Gas compressor blade flow field prediction method based on multi-scale characteristics and attention enhancement

Abstract

The invention relates to a gas compressor blade flow field prediction method based on multi-scale features and attention enhancement. The method comprises the following steps: firstly, acquiring geometric data of a compressor blade, and carrying out gridding preprocessing; then, K-hop neighborhood sampling is carried out based on the geometric data after gridding preprocessing, and a graph neural network input data structure is constructed based on a point set and edge connection relation; then, a multi-branch flow field prediction model is constructed, and the multi-branch flow field prediction model comprises an encoder, a geometric branch module, a flow field branch module, a feature fusion module and a decoder; then, training the multi-branch flow field prediction model by adopting the graph neural network input data structure; and finally, inputting geometric data of a to-be-predicted compressor blade into the trained multi-branch flow field prediction model to obtain a flow field prediction result. The overall precision and the boundary fitting capability of flow field prediction are remarkably improved, and meanwhile good performance is shown in the aspects of calculation efficiency and model expandability.

Description

Technical Field

[0001] This application relates to the field of computational fluid dynamics, and in particular to a method for predicting the flow field of compressor blades based on multi-scale features and attention enhancement. Background Technology

[0002] In aerospace and other fields, compressors are core components of power plants such as aero engines, and accurate prediction of the blade flow field is crucial for blade shape design. However, the flow field of compressor blades is extremely complex, containing numerous turbulent features ranging from macroscopic mainstream structures to small-scale vortices. These vortex structures span a wide range in space, and the boundary layer in the near-wall region of the blade is thin but has a sharp gradient, making it difficult for traditional prediction methods to achieve high-precision predictions.

[0003] In recent years, with the development of deep learning technology, it has also shown great application potential in the field of computational fluid dynamics (CFD). Deep neural networks have the ability to automatically extract features from complex data and can effectively model nonlinear flow laws. Compared with traditional numerical methods that rely on iterative solutions to partial differential equations, deep learning models can achieve rapid prediction of flow field states in an end-to-end manner, significantly reducing computational costs and providing new ideas and paths for accelerating and intelligentizing CFD simulations.

[0004] In computational fluid dynamics (CFD) research, most deep learning methods are represented by Physically Informed Neural Networks (PINNs). The core idea is to introduce the residuals of partial differential equations controlling flow behavior as part of the loss function during network training. This allows for model training without requiring large amounts of flow field data, making it suitable for situations where physical laws are known but data is scarce. However, for the high-precision prediction problem of compressor blade flow fields—characterized by complex geometry, multi-scale vortex structures, and dramatic boundary layer changes—traditional physically constrained deep learning methods face challenges such as high computational complexity and unstable training during modeling, making efficient deployment in practical engineering difficult.

[0005] Therefore, there is an urgent need in related technologies for a way to improve the accuracy of compressor blade flow field prediction and boundary fitting capability. Summary of the Invention

[0006] Therefore, it is necessary to provide a compressor blade flow field prediction method based on multi-scale features and attention enhancement, which can improve the accuracy and boundary fitting ability of compressor blade flow field prediction, in order to address the above-mentioned technical problems.

[0007] Firstly, this application provides a method for predicting the flow field of compressor blades based on multi-scale features and attention enhancement. The method includes: Obtain compressor blade geometry data and perform mesh preprocessing; K-hop neighborhood sampling is performed on the pre-processed geometric data, and the input data structure of the graph neural network is constructed based on the connection relationship between the point set and the edge. A multi-branch flow field prediction model is constructed, including an encoder, a geometric branch module, a flow field branch module, a feature fusion module, and a decoder; The multi-branch flow field prediction model is trained using the graph neural network input data structure. The geometric data of the compressor blade to be predicted are input into the trained multi-branch flow field prediction model to obtain the flow field prediction results.

[0008] Optionally, in one embodiment of this application, the meshing preprocessing includes: Based on the generation of a symbolic distance function field from geometric points, the geometric distance characteristics in discrete space are described.

[0009] Optionally, in one embodiment of this application, the K-hop neighborhood sampling includes: Seed nodes are selected by random sampling; We employ K-hop breadth-first search to construct local subgraphs and learn complete local representations based on graph message passing mechanisms.

