain the prediction error distribution result;

[0009] Based on the predicted error distribution results, the power error scenario is sampled using the first sampling method;

[0010] Historical power data is acquired and combined with the power error scenario to estimate the power scenario and generate the future power scenario.

[0011] As a preferred embodiment of the scene generation method for power prediction error distribution according to the present invention, the method for nonparametric estimation of power prediction error distribution using a first estimation method includes:

[0012] The error data between the predicted and actual power values ​​is used as sample data;

[0013] Calculate the cumulative distribution function and probability density function of the sample data, and select the first kernel function;

[0014] Calculate the cumulative distribution function and probability density function of the sample data;

[0015] Calculate the kernel density estimate based on the cumulative distribution function, probability density function, and first kernel function of the sample data;

[0016] The power prediction error distribution is estimated using kernel density estimation.

[0017] The beneficial effect of this preferred technical solution is that, by using a non-parametric estimation method, the accuracy and adaptability of power prediction error estimation are improved without the need to preset the error distribution form.

[0018] As a preferred embodiment of the scene generation method for power prediction error distribution described in this invention, the calculation of kernel density estimation includes:

[0019] The kernel density estimation is achieved by convolving the empirical distribution function with the first kernel function, and the smoothness of the estimation is controlled by the first parameter optimization operation.

[0020] As a preferred embodiment of the power prediction error distribution scene generation method described in this invention, the method of sampling the power error scene using a first sampling method includes:

[0021] Obtain the power output scenario data dimension, power output scenario data, and number of power output scenarios, and obtain the cumulative probability distribution function corresponding to the power output scenario data dimension;

[0022] Based on the number of power output scenarios and the cumulative probability distribution function, the second sampling method is used to obtain the sampled values ​​of the power output scenario data;

[0023] Based on the sampled values, the random output values ​​are randomly arranged into a row in the matrix to obtain the sampling matrix.

[0024] The beneficial effect of this preferred technical solution is that it efficiently generates comprehensive power error scenarios through scientific sampling methods, thereby enhancing the representativeness of the prediction scenarios.

[0025] As a preferred embodiment of the scene generation method for power prediction error distribution described in this invention, it further includes:

[0026] Sort the positions of the elements in the sampling matrix to obtain the permutation matrix;

[0027] Sort the positions of each element in the permutation matrix to obtain the initial permutation matrix.

[0028] As a preferred embodiment of the scene generation method for power prediction error distribution described in this invention, sorting the position of each element in the permutation matrix includes:

[0029] The correlation coefficient matrix is ​​obtained by calculating the correlation coefficient between each row of the initial permutation matrix using the covariance and standard deviation of each row.

[0030] The correlation coefficient matrix is ​​decomposed into the product of a lower triangular matrix and its corresponding transpose.

[0031] As a preferred embodiment of the scene generation method for power prediction error distribution described in this invention, it further includes:

[0032] A new matrix with reduced column correlation is calculated based on the lower triangular matrix and the initial permutation matrix;

[0033] Rearrange the elements of each row of the initial permutation matrix according to the order of the elements in each row of the new matrix;

[0034] Arrange the sampled values ​​in the sampling matrix according to the element order of the rearranged initial permutation matrix;

[0035] Repeat the steps of correlation coefficient calculation, matrix decomposition and rearrangement until the row correlation coefficient of the sampling matrix meets the preset correlation requirements to obtain the power error scenario;

[0036] By combining historical power data with power error scenarios, power scenarios are estimated to generate future power scenarios.

[0037] Secondly, the present invention provides a scene generation system for power prediction error distribution, comprising:

[0038] The estimation module is used to obtain the predicted power value and the actual power value, calculate sample data based on the predicted power value and the actual power value, and perform nonparametric estimation of the power prediction error distribution based on the sample data using the first estimation method to obtain the prediction error distribution result.

[0039] The sampling module is used to sample the power error scenario based on the prediction error distribution result using a first sampling method.

