CN-121981855-A - ARMA-Copula-based source load uncertainty scene generation method

CN121981855ACN 121981855 ACN121981855 ACN 121981855ACN-121981855-A

Abstract

The invention relates to a source load uncertainty scene generation method based on ARMA-Copula, belongs to the technical field of comprehensive energy system planning, and solves the problem that the scene generation deviation is large due to the fact that the existing method is difficult to accurately describe the time sequence fluctuation and the space cross correlation of renewable energy output and multiple loads. The method comprises the steps of calculating edge distribution parameters at each moment by using maximum likelihood estimation, mapping the edge distribution parameters to a standard normal space by probability integral transformation, constructing an autoregressive moving average model, stripping time autocorrelation to extract white noise residual errors, decomposing and reconstructing a spatial cross-correlation structure of the residual errors based on a Gaussian Copula function and Cholesky, generating an initial scene set by forward recursion and inverse transformation of the model, and finally carrying out double-objective optimization reduction by adopting an improved non-dominant ranking genetic algorithm to screen an optimal typical scene. The invention can restore probability distribution and space-time coupling characteristics of source load data with high precision.

Inventors

LI XIANG
LI GUOQUAN
ZHU HONGYU
LIN JINCHAO
PANG YU

Assignees

重庆邮电大学

Dates

Publication Date: 20260505
Application Date: 20260122

Claims (9)

1. A source load uncertainty scene generation method based on ARMA-Copula is characterized by comprising the following steps: the data preprocessing and clustering step, carrying out normalization processing on the source load historical data and dividing a typical daily data set with different climate characteristics by adopting a clustering algorithm; a distribution parameter estimation and normal transformation step, namely calculating edge probability distribution parameters of various source charges at different moments by using a maximum likelihood estimation MLE and mapping physical space data into a standard normal time sequence by probability integral transformation PIT; A time sequence autocorrelation decoupling step, namely constructing an autoregressive moving average ARMA model under a standard normal space, and stripping time autocorrelation components by utilizing a reverse filtering technology to extract independent white noise residual sequences with the same distribution; space cross-correlation reconstruction and scene synthesis, namely performing Cholesky decomposition on a covariance matrix of a residual sequence based on a Gaussian Copula theory to reconstruct a space cross-correlation structure and generate an associated residual matrix, and restoring physical characteristics through forward recursion and inverse probability integral transformation of an ARMA model to generate an initial scene set; and a double-target scene optimization step of establishing a double-target optimization model for minimizing the error and the probability distribution distance of the spatial cross-correlation matrix and adopting an improved non-dominant ranking genetic algorithm NSGA-II to perform scene reduction and optimization so as to screen out an optimal typical scene set.
2. The ARMA-Copula-based source load uncertainty scene generating method as claimed in claim 1, wherein in the data preprocessing and clustering step, a K-means clustering algorithm is adopted and the optimal clustering number is combined with Davies-Bouldin index DBI to divide a typical daily data set.
3. The ARMA-Copula-based source load uncertainty scene generating method according to claim 1, wherein in the distribution parameter estimation and normal transformation steps, the edge probability distribution parameters are calculated by adopting Weibull distribution for wind power output, beta distribution for photovoltaic output and normal distribution for load, and data are mapped to a standard normal space through probability integral transformation.
4. The method for generating a source load uncertainty scene based on ARMA-Copula as claimed in claim 1, wherein in the step of time sequence autocorrelation decoupling, the ARMA model is expressed as Wherein Is the first The standard normal sequence value of the class source load at the time t, As a result of the autoregressive coefficients, In order to be a coefficient of a sliding average, And The number of the model orders is respectively that of the model, Is a white noise residual.
5. The ARMA-Copula-based source load uncertainty scene generation method as claimed in claim 1, wherein in the spatial cross-correlation reconstruction and scene synthesis steps, a residual covariance matrix is decomposed by Cholesky Obtaining a lower triangular matrix And using linear transformation Generating an associated residual matrix, wherein In order to simulate the residual matrix, Is an independent standard normal random matrix.
6. The ARMA-Copula-based source load uncertainty scene generating method as claimed in claim 1, wherein in the dual-objective scene optimization step, the objective function of the dual-objective optimization model includes Frobenius norm error of minimizing cross-correlation coefficient matrix And Wasserstein distance to minimize probability distribution Wherein For the matrix of historical cross-correlation coefficients, For a matrix of scene subset cross-correlation coefficients, For the distribution of the experience of the history, Is empirically distributed for a subset of scenes.
7. The ARMA-Copula-based source load uncertainty scene generation method as claimed in claim 1, wherein the improved NSGA-II algorithm adopts fixed length coding and greedy repair strategy, wherein the cardinality constraint is satisfied when initializing population , For binary coded vectors, S is the initial scene total number, and K is the target cut scene number.
8. The ARMA-Copula-based source load uncertainty scene generation method of claim 7, wherein the greedy repair strategy eliminates scenes which contribute worst to probability distribution distance margin when the scene number exceeds K, and adds scenes which minimize probability distribution distance when the scene number is less than K.
9. The method for generating a source load uncertainty scene based on ARMA-Copula according to claim 1, wherein the dual-target scene optimization step finally selects a scene set with minimum weighted Euclidean distance as output through pareto front screening and adopting an ideal point decision method.