[0010] Optionally, in one embodiment of this application, the geometric branch module is a multi-scale feature extraction module based on the U-Net architecture, including an encoder, a decoder, and cross-layer connections.

[0011] Optionally, in one embodiment of this application, the flow field branching module is a local modeling and attention enhancement module based on graph neural networks and Transolver.

[0012] Secondly, this application also provides a compressor blade flow field prediction device based on multi-scale features and attention enhancement. The device includes: The geometric modeling and mesh generation module is used to acquire compressor blade geometric data and perform mesh preprocessing. The graph structure representation and data encoding module is used to perform K-hop neighborhood sampling based on the pre-processed gridded geometric data, and to construct the input data structure of the graph neural network based on the connection relationship between the point set and the edge. The prediction model building module is used to build a multi-branch flow field prediction model, including an encoder, a geometric branch module, a flow field branch module, a feature fusion module, and a decoder; A prediction model training module is used to train the multi-branch flow field prediction model using the graph neural network input data structure. The flow field prediction module is used to input the geometric data of the compressor blade to be predicted into the trained multi-branch flow field prediction model to obtain the flow field prediction results.

[0013] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the steps of the methods described in the various embodiments above.

[0014] Fourthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the steps of the methods described in the various embodiments above.

[0015] The aforementioned compressor blade flow field prediction method based on multi-scale features and attention enhancement first acquires compressor blade geometric data and performs gridded preprocessing. Then, K-hop neighborhood sampling is performed on the preprocessed geometric data, and a graph neural network input data structure is constructed based on the point set and edge connection relationship. Next, a multi-branch flow field prediction model is constructed, including an encoder, a geometric branch module, a flow field branch module, a feature fusion module, and a decoder. Then, the multi-branch flow field prediction model is trained using the graph neural network input data structure. Finally, the compressor blade geometric data to be predicted is input into the trained multi-branch flow field prediction model to obtain the flow field prediction result. In other words, by introducing a graph structure perception mechanism based on K-hop neighborhood sampling, combined with U-Net to effectively extract multi-scale flow features, and integrating the Transolver model to enhance boundary layer attention and reduce attention to the far field, a high-precision prediction model suitable for compressor blade flow fields is constructed. This not only significantly improves the overall accuracy and boundary fitting ability of flow field prediction but also demonstrates good performance in terms of computational efficiency and model scalability. This provides a feasible path for data-driven CFD modeling and offers a practical technical solution for high-precision prediction of complex engineering flow fields. Attached Figure Description

[0016] Figure 1 This is an application environment diagram of a compressor blade flow field prediction method based on multi-scale features and attention enhancement in one embodiment; Figure 2 This is a flowchart illustrating a compressor blade flow field prediction method based on multi-scale features and attention enhancement in one embodiment. Figure 3 This is a schematic diagram of the K-hop sampling process in one embodiment; Figure 4 This is a schematic diagram of the structure of a multi-branch flow field prediction model in one embodiment; Figure 5This is a schematic diagram of the structure of U-net in one embodiment; Figure 6 This is a flowchart illustrating the specific steps of a compressor blade flow field prediction method based on multi-scale features and attention enhancement in one embodiment. Figure 7 This is a block diagram of a compressor blade flow field prediction device based on multi-scale features and attention enhancement in one embodiment. Figure 8 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0018] The compressor blade flow field prediction method based on multi-scale features and attention enhancement provided in this application can be applied to, for example... Figure 1 The application environment is illustrated. The terminal communicates with the server via a network. The data storage system stores the data the server needs to process. This system can be integrated onto the server, or it can be hosted in the cloud or on other network servers. The terminal can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart in-vehicle systems, etc. Portable wearable devices can include smartwatches, smart bracelets, head-mounted devices, etc. The server can be implemented using a standalone server or a server cluster consisting of multiple servers.

[0019] In one embodiment, such as Figure 2 As shown, a method for predicting compressor blade flow field based on multi-scale features and attention enhancement is presented, and this method is applied to... Figure 1 Taking the server in the example, the following steps are included: S201: Obtain compressor blade geometry data and perform gridding preprocessing.