[0040] The scene generation module is used to acquire historical power data and, in conjunction with the power error scene, estimate the power scene to generate the future power scene.

[0041] Thirdly, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, characterized in that the processor executes the computer program to implement the steps of the scene generation method for a power prediction error distribution.

[0042] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that, when the computer program is executed by a processor, it implements the steps of the scene generation method for power prediction error distribution.

[0043] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention estimates the power prediction error distribution based on kernel density estimation, obtaining the prediction error distribution result; based on the prediction error distribution result, Latin hypercube sampling is used to extract power error scenarios; based on the power error scenarios and historical data, the power scenarios are estimated to generate future scenarios. This invention overcomes the technical barriers of traditional power prediction methods, taking a different approach. Instead of directly using historical data features for power estimation, it constructs power scenarios for a future period through the estimation of power prediction errors, forming a power prediction for the future period, thus providing a better estimate of power. Attached Figure Description

[0044] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 A schematic diagram of the overall process logic of a scene generation method for power prediction error distribution provided in an embodiment of the present invention;

[0046] Figure 2This is a schematic diagram illustrating the technical concept of kernel density estimation in a scene generation method for power prediction error distribution according to an embodiment of the present invention. Detailed Implementation

[0047] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0048] Example 1, referring to Figures 1-2 As an embodiment of the present invention, a method for generating a scene of power prediction error distribution is provided, comprising:

[0049] S100: Obtain the predicted and actual power values, calculate sample data based on the predicted and actual power values, and use the first estimation method to perform nonparametric estimation of the power prediction error distribution based on the sample data to obtain the prediction error distribution result;

[0050] In an optional embodiment, the first estimation method can be K-nearest neighbor estimation, which determines the number of nearest neighbor points K, calculates the distance to all points in the sample data for each target point, finds the K sample points closest to the target point, and estimates the probability density of the target point based on the distribution of these K nearest neighbor points.

[0051] In another alternative embodiment, the first estimation method may also be local weighted regression, which determines the bandwidth parameter of local weighted regression, calculates the distance between each target point and all points in the sample data, assigns different weights according to the distance, fits a weighted regression model in the local region, estimates the regression coefficients using weighted least squares, and estimates the probability density of the target point based on the fitted regression model.

[0052] In this embodiment of the invention, the first estimation method includes kernel density estimation;

[0053] The first kernel function is obtained based on the cumulative distribution function and probability density function of the sample data;

[0054] Based on the first kernel function, the kernel density estimate is obtained, and based on the kernel density estimate, the power prediction error distribution is estimated.

[0055] It should be noted that kernel density estimation eliminates the need to pre-define the error distribution, accurately characterizing the power prediction error distribution and improving prediction accuracy.

[0056] S200: Based on the prediction error distribution results, the power error scenario is sampled using the first sampling method;

[0057] In an optional embodiment, the first sampling method can be Monte Carlo sampling. Based on the power prediction error distribution result, the cumulative probability distribution function is determined, a set of uniformly distributed random numbers is generated, and the sampled value corresponding to each random number is calculated through the inverse function of the cumulative probability distribution function. All sampled values ​​are combined into a matrix to obtain the sampling matrix.

[0058] In another alternative embodiment, the first sampling method can also be stratified sampling. Based on the power prediction error distribution results, the error range is divided into several layers. For each layer, the cumulative probability distribution function is calculated. Within each layer, uniformly distributed random numbers are generated, and the corresponding sampled values ​​are calculated. The sampled values ​​of each layer are combined into a matrix, with each row representing a power error scenario.

[0059] In this invention, the first sampling method includes Latin hypercube sampling;

[0060] Specifically, Latin hypercube sampling is a stratified random sampling method that can efficiently sample from the distribution range of variables. Suppose we have k variables x1, x2, x3, ..., x... k Now, if we want to take N samples from a specified interval, the cumulative distribution of each variable is divided into the same N small intervals. A value is randomly selected from each interval, and the N values ​​of each variable are randomly combined with the values ​​of other variables, which is different from random sampling.

