A method, system and medium for extracting key time steps of unsteady flow field based on multi-scale information
Through multi-scale information analysis and multi-level point cloud automatic encoder extraction of flow field features, the problem of storage pressure and key evolution loss in numerical simulation of non-stable flow field is solved, and efficient and economical flow field simulation and analysis are achieved, suitable for Multiple grid types.
Patent Information
- Application Number
- CN202411261250.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-10
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2044-09-10
AI Technical Summary
Prior Art In numerical simulation of non-constant flow field, storage pressure and key evolution loss problems caused by fixed time steps, and existing methods are cost-effective in computing, are sensitive to parameters and noise, and are unable to adapt to non-structural grids or multi-blocks. Grid.
Using a non-stable flow field key time step extraction method based on multi-scale information, we intelligently analyze the similarity of adjacent time step flow fields, use a multi-level point cloud automatic encoder to extract the flow field features, calculate the feature distance and set the threshold. Automatically identify critical time steps and discard redundant data.
The storage resource management is optimized, the description and reconstruction efficiency of flow field dynamic evolution is improved, the calculation complexity is reduced, and it is suitable for a wider range of flow field dynamic variations, which significantly improves the algorithm execution efficiency.
Smart Images

Figure CN118780212B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of unsteady flow fields, and in particular to a method, system and medium for extracting key time steps of unsteady flow fields based on multi-scale information. Background Art
[0002] Unsteady flow fields specifically refer to those flow phenomena in which the flow characteristics (such as velocity, pressure, etc.) at each point in the flow field show obvious time-varying characteristics during the time evolution. When performing numerical simulations of unsteady flow fields, in order to accurately capture this dynamic process, it is usually necessary to continuously calculate and save a series of spatial flow field data at predetermined time intervals. However, with the improvement of the accuracy of numerical simulation technology, the amount of flow field data generated at each time step has increased dramatically, especially when the simulation covers a longer time range, the accumulated data storage requirements have become extremely large, which has brought a huge burden to storage resources.
[0003] In the prior art, Patent ZL202110787112.6 proposes a method for extracting and reconstructing key time steps of the flow field based on multivariable fusion, which has the following three limitations: First, the main scheme of this method is to simultaneously fuse and splice multiple physical quantity characteristics of the flow field without considering multi-scale characteristics; Second, this method uses mutual information to calculate the similarity between adjacent frames, which has high computational cost and is sensitive to parameters and noise; Third, this method uses dynamic programming to extract key time steps, which is not concise enough and has a certain impact on performance.
[0004] In the prior art, the paper "Key time steps selection for CFD data based on deepmetric learning" proposed an intelligent extraction method for key time steps in unsteady flow fields. It adopts the idea of contrastive learning and is a supervised method, that is, the training process of the intelligent extraction method requires expert-defined label guidance. However, there are very few expert-defined label data, which is not conducive to the promotion and application of this method. In addition, the method in this paper is only applicable to regular grids, and cannot be applied to unstructured grids or multi-block grids.
[0005] Therefore, given the limitations of existing technologies, how to maintain a reasonable description and reconstruction of the dynamic evolution process of the flow field while optimizing storage resource management has become a bottleneck problem that technical personnel in this field want to break through. Summary of the invention
[0006] Based on the current status of the background technology, the purpose of the present invention is to solve the storage pressure and key evolution loss problems caused by fixed time steps, and propose a method, system and medium for extracting key time steps of unsteady flow fields based on multi-scale information. The present invention can obtain the key time steps of the flow field changing over time, and applies the "key frame" extraction strategy; this strategy automatically identifies and selects the most representative "key" flow field states for storage by intelligently analyzing the similarity of the flow fields between adjacent time steps, while discarding those flow field data with high information redundancy. Therefore, the present invention not only optimizes storage resource management, but also promotes efficient and economical long-term simulation and analysis of complex unsteady flow fields.
