A heliostat aerodynamic coefficient calculation method and system fusing geometric modal features

Abstract

The application discloses a heliostat aerodynamic coefficient calculation method and system fusing geometric modal characteristics, and belongs to the technical field of heliostat simulation. The existing heliostat aerodynamic coefficient prediction method does not consider the mutual influence and interaction between multi-modal variables, resulting in a large deviation between the aerodynamic coefficient prediction result and the actual situation. The heliostat aerodynamic coefficient calculation method fusing geometric modal characteristics fully considers the mutual influence and interaction between the multi-modal variables of the heliostat, mines the interaction relationship between the modal characteristics of the heliostat, introduces a graph convolution operation, represents the modal coupling data as a graph structure, represents the dependency relationship between the modal characteristics, and obtains accurate heliostat aerodynamic coefficients, so that the multi-physical field interaction of the heliostat can be effectively modeled, the aerodynamic characteristics of the heliostat in a complex environment can be accurately reflected, and the heliostats of different design schemes can be quickly evaluated and optimized.

Description

Technical Field

[0001] This invention relates to a method and system for calculating the aerodynamic coefficients of a heliostat by integrating geometric modal characteristics, belonging to the field of heliostat simulation technology. Background Technology

[0002] In existing technologies, methods for calculating the aerodynamic coefficients of heliostats mainly fall into two categories:

[0003] First, wind tunnel testing is used for calculation. In order to obtain accurate heliostat aerodynamic coefficients, a 1:1 scale heliostat model needs to be built, which is extremely costly. Moreover, the entire cycle from model building to test completion takes 3 to 6 months, which seriously limits the rapid evaluation and optimization of different design schemes.

[0004] Second, computational fluid dynamics (CFD) numerical simulations are used, but even with the simplest turbulence model k-ωSST, a single simulation on an NVIDIA V100 GPU still takes 8 to 12 hours. The long simulation process makes it difficult to respond quickly to design changes and operating conditions in actual engineering, and cannot meet the real-time requirements.

[0005] Furthermore, most existing methods for predicting the aerodynamic coefficients of heliostats only consider a single type of data and do not take into account the mutual influence and interaction between multimodal variables. This results in a large deviation between the predicted aerodynamic coefficients and the actual situation, making it difficult to accurately reflect the aerodynamic characteristics of heliostats in complex environments.

[0006] The information disclosed in this background section is only for understanding the background of the inventive concept, and therefore may include information that does not constitute prior art. Summary of the Invention

[0007] To address the aforementioned problems, or one of them, the present invention aims to provide a method and system for calculating the aerodynamic coefficients of a heliostat by integrating geometric modal features. This method fully considers the mutual influence and interaction between multimodal variables of the heliostat. By mining the interaction relationships between the modal features of the heliostat, modal coupling data is obtained. A graph convolution operation is introduced to represent the modal coupling data as a graph structure, and message propagation is performed on the graph structure to characterize the dependencies between modal features, forming graph embedding representation information. This graph embedding representation information is then aggregated and mapped to obtain the predicted aerodynamic coefficients of the heliostat. This allows for accurate prediction of the aerodynamic coefficients of the heliostat, thus enabling effective modeling of the multiphysics interactions of the heliostat and accurately reflecting its aerodynamic characteristics in complex environments. The predicted aerodynamic coefficients are close to reality. Furthermore, simulation calculations using this algorithm model effectively reduce modeling costs and shorten simulation time, facilitating rapid evaluation and optimization of heliostats with different design schemes and meeting real-time requirements.

[0008] To address the aforementioned problems or one of them, the second objective of this invention is to provide a method and system for calculating the aerodynamic coefficient of a heliostat by integrating geometric modal features. By integrating the geometric and physical modal features of the heliostat, the heliostat is modeled, enabling accurate and efficient prediction of the heliostat's aerodynamic coefficient. This provides a reliable basis for the stable operation and optimized design of tower solar thermal power plants, effectively solving the problems of low accuracy, poor real-time performance, and inability to effectively utilize multimodal data in existing heliostat aerodynamic coefficient prediction models.

[0009] To achieve one of the above objectives, the first technical solution of the present invention is as follows:

[0010] A method for calculating the aerodynamic coefficients of a heliostat by incorporating geometric modal characteristics includes the following steps:

[0011] Step 1: Process the geometric and atmospheric data of the heliostat to generate multiple modal features;

[0012] Step 2: Mining the interaction relationships between modal features to obtain modal coupling data;

[0013] Step 3: Introduce graph convolution operations to represent modal coupling data as a graph structure, and perform message propagation on the graph structure to characterize the dependencies between modal features and form graph embedding representation information;

[0014] Step four involves aggregating and mapping the graph embedding representation information to obtain the predicted heliostat aerodynamic coefficients, thus realizing the calculation of heliostat aerodynamic coefficients by fusing geometric modal features.

[0015] This invention fully considers the mutual influence and interaction among the multimodal variables of a heliostat. By mining the interaction relationships between the modal features of the heliostat, modal coupling data is obtained. Graph convolution operation is introduced to represent the modal coupling data as a graph structure, and message propagation is performed on the graph structure to characterize the dependencies between modal features, forming graph embedding representation information. The graph embedding representation information is then aggregated and mapped to obtain the predicted aerodynamic coefficients of the heliostat, thus enabling accurate prediction of the aerodynamic coefficients. Therefore, it can effectively model the multiphysics interaction of the heliostat, accurately reflect the aerodynamic characteristics of the heliostat in complex environments, and make the predicted aerodynamic coefficients close to the actual situation. Furthermore, simulation calculations using this algorithm model can effectively reduce modeling costs and shorten simulation time, facilitating rapid evaluation and optimization of heliostats with different design schemes, meeting real-time requirements. The solution is scientific, reasonable, and feasible.

[0016] Furthermore, this invention integrates the geometric and physical modal characteristics of heliostats to model them, enabling accurate and efficient prediction of heliostat aerodynamic coefficients. This provides a reliable basis for the stable operation and optimized design of tower solar thermal power plants, effectively solving the problems of low accuracy, poor real-time performance, and inability to effectively utilize multimodal data in existing heliostat aerodynamic coefficient prediction models.

[0017] Furthermore, this invention has good versatility and is applicable to heliostats of different structural forms (such as single-piece, spliced, curved mirrors, etc.) and different wind field environmental conditions (such as various wind speeds, wind directions, turbulence intensities, etc.), significantly improving the adaptability and practicality of the model in actual engineering.