[0061] It should be noted that by maximizing the layering of each edge distribution, full coverage of the range of each variable can be guaranteed.

[0062] S300: Acquire historical power data and combine it with power error scenarios to estimate power scenarios and generate future power scenarios.

[0063] It should be noted that the nonparametric estimation method accurately captures the power prediction error distribution without requiring a pre-defined error distribution form, thus improving the adaptability and accuracy of the estimation. Scientific sampling methods are used to efficiently generate comprehensive power error scenarios, enhancing the representativeness of the prediction scenarios. By combining historical power data, future power scenarios are generated, providing a scientific basis for power system scheduling and operation, effectively addressing the randomness, volatility, and intermittency of power, improving the dispatchability of wind power, and ensuring the safe and stable operation of the power system.

[0064] In this embodiment of the invention, step S100 includes the following sub-steps A1-A3;

[0065] In A1: The error data between the predicted power value and the actual power value is used as sample data;

[0066] In A2: Calculate the cumulative distribution function and probability density function of the sample data, and select the first kernel function;

[0067] In A3: Calculate the kernel density estimate based on the cumulative distribution function and probability density function of the sample data, as well as the first kernel function;

[0068] In A4: The power prediction error distribution is estimated using kernel density estimation.

[0069] In one optional embodiment, the first kernel function can be a uniform kernel function, which maintains a constant value within a certain interval and is zero outside of that interval. When the absolute value of the sample data is less than or equal to 1, the kernel function is: When the absolute value of the sample data is greater than 1, the kernel function is 0;

[0070] In another optional embodiment, the first kernel function can also be a triangular kernel function, which decreases linearly with the increase of distance. When the absolute value of the sample data is less than or equal to 1, the kernel function is 1 minus the absolute value of the sample data, and when the sample data is greater than 1, the kernel function is 0.

[0071] In this embodiment of the invention, the first kernel function includes a Gaussian function;

[0072] Specifically, the kernel function is represented as:

[0073]

[0074] Where K(x) is the kernel function and x is the sample data.

[0075] The kernel function is the probability density distribution function, and the kernel function estimate is the probability density estimate of the sample set. Each sample data in the sample set corresponds to a probability density function. The probability density estimate of the sample set is obtained by averaging the probability density functions of all samples in the sample set.

[0076] It should be noted that the kernel density model is a typical nonparametric model. Directly predicting power makes it difficult to estimate power from the perspective of error. Predicting power based on the prediction error can provide a more comprehensive estimate of power.

[0077] Nonparametric estimation methods offer better fitting results than classical hypothetical models and provide higher accuracy in inferring past production phenomena. The parameter values ​​or number of parameters in nonparametric estimation change with the original data. Kernel density estimation is a data-driven method for probability density estimation that requires no prior knowledge of the distribution or any assumptions about the probability distribution form, instead focusing on the characteristics of the data distribution from the data sample itself.

[0078] In this embodiment of the invention, after completing steps A1-A3, step S100 further includes step A4;

[0079] In A4: the kernel density estimation is achieved by convolving the empirical distribution function with the first kernel function, and the smoothness of the estimation is controlled by the first parameter optimization operation.

[0080] In one optional embodiment, the first parameter optimization can be to optimize the sample weights by assigning initial weights to each sample data, selecting an optimization objective function, using an optimization algorithm to adjust the sample weights to minimize the optimization objective function, verifying the effect of the optimized weights through methods such as cross-validation, and adjusting them as needed.

[0081] In another optional embodiment, the first parameter optimization can also be the shape parameter of the optimization kernel function. A kernel function with shape parameters is selected, an optimization objective function is selected, the shape parameter of the kernel function is adjusted using an optimization algorithm to minimize the optimization objective function, the effect of the optimized shape parameter is verified by methods such as cross-validation, and adjustments are made as needed.

