CN-121997742-A - Wind field data joint distribution modeling method and system based on standardized flow model

CN121997742ACN 121997742 ACN121997742 ACN 121997742ACN-121997742-A

Abstract

The application discloses a wind field data joint distribution modeling method and system based on a standardized flow model, which relate to the field of wind engineering and artificial intelligence, and specifically comprise the steps of acquiring wind field monitoring data and preprocessing; the method comprises the steps of constructing a standardized flow model frame, establishing bidirectional reversible mapping between an original data space and a hidden variable space through a training module, training the standardized flow model frame by utilizing preprocessed data to obtain a trained standardized flow model, and calculating multidimensional joint distribution probability density of any point in the original data space through forward transformation and variable transformation formulas of the bidirectional reversible mapping based on the standardized flow model. According to the method, the related structures among the variables are not needed to be assumed in advance, the inherent probability dependence characteristic is directly learned based on the actual monitoring data, the characteristic capability of the joint probability model on the random characteristic of the actual wind field is greatly improved, and a more accurate uncertainty quantification basis is provided for the subsequent wind load and engineering structure reliability analysis.

Inventors

ZHAO ZHAO
XU TENGFEI
WAN QI
LI PEIPEI

Assignees

西南交通大学

Dates

Publication Date: 20260508
Application Date: 20260127

Claims (8)

1. A wind field data joint distribution modeling method based on a standardized flow model is characterized by comprising the following specific steps: acquiring wind field monitoring data, and preprocessing the wind field monitoring data; Constructing a standardized flow model framework, wherein the standardized flow model framework establishes bidirectional reversible mapping between an original data space and a hidden variable space through a training module, the original data space is used for representing real multidimensional joint distribution of wind field monitoring data, the priori distribution of the hidden variable space is assumed to be normal distribution, the training module comprises n layers of coupling layers, and each layer of coupling layer comprises an affine coupling block and a random substitution layer; Training the standardized flow model framework by utilizing the preprocessing data to obtain a trained standardized flow model; Based on the normalized flow model, a sample of the original data space is generated by sampling from the hidden variable space and inverse transforming through the bidirectional reversible mapping, and/or a multi-dimensional joint distribution probability density of any point in the original data space is calculated by a forward transformation of the bidirectional reversible mapping and a variable transformation formula.
2. The wind field data joint distribution modeling method based on the standardized flow model of claim 1, wherein the training step of the standardized flow model framework is as follows: acquiring and preprocessing the wind field monitoring data to obtain an original training data matrix; And constructing the standardized flow model framework, training the standardized flow model framework by utilizing the original training data matrix, and optimizing model parameters based on a loss function to obtain the trained standardized flow model.
3. The wind field data joint distribution modeling method based on the standardized flow model as claimed in claim 2, wherein the step of obtaining the original training data matrix is: Acquiring real-time wind field monitoring data; Preprocessing the wind field monitoring data, and converting the preprocessed data with uniform format into a data matrix to obtain the original training data matrix.
4. The wind field data joint distribution modeling method based on the standardized flow model of claim 1, wherein the training module comprises a plurality of affine coupling blocks and random substitution layers which are sequentially connected; Dividing input data into first partial data and second partial data, performing direct copy operation on the first partial data, and performing reversible affine transformation operation on the second partial data; The random permutation layer is configured with a random scrambling of the dimensional order of the output data of the affine coupling block.
5. The wind field data joint distribution modeling method based on the standardized flow model of claim 4, wherein the expression of the affine coupling block is: ; in the formula, Is a first portion of data; Is the second part of data; and (3) with Real value output of the neural network; is a neural network parameter.
6. The wind field data joint distribution modeling method based on the standardized flow model of claim 5, wherein the expression of the reversible affine transformation operation is: positive transformation: ; and (3) inverse transformation: ; in the formula, Outputting data for a second portion of the positive transform; Is a neural network parameter; is Hadamard product.
7. The wind field data joint distribution modeling method based on the standardized flow model of claim 1, wherein the expression of the multidimensional joint distribution probability density is: ; in the formula, Probability density for the original data space; is input data; Is a neural network parameter; a positive transformation function for the model; Probability density for hidden variable space; Is a jacobian matrix.
8. A wind farm data joint distribution modeling system based on a standardized flow model, comprising: the data acquisition module is used for acquiring wind field monitoring data and preprocessing the wind field monitoring data; The system comprises a model construction module, a model replacement module and a model replacement module, wherein the model construction module is used for constructing a standardized flow model framework, the standardized flow model framework is used for establishing bidirectional reversible mapping between an original data space and a hidden variable space through a training module, the original data space is used for representing real multidimensional joint distribution of wind field monitoring data, the prior distribution of the hidden variable space is assumed to be normal distribution, the training module comprises n layers of coupling layers, and each layer of coupling layer comprises an affine coupling block and a random replacement layer; the model training module is used for training the standardized flow model framework by utilizing the preprocessing data to obtain a trained standardized flow model; A data processing module, configured to generate samples of the original data space by sampling from the hidden variable space and performing inverse transformation of the bidirectional reversible mapping, and/or calculate a multi-dimensional joint distribution probability density of any point in the original data space by using a forward transformation of the bidirectional reversible mapping and a variable transformation formula, based on the normalized flow model.