[0018] As a preferred technical measure:

[0019] Step one involves processing the geometric and atmospheric data of the heliostat to generate multiple modal features, as follows:

[0020] Acquire the geometric data of the heliostat, including the three-dimensional point cloud data and the elevation angle data of the heliostat;

[0021] Acquire air data, including air density data, and combine it with historical meteorological data to construct air density data;

[0022] The three-dimensional point cloud data is processed to extract its geometric features, thus obtaining the first modal features;

[0023] Angle identification and encoding were performed on the heliostat elevation angle data to obtain the second modal features;

[0024] The air density data is normalized to obtain the third modal features;

[0025] The first modal feature, the second modal feature, and the third modal feature are fused to form multiple modal features.

[0026] As a preferred technical measure:

[0027] Step two involves mining the interaction relationships between modal features to obtain modal coupling data. The method is as follows:

[0028] Multiple modal features are obtained, including at least a first modal feature and a second modal feature. Each modal feature includes several positions, and each position is a d-dimensional feature vector.

[0029] Considering the differences between modal features and physical consistency constraints, a learnable weight matrix is ​​set for each modal feature; the learnable weight matrix is ​​obtained by training a deep cross-attention network, and the training objective is to minimize the physical constraint loss function.

[0030] Based on the learnable weight matrix, a linear transformation is performed on the modal features to obtain the query matrix, key matrix, and value matrix;

[0031] Calculate the attention score between the first modality feature and the second modality feature based on the query matrix and the key matrix;

[0032] The first fused feature of the first modality is obtained by weighted summation of the value matrix of the second modality feature based on the attention score; the second fused feature of the second modality feature is obtained by weighted summation of the value matrix of the first modality feature based on the attention score.

[0033] Based on the first and second fusion features, calculate the new query matrix, key matrix, and value matrix;

[0034] Based on the new query matrix and key matrix, calculate the attention score between the first fusion feature and the second fusion feature;

[0035] The attention score and fusion features are continuously updated iteratively to achieve deep interaction between modal features at multiple semantic levels. Finally, the first and second fusion features are summarized to obtain modal coupling data.

[0036] As a preferred technical measure:

[0037] The method for training a deep cross-attention network is as follows:

[0038] Obtain a predicted heliostat aerodynamic coefficient and compare it with a reference aerodynamic coefficient value, then calculate the mean square error between the two.

[0039] Based on mean squared error, weighting coefficients, and penalty coefficients, a physical constraint loss function is constructed with the loss value as the objective.

[0040] Global interaction features between different modalities are extracted using a deep cross-attention network;

[0041] Subsequently, a graph structure is constructed among modal features, and local dependencies and topological associations are modeled through graph convolution;

[0042] Then, based on the deep cross-attention network and graph structure, the aerodynamic coefficient prediction results are output, and the physical constraint loss function is used as the optimization objective to guide the entire deep cross-attention network to maintain physical consistency while fitting the aerodynamic coefficients; an end-to-end joint learning network is constructed through variable dependency and backpropagation mechanisms.

[0043] As a preferred technical measure:

[0044] Step 3: Introduce graph convolution operations to represent modal coupling data as a graph structure, and perform message propagation on the graph structure to characterize the dependencies between modal features. The method for forming graph embedding representation information is as follows:

[0045] Based on the observation points on the surface of the heliostat and their corresponding location information, several nodes are constructed;

[0046] Based on the modal coupling data and the corresponding location information, the nodes are assigned values ​​so that each node has a geometric mode and a physical mode; and the geometric mode and physical mode are fused to form a node feature vector;

[0047] Based on modal coupling data, the relationships between features are obtained; geometric edges are constructed based on the relationships between features.

[0048] Draw the graph structure based on geometric edges, nodes, and node feature vectors;

[0049] Based on deep cross-attention networks and graph convolutional networks, and combined with the node feature vectors of neighboring nodes, the node feature vectors are updated, so that local modal features can be propagated to the global level and can represent the dependencies and structural associations between modal features.

[0050] As a preferred technical measure:

[0051] The relationships between features include geometric adjacency, feature similarity, and physical prior relationships;

[0052] Geometric adjacency relationships are established based on the location of point clouds or structural elements in space and using a nearest neighbor algorithm to reflect local physical interactions.

[0053] Feature similarity relationships are established by calculating the cosine similarity of modal features, and are used to capture coupling characteristics that are far apart but have similar behaviors.

[0054] Physical a priori relationships, established based on structural connections and the relationships between airflow paths and heat transfer pathways, are used to enhance the physical interpretability of the graph structure.

[0055] As a preferred technical measure:

[0056] The method for updating node feature vectors based on deep cross-attention networks and graph convolutional networks, combined with the node feature vectors of neighboring nodes, is as follows:

[0057] First, through a deep cross-attention network, the node feature vectors are cross-attentioned layer by layer, and combined with the attention mechanism, multimodal fusion features are generated, which include several fusion feature units.

[0058] Multimodal fusion features are used as input nodes of a graph convolutional network to construct a graph structure, which includes several nodes and multiple edges. Nodes represent fusion feature units, and edges represent physical dependencies, spatial connections, or structural similarities between feature units.

[0059] Based on the node feature vectors of neighboring nodes, construct the adjacency matrix;

[0060] In graph convolutional networks, the message propagation mechanism controlled by the adjacency matrix enables the layer-by-layer updating of node feature vectors, allowing node feature vectors to interact with local modal features and incorporate topological information from the global physical field.

[0061] As a preferred technical measure:

[0062] Step four involves aggregating and mapping the graph embedding representation information to obtain the predicted heliostat aerodynamic coefficients, as follows:

[0063] A formula for calculating coefficients is constructed based on the weight matrix, bias vector, and activation function.

[0064] Based on the coefficient calculation formula, multiple hidden layers are established;

[0065] The graph embedding representation information is input into multiple hidden layers for multi-layer computation to obtain the final node features;

[0066] The final features of the nodes are aggregated and weighted, and then input into the regression function for calculation to obtain the predicted heliostat aerodynamic coefficients; the heliostat aerodynamic coefficients include drag coefficients and pressure coefficients.