[0082] In this embodiment of the invention, the first parameter optimization includes adjusting the bandwidth parameter;

[0083] Specifically, such as Figure 2 As shown, if a sample is observed to exist in the sample set, then this sample can be considered a maximum point of probability density. Numbers closer to this sample have a higher probability of appearing, while numbers farther away from the sample have a lower probability of appearing. This relationship can be expressed using the probability density function of a normal distribution, which is called the kernel function. If a similar probability density function is formed for each sample, then the estimate of the probability density of the entire sample set is the average of all these probability density functions.

[0084] There are n sample data, using x1, x2, x3, ..., x n express;

[0085] Use separately The cumulative distribution function and probability density function of the sample data are expressed as follows:

[0086]

[0087] Where x represents the sample data and h represents the bandwidth;

[0088] make Let the empirical distribution function be expressed as:

[0089]

[0090] in, Let x1, x2, x3, ..., x n The number of elements less than t;

[0091] F n Substituting (t) into the equation, we get:

[0092]

[0093] When the bandwidth is a constant value Transformed into:

[0094]

[0095] The kernel function K(t) is expressed as:

[0096]

[0097] And it satisfies:

[0098]

[0099] Where K is the kernel function, and has the following properties:

[0100] ∫K(x)dx=1∫xK(x)dx=0∫x 2 K(x)dx≥0

[0101] By relaxing the kernel function, a general kernel density estimate is obtained:

[0102]

[0103] in, To select a symmetric unimodal probability density function centered at 0.

[0104] The kernel function is a continuous bounded even function with an integral of 1. Once the sample is determined, the kernel function and bandwidth jointly determine the quality of the nonparametric kernel density estimate. Common kernel functions include uniform functions, Gaussian functions, trigonometric functions, etc. The choice of kernel function is usually not the most critical factor in density estimation, because any kernel function can guarantee stable consistency of the density estimate, as long as it is symmetric and unimodal. It can be proven that when the bandwidth coefficient h is optimally chosen, different kernel functions are equivalent. Therefore, any kernel function that meets certain conditions is generally considered suitable. This project uses the Gaussian function as the kernel function; the Gaussian function is the standard normal distribution function.

[0105] It should be noted that nonparametric estimation methods have better fitting effects than classical hypothetical models and higher accuracy in inferring past production activities. The parameter values ​​or number of parameters in nonparametric estimation will change with the original data. Kernel density estimation is a data-driven method for probability density estimation that does not require any prior knowledge of the distribution or any assumptions about the probability distribution form, but rather studies the distribution characteristics of the data sample itself.

[0106] In this embodiment of the invention, step S200 includes the following sub-steps B1-B3;

[0107] In B1: Obtain the output scenario data dimension, output scenario data, and number of output scenarios to obtain the cumulative probability distribution function corresponding to the output scenario data dimension;

[0108] In B2: Based on the number of power output scenarios and the cumulative probability distribution function, the second sampling method is used to obtain the sampled values ​​of the power output scenario data;

[0109] In B3: Based on the sampled values, the random output values ​​are randomly arranged into a row in the matrix to obtain the sampling matrix.

[0110] In this embodiment of the invention, the second sampling method is an inverse function;

[0111] Specifically, the cumulative probability distribution function corresponding to the data dimension of the output scenario is expressed as:

[0112] Y k =F k (X k )

[0113] Among them, Y k X is the cumulative probability distribution function corresponding to the data dimension of the output scenario. k (k = 1, 2, ..., K) represents the output scenario data for any one of the power levels, and K represents the dimension of the output scenario data for each required power level.