[0007] The present invention adopts the following technical solutions to achieve the purpose:
[0008] A method for extracting key time steps of an unsteady flow field based on multi-scale information, the method comprising the following steps:
[0009] S1. Defining data of the unsteady flow field and determining multiple flow fields corresponding to the unsteady flow field at different time steps;
[0010] S2, dividing the unsteady flow field into a spatial field region and a near-wall region, and extracting multi-scale point cloud data of the near-wall region;
[0011] S3, train multiple point cloud autoencoders to extract features from multi-scale point cloud data;
[0012] S4. Calculate the characteristic distance of the adjacent spatial flow fields based on the extracted features, and accumulate the characteristic distance values between the flow fields at each time step and the flow field at the first time step;
[0013] S5. Divide the accumulated feature distance value into multiple parts according to the preset number of key frames, as multiple key thresholds;
[0014] S6. Based on the key threshold and the characteristic distance value, multiple flow fields with the same number as the key threshold are selected; the selected flow fields are multiple key flow fields in the unsteady flow field, and the time step of the key flow field is the key time step of the unsteady flow field.
[0015] Furthermore, after the near-wall area is determined, the process of extracting multi-scale point cloud data is as follows:
[0016] For a flow field, set multiple sampling areas, respectively , and satisfy it Constraints, Indicates the original number of grid points; use Indicates the first The number of sampling regions for the layer, Indicates the total number of layers;
[0017] For a flow field layer, automatically divide the near-wall area of the layer, and thus generate sampling areas; collect 1 point in each sampling area, and obtain grid points and the corresponding flow field attribute values; repeat the sampling operation until it is completed Layer to All samples of the layer, at this time we get a flow field Point cloud data represented at different levels;
[0018] The above-mentioned sampling area quantity setting and repeated sampling operation are performed on the multiple flow fields determined in step S1, so that the respective sampling areas of all flow fields are obtained. Point cloud data represented at different levels.
[0019] Further, in step S3, design An autoencoder that accepts point cloud data as input uses the obtained flow field Point cloud data with different levels of representation are trained separately using unsupervised training. An autoencoder;
[0020] After the training operation is executed, The decoding part of each autoencoder is discarded, and only the encoding part is retained as the trained An autoencoder;
[0021] For the definition operation of step S1, determine flow fields; all flow fields Point cloud data with different levels of representation are input into the trained After the autoencoder, we get group feature vector;
[0022] The same flow field The groups of feature vectors are concatenated or added to obtain a group of global-local feature vectors corresponding to the flow field; After splicing or adding each flow field, the obtained The group global-local eigenvector is denoted by , complete the feature extraction of multi-scale point cloud data.
[0023] Further, in step S4, according to the defined flow field time sequence, the determined flow field, calculate the The global-local eigenvectors and Euclidean distance between group global-local eigenvectors ,get The group feature distance is denoted as ;
[0024] Will The group feature distances are updated to the accumulated feature distance values. , where the characteristic distance value is Indicated in In the flow field The characteristic distance between the flow field at the first time step and the flow field at the first time step.
[0025] The present invention also provides a computer device system, including a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the aforementioned method for extracting key time steps of an unsteady flow field based on multi-scale information.
[0026] The present invention also provides a computer-readable storage medium on which a computer program / instruction is stored. When the computer program / instruction is executed by a processor, the steps of the aforementioned method for extracting key time steps of an unsteady flow field based on multi-scale information are implemented.
[0027] In summary, due to the adoption of this technical solution, the beneficial effects of the present invention are as follows:
[0028] The core of the present invention is the innovative use of a multi-scale analysis method, which not only captures the overall dynamics of the flow field from a macroscopic global perspective, but also delicately explores the microscopic local fluid behavior. Through the advanced artificial intelligence technology of three-dimensional autoencoder, the present invention can efficiently extract flow field features from multiple levels, each level focusing on the dynamic performance of different spatial scales, thereby achieving a comprehensive and detailed description of flow field characteristics. This method transcends the limitations of traditional single-scale analysis, making the expression of flow field characteristics richer and more in-depth.
[0029] The present invention uses the distance calculation between multi-level flow field features to evaluate the degree of change of the flow field at adjacent time steps from a more comprehensive perspective. By flexibly setting the threshold, the present invention can adapt to a wider range of dynamic changes in the flow field and accurately distinguish between critical and non-critical time steps; compared with traditional methods that only rely on dynamic programming, the present invention not only improves the accuracy of feature extraction, but also greatly simplifies the key frame selection process and significantly improves the algorithm execution efficiency.
[0030] At the same time, compared with the prior art, the present invention abandons the complex dynamic programming strategy and instead adopts a direct and efficient threshold division method to automatically extract key frames, which greatly simplifies the implementation steps, reduces the computational complexity, and makes the analysis and processing of unsteady flow fields faster, which is conducive to the processing of large-scale data sets and long-term simulation research, effectively reduces the pressure on computing resources, and provides a more convenient and efficient tool for scientific researchers and engineers. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] Figure 1 The figure is a simplified diagram of the overall process of the method of the present invention. DETAILED DESCRIPTION
[0032] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.