[0067] To achieve one of the above objectives, the second technical solution of the present invention is as follows:

[0068] A method for calculating the aerodynamic coefficients of a heliostat by incorporating geometric modal characteristics, comprising the following:

[0069] Using a pre-built multi-source data processing model, the geometric and atmospheric data of the heliostat are processed to generate multiple modal features;

[0070] Based on a pre-built feature capture model, the interaction relationships between modal features are mined to obtain modal coupling data;

[0071] Based on the pre-built feature propagation model, graph convolution operation is introduced to represent modal coupling data as a graph structure, and message passing is performed on the graph structure to characterize the dependencies between modal features and form graph embedding representation information;

[0072] By using a pre-built aerodynamic simulation model, the graph embedding representation information is aggregated and mapped to obtain the predicted heliostat aerodynamic coefficients, thus realizing the calculation of heliostat aerodynamic coefficients by fusing geometric modal features.

[0073] This invention integrates the geometric and physical modal characteristics of heliostats to model them, achieving accurate and efficient prediction of heliostat aerodynamic coefficients. This provides a reliable basis for the stable operation and optimized design of tower solar thermal power plants, effectively solving the problems of low accuracy, poor real-time performance, and ineffective utilization of multimodal data in existing heliostat aerodynamic coefficient prediction models. Furthermore, this invention has good versatility, applicable to heliostats of different structural forms (such as monolithic, modular, and curved mirrors) and different wind field environmental conditions (such as various wind speeds, directions, and turbulence variations), significantly improving the model's adaptability and practicality in actual engineering projects.

[0074] To achieve one of the above objectives, the third technical solution of the present invention is as follows:

[0075] A system for calculating the aerodynamic coefficients of a heliostat that integrates geometric modal features includes a multi-source data processing module, a feature capture module, a feature propagation module, and an aerodynamic simulation module.

[0076] The multi-source data processing module is used to process the geometric and atmospheric data of the heliostat to generate multiple modal features;

[0077] The feature capture module is used to mine the interaction relationships between modal features to obtain modal coupling data;

[0078] The feature propagation module is used to introduce graph convolution operations, represent modal coupling data as a graph structure, and perform message passing on the graph structure to characterize the dependencies between modal features and form graph embedding representation information;

[0079] The aerodynamic simulation module is used to aggregate and map graph embedding representation information to obtain the predicted heliostat aerodynamic coefficients, realizing the calculation of heliostat aerodynamic coefficients by fusing geometric modal features.

[0080] Compared with existing technical solutions, the present invention has the following beneficial effects:

[0081] This invention fully considers the mutual influence and interaction among the multimodal variables of a heliostat. By mining the interaction relationships between the modal features of the heliostat, modal coupling data is obtained. Graph convolution operation is introduced to represent the modal coupling data as a graph structure, and message propagation is performed on the graph structure to characterize the dependencies between modal features, forming graph embedding representation information. The graph embedding representation information is then aggregated and mapped to obtain the predicted aerodynamic coefficients of the heliostat, thus enabling accurate prediction of the aerodynamic coefficients. Therefore, it can effectively model the multiphysics interaction of the heliostat, accurately reflect the aerodynamic characteristics of the heliostat in complex environments, and make the predicted aerodynamic coefficients close to the actual situation. Furthermore, simulation calculations using this algorithm model can effectively reduce modeling costs and shorten simulation time, facilitating rapid evaluation and optimization of heliostats with different design schemes, meeting real-time requirements. The solution is scientific, reasonable, and feasible.

[0082] Furthermore, this invention models the heliostat by integrating its geometric and physical modal characteristics, thereby accurately capturing the implicit correlations between different modes and fully considering the multi-physics interaction characteristics. This enables accurate and efficient prediction of the heliostat's aerodynamic coefficient, providing a reliable basis for the stable operation and optimized design of tower solar thermal power plants. It can effectively solve the problems of low accuracy, poor real-time performance, and inability to effectively utilize multi-modal data in existing heliostat aerodynamic coefficient prediction models.

[0083] Furthermore, this invention has good versatility and is applicable to heliostats of different structural forms (such as single-piece, spliced, curved mirrors, etc.) and different wind field environmental conditions (such as various wind speeds, wind directions, turbulence intensities, etc.), significantly improving the adaptability and practicality of the model in actual engineering. Attached Figure Description

[0084] Figure 1 This is a schematic flowchart of the method for calculating the aerodynamic coefficient of a heliostat according to the present invention;

[0085] Figure 2 This is a schematic diagram of a framework for the heliostat aerodynamic coefficient calculation system of the present invention;

[0086] Figure 3 A flowchart illustrating data processing using the deep cross-attention network of this invention;

[0087] Figure 4 This is a flowchart illustrating one aspect of graph convolution processing in this invention. Detailed Implementation

[0088] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this application. This invention covers any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of the invention as defined by the claims.

[0089] like Figure 1 As shown, this invention provides a first specific embodiment of the method for calculating the aerodynamic coefficients of a heliostat by integrating geometric modal characteristics:

[0090] A method for calculating the aerodynamic coefficients of a heliostat by incorporating geometric modal characteristics includes the following steps:

[0091] Step 1: Process the geometric and atmospheric data of the heliostat to generate multiple modal features;

[0092] Step 2: Mining the interaction relationships between modal features to obtain modal coupling data;

[0093] Step 3: Introduce graph convolution operations to represent modal coupling data as a graph structure, and perform message propagation on the graph structure to characterize the dependencies between modal features and form graph embedding representation information;

[0094] Step four involves aggregating and mapping the graph embedding representation information to obtain the predicted heliostat aerodynamic coefficients, thus realizing the calculation of heliostat aerodynamic coefficients by fusing geometric modal features.

[0095] A second specific embodiment of the heliostat aerodynamic coefficient calculation method integrating geometric modal characteristics of the present invention:

[0096] A method for calculating the aerodynamic coefficients of a heliostat by incorporating geometric modal characteristics, comprising the following:

[0097] Using a pre-built multi-source data processing model, the geometric and atmospheric data of the heliostat are processed to generate multiple modal features;

[0098] Based on a pre-built feature capture model, the interaction relationships between modal features are mined to obtain modal coupling data;

[0099] Based on the pre-built feature propagation model, graph convolution operation is introduced to represent modal coupling data as a graph structure, and message passing is performed on the graph structure to characterize the dependencies between modal features and form graph embedding representation information;

[0100] By using a pre-built aerodynamic simulation model, the graph embedding representation information is aggregated and mapped to obtain the predicted heliostat aerodynamic coefficients, thus realizing the calculation of heliostat aerodynamic coefficients by fusing geometric modal features.