[0114] If the number of scenarios requiring power output is N, the graph of the cumulative probability distribution function is divided into N intervals, with the N intervals arranged consecutively and without overlap, and all intervals having an equal length of 1 / N. The midpoint value of each of the divided intervals is taken as the corresponding interval Y. k And X is derived using the inverse function. k The value of

[0115] The sampled values ​​of the output scenario data are represented as follows:

[0116]

[0117] After sampling for each output scenario, the random output values ​​are randomly arranged into a row in the matrix. The resulting sampled values ​​are a K×N matrix, represented as follows:

[0118]

[0119] Wherein, the sampling matrix R KN Each column represents a set of sampled values ​​from K variables, and the sampling matrix R KN Each row represents N sample values ​​of a certain variable, and they are all arranged in ascending order. K is the output scenario data dimension for each power, and N is the number of output scenarios required.

[0120] It should be noted that by scientifically dividing the cumulative probability distribution interval using the inverse function method, the sampled values ​​of the power output scenario data can be accurately obtained, and a structured sampling matrix can be generated, effectively improving the accuracy and representativeness of power error scenario sampling.

[0121] In this embodiment of the invention, after completing steps B1-B3, step S200 also includes steps B4-B5.

[0122] In B4: Sort the positions of the elements in the sampling matrix to obtain the permutation matrix;

[0123] In B5: Sort the positions of each element in the permutation matrix to obtain the initial permutation matrix.

[0124] In one alternative embodiment, the sorting can be random sorting, simulated annealing algorithm, genetic algorithm, Cholesky decomposition method, Gram-Schmidt sequence orthogonalization method, etc.

[0125] The random sorting process includes randomly shuffling each row of the sampling matrix to ensure that the order of elements in each row is randomized, calculating the inter-row correlation coefficient of the shuffled sampling matrix, and checking whether it meets the set correlation requirements. If the inter-row correlation does not meet the requirements, the above steps are repeated until the requirements are met.

[0126] In another alternative embodiment, the sorting can also be a simulated annealing algorithm, setting an initial temperature and cooling rate, randomly selecting two rows to swap, generating a new permutation matrix, calculating the inter-row correlation coefficient matrix of the new permutation matrix, and calculating the objective function value. If the objective function value of the new state is better than the current state, the new state is accepted; if the objective function value of the new state is not better than the current state, the new state is accepted with a probability value. The temperature is multiplied by the cooling rate, and the steps are repeated until the temperature drops to a very small value or the preset number of iterations is reached.

[0127] In this embodiment of the invention, the sorting includes the Cholesky decomposition method, which is the most computationally efficient sorting method.

[0128] Specifically, the positions of each element in the sampling matrix are sorted using the Cholesky decomposition method, constructing a permutation matrix Q with dimension K×N. KN Permutation matrix Q KN The element in the k-th row (1≤k≤K) represents the permutation matrix Q. KN The position of the N sampled values ​​of the k-th variable.

[0129] In this embodiment of the invention, after completing steps B4-B5, step S200 also includes steps B6-B12.

[0130] In B6: The correlation coefficient matrix is ​​obtained by calculating the correlation coefficient between each row of the initial permutation matrix using the covariance and standard deviation of each row;

[0131] In B7: The correlation coefficient matrix is ​​decomposed into the product of a lower triangular matrix and its corresponding transpose.

[0132] In B8: A new matrix with reduced column correlation is calculated based on the lower triangular matrix and the initial permutation matrix;

[0133] In B9: The elements of each row of the initial permutation matrix are rearranged according to the order of the elements in each row of the new matrix;

[0134] In B10: The sampled values ​​in the sampling matrix are arranged according to the element order of the rearranged initial permutation matrix;

[0135] In B11: Repeat the steps of correlation coefficient calculation, matrix decomposition and rearrangement until the row correlation coefficient of the sampling matrix meets the preset correlation requirements to obtain the power error scenario;

[0136] In B12: Historical power data is combined with power error scenarios to estimate power scenarios and generate future power scenarios.