[0033] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0034] Example 1
[0035] A method for extracting key time steps in unsteady flow fields based on multi-scale information. Figure 1 The overall process of the method is briefly described, and the overall steps of the method are summarized as follows:
[0036] S1. Defining data of the unsteady flow field and determining multiple flow fields corresponding to the unsteady flow field at different time steps;
[0037] S2, dividing the unsteady flow field into a spatial field region and a near-wall region, and extracting multi-scale point cloud data of the near-wall region;
[0038] S3, train multiple point cloud autoencoders to extract features from multi-scale point cloud data;
[0039] S4. Calculate the characteristic distance of the adjacent spatial flow fields based on the extracted features, and accumulate the characteristic distance values between the flow fields at each time step and the flow field at the first time step;
[0040] S5. Divide the accumulated feature distance value into multiple parts according to the preset number of key frames, as multiple key thresholds;
[0041] S6. Based on the key threshold and the characteristic distance value, multiple flow fields with the same number as the key threshold are selected; the selected flow fields are multiple key flow fields in the unsteady flow field, and the time step of the key flow field is the key time step of the unsteady flow field.
[0042] This embodiment introduces the specific contents of each step in detail according to the above step sequence. Before the specific introduction, the relevant terms in the description of step S1 are explained as follows:
[0043] Unsteady flow field: In fluid mechanics, flow fields can be divided into steady flow fields and unsteady flow fields. A steady flow field refers to a fluid whose flow parameters (such as velocity, pressure, etc.) may change in space but not in time. On the contrary, an unsteady flow field has flow parameters that change not only with spatial position but also with time. This means that in an unsteady flow field, the state of the fluid (such as velocity, pressure) at any given point will be different at different points in time.
[0044] Different time steps: A time step is a process of dividing time into a series of discrete time intervals during a numerical simulation or analysis. Each time step represents a moment or time period in the simulation. When studying unsteady flow fields, by considering different time steps, it is possible to observe and analyze how the flow characteristics evolve over time.
[0045] Corresponding multiple flow fields: The "multiple flow fields" mentioned here means that at different time steps, the state of the fluid system is different, so it can be considered as different flow fields. Each time step has its own unique flow field characteristics such as velocity distribution and pressure distribution. By analyzing these flow fields captured at different time steps, researchers can understand the dynamic process of fluid motion, such as the generation and dissipation of vortices, the propagation of pressure waves, etc.
[0046] In step S1 of this embodiment, the unsteady flow field uniformly sampled in time is defined as ,use Indicates The flow field of time steps, represents the total number of time steps. Therefore, after the definition operation, A flow field.
[0047] In step S2, the unsteady flow field is first divided. The division method is: for multiple flow fields corresponding to the unsteady flow field at different time steps, the distance between the grid point or unit center of each flow field and the wall is calculated, and a distance threshold is preset; if the calculated distance is greater than the distance threshold, the corresponding grid point or unit center belongs to the space field area, otherwise it belongs to the near-wall area.
[0048] Here, the distance is calculated based on the geometric structure and grid distribution of the current flow field, and the selection of the distance threshold is usually based on the flow characteristics, grid size, and the required simulation accuracy. In the divided area, the spatial field area is usually far away from the wall, and the flow characteristics may be more complex, but the requirements for grid resolution are relatively low relative to the near-wall area. The fluid behavior in the near-wall area is significantly affected by the wall effect, and a higher resolution grid is often required to accurately capture the flow details in the boundary layer. The core of this embodiment focuses on the processing of the near-wall area, because the near-wall area usually uses a more complex and dense grid, which requires high computational costs.
[0049] In this embodiment, after the near-wall area is determined, the process of extracting multi-scale point cloud data is as follows:
[0050] For a flow field, set multiple sampling areas, respectively , and satisfy it Constraints, Indicates the original number of grid points; use Indicates the first The number of sampling regions for the layer, Indicates the total number of layers. This design makes the number of sampling areas increase continuously, indicating that the number of sampling points gradually increases, so the area represented by each sampling point becomes smaller, which is a typical pyramid sampling method.