[0020] In this embodiment, firstly, the compressor blade geometry data is acquired. The blade geometry is stored in a standard .ply file format, preserving the surface structure information. For the compressor blade geometry, a meshing preprocessing operation is first performed, and the corresponding computational fluid dynamics (CFD) data is calculated using a traditional numerical iteration method.

[0021] In one embodiment of this application, the meshing preprocessing includes: Based on the generation of a symbolic distance function field from geometric points, the geometric distance characteristics in discrete space are described.

[0022] In one embodiment of this application, a regular mesh domain (e.g., 64×64 or 128×128 resolution) is constructed at the two-dimensional cross-sectional level to generate a Signed Distance Function (SDF) field to describe the distance characteristics of the geometry in discrete space. This SDF can effectively characterize the shape contours of complex geometries and serves as input to subsequent feature extraction networks. The specific method for calculating the signed distance function is as follows: First, traverse all points to be calculated. Initialize the shortest distance from the current point to the surface d_min = ∞; then, traverse each triangle tj in the triangular mesh, and for each tj, calculate the Euclidean distance from point pi to tj (usually the shortest distance from a point to a triangle). If this distance is less than the current minimum value d_min, update d_min; then, determine the current point... Whether it is inside the geometry: This is determined using methods such as ray casting (InsideOutsideTest). S) judgment Whether it is inside the geometry S; if it is inside, the sign is -1, otherwise +1; finally, the signed distance value is obtained. =sign × d_min. The final output is the set of signed distance values ​​for all points.

[0023] S203: K-hop neighborhood sampling is performed based on the preprocessed gridded geometric data, and the input data structure of the graph neural network is constructed based on the connection relationship between the point set and the edge.

[0024] In this embodiment, in the flow field graph structure generated by CFD simulation, nodes represent the physical state of the control volume unit, and edges reflect the spatial coupling relationship between physical variables. The flow field problem for compressor blades is often very complex, requiring a massive number of CFD data nodes for training. Directly loading the CFD data into the neural network, or loading the entire graph into the training neural network, would lead to memory overflow and computational inefficiency; the large data scale conflicts with limited computing resources. However, when the computational field of view of the input neural network is limited to a very small area, good inference performance cannot be guaranteed. To alleviate this problem, this invention introduces a K-hop neighborhood sampling strategy. K-hop sampling is a local subgraph extraction algorithm based on graph structure, used to find all nodes and their connecting edges within a range of no more than K hops (i.e., K-order adjacency) starting from a seed node in the graph. K-hop sampling can effectively preserve local physical continuity and spatial coupling characteristics without loading the complete graph structure, supporting efficient training and inference of graph neural networks. This method is particularly suitable for flow field modeling under the complex geometry of compressor blades. It helps neural networks accurately capture key features such as the boundary layer under limited memory conditions, and significantly improves the accuracy and robustness of flow field prediction.

[0025] In the sampled data, the flow field mesh on or around the blade surface is represented using a graph structure of points and edge indices. Each node represents a grid point, and edges represent the spatial connections between nodes, forming the input data structure for a graph neural network in a non-Euclidean domain. This graph structure can flexibly adapt to complex boundaries and unstructured meshes, which is beneficial for capturing the non-uniform characteristics of spatial flow.

[0026] In one embodiment of this application, the K-hop neighborhood sampling includes: Seed nodes are selected by random sampling; We employ K-hop breadth-first search to construct local subgraphs and learn complete local representations based on graph message passing mechanisms.