[0101] A third specific embodiment of the heliostat aerodynamic coefficient calculation method integrating geometric modal characteristics of the present invention:

[0102] A method for calculating the aerodynamic coefficients of a heliostat by incorporating geometric modal characteristics includes the following steps:

[0103] Step 1. Fuse the multi-source data from the heliostat to obtain multiple modal features, including the following:

[0104] A fused feature vector is obtained by processing and extracting features from multi-source data from a heliostat. This multi-source data includes 3D point cloud data, air density data, and heliostat elevation angle data. The 3D point cloud data is processed to extract its geometric features; the air density data is normalized; and the heliostat elevation angle data is identified and encoded. The processed modal features are then fused to form a fused feature vector containing rich information, providing comprehensive and crucial input data for subsequent aerodynamic coefficient prediction.

[0105] Step 2. Mine the interaction relationships between modal features to obtain modal coupling data; then introduce graph convolution operation to represent the modal coupling data as a graph structure, and perform message propagation on the graph structure to characterize the dependencies between modal features and form graph embedding representation information.

[0106] In this embodiment, the method for mining the interaction relationships between modal features to obtain modal coupling data is as follows:

[0107] A deep cross-attention network is constructed to achieve mutual attention between different modalities at multiple levels. This network iterates multiple times, calculating attention scores between features of different modalities and assigning dynamically changing weights to each feature.

[0108] A deep cross-attention network is constructed based on the cross-attention mechanism to more deeply explore the interaction relationships between different modalities of data. Let the input multimodal features be as follows: Each of them The data representing the i-th modal feature (e.g., point cloud, air density, elevation angle, etc.) consists of M modalities, each containing There are positions (or nodes), each position being... 3D eigenvectors.

[0109] First, a linear transformation is performed on each modal feature to obtain the query matrix, key matrix, and value matrix. For the first... Modal features Its query matrix Key matrix Value matrix ,in , The weight matrix is ​​a learnable matrix. For the attention dimension.

[0110] The learnable weight matrix in this embodiment , All features were automatically optimized during the training phase of a deep cross-attention network, with the training objective being to minimize the designed physical constraint loss function. This training process fully considers the modal differences and physical consistency constraints of the input features, ensuring that the fused features retain physical interpretability while possessing strong fitting and generalization abilities.

[0111] Then, the attention scores between different modal features are calculated, for modal feature i and modal feature i. Attention score The calculation formula is as follows:

[0112]

[0113] in, It is the dimension of the key vector. For activation function, Modal features The query matrix, Modal features The bond matrix.

[0114] Modal features refer to data types or information sources with different physical meanings, such as 3D point cloud data, air density, heliostat elevation angle, and structural parameters. These modal features, due to differences in acquisition methods, physical dimensions, and semantic interpretations, need to be modeled and fused separately in the model; therefore, they are collectively referred to as multimodal data. A key matrix is ​​a matrix formed by combining all key vectors row-wise. The dimension of the key vectors... The feature dimension of each key vector is determined by the feature length of each feature point after linear mapping. In this embodiment... The value is set to 64. Experiments have shown that 64 is more suitable for fine-grained matching and attention calculation between point clouds and physical quantities.

[0115] Activation function The expression for row-wise calculation is as follows:

[0116]

[0117] in, The output value of the i-th node, where C is the number of output nodes, i.e., the number of categories; and e is the natural constant. The output value of the c-th node.

[0118] Then, the value matrix of modality feature j is weighted and summed according to the attention weights to obtain the fused feature of modality feature i, and the calculation formula is as follows:

[0119]

[0120] in, Modal features Each location feature is the result of a weighted fusion of all location features of modality feature j.

[0121] In deep cross-attention networks, multi-layer attention computation is performed. Let the first... Layer modal features The fusion characteristics are Then the first The query, key, and value matrices of the layer are further generated based on the fused features of the previous layer, and their calculation formula is as follows:

[0122]

[0123] Similarly defined and By continuously iterating and updating attention scores and fusion features, deep interactions between modalities can be achieved at multiple semantic levels, thereby improving the ability to model complex physical coupling relationships.

[0124] In this embodiment, graph convolution operations are introduced to represent modal coupling data as a graph structure, and message propagation is performed on the graph structure to characterize the dependencies between modal features. The method for forming graph embedding representation information is as follows:

[0125] Data from different modalities is represented as a graph structure, with nodes representing features and edges representing relationships between features. Graph convolution operations are introduced to perform message passing and feature aggregation on the graph. By updating the feature representations of nodes through a graph convolutional network, the influence of multiphysics on the heliostat's aerodynamic coefficients can be effectively captured, enhancing the ability to model physical relationships between modalities.

[0126] Representing data of different modalities as a graph structure ,in A set of nodes represents the features of multimodal data; Let be the set of edges, which represents the relationships between features.

[0127] In this embodiment, the node set V represents the observation points on the surface of the heliostat. Each node contains the geometric modal features (such as coordinates, normal vector, curvature, etc.) and physical modal features (such as local velocity, pressure, etc.) of its corresponding position. After fusion, they form a node feature vector. The edge set E represents the adjacency relationship between nodes (such as Euclidean distance, flow field correlation), which is used for information propagation and aggregation in the subsequent graph neural network.

[0128] In this invention, the relationships between features are modeled by constructing edges in a graph structure. The relationships between features include geometric adjacency relationships, feature similarity relationships, and physical prior relationships.

[0129] Geometric adjacency relationships include the following:

[0130] Based on the location of point clouds or structural elements in space, the K-nearest neighbor algorithm is used to establish edges between nodes, reflecting local physical interactions.

[0131] Feature similarity relationships include the following:

[0132] Calculate the cosine similarity of modal features between nodes, establish connections between nodes with high similarity, and capture the coupling characteristics of nodes that are far apart but have similar behaviors.

[0133] Physical a priori relations include the following:

[0134] By combining computational fluid dynamics (CFD) simulation data or structural connection relationships, physical dependencies such as airflow paths and heat transfer pathways are defined to enhance the physical interpretability of the graph structure.

[0135] Ultimately, the three relationships are fused to construct a graph structure G=(V,E), providing basic topological support for graph neural networks and enabling deep intermodal fusion and physical information propagation.

[0136] Further, graph convolution operations are introduced for nodes. The feature update is performed using the following formula:

[0137]

[0138] in: For nodes The set of neighboring nodes; , They are nodes The degree of v; For the first The learnable weight matrix of the layer; For example, activation functions (such as ReLU); The updated features are those derived from integrating neighbor information; For neighbor information.