[0051] For a flow field layer, automatically divide the near-wall area of the layer, and thus generate sampling areas; collect 1 point in each sampling area, and obtain grid points and the corresponding flow field attribute values; repeat the sampling operation until it is completed Layer to All samples of the layer, at this time we get a flow field Point cloud data represented at different levels;
[0052] The above-mentioned sampling area quantity setting and repeated sampling operation are performed on the multiple flow fields determined in step S1, so that the respective sampling areas of all flow fields are obtained. Point cloud data represented at different levels.
[0053] In step S3, the point cloud autoencoder is first trained. An autoencoder that accepts point cloud data as input uses the obtained flow field Point cloud data with different levels of representation are trained separately using unsupervised training. After the training operation is executed, The decoding part of each autoencoder is discarded, and only the encoding part is retained as the trained An autoencoder.
[0054] Each autoencoder itself is a structure consisting of an encoding part and a decoding part, which constitutes an end-to-end learning framework. Unsupervised learning does not require any labeled information during the training process. The autoencoder will try to discover the inherent structure and pattern of the input data by itself. Through iterative optimization, the autoencoder can learn a set of features that can efficiently represent the original data. These features are often more concise than the original data, while also retaining important information.
[0055] After training is complete, the decoding portion of each autoencoder is discarded, and only the encoding portion is retained. This is because the encoder has learned how to convert flow field data into a compact feature representation, and the decoder portion is no longer needed in subsequent analysis or applications. The retained encoding portion allows the autoencoder to be used as a feature extraction tool, converting complex flow field data into easy-to-process feature vectors; these feature vectors can then be used for a variety of downstream tasks, such as flow field classification, anomaly detection, etc.
[0056] Therefore, all flow fields Point cloud data with different levels of representation are input into the trained After the autoencoder, we get group of eigenvectors; The groups of feature vectors are concatenated or added to obtain a group of global-local feature vectors corresponding to the flow field; After splicing or adding each flow field, the obtained The group global-local eigenvector is denoted by , complete the feature extraction of multi-scale point cloud data.
[0057] In step S4, according to the defined flow field time sequence, flow field, calculate the The global-local eigenvectors and Euclidean distance between group global-local eigenvectors ,get The group feature distance is denoted as .
[0058] In order to ensure that the final flow field still maintains the order defined in step S1, The group feature distances are updated to the accumulated feature distance values. , the characteristic distance value here Indicated in In the flow field The characteristic distance between the flow field at the first time step and the flow field at the first time step.
[0059] In step S5, according to the preset number of key frames , will The characteristic distance value between the first flow field and the first flow field Divided into Get A critical threshold, denoted as .
[0060] In step S6, according to The key thresholds are Select from the flow fields flow field; the selection method is: for the Key thresholds ,as well as In the flow field The characteristic distance value of the flow field after accumulation of time steps , traverse All the accumulated characteristic distance values in the flow field , when The feature distance value is greater than the critical threshold for the first time When , the feature distance value is recorded as , its corresponding Key Flow Field ; The above traversal and recording operations are performed for each key threshold, and a corresponding key flow field is obtained for each key threshold, and finally A key flow field.
[0061] In summary, the present invention not only enhances the depth and breadth of understanding of the complex dynamics of unsteady flow fields by integrating multi-scale feature extraction, optimized flow field distance calculation and a simple threshold screening strategy, but also significantly improves processing efficiency and practicality, bringing innovative progress to the field of fluid mechanics research and engineering applications.
[0062] Example 2
[0063] Based on Example 1, this embodiment provides a computer device system, including a memory, a processor, and a computer program stored in the memory, and the processor executes the computer program to implement the steps of the method for extracting key time steps of unsteady flow fields based on multi-scale information described in Example 1.
[0064] This embodiment also provides a computer-readable storage medium on which a computer program / instruction is stored. When the computer program / instruction is executed by a processor, the steps of the method for extracting key time steps of an unsteady flow field based on multi-scale information described in Example 1 are implemented.
[0065] In this embodiment, these computer programs / instructions may be stored in a computer-readable memory that can direct a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device that implements the functions specified in one or more steps in the method.