Description

ARMA-Copula-based source load uncertainty scene generation method Technical Field The invention belongs to the technical field of comprehensive energy system planning, and relates to a source load uncertainty scene generation method based on ARMA-Copula. Background The regional comprehensive energy system realizes the efficient interconnection and mutual utilization of energy through the coupling utilization of multiple energy sources such as electricity, gas, cold, heat and the like. In medium-short term planning, the running state of the system is greatly influenced by renewable energy output such as wind power, photovoltaic and the like and multi-element load fluctuation. However, due to the significant randomness, volatility, and complex multidimensional coupling characteristics of these source-to-load variables, how to obtain high-precision multi-energy-source-to-load scene simulation results becomes a key challenge. In planning scenarios, renewable energy output and user load exhibit complex time-series variations and correlation features. Existing conventional scene generation algorithms for single energy or deterministic planning cannot accurately describe these dynamic changes. The probability distribution model method is widely used for describing source load uncertainty, and the uncertainty analysis is reconstructed into a parameter fitting and random sampling problem. The traditional Monte Carlo simulation (MonteCarlo Sampling, MCS) or autoregressive moving average (Auto-REGRESSIVE MOVING AVERAGE, ARMA) model is widely applied, but often ignores the time sequence difference of probability distribution parameters at different moments, and has difficulty in simultaneously considering the autocorrelation inside source load data and the cross correlation between different source loads. The lack of the analysis of the correlation influence mechanism leads to the fact that the generated scene set is similar in statistical distribution, but has larger deviation from the historical actual data in the aspects of time sequence fluctuation law and multi-energy coupling characteristics, the phenomenon of under fitting or over fitting is easy to occur, and the requirement of fine planning under high-proportion renewable energy access cannot be met. In addition, most of the existing scene reduction technologies are based on geometric distance clustering, and lack a targeted optimization mechanism for maintaining cross-correlation features in generated scenes. In summary, in the prior art, it is difficult to simultaneously and precisely reserve the time sequence autocorrelation characteristic and the spatial cross correlation structure of the source-load variables in the generated scene, which restricts the reliability of the regional comprehensive energy system planning scheme. Therefore, a new method capable of jointly modeling spatio-temporal correlations and achieving a fine scene cut is urgently needed. Disclosure of Invention In view of the above, the present invention aims to provide a source load uncertainty scene generation method based on ARMA-Copula. In order to achieve the above purpose, the present invention provides the following technical solutions: A source load uncertainty scene generation method based on ARMA-Copula comprises the following steps: the data preprocessing and clustering step, carrying out normalization processing on the source load historical data and dividing a typical daily data set with different climate characteristics by adopting a clustering algorithm; A distribution parameter estimation and normal transformation step, namely calculating edge probability distribution parameters of various source charges at different moments by using maximum likelihood estimation (Maximum Likelihood Estimation, MLE) and mapping physical space data into a standard normal time sequence by probability integral transformation (Probability Integral Transform, PIT); A time sequence autocorrelation decoupling step, namely constructing an autoregressive moving average (Auto-REGRESSIVE MOVING AVERAGE, ARMA) model under a standard normal space and stripping time autocorrelation components by utilizing a reverse filtering technology to extract independent white noise residual sequences with the same distribution; space cross-correlation reconstruction and scene synthesis, namely performing Cholesky decomposition on a covariance matrix of a residual sequence based on a Gaussian Copula theory to reconstruct a space cross-correlation structure and generate an associated residual matrix, and restoring physical characteristics through forward recursion and inverse probability integral transformation of an ARMA model to generate an initial scene set; and a double-objective scene optimization step of establishing a double-objective optimization model for minimizing the error and the probability distribution distance of the spatial cross-correlation matrix and adopting an improved Non-dominant ranking genetic