[0139] In this embodiment, the nodes in the graph structure represent spatial observation points on the surface of the heliostat. Their feature vectors consist of the features of that point in multiple modalities, including geometric features (such as coordinates, normal vectors, and curvature) and physical features (such as local wind speed, air pressure, and density). These modal features are transformed linearly to obtain the query matrix, key matrix, and value matrix required by the attention network, and then cross-fused and propagated through graph convolution in the graph structure.

[0140] Furthermore, by performing multiple graph convolution operations and incorporating information from neighboring nodes, the feature representations of nodes are continuously updated, further enhancing the ability to model the physical relationships between modalities and enabling the propagation of local modal features to the global model. Simultaneously, deep cross-attention networks and graph convolutional networks are integrated to model the interaction relationships and structural associations between multimodal features. Specifically, the following steps are included:

[0141] First, a deep cross-attention network is used to perform layer-by-layer cross-attention on the original modal features (such as point clouds, air density, heliostat elevation angle, etc.) to generate a multimodal fusion feature representation F. The corresponding attention mechanism is as follows:

[0142]

[0143] After obtaining the fused modal features F, this embodiment further uses them as input nodes of a graph neural network to construct a graph structure G=(V,E), where node V represents the fused feature unit and edge E represents the physical dependency, spatial connection or structural similarity between feature units.

[0144] In Graph Convolutional Networks (GNNs), the message propagation mechanism controlled by the adjacency matrix is ​​used to update node features layer by layer. The update formula is as follows:

[0145]

[0146] This update mechanism ensures that the fused features, based on local modal interactions, further incorporate the topological information of the global physical field, thereby improving the model's prediction accuracy and physical consistency for heliostat aerodynamic coefficients.

[0147] Step 3. Train and optimize the model network, which includes the following:

[0148] The physical constraint loss function is defined as follows:

[0149]

[0150]

[0151] Where L represents the final loss value used for training optimization. and The weighting coefficients and + =1, satisfying the normalization constraint, MSE( ) and MSE ( , where are the mean square errors of the drag coefficient and pressure coefficient, respectively, and η and γ are penalty coefficients. These are determined through hyperparameter grid search. =0.6, =0.4, η=0.5, γ=0.3, guiding the model network to learn features that conform to physical laws; This represents the predicted value of the drag coefficient corresponding to the i-th sample or prediction point; This represents the predicted pressure coefficient value corresponding to the i-th sample or prediction point.

[0152] In this framework, a deep cross-attention network is first used to extract global interaction features between different physical modalities. Then, a graph structure is constructed between these modal features, and local dependencies and topological associations are further modeled through graph convolution. Finally, the aerodynamic coefficient prediction results are output, and a physical constraint loss function is used as the optimization objective to guide the entire model to maintain physical consistency while fitting labeled data. These three components, through variable dependencies and backpropagation mechanisms, constitute a joint learning system that can be trained end-to-end.

[0153] The true values ​​(benchmark values) of the drag and pressure coefficients are provided by computational fluid dynamics (CFD) simulation data as monitoring signals. The mean squared error (MSE) between the predicted values ​​output by the model and these true labels is incorporated into the physical constraint loss function to ensure that the constructed model not only fits well at the data level but also meets basic physical consistency requirements.

[0154] Step 4. Make predictions using the trained model network, which includes the following:

[0155] A fully connected neural network is used as the drag coefficient prediction head, with the fused feature vector as input. The network contains multiple hidden layers. The calculation formula for each layer is as follows:

[0156]

[0157] in, It is a weight matrix. σ is the bias vector, and σ is the activation function.

[0158] After multiple calculations, the predicted drag coefficient is output. .

[0159] bias vector The specific value is automatically learned during model training, and the learning process is as follows:

[0160] S1. During the initialization phase, i.e., when the network is built, the bias vector of each layer is assigned all zeros to complete the initialization operation. S2. During the forward propagation phase, the bias is simply a vector with the same dimension as the output, which is added to the output variable of each node. S3. During the backpropagation phase, i.e., during training, the gradient of the loss with respect to the bias is calculated using the chain rule based on the loss between the network output and the true target; these gradients are updated by the optimizer Adam. After multiple iterations of forward and backpropagation, the values ​​of the bias vector gradually converge to the set of parameter values ​​that minimize the loss function.

[0161] In this embodiment, the multi-layer calculation method is as follows:

[0162] Step 1: Obtain the input data. The input data consists of multiple physically meaningful modal feature vectors, and their expressions are as follows:

[0163]

[0164] Each of them Indicates the first Data features of each modality, for example: The shape characteristics of point clouds, This is the air density value. For elevation angle parameters, This refers to wind field characteristics.

[0165] Step 2: Perform deep cross-attention fusion on the input data, which includes the following:

[0166] (1) For each modal feature Learnable matrices are used to transform query matrices, key matrices, and value matrices. The expression for a learnable matrix is ​​as follows:

[0167]

[0168] (2) For modal and Calculate the cross-attention score, i.e., the attention weight, the first... The formula for calculating the layer attention weights is as follows:

[0169]

[0170] (3) Perform feature fusion to obtain fused features. Its expression is as follows:

[0171]

[0172] go through The layer attention mechanism iteratively updates to obtain the fused feature representation. This refers to modal coupling data.

[0173] Step 3: Model the graph structure and perform graph convolution processing, which includes the following:

[0174] (1) Construct a graph structure G=(V,E), where the set of nodes is... Each node represents a feature vector obtained after modality fusion (i.e., the one obtained in the previous step). (Each part). The edge set E is obtained from the physical correlation or Euclidean distance / structural dependency between features, and the edge weights can be set based on correlation coefficients or prior knowledge.

[0175] (2) Perform graph convolution operation on each layer. nodes Graph embedding represents information To perform the update, the expression is as follows:

[0176]

[0177] in: It is the set of adjacent nodes; For the first Layered graph convolution can learn matrices; The initial input for the graph convolution is the output of the cross-attention step from the previous step. For the first The fusion characteristics of layer node i; The activation function is ReLU.

[0178] go through After the layer graph convolution operation, the final graph embedding representation information is obtained. .

[0179] Step 4: Map the final features of all nodes (graph embedding representation information) to predicted drag coefficient values ​​through aggregation and a nonlinear regression layer. Its expression is as follows:

[0180]

[0181] in: The feature aggregation method is calculated using a weighted summation method; The regression function can be a fully connected network or a linear layer.