Description

Wind field data joint distribution modeling method and system based on standardized flow model Technical Field The application relates to the field of wind engineering and artificial intelligence, in particular to a wind field data joint distribution modeling method and system based on a standardized flow model. Background At present, the accurate modeling of environmental variables such as wind direction, wind speed and the like in wind engineering and structure reliability evaluation mainly depends on various univariate and bivariate probability distribution methods. The univariate modeling generally adopts Gumbel, weibull or normal distribution for wind speed and von Mises distribution for wind direction, and the method is simple and feasible, but fails to consider the dependency relationship between wind speed and wind direction, so that a wind load evaluation result may have significant deviation. To characterize the correlation between variables, dual-variable joint distribution models, such as the Angular-linear model applicable to circular-linear variable combinations, and construction methods based on Copula functions or finite-mix distributions have been developed in the prior art. However, the Angular-linear model often faces the problems of high calculation requirements and discontinuous probability density functions in large-scale data application, while the Copula method and the mixed model can improve the characterization precision, but are still limited by strict assumptions on edge distribution or mixed components, the parameter identification process is complex, the selection of the Copula function directly influences the performance of the model, and the general rule is lacked. In addition, when the dimension of the variable is increased and the data presents a complex nonlinear related structure, the modeling difficulty of the traditional method is increased sharply, and the traditional method is difficult to effectively expand to a high-dimensional wind field and other environment data sets. Therefore, how to construct a wind field joint probability distribution modeling method that can be flexibly extended to a high-dimensional situation is a problem that needs to be solved by those skilled in the art. Disclosure of Invention In view of the above, the present invention provides a wind field data joint distribution modeling method and system based on a standardized flow model, which overcomes the above drawbacks. In order to achieve the above object, the present application provides the following solutions: In a first aspect, the application provides a wind field data joint distribution modeling method based on a standardized flow model, which comprises the following specific steps: acquiring wind field monitoring data, and preprocessing the wind field monitoring data; Constructing a standardized flow model framework, wherein the standardized flow model framework establishes bidirectional reversible mapping between an original data space and a hidden variable space through a training module, the original data space is used for representing real multidimensional joint distribution of wind field monitoring data, the priori distribution of the hidden variable space is assumed to be normal distribution, the training module comprises n layers of coupling layers, and each layer of coupling layer comprises an affine coupling block and a random substitution layer; Training the standardized flow model framework by utilizing the preprocessing data to obtain a trained standardized flow model; Based on the normalized flow model, a sample of the original data space is generated by sampling from the hidden variable space and inverse transforming through the bidirectional reversible mapping, and/or a multi-dimensional joint distribution probability density of any point in the original data space is calculated by a forward transformation of the bidirectional reversible mapping and a variable transformation formula. Optionally, the training step of the standardized flow model framework is as follows: acquiring and preprocessing the wind field monitoring data to obtain an original training data matrix; And constructing the standardized flow model framework, training the standardized flow model framework by utilizing the original training data matrix, and optimizing model parameters based on a loss function to obtain the trained standardized flow model. Optionally, the step of obtaining the original training data matrix includes: Acquiring real-time wind field monitoring data; Preprocessing the wind field monitoring data, and converting the preprocessed data with uniform format into a data matrix to obtain the original training data matrix. Optionally, the training module includes a plurality of affine coupling blocks and random permutation layers connected in sequence; Dividing input data into first partial data and second partial data, performing direct copy operation on the first partial data, and perf