An adaptive method and system for simulating typhoon wind fields by enhancing the influence of complex terrain and the linkage effect between the typhoon and the monitoring stations.

Abstract

This invention relates to an adaptive method and system for simulating typhoon wind fields by enhancing the influence of complex terrain and the linkage effect between stations. The method involves acquiring meteorological observation station data, low-resolution typhoon wind field data, and high-resolution terrain data through a data acquisition and standardization preprocessing module, followed by standardization processing. Based on the observation station data, a spatial location map and a feature correlation map are constructed. A multi-scale wind-terrain feature fusion module extracts neighborhood wind field and terrain information. A spatial linkage network prediction module combines the fused features with the observed values ​​for prediction. A typhoon wind field model training module optimizes parameters through backpropagation, minimizing the L2 loss between predicted and actual noise. In the typhoon wind field reconstruction module, target grid points are inserted into the station map, edges are constructed, and wind components are calculated to reconstruct the typhoon wind field. This method addresses the problem of insufficient representation of complex terrain and the linkage effect between stations in existing typhoon wind field simulation methods.

Description

Technical Field

[0001] This invention belongs to the field of typhoon wind field reconstruction and simulation technology, and relates to a method and system for typhoon wind field simulation that adaptively enhances the influence of complex terrain and the linkage effect of stations. In particular, it relates to a method and system for typhoon fine wind field simulation that adaptively enhances the influence of complex terrain and the linkage effect of stations under the influence of mountainous terrain. Background Technology

[0002] With the increasing frequency of extreme weather events, the demand for accuracy in typhoon wind field simulation and reconstruction is constantly rising. Related research has made positive progress in physical modeling, statistical inference, numerical simulation, and terrain correction. Existing methods mainly improve the physical consistency of wind fields and their terrain responsiveness by constructing empirical formulas, fusing reanalysis data, and introducing Bayesian inference or deep learning techniques. However, existing technologies generally suffer from the following shortcomings: First, the models have limited ability to express the nonlinear coupling relationship between complex terrain and wind fields, relying on fixed formulas or preset parameters and lacking adaptive adjustment mechanisms; second, their spatiotemporal resolution and regional generalization ability are insufficient, making it difficult to achieve high-precision local simulations while maintaining wide-area adaptability; and third, some methods have high computational costs, making it difficult to meet the practical needs of rapid prediction and real-time assessment.

[0003] Regarding terrain correction, although existing studies have improved the accuracy of microscale wind fields through fluid dynamics simulations or deep learning methods, these methods still suffer from weak generalization and limited application scenarios due to reliance on pre-simulated data training or highly complex numerical solutions. Furthermore, the collaborative modeling of multi-site wind speed observations and regional wind field structures is insufficient, limiting a comprehensive description of the spatial continuity and physical consistency of wind fields. Therefore, how to construct an efficient typhoon wind field simulation method with adaptive terrain modulation capabilities and site-linked feature modeling capabilities, while ensuring physical rationality and local detail representation, has become a critical issue that urgently needs to be addressed.

[0004] Chinese invention patent CN115952690B discloses a method and system for constructing long return period typhoon wind fields, which relates to the field of typhoon wind field construction technology. The invention considers the differentiated nonlinear relationship of the main intensity characteristic elements of typhoons under different typhoon path types to construct long return period typhoon wind fields, which is applicable to the calculation of long return period wave elements in different directions.

[0005] Chinese invention patent CN116595913B discloses a typhoon wind field simulation system and method for improving the accuracy of ocean models, belonging to the field of meteorological data fusion and processing technology. The method is characterized by filling in areas far from the typhoon center in the typhoon wind field constructed based on empirical formulas using ERA5 wind fields, and smoothly transitioning between cyclone wind fields and ERA5 wind fields, thereby expanding the typhoon wind field simulation area.

[0006] Chinese invention patent application CN118898196A discloses a method, apparatus, equipment, and medium for reconstructing typhoon wind fields based on a Bayesian model. Its key features include constructing an initial typhoon wind field model based on a dataset of optimal typhoon paths, converting the initial typhoon wind field model into an average wind speed model based on each quadrant, and determining weight parameters using Bayesian hierarchical models, Monte Carlo methods, MAP methods, and feedforward neural network models. This addresses the problem of inaccurate wind field hazard assessment caused by inconsistencies between physical characteristics of existing typhoon wind field models.

[0007] Chinese invention patent application CN119203816A discloses a typhoon wind field prediction method driven by the fusion of model and measured data. Its key feature is the use of the mean and standard deviation of wind speed from measured typhoon wind field data to calibrate the turbulence intensity of the typhoon wind field, constructing a turbulence intensity model, and recalculating the complete time-series wind speed data of the typhoon wind field during the typhoon period to complete the prediction of the typhoon wind field. Because this method uses marine observation data for correction, it improves the accuracy of the typhoon wind speed time-series data at single points, thereby enhancing the accuracy of numerical simulations of extreme operating conditions for offshore wind turbines.

[0008] Chinese invention patent application CN118734725A discloses a method for correcting the influence of complex micro-topography on parametric wind fields of typhoons. Its main features include: wind field data considering micro-topography influence obtained from numerical model simulation; wind field data not considering micro-topography influence obtained from parametric wind field model simulation; and geographical environmental data characterizing micro-topography. The method involves: S2, constructing a micro-topography correction model: processing the training data for the micro-topography correction model; constructing and training the micro-topography correction model based on a deep learning network; capturing the spatial characteristics of micro-topography and wind field, as well as the coupling relationship between micro-topography and wind field; and S3, inputting the wind field data not considering micro-topography influence and the geographical environmental data into the trained micro-topography correction model to obtain a parametric wind field considering micro-topography influence. This invention effectively considers the influence of complex micro-topography and can provide a more reliable parametric wind field for typhoons for engineering applications.