[0182] Step 5: Supervised training of the loss function. During the training phase, the predicted values ​​are... Compared with the true value The mean squared error is calculated by comparing the two values, and the formula is as follows:

[0183]

[0184] Among them, pressure coefficient The prediction process and drag coefficient The prediction process is the same, and the same process calculates... Finally, it is combined with the physical constraint loss function, and its calculation formula is as follows:

[0185]

[0186] In this embodiment, the pressure coefficient prediction head is also a fully connected neural network, with a structure similar to the drag coefficient prediction head. It takes a fused feature vector as input, performs multi-layer calculations, and outputs the predicted pressure coefficient. .

[0187] This invention utilizes multimodal fusion feature input to uniformly encode and deeply fuse multi-source data from heliostats, including 3D point clouds, air density, elevation angle, structural parameters, and wind field characteristics, significantly improving the accuracy and stability of aerodynamic coefficient prediction. Furthermore, by jointly modeling the heliostat using a deep cross-attention network and a graph convolutional network, the model can accurately capture the implicit correlations between different modes and fully consider multiphysics interaction features.

[0188] The introduced structural and wind field parameter modes (modal variables) enable the model to adapt to various heliostat designs and changing environmental conditions, exhibiting good generalization ability. Experiments show that the prediction accuracy remains high under different mirror configurations and wind field combinations, and it can quickly respond to design changes and actual operating conditions, meeting the real-time and scalability requirements of engineering projects.

[0189] Furthermore, this invention can fully explore and utilize real-time and historical data resources from the heliostat design and flow field simulation process, maximizing data value. A unified feature encoding and fusion standard enables seamless integration of multimodal data and prediction models, significantly reducing system complexity and deployment costs.

[0190] like Figure 2 As shown, a specific embodiment of the heliostat aerodynamic coefficient calculation system integrating geometric modal characteristics of the present invention is presented:

[0191] A system for calculating the aerodynamic coefficients of a heliostat that integrates geometric modal features includes a multi-source data processing module, a feature capture module, a feature propagation module, and an aerodynamic simulation module.

[0192] The multi-source data processing module is used to process the geometric and atmospheric data of the heliostat to generate multiple modal features;

[0193] The feature capture module is used to mine the interaction relationships between modal features to obtain modal coupling data;

[0194] The feature propagation module is used to introduce graph convolution operations, represent modal coupling data as a graph structure, and perform message passing on the graph structure to characterize the dependencies between modal features and form graph embedding representation information;

[0195] The aerodynamic simulation module is used to aggregate and map graph embedding representation information to obtain the predicted heliostat aerodynamic coefficients, realizing the calculation of heliostat aerodynamic coefficients by fusing geometric modal features.

[0196] like Figure 3 and Figure 4 As shown, a specific embodiment of the present invention is used to calculate the aerodynamic coefficient of a heliostat in a tower solar thermal power plant:

[0197] The heliostat aerodynamic coefficient calculation method based on the fusion of geometric modal features of this invention is used to calculate the aerodynamic coefficient of a heliostat in a tower solar thermal power plant, which includes the following:

[0198] An energy company plans to construct a large-scale tower solar thermal power plant. Heliostats, as the core equipment of the plant, directly affect the plant's efficiency and stability due to their aerodynamic performance. Traditional wind tunnel experiments and CFD numerical simulations are costly and time-consuming, failing to meet the project's need for rapid evaluation and optimization of heliostat design. Therefore, the energy company decided to adopt the heliostat aerodynamic coefficient calculation method based on the integration of geometric modal characteristics, as described in this invention, to predict the heliostat's aerodynamic coefficients. This method includes the following:

[0199] Step 1: First, collect multi-source data, which includes the following:

[0200] A high-precision laser scanner was used to scan heliostats with different design schemes to obtain three-dimensional geometric information of the heliostats and obtain three-dimensional point cloud data.

[0201] Meteorological monitoring equipment was installed in the power plant site area to collect air density data in real time. At the same time, historical meteorological data was combined to construct an air density dataset.

[0202] The heliostat's control system records its elevation angle information at different times, thus obtaining the heliostat's elevation angle data.

[0203] The collected data is then preprocessed, including the following:

[0204] The 3D point cloud data obtained from the scan is denoised, filtered, and feature-extracted to extract the geometric features of the heliostat, such as area and shape. The collected air density data is then normalized, mapping the data to the [0,1] interval to eliminate the influence of dimensions. Finally, the heliostat elevation angle data is identified and encoded, converting the angle information into numerical features to obtain the encoded data of the heliostat elevation angle.

[0205] Step 2, building and training the model, includes the following:

[0206] The first step is to fuse the processed 3D point cloud features, air density features, and heliostat elevation angle features to form a fused feature vector (multiple modal features) containing rich information, which is then used as the input to the model.

[0207] The second step involves constructing a deep cross-attention network to process the input multimodal fusion features, specifically including the following:

[0208] First, a linear transformation is performed on the features of each modality to obtain the query matrix, key matrix, and value matrix. Then, attention scores between different modalities are calculated. The value matrices are then weighted and summed using these attention scores to obtain the fused features (interaction features). In the deep cross-attention network, multiple layers of attention calculation are performed, and the attention scores and fused features are continuously updated iteratively to more accurately capture the potential connections between modalities.

[0209] Further network enhancement is achieved using graph convolution, representing data from different modalities as a graph structure, where nodes represent features and edges represent relationships between features. Graph convolution operations are introduced to perform message passing and feature aggregation on the graph. Through multiple graph convolution operations, the feature representations of nodes are continuously updated, enhancing the ability to model the physical relationships between modalities.

[0210] Step 3: Design the physical constraint loss function to optimize the model. The expression for the physical constraint loss function is as follows:

[0211]

[0212] in =0.6, =0.4, η=0.5, γ=0.3.

[0213] The model is trained using collected multi-source data, and the optimal model parameters are determined through hyperparameter grid search to guide the model to learn features that conform to physical laws.

[0214] Step 4: Use the trained model to predict the drag and pressure coefficients of heliostats with different design schemes. Input the fused feature vectors into the drag coefficient prediction head and the pressure coefficient prediction head, and output the predicted drag coefficients respectively. and pressure coefficient .