Claims
1. A method for extracting key time steps of unsteady flow fields based on multi-scale information, characterized in that: The method comprises the following steps: S1. Defining data of the unsteady flow field and determining multiple flow fields corresponding to the unsteady flow field at different time steps; S2, dividing the unsteady flow field into a spatial field region and a near-wall region, and extracting multi-scale point cloud data of the near-wall region; S3, train multiple point cloud autoencoders to extract features from multi-scale point cloud data; S4. Calculate the characteristic distance of the adjacent spatial flow fields based on the extracted features, and accumulate the characteristic distance values between the flow fields at each time step and the flow field at the first time step; S5. Divide the accumulated feature distance value into multiple parts according to the preset number of key frames, as multiple key thresholds; S6. Select multiple flow fields with the same number as the key threshold value according to the key threshold value and the characteristic distance value; the selected flow fields are multiple key flow fields in the unsteady flow field, and the time step of the key flow field is the key time step of the unsteady flow field; In step S3, design An autoencoder that accepts point cloud data as input uses the obtained flow field Point cloud data with different levels of representation are trained separately using unsupervised training. An autoencoder; After the training operation is executed, The decoding part of each autoencoder is discarded, and only the encoding part is retained as the trained An autoencoder; For the definition operation of step S1, determine flow fields; all flow fields Point cloud data with different levels of representation are input into the trained After the autoencoder, we get group feature vector; The same flow field The groups of feature vectors are concatenated or added to obtain a group of global-local feature vectors corresponding to the flow field; After splicing or adding each flow field, the obtained The group global-local eigenvector is denoted by , complete the feature extraction of multi-scale point cloud data; For the definition operation of step S1, determine After the flow field is formed, in step S5, according to the preset number of key frames, , will The characteristic distance value between the first flow field and the first flow field Divided into Get A critical threshold, denoted as ; In step S6, according to The key thresholds are Select from the flow fields flow field; The selection method is: for the Key thresholds ,as well as In the flow field The characteristic distance value of the flow field after accumulation of time steps , traverse All the accumulated characteristic distance values in the flow field , when The feature distance value is greater than the critical threshold for the first time When , the feature distance value is recorded as , its corresponding Key Flow Field ; The above traversal and recording operations are performed for each key threshold, and a corresponding key flow field is obtained for each key threshold, and finally A key flow field.
2. The method for extracting key time steps of unsteady flow fields based on multi-scale information according to claim 1 is characterized in that: In step S1, the unsteady flow field uniformly sampled in time is defined as ,use Indicates The flow field of time steps, Represents the total number of time steps.
3. The method for extracting key time steps of unsteady flow fields based on multi-scale information according to claim 1 is characterized in that: In step S2, for multiple flow fields corresponding to the unsteady flow field at different time steps, the distance between the grid point or unit center of each flow field and the wall is calculated, and a distance threshold is preset; if the calculated distance is greater than the distance threshold, the corresponding grid point or unit center belongs to the space field area, otherwise it belongs to the near-wall area.
4. The method for extracting key time steps of unsteady flow fields based on multi-scale information according to claim 3 is characterized in that: After the near-wall area is determined, the process of extracting multi-scale point cloud data is as follows: For a flow field, set multiple sampling areas, respectively , and satisfy it Constraints, Indicates the original number of grid points; use Indicates the first The number of sampling regions for the layer, Indicates the total number of layers; For a flow field layer, automatically divide the near-wall area of the layer, and thus generate sampling areas; collect 1 point in each sampling area, and obtain grid points and the corresponding flow field attribute values; repeat the sampling operation until it is completed Layer to All samples of the layer, at this time we get a flow field Point cloud data represented at different levels; The above-mentioned sampling area quantity setting and repeated sampling operation are performed on the multiple flow fields determined in step S1, so that the respective sampling areas of all flow fields are obtained. Point cloud data represented at different levels.
5. The method for extracting key time steps of unsteady flow fields based on multi-scale information according to claim 4 is characterized in that: In step S4, according to the defined flow field time sequence, flow field, calculate the The global-local eigenvectors and Euclidean distance between group global-local eigenvectors ,get The group feature distance is denoted as ; Will The group feature distances are updated to the accumulated feature distance values. , the characteristic distance value here Indicated in In the flow field The characteristic distance between the flow field at the first time step and the flow field at the first time step.
6. A computer device system, comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method for extracting key time steps of an unsteady flow field based on multi-scale information as described in any one of claims 1 to 5.
7. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instruction is executed by a processor, the steps of the method for extracting key time steps of an unsteady flow field based on multi-scale information as described in any one of claims 1 to 5 are implemented.