[0009] Chinese invention patent application CN116882310A discloses a method and apparatus for calculating typhoon wind fields based on fluid dynamics. This invention simulates the fluid dynamic distribution of a typical mountain in various directions by constructing a mountain terrain model and a fluid dynamics calculation model, thereby obtaining simulated wind speed results. Further quantitative analysis of these wind speed simulation results extracts the wind speed variation characteristics of typical mountains. Based on the extracted wind speed variation characteristics, a large-scale terrain correction factor calculation model suitable for typhoon disasters is constructed. Combining the mountain model, the fluid dynamics calculation model, and the terrain correction factor model enables refined simulation of typhoon wind fields.

[0010] Chinese invention patent CN116644676A proposes a method for simulating typhoon wind fields considering the aerodynamic effects of undulating terrain. This method simplifies the terrain to a two-dimensional ideal terrain, combines CFD numerical simulation and a gradient wind field model to quantitatively calculate the wind pressure and terrain drag effects at different slopes, introduces a terrain drag term into the original wind field model, and iteratively corrects it to obtain a more realistic typhoon gradient equilibrium wind field. Finally, a boundary layer model is used to transform it into a near-surface wind field. This research is of great significance in improving the accuracy of typhoon wind field simulation in complex terrain areas.

[0011] Chinese invention patent CN115017670A proposes a method for constructing a typhoon wind field model that considers complex underlying surfaces. Based on a pre-defined typhoon wind field model, this method constructs multiple components, including tangential wind speed, geostrophic wind speed, time-distance wind speed, and vertical wind speed, to form a typhoon wind field model suitable for complex underlying surface conditions. The reliability of the model is verified by comparison with time-history wind speed data measured at meteorological stations, demonstrating that it is applicable not only to complex terrain conditions but also to typhoon wind field reconstruction and analysis in high-latitude regions.

[0012] Current research on the construction and simulation of typhoon wind fields has made some progress. Related methods mainly include physical model construction, statistical modeling, numerical simulation, and terrain correction. Among them, CN115952690B establishes a nonlinear intensity element relationship model to estimate long-return-period wind fields by distinguishing typhoon path types, but its ability to model the spatiotemporal distribution of wind fields is limited. CN116595913B uses an empirical model and ERA5 reanalysis data to expand the wind field simulation range, but the model relies too heavily on data smoothing and struggles to capture high-resolution local features. CN118898196A reconstructs the wind field based on a Bayesian hierarchical model, enhancing the model's ability to handle parameter uncertainties, but suffers from computational complexity and low efficiency. CN119203816A introduces observational data to correct wind field turbulence intensity, improving the accuracy of single-point time-series wind speeds, but it lacks sufficient assurance of large-scale spatial consistency.

[0013] In addition, some studies have focused on topographic and underlying surface correction: for example, CN118734725A uses deep learning methods to correct the impact of micro-topography, which has achieved good results at the micro scale, but relies on pre-simulation results as training data, and the model's generalization ability is still limited; CN116882310A and CN116644676A are based on fluid dynamics and CFD to simulate the aerodynamic effects of topography, which improves the accuracy of typhoon wind field response to complex topography, but the simulation cost is high and the applicability is limited; CN115017670A models the impact of complex underlying surface through multiple wind speed components to enhance the model's adaptability, but the model structure is fixed and it is difficult to adapt to the complex changes in the structure of different types of typhoons.

[0014] In summary, existing methods have achieved certain results in improving physical rationality and local terrain response, but they still generally have the following shortcomings: (1) Model parameters rely on empirical rules or preset formulas and lack the ability to adaptively express complex nonlinear relationships; (2) The simulation results are insufficient in terms of spatiotemporal accuracy or generalization ability, and it is difficult to simultaneously take into account both large-scale adaptability and local refinement; (3) Most methods have low computational efficiency and are difficult to meet the needs of rapid prediction or real-time evaluation. Summary of the Invention

[0015] In view of this, in order to address the problems that typhoon wind field simulation methods generally rely on physical empirical formulas or preset model parameters, making it difficult to effectively express the nonlinear coupling relationship between complex terrain features and wind fields; and that the lack of collaborative modeling capability between multi-site wind speed observation data and regional wind field structure leads to deviations in spatiotemporal resolution and physical consistency of the wind field, thus limiting its applicability in actual wind disaster pre-assessment and engineering applications, this invention provides an adaptive method and system for typhoon wind field simulation that enhances the influence of complex terrain and the linkage effect of stations.

[0016] To achieve the above objectives, the present invention provides the following technical solution:

[0017] An adaptive method for simulating typhoon wind fields that enhances the influence of complex terrain and the interaction between the wind farm and the site, comprising the following steps:

[0018] S1. Data Acquisition and Standardization Preprocessing; Acquiring meteorological observation station data and low-resolution typhoon wind field data. and high-resolution terrain data The acquired wind field data and terrain data from the engineering typhoon model are preprocessed to generate standardized data.

[0019] S2. Geographic location modeling: Based on the meteorological observation station data from step S1, establish a spatial location map of the stations and a correlation map of station characteristics;

[0020] S3, Multi-scale wind-topography feature fusion; such as Figure 2 As shown, taking each node in step S1 as the center, the neighborhood wind field and terrain spatial information are extracted, and the standardized wind field and terrain data are input into the multi-scale wind-terrain spatial correlation extraction module.

[0021] S4. Spatial Linkage Network Prediction: Flatten the multi-scale wind-topography fusion features of each graph node in step S