CN-121327529-B - Data-driven complex flow evolution analysis method and system

CN121327529BCN 121327529 BCN121327529 BCN 121327529BCN-121327529-B

Abstract

The invention provides a data-driven complex flow evolution analysis method and a data-driven complex flow evolution analysis system, which are used for accurately identifying the causal relationship between each typical dynamic mode in a flow field by combining an orthogonal mode decomposition and entropy transfer strategy, so as to mine a core mode combination with obvious influence on flow evolution and system performance, and finally realize analysis and control of complex flow field evolution. The causal network relation among the modes is identified through the system, so that the degree of dependence on high-dimensional original flow field data can be obviously reduced, and analysis emphasis is focused on key modes with strong dynamics dominance and obvious interactivity. The shortcomings of traditional modal energy sequencing are supplemented through causality quantification, so that the modal screening not only considers the contribution degree, but also considers dynamic dominance and interaction, and a more targeted and reliable basis is provided for quantitative modeling and prediction of complex flow behaviors.

Inventors

LI BINGHUA
ZHENG NAN
CHEN XIAOFENG
CHEN ZEYU
WU QILONG
ZHU MEIYIN
YE ZHOUTENG
Li Kenlong

Assignees

杭州市北京航空航天大学国际创新研究院(北京航空航天大学国际创新学院)

Dates

Publication Date: 20260508
Application Date: 20251212

Claims (7)

1. A method of data-driven complex flow evolution analysis, the method comprising: step a, obtaining flow field snapshots of a target flow field in continuous time; Extracting an orthogonal mode of a flow field by adopting an orthogonal mode decomposition method to obtain a time coefficient matrix and a space mode matrix, wherein the time coefficient matrix comprises Time coefficient sequences of the corresponding modes; Step c, calculating transfer entropy on lag time for the time coefficient sequences of all mode pairs, and obtaining a causal influence matrix between the mode pairs; step d, carrying out structural analysis on the causal influence matrix to obtain a similarity matrix; step e, obtaining a plurality of causal modal clusters through cluster analysis based on the causal influence matrix and the similarity matrix; F, reconstructing the causal mode cluster to obtain a reconstructed time coefficient and a spatial projection of a cluster mode; Step g, calculating by adopting transfer entropy according to the reconstruction time coefficient of the cluster mode to obtain a causal evolution network, fine-tuning the number of clusters involved in the cluster analysis according to the complexity of the causal evolution network, and obtaining a key causal evolution network of the flow field through multiple iterations; the step g specifically comprises the following steps: Calculating a cluster time coefficient of each causal mode cluster; Calculating transfer entropy on lag time for the cluster time coefficient to obtain causal strength among causal mode clusters; Organizing all the causal intensities into a cluster-level causal matrix; constructing a directed weighted graph among clusters based on the cluster-level cause and effect matrix to obtain the cause and effect evolution network; And carrying out granularity analysis according to the obtained causal evolution network, readjusting the clustering quantity, and realizing complexity adjustment of the causal evolution network to obtain the key causal evolution network.
2. The data-driven complex flow evolution analysis method of claim 1, wherein step b specifically comprises: Arranging the flow field snapshots in time sequence to form a data matrix; performing mean value removal processing on the data matrix to obtain a centralized matrix; and carrying out orthonormal modal decomposition on the centralized matrix to obtain the time coefficient matrix and the space modal matrix.
3. The data-driven complex flow evolution analysis method of claim 1, wherein step c specifically comprises: Discretizing the time coefficient sequence, and selecting the maximum time lag parameter For any of the modality pairs, calculate the time-lag Taking the maximum value as the final causal influence intensity between the modal pairs; Summarizing the transfer entropy values between all the modality pairs into the causal influence matrix.
4. The data-driven complex flow evolution analysis method of claim 1, wherein, The step d specifically comprises the following steps: the causal influence matrix comprises k rows, and each row is a causal influence vector of a corresponding mode; normalizing each causal influence vector; calculating the similarity of causal behaviors between each modal pair; And constructing the similarity matrix according to the similarity calculation result.
5. The data-driven complex flow evolution analysis method of claim 4, wherein, The step e specifically comprises the following steps: Taking the causal influence vector as an input characteristic of cluster analysis; And clustering all modes by using a K-means clustering algorithm in unsupervised learning, wherein the clustering algorithm automatically groups the modes with similar causal behaviors according to the similarity matrix, and a plurality of causal mode clusters are obtained according to the selected clustering quantity.
6. The data-driven complex flow evolution analysis method of claim 1, wherein, The step f specifically comprises the following steps: Reconstructing a flow field estimation corresponding to the causal mode cluster by superposing contributions of all modes in the causal mode cluster, wherein the reconstruction time coefficient is a weighted average or energy weighted sum of the mode time coefficients of the causal mode cluster, and the space projection is an average flow structure of the causal mode cluster in the whole time period.
7. A data-driven complex flow evolution analysis system, comprising The preprocessing module is used for acquiring flow field snapshots of the target flow field in continuous time; The modal causal analysis module is used for extracting orthogonal modes of a flow field by adopting an orthogonal mode decomposition method to obtain a time coefficient matrix and a spatial mode matrix, wherein the time coefficient matrix comprises r time coefficient sequences of corresponding modes, and transfer entropy on lag time is calculated on the time coefficient sequences of all mode pairs to obtain a causal influence matrix between the mode pairs; The causal evaluation module is used for carrying out structural analysis on the causal influence matrix to obtain a similarity matrix; the modal clustering module is used for obtaining a plurality of causal modal clusters through clustering analysis based on the causal influence matrix and the similarity matrix; The result visualization module is used for reconstructing the causal modal cluster to obtain a reconstructed time coefficient and a space projection, calculating a causal evolution network by adopting a transfer entropy according to the reconstructed time coefficient, and fine-tuning the number of clusters related to the cluster analysis according to the complexity of the causal evolution network to obtain a key causal evolution network of the flow field through multiple iterations, and specifically comprises the following steps: Calculating a cluster time coefficient of each causal mode cluster; Calculating transfer entropy on lag time for the cluster time coefficient to obtain causal strength among causal mode clusters; Organizing all the causal intensities into a cluster-level causal matrix; constructing a directed weighted graph among clusters based on the cluster-level cause and effect matrix to obtain the cause and effect evolution network; And carrying out granularity analysis according to the obtained causal evolution network, readjusting the clustering quantity, and realizing complexity adjustment of the causal evolution network to obtain the key causal evolution network.