[0027] In one embodiment of this application, such as Figure 3As shown, it is a schematic diagram of the K-hop sampling process. First, initialize an empty list Subgraphs to store the local subgraphs of all seed points. For each seed point s ∈ S, perform the following operations: Initialize the set visited = {}, representing the visited nodes; Initialize the set frontier = {s}, representing the boundary nodes currently being explored; Initialize the variable depth = 0, representing the current traversal depth. Then, enter the K-hop local search loop, with the conditions depth < K and frontier not being empty: Initialize an empty set next_frontier = {} to store the nodes to be visited in the next hop; Traverse each node v in the current frontier: Add v to the visited set; Traverse all neighbor nodes u ∈ Neighbor(v) of v: If u is not in visited, add u to next_frontier. Update frontier = next_frontier and depth = depth + 1. When the K-hop sampling is completed, construct the subgraph corresponding to this seed point: The set of subgraph nodes is: sub_nodes = visited ∪ frontier, and the set of subgraph edges is: sub_edges = {(u, v) ∈ E | u, v ∈ sub_nodes}. Add the subgraph Graph(sub_nodes, sub_edges) to the Subgraphs list. Repeat the above steps until all seed points are traversed, and return the set of all subgraphs Subgraphs.

[0028] In this embodiment, local subgraphs with compact structures and complete information are extracted through K-hop sampling, and distributed or batch graph network training can be performed while retaining the physical flow characteristics.

[0029] S205: Construct a multi-branch flow field prediction model, including an encoder, a geometric branch module, a flow field branch module, a feature fusion module, and a decoder.

[0030] In the embodiment of the present application, to effectively fuse geometric perception and fluid physical characteristics, a multi-branch flow field prediction model is constructed, including an encoder, a geometric branch module, a flow field branch module, a feature fusion module, and a decoder, as Figure 4The diagram shows the structure of a multi-branch flow field prediction model. The main body of the model is Transolver, which includes an encoder and a decoder. An intermediate enhancement module (Processor) is added between the encoder and decoder for improvement. The geometry branch module uses U-Net to extract multi-scale features specifically for the signed distance function (SDF), enhancing the global and local geometric perception of the complex shape of the blade. The flow field branch focuses on the topology and physical coupling characteristics of the flow field, using GNN and Transolver to deeply mine the flow details. The feature fusion module fuses the outputs of the geometry branch and the flow field branch through feature concatenation. Finally, the decoder obtains the predicted key physical quantities such as velocity and pressure around the blade.

[0031] Specifically, in one embodiment of this application, the geometric branch module is a multi-scale feature extraction module based on the U-Net architecture, including an encoder, a decoder, and cross-layer connections.

[0032] In one embodiment of this application, U-Net is a classic fully convolutional neural network (FCN) architecture, primarily used for biomedical image segmentation tasks. Its name derives from the symmetrical "U"-shaped layout of its network structure. Figure 5 As shown, the U-Net structure is characterized by: downsampling the input feature map layer by layer by the encoder to obtain global contextual information, then upsampling layer by layer by the decoder to restore spatial resolution, and preserving local details through skip connections, thus establishing an effective bridge between high-level semantic representation and low-level spatial information. In complex fluid dynamics simulation tasks, the flow field has obvious multi-scale characteristics—from micro-scale boundary layer perturbations to macro-scale overall flow organization, which poses a modeling requirement for neural networks to "simultaneously focus on both local and global aspects." The multi-scale structure of U-Net perfectly meets this requirement: the encoder path learns the large-scale features of the global flow field, the decoder path preserves fine-grained local information, and skip connections provide a bridge between the original high-resolution information and deep semantic features, enhancing the fitting ability for high-frequency phenomena such as turbulence.

[0033] This invention introduces a multi-scale feature extraction module based on the U-Net architecture. During encoding, features are compressed layer by layer to extract global flow patterns, and during decoding, spatial details are gradually restored. Key local information is preserved through a cross-layer connection mechanism. This structure effectively achieves the fusion of global and local features, enhancing the model's ability to perceive and express flow behavior at different scales. It exhibits higher accuracy and robustness in handling complex flows within blade channels and strengthening boundary layer fitting, providing crucial support for achieving high-resolution, high-fidelity compressor flow field prediction.

[0034] Specifically, in one embodiment of this application, the flow field branching module is a local modeling and attention enhancement module based on graph neural networks and Transolver.

[0035] In one embodiment of this application, a graph neural network (GNN) is used to model local topological features in the flow field. Transformer introduces an attention mechanism to dynamically focus on feature learning in key regions such as the boundary layer. While maintaining the global perception advantage of Transformer, it integrates a position perception mechanism and a local attention module, which enhances the model's ability to identify key physical regions such as the boundary layer and vortex region.