[0215] Based on the predicted drag and pressure coefficients, heliostats of different designs (heliostats of different shapes) are evaluated and optimized. The design with the optimal aerodynamic performance is selected. and We will manufacture and install the smallest possible heliostat design to improve the efficiency and stability of the power plant.

[0216] Step 5: Compare the predicted drag coefficient and pressure coefficient with the actual measured values, and calculate evaluation indicators such as mean square error (MSE) and mean absolute error (MAE) to assess the model's prediction accuracy. The results show that the prediction accuracy of the model in this invention is significantly higher than that of traditional prediction methods.

[0217] Experiments have shown that, compared with traditional wind tunnel testing and CFD numerical simulation methods, the method of this invention significantly reduces costs and time. Wind tunnel testing requires constructing a 1:1 scale heliostat model, which is extremely costly and takes 3-6 months; while the method of this invention can predict the aerodynamic coefficients of the heliostat in a short time, providing strong support for the rapid advancement of projects. The above application examples demonstrate that the method of this invention has achieved significant results in the design optimization of heliostats in tower solar thermal power plants, providing crucial technical support for the construction and operation of the power plant.

[0218] A server embodiment applying the method of the present invention:

[0219] A server comprising:

[0220] One or more processing units;

[0221] Storage device for storing one or more programs;

[0222] When the one or more programs are executed by the one or more processing units, the one or more processing units implement the above-described method for calculating the aerodynamic coefficients of a heliostat by incorporating geometric modal features.

[0223] The storage device can be internal memory, external memory, cache memory, or other special memory. The processing unit has signal processing capabilities and can be a general-purpose processor, a digital signal processor, an application-specific integrated circuit, an off-the-shelf programmable gate array, or other programmable logic device.

[0224] An embodiment of a device applying the method of the present invention:

[0225] An electronic device is provided with a computer-readable storage medium storing a computer program, which, when executed by a processing unit, implements the above-described method for calculating the aerodynamic coefficient of a heliostat by incorporating geometric modal features.

[0226] Computer-readable storage media refers to physical carriers capable of storing computer-recognizable data, instructions, or programs. These media must meet the core characteristic of being "readable by a computer" (i.e., the data exists in the form of electrical, magnetic, or optical signals and can be converted into binary information that a computer can process through appropriate devices). The physical carrier can be a magnetic storage medium, optical storage medium, semiconductor storage medium, or other storage media.

[0227] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, optical storage, etc.) containing computer-usable program code.

[0228] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features within the scope of the technology disclosed in the present invention; and these modifications or substitutions will not cause the substance of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. Any modifications or equivalent substitutions that do not deviate from the spirit and scope of the present invention should be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for calculating the aerodynamic coefficients of a heliostat by fusing geometric modal features, comprising the following steps: Step 1: processing the geometric data and air data of the heliostat to generate multiple modal features; Step 2: mining the interaction between modal features to obtain modal coupling data; Step 3: introducing graph convolution operations to represent the modal coupling data as a graph structure and performing message propagation on the graph structure to represent the dependency between modal features and form graph embedding representation information; Step 4: aggregating and mapping the graph embedding representation information to obtain the predicted aerodynamic coefficients of the heliostat, realizing the calculation of the aerodynamic coefficients of the heliostat by fusing geometric modal features, which includes the following contents: constructing a coefficient calculation formula based on a weight matrix, a bias vector, and an activation function; establishing multiple hidden layers according to the coefficient calculation formula; inputting the graph embedding representation information into the multiple hidden layers for multi-layer calculation to obtain node final features; performing feature aggregation on the node final features, then performing weighted summation, and then inputting into a regression function for calculation to obtain the predicted aerodynamic coefficients of the heliostat; the aerodynamic coefficients of the heliostat include the drag coefficient and the pressure coefficient.

2. The method for calculating the aerodynamic coefficients of a heliostat by fusing geometric modal features according to claim 1, wherein: Step 1: the method for generating multiple modal features by processing the geometric data and air data of the heliostat is as follows: obtain the geometric data of the heliostat, including the three-dimensional point cloud data of the heliostat and the elevation angle data of the heliostat; obtain the air data, including the air density data, and construct the air density data by combining historical meteorological data; process the three-dimensional point cloud data to extract its geometric features and obtain the first modal feature; perform angle recognition and coding on the elevation angle data of the heliostat to obtain the second modal feature; perform normalization processing on the air density data to obtain the third modal feature; fuse the first modal feature, the second modal feature, and the third modal feature to form multiple modal features.

3. The method for calculating the aerodynamic coefficients of a heliostat by fusing geometric modal features according to claim 1, wherein: Step 2: the method for mining the interaction between modal features to obtain modal coupling data is as follows: obtain multiple modal features, including at least the first modal feature and the second modal feature, each modal feature including a plurality of positions, each position being a d-dimensional feature vector; considering the difference between modal features and the physical consistency constraint, set a learnable weight matrix corresponding to each modal feature; the learnable weight matrix is obtained by training a deep cross-attention network, and the training target is to minimize the physical constraint loss function; based on the learnable weight matrix, perform linear transformation on the modal features to obtain a query matrix, a key matrix, and a value matrix; calculate the attention score between the first modal feature and the second modal feature according to the query matrix and the key matrix; based on the attention score, perform weighted summation on the value matrix of the second modal feature to obtain the first fusion feature of the first modal feature; based on the attention score, perform weighted summation on the value matrix of the first modal feature to obtain the second fusion feature of the second modal feature; ​ According to the first fusion feature and the second fusion feature, a new query matrix, a key matrix and a value matrix are calculated; Based on the new query matrix and the key matrix, the attention score between the first fusion feature and the second fusion feature is calculated; The attention score and the fusion feature are continuously updated, so as to realize the deep interaction between the modal features at multiple semantic levels, and finally the first fusion feature and the second fusion feature are summarized to obtain the modal coupling data.

4. The heliostat aerodynamic coefficient calculation method of fusing geometric modal features according to claim 3, characterized in that: The method for training the deep cross-attention network is as follows: Obtain a predicted heliostat aerodynamic coefficient, and compare it with the benchmark aerodynamic coefficient value to calculate the mean square error of the two; Based on the mean square error, the weighted coefficient and the penalty coefficient, a physical constraint loss function is constructed with the loss value as the target; Global interaction features between different modal features are extracted through the deep cross-attention network; Then, a graph structure between modal features is constructed, and a local dependency relationship and topological association are modeled through graph convolution; Then, based on the deep cross-attention network and the graph structure, an aerodynamic coefficient prediction result is output, and the physical constraint loss function is used as an optimization target to guide the entire deep cross-attention network to maintain physical consistency while fitting the aerodynamic coefficient; and a variable-dependent and back-propagation mechanism constitutes an end-to-end training joint learning network.