Description

Data-driven complex flow evolution analysis method and system Technical Field The invention relates to the technical field of image processing, in particular to a data-driven complex flow evolution analysis method and system. Background In complex flow fields, because the flow states exhibit highly nonlinear, multi-scale, and non-stationary characteristics, traditional analysis often relies on global statistics or local empirical features, and it is difficult to deeply reveal interaction mechanisms between different dynamic structures. In engineering practice, the evolution characteristic of a complex flow field easily influences the system stability, resistance characteristic or key modes of energy transmission, and the prior art means has the problems of huge calculation resource consumption and high analysis cost in the aspect of identifying and monitoring few key modes with obvious causal influence on the system state. Disclosure of Invention The invention has been completed in view of the above-mentioned existing situation, and an object of the invention is to provide a data-driven complex flow evolution analysis method and system, which can develop a high-efficiency and interpretable flow field modal causal identification technology by deeply fusing an advanced modal decomposition method with an information theory causal analysis, and can be widely applied in the fields of aerospace, energy engineering, environmental fluid and the like. In order to achieve the above object, the present invention provides the following technical solutions: The invention provides a data-driven complex flow evolution analysis method, which comprises the following steps: a) Acquiring flow field snapshots of a target flow field in continuous time; b) Extracting an orthogonal mode of the flow field by adopting an orthogonal mode decomposition method to obtain a time coefficient matrix and a space mode matrix, wherein the time coefficient matrix comprises r time coefficient sequences of corresponding modes; c) Calculating transfer entropy on lag time for the time coefficient sequences of all modal pairs, and obtaining a causal influence matrix between the modal pairs; d) Carrying out structural analysis on the causal influence matrix to obtain a similarity matrix; e) Based on the causal influence matrix and the similarity matrix, a plurality of causal modal clusters are obtained through cluster analysis; f) Reconstructing the causal modal cluster to obtain a reconstructed time coefficient and a spatial projection; g) And calculating by adopting a transfer entropy according to the reconstruction time coefficient to obtain a causal evolution network, adjusting the number of clusters related to the cluster analysis according to the causal evolution network complexity, and iterating to obtain a key causal evolution network of a flow field. Under the condition, by combining the strategies of orthogonal modal decomposition and entropy transfer, the causal relationship between each typical dynamic mode in the flow field is accurately identified, and then a core mode combination with obvious influence on the flow evolution and the system performance is excavated, so that analysis of the complex flow field evolution is realized. Wherein, the step b) specifically includes: Arranging the flow field snapshots in time sequence to form a data matrix; performing mean value removal processing on the data matrix to obtain a centralized matrix; and carrying out orthonormal modal decomposition on the centralized matrix to obtain the time coefficient matrix and the space modal matrix. Under the condition, through the orthogonal modal decomposition step, the original complex flow field snapshot set is successfully reduced to be a group of modal representations which have energy ordering, space-time decoupling and are mutually orthogonal, and high-quality input features are provided for subsequent causal relationship identification. Wherein, the step c) specifically includes: Discretizing the time coefficient sequence, and selecting the maximum time lag parameter For any of the modality pairs, calculate the time-lagTaking the maximum value as the final causal influence intensity between the modal pairs as the transfer entropy value: Summarizing the transfer entropy values between all the modality pairs into the causal influence matrix. Under the condition, the causal coupling strength among the modes can be quantitatively captured through the steps, the causal coupling strength is obviously superior to a method based on correlation analysis, causal confusion and erroneous judgment are effectively avoided, and a causal basis is laid for subsequent mode screening, clustering and control strategy design. Wherein, the step d) specifically includes: the causal influence matrix comprises k rows, and each row is a causal influence vector of a corresponding mode; normalizing each causal influence vector; calculating the similarity of causal behaviors bet