[0036] S207: The multi-branch flow field prediction model is trained using the graph neural network input data structure.

[0037] In this embodiment, a multi-branch flow field prediction model is trained using the obtained graph neural network input data structure, and the corresponding computational fluid dynamics (CFD) data calculated using traditional numerical iteration methods is used as label data for supervision. Specifically, the model is configured with a 128-dimensional hidden layer, and node feature interaction is achieved through 8 message passes to enhance the continuity of flow field information transmission; 4 attention blocks are introduced to improve the model's ability to learn complex coupling relationships. The initial learning rate is set to 0.001, and the FixLR fixed learning rate scheduler is used to maintain training stability and avoid interference from dynamic learning rates on the geometry-flow field mapping; the regularization parameter is set to 1.0 to suppress the risk of overfitting and ensure the model's generalization ability. A total of 600 training rounds are executed to fully fit the flow field distribution law. At the same time, since this method achieves model training constraints through a pure data-driven approach, the loss function is based on the CFD flow field calculation results obtained through numerical iteration. The flow field data predicted by the neural network is compared with it, and the loss function is constructed by calculating the mean square error of the two, thereby constraining the model and evaluating the model training effect.

[0038] S209: Input the geometric data of the compressor blade to be predicted into the trained multi-branch flow field prediction model to obtain the flow field prediction results.

[0039] In this embodiment of the application, after the model training is completed, the initial geometry of the new tablet press blade to be predicted is input, and the model quickly generates the flow field prediction result without the need for an iterative solution process, thereby significantly improving the prediction efficiency and engineering practicality.

[0040] like Figure 6 The figure shows the specific implementation steps of a compressor blade flow field prediction method based on multi-scale features and attention enhancement in a specific embodiment.

[0041] The aforementioned compressor blade flow field prediction method based on multi-scale features and attention enhancement first acquires compressor blade geometric data and performs gridded preprocessing. Then, K-hop neighborhood sampling is performed on the preprocessed geometric data, and a graph neural network input data structure is constructed based on the point set and edge connection relationship. Next, a multi-branch flow field prediction model is constructed, including an encoder, a geometric branch module, a flow field branch module, a feature fusion module, and a decoder. Then, the multi-branch flow field prediction model is trained using the graph neural network input data structure. Finally, the compressor blade geometric data to be predicted is input into the trained multi-branch flow field prediction model to obtain the flow field prediction result. In other words, by introducing a graph structure perception mechanism based on K-hop neighborhood sampling, combined with U-Net to effectively extract multi-scale flow features, and integrating the Transolver model to enhance boundary layer attention and reduce attention to the far field, a high-precision prediction model suitable for compressor blade flow fields is constructed. This not only significantly improves the overall accuracy and boundary fitting ability of flow field prediction but also demonstrates good performance in terms of computational efficiency and model scalability. This provides a feasible path for data-driven CFD modeling and offers a practical technical solution for high-precision prediction of complex engineering flow fields.

[0042] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0043] Based on the same inventive concept, this application also provides a compressor blade flow field prediction device based on multi-scale features and attention enhancement for implementing the compressor blade flow field prediction method based on multi-scale features and attention enhancement described above. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the compressor blade flow field prediction device based on multi-scale features and attention enhancement provided below can be found in the limitations of the compressor blade flow field prediction method based on multi-scale features and attention enhancement described above, and will not be repeated here.

[0044] In one embodiment, such as Figure 7 As shown, a compressor blade flow field prediction device 700 based on multi-scale features and attention enhancement is provided, including: a geometric modeling and mesh generation module 701, a graph structure representation and data encoding module 703, a prediction model construction module 705, a prediction model training module 707, and a flow field prediction module 709, wherein: The geometric modeling and mesh generation module 701 is used to acquire compressor blade geometric data and perform mesh preprocessing.

[0045] The graph structure representation and data encoding module 703 is used to perform K-hop neighborhood sampling based on the preprocessed gridded geometric data and to construct the input data structure of the graph neural network based on the connection relationship between the point set and the edge.