5. The heliostat aerodynamic coefficient calculation method of fusing geometric modal features according to claim 1, characterized in that: Step three, introduce graph convolution operation, represent the modal coupling data as a graph structure, and perform message propagation on the graph structure to represent the dependency relationship between modal features, and the method for forming graph embedding representation information is as follows: Based on the observation points on the surface of the heliostat and the corresponding position information, a plurality of nodes are constructed; According to the modal coupling data and the corresponding position information, the nodes are valued, so that each node has geometric modal and physical modal; And fuse the geometric modal and the physical modal to form a node feature vector; Based on the modal coupling data, the relationship between the features is obtained; according to the relationship between the features, a geometric edge is constructed; According to the geometric edge, the node and the node feature vector, a graph structure is drawn; Based on the deep cross-attention network and the graph convolution network, and combined with the node feature vectors of the neighbor nodes, the node feature vectors are updated, so that the local modal features can be propagated globally and can represent the dependency relationship and structural association between modal features.

6. The heliostat aerodynamic coefficient calculation method of fusing geometric modal features according to claim 5, characterized in that: The relationship between the features includes geometric adjacency relationship, feature similarity relationship and physical prior relationship; The geometric adjacency relationship is established according to the positions of the point cloud or structural elements in space and using a nearest neighbor algorithm, and is used to reflect local physical interaction; The feature similarity relationship is established by calculating the cosine similarity of the modal features, and is used to capture the coupling characteristics of similar behavior at a distance; The physical prior relationship is established based on the relationship between the structural connection relationship and the airflow path and heat transfer path, and is used to enhance the physical interpretability of the graph structure.

7. The method according to claim 6, wherein the method comprises the following steps: Based on the deep cross-attention network and the graph convolution network, and combined with the node feature vectors of the neighbor nodes, the method for updating the node feature vectors is as follows: First, through the deep cross-attention network, the node feature vectors are cross-attended layer by layer, and the multi-modal fusion features are generated by combining the attention mechanism, which include a plurality of fusion feature units; The multi-modal fusion features are used as input nodes of the graph convolution network, and a graph structure is constructed, which includes a plurality of nodes and a plurality of edges, wherein the nodes represent the fusion feature units, and the edges represent the physical dependency relationship, spatial connection or structural similarity between the feature units; According to the node feature vectors of the neighbor nodes, an adjacency matrix is established; In the graph convolution network, the message propagation mechanism controlled by the adjacency matrix is used to update the node feature vectors layer by layer, so that the node feature vectors can interact with the local modal features and can be integrated into the topological information of the global physical field.

8. The method according to claim 6, wherein the method comprises the following steps: Using a pre-constructed multi-source data processing model, the geometric data and air data of the heliostat are processed to generate a plurality of modal features; Based on a pre-constructed feature capturing model, the interaction relationship between the modal features is mined to obtain modal coupling data; The method for obtaining the modal coupling data is as follows: Obtain a plurality of modal features, which at least include a first modal feature and a second modal feature, each modal feature includes a plurality of positions, and each position is a d-dimensional feature vector; Considering the difference between the modal features and the physical consistency constraint, a learnable weight matrix corresponding to each modal feature is set; the learnable weight matrix is obtained by training the deep cross-attention network, and the training target is to minimize the physical constraint loss function; Based on the learnable weight matrix, the modal features are linearly transformed to obtain a query matrix, a key matrix and a value matrix; According to the query matrix and the key matrix, the attention score between the first modal feature and the second modal feature is calculated; Based on the attention score, the value matrix of the second modal feature is weighted and summed to obtain the first fusion feature of the first modal feature; According to the attention score, the value matrix of the first modal feature is weighted and summed to obtain the second fusion feature of the second modal feature; According to the first fusion feature and the second fusion feature, new query matrix, key matrix and value matrix are calculated; Based on the new query matrix and the key matrix, the attention score between the first fusion feature and the second fusion feature is calculated; The attention score and the fusion feature are continuously updated, so as to realize the deep interaction between the modal features at multiple semantic levels, and finally the first fusion feature and the second fusion feature are summarized to obtain the modal coupling data; According to the pre-constructed feature propagation model, the graph convolution operation is introduced to represent the modal coupling data as a graph structure, and the message passing is performed on the graph structure to represent the dependency relationship between the modal features and form the graph embedding representation information. ​ The pre-constructed aerodynamic simulation model is used to aggregate and map the graph embedding representation information, so as to obtain the predicted heliostat aerodynamic coefficient, and realize the calculation of the heliostat aerodynamic coefficient fusing the geometric modal characteristics.

9. A system for calculating a heliostat aerodynamic coefficient fusing geometric modal characteristics, characterized in that: It comprises a multi-source data processing module, a feature capturing module, a feature propagation module and an aerodynamic simulation module. The multi-source data processing module is used to process the geometric data and air data of the heliostat, and generate a plurality of modal characteristics. The feature capturing module is used to mine the interaction relationship between the modal characteristics, and obtain modal coupling data. The feature propagation module is used to introduce a graph convolution operation, represent the modal coupling data as a graph structure, and perform message passing on the graph structure to represent the dependency relationship between the modal characteristics, and form graph embedding representation information. The method for forming the graph embedding representation information is as follows: Based on the observation points on the surface of the heliostat and the corresponding position information, a plurality of nodes are constructed. According to the modal coupling data and the corresponding position information, the nodes are valued, so that each node has a geometric modal and a physical modal. The geometric modal and the physical modal are fused to form a node feature vector. Based on the modal coupling data, the relationship between the features is obtained. According to the relationship between the features, a geometric edge is constructed. According to the geometric edge, the node and the node feature vector, a graph structure is drawn. Based on the deep cross-attention network and the graph convolution network, and combined with the node feature vectors of the neighbor nodes, the node feature vectors are updated, so that the local modal characteristics can be propagated to the global, and the dependency relationship and the structural association between the modal characteristics can be represented. The aerodynamic simulation module is used to aggregate and map the graph embedding representation information, so as to obtain the predicted heliostat aerodynamic coefficient, and realize the calculation of the heliostat aerodynamic coefficient fusing the geometric modal characteristics.

Images