[0137] Specifically, the Cholesky decomposition method is used to decompose the permutation matrix Q. KN Sort the positions of each element to generate an initial permutation matrix Q = [Q1, Q2, ..., QN] of dimension K×N. K ] T Each row of the initial permutation matrix is ​​a random permutation of integers 1, ..., N;

[0138] Calculate the correlation coefficients between each row of the initial permutation matrix Q to obtain the K×K dimensional correlation coefficient matrix P. KK , is represented as:

[0139] P KK ={ρ ij |i=1,2,…,Kandj=1,2,…,K}

[0140] Where, ρ ij Let matrix R KN The correlation coefficient between the i-th row and the j-th row;

[0141] The correlation coefficient is expressed as:

[0142]

[0143] Using the correlation coefficient matrix P KK The root mean square (RMS) is used to measure the correlation between different sampling matrices, expressed as:

[0144]

[0145] Correlation coefficient matrix P KK P is a symmetric positive definite matrix. Using the Cholesky decomposition method, P... KK Decomposed into a lower triangular matrix L and its transpose L T The product of these can be expressed as:

[0146] P KK =LL T

[0147] Based on the correlation coefficient matrix and the initial permutation matrix, the matrix with the minimum column correlation is obtained, represented as follows:

[0148] Q KN =L -1 Q

[0149] Matrix Q KN The correlation coefficient matrix between rows is a K×K dimensional identity matrix, indicating that matrix Q... KN There is no correlation between the rows of matrix Q. KN The elements in the sample matrix R are not necessarily all positive integers, so the sample matrix R cannot be directly processed. KN The sampled values ​​in the matrix Q need to be arranged, therefore it is necessary to do so based on the matrix Q. KN The elements of each row in the initial permutation matrix Q are arranged in ascending order. Based on this, the sampling matrix R is then processed according to the element order of the initial permutation matrix Q. KN Arrange the sampled values. Repeat the above steps until the sampling matrix R is obtained. KN row correlation ρ rms It meets the preset correlation requirement, that is, it is less than 0.8.

[0150] Kernel density estimation is used to estimate the power prediction error distribution. Then, Latin hypercube sampling is used to extract power error scenarios based on the obtained prediction error distribution results. These scenarios are then used to generate future scenarios based on historical data and the error distribution. By defining scenarios from the perspective of error, the prediction results are made more accurate.

[0151] It should be noted that optimizing the permutation matrix using the Cholesky decomposition method reduces inter-row correlation in the sampling matrix and improves sample independence. Combining kernel density estimation and Latin hypercube sampling, the scenario is defined from an error perspective, significantly improving the accuracy and reliability of power prediction.

[0152] The above is an illustrative scheme of a scene generation method for power prediction error distribution according to this embodiment. It should be noted that the technical solution of this scene generation system for power prediction error distribution belongs to the same concept as the technical solution of the scene generation method for power prediction error distribution described above. For details not described in detail in the technical solution of the scene generation system for power prediction error distribution in this embodiment, please refer to the description of the technical solution of the scene generation method for power prediction error distribution described above.

[0153] The scene generation system for power prediction error distribution in this embodiment includes:

[0154] The estimation module is used to obtain the predicted power value and the actual power value, calculate sample data based on the predicted power value and the actual power value, and perform nonparametric estimation of the power prediction error distribution based on the sample data using the first estimation method to obtain the prediction error distribution result.

[0155] The sampling module is used to sample the power error scenario based on the prediction error distribution result using a first sampling method.

[0156] The scene generation module is used to acquire historical power data and, in conjunction with the power error scene, estimate the power scene to generate the future power scene.

[0157] This embodiment also provides a computer device suitable for generating scenarios of power prediction error distribution, including:

[0158] The system includes a memory and a processor. The memory stores computer-executable instructions, and the processor executes these instructions to implement a method for generating a power prediction error distribution scenario as proposed in the above embodiments.

[0159] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a method for generating a power prediction error distribution scenario as proposed in the above embodiments.