[0046] The prediction model building module 705 is used to build a multi-branch flow field prediction model, including an encoder, a geometric branch module, a flow field branch module, a feature fusion module, and a decoder.

[0047] The prediction model training module 707 is used to train the multi-branch flow field prediction model using the graph neural network input data structure.

[0048] The flow field prediction module 709 is used to input the geometric data of the compressor blade to be predicted into the trained multi-branch flow field prediction model to obtain the flow field prediction results.

[0049] In one embodiment of this application, the geometric modeling and mesh generation module is further used for: Based on the generation of a symbolic distance function field from geometric points, the geometric distance characteristics in discrete space are described.

[0050] In one embodiment of this application, the graph structure representation and data encoding module is further configured to: Seed nodes are selected by random sampling; We employ K-hop breadth-first search to construct local subgraphs and learn complete local representations based on graph message passing mechanisms.

[0051] In one embodiment of this application, the geometric branch module is a multi-scale feature extraction module based on the U-Net architecture, including an encoder, a decoder, and cross-layer connections.

[0052] In one embodiment of this application, the flow field branching module is a local modeling and attention enhancement module based on graph neural networks and Transolver.

[0053] The modules in the aforementioned compressor blade flow field prediction device based on multi-scale features and attention enhancement can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the computer device's memory as software, so that the processor can call and execute the corresponding operations of each module.

[0054] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 8 As shown, the computer device includes a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When executed by the processor, the computer program implements a compressor blade flow field prediction method based on multi-scale features and attention enhancement. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device casing, or an external keyboard, touchpad, or mouse.

[0055] Those skilled in the art will understand that Figure 8 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0056] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0057] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0058] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0059] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0060] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0061] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0062] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for predicting the flow field of compressor blades based on multi-scale features and attention enhancement, characterized in that, The method includes: Obtain compressor blade geometry data and perform mesh preprocessing; K-hop neighborhood sampling is performed on the pre-processed geometric data, and the input data structure of the graph neural network is constructed based on the connection relationship between the point set and the edge. A multi-branch flow field prediction model is constructed, including an encoder, a geometric branch module, a flow field branch module, a feature fusion module, and a decoder; The multi-branch flow field prediction model is trained using the graph neural network input data structure. The geometric data of the compressor blade to be predicted are input into the trained multi-branch flow field prediction model to obtain the flow field prediction results.

2. The compressor blade flow field prediction method based on multi-scale features and attention enhancement according to claim 1, characterized in that, The meshing preprocessing includes: Based on the generation of a symbolic distance function field from geometric points, the geometric distance characteristics in discrete space are described.

3. The compressor blade flow field prediction method based on multi-scale features and attention enhancement according to claim 1, characterized in that, The K-hop neighborhood sampling includes: Seed nodes are selected by random sampling; We employ K-hop breadth-first search to construct local subgraphs and learn complete local representations based on graph message passing mechanisms.

4. The compressor blade flow field prediction method based on multi-scale features and attention enhancement according to claim 1, characterized in that, The geometric branch module is a multi-scale feature extraction module based on the U-Net architecture, including an encoder, a decoder, and cross-layer connections.

5. The compressor blade flow field prediction method based on multi-scale features and attention enhancement according to claim 1, characterized in that, The flow field branching module is a local modeling and attention enhancement module based on graph neural networks and Transolver.

6. A compressor blade flow field prediction device based on multi-scale features and attention enhancement, characterized in that, The device includes: The geometric modeling and mesh generation module is used to acquire compressor blade geometric data and perform mesh preprocessing. The graph structure representation and data encoding module is used to perform K-hop neighborhood sampling based on the pre-processed gridded geometric data, and to construct the input data structure of the graph neural network based on the connection relationship between the point set and the edge. The prediction model building module is used to build a multi-branch flow field prediction model, including an encoder, a geometric branch module, a flow field branch module, a feature fusion module, and a decoder; A prediction model training module is used to train the multi-branch flow field prediction model using the graph neural network input data structure. The flow field prediction module is used to input the geometric data of the compressor blade to be predicted into the trained multi-branch flow field prediction model to obtain the flow field prediction results.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 5.

8. 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 method according to any one of claims 1 to 5.

Images