[0160] The storage medium proposed in this embodiment and the scene generation method for realizing power prediction error distribution proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0161] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computing device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0162] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for generating a scene of power prediction error distribution, characterized in that, include: Obtain the predicted power value and the actual power value, calculate sample data based on the predicted power value and the actual power value, and perform nonparametric estimation of the power prediction error distribution based on the sample data using the first estimation method to obtain the prediction error distribution result; Based on the predicted error distribution results, the power error scenario is sampled using the first sampling method; Historical power data is acquired and combined with the power error scenario to estimate the power scenario and generate the future power scenario.

2. The method for generating a scene of power prediction error distribution as described in claim 1, characterized in that, The nonparametric estimation of the power prediction error distribution using the first estimation method includes: The error data between the predicted and actual power values ​​is used as sample data; Calculate the cumulative distribution function and probability density function of the sample data, and select the first kernel function; Calculate the kernel density estimate based on the cumulative distribution function, probability density function, and first kernel function of the sample data; The power prediction error distribution is estimated using kernel density estimation.

3. The method for generating a scene of power prediction error distribution as described in claim 1 or 2, characterized in that, The calculation of kernel density estimation includes: The kernel density estimation is achieved by convolving the empirical distribution function with the first kernel function, and the smoothness of the estimation is controlled by the first parameter optimization operation.

4. The method for generating a scene of power prediction error distribution as described in claim 3, characterized in that, Sampling for power error scenarios using the first sampling method includes: Obtain the power output scenario data dimension, power output scenario data, and number of power output scenarios, and obtain the cumulative probability distribution function corresponding to the power output scenario data dimension; Based on the number of power output scenarios and the cumulative probability distribution function, the second sampling method is used to obtain the sampled values ​​of the power output scenario data; Based on the sampled values, the random output values ​​are randomly arranged into a row in the matrix to obtain the sampling matrix.

5. The method for generating a scene of power prediction error distribution as described in claim 4, characterized in that, Also includes: Sort the positions of the elements in the sampling matrix to obtain the permutation matrix; Sort the positions of each element in the permutation matrix to obtain the initial permutation matrix.

6. The method for generating a scene of power prediction error distribution as described in claim 5, characterized in that, Sort the position of each element in the permutation matrix as follows: The correlation coefficient matrix is ​​obtained by calculating the correlation coefficient between each row of the initial permutation matrix using the covariance and standard deviation of each row. The correlation coefficient matrix is ​​decomposed into the product of a lower triangular matrix and its corresponding transpose.

7. The method for generating a scene of power prediction error distribution as described in claim 6, characterized in that, Also includes: A new matrix with reduced column correlation is calculated based on the lower triangular matrix and the initial permutation matrix; Rearrange the elements of each row of the initial permutation matrix according to the order of the elements in each row of the new matrix; Arrange the sampled values ​​in the sampling matrix according to the element order of the rearranged initial permutation matrix; Repeat the steps of correlation coefficient calculation, matrix decomposition and rearrangement until the row correlation coefficient of the sampling matrix meets the preset correlation requirements to obtain the power error scenario; By combining historical power data with power error scenarios, power scenarios are estimated to generate future power scenarios.

8. A scene generation system for power prediction error distribution, employing the scene generation method for power prediction error distribution as described in any one of claims 1-7, characterized in that, include: The estimation module is used to obtain the predicted power value and the actual power value, calculate sample data based on the predicted power value and the actual power value, and perform nonparametric estimation of the power prediction error distribution based on the sample data using the first estimation method to obtain the prediction error distribution result. The sampling module is used to sample the power error scenario based on the prediction error distribution result using a first sampling method. The scene generation module is used to acquire historical power data and, in conjunction with the power error scene, estimate the power scene to generate the future power scene.

9. A computer device, characterized in that, include: A memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of a scene generation method for a power prediction error distribution according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores a computer program, which, when executed by a processor, implements the steps of a scene generation method for a power prediction error distribution according to any one of claims 1 to 7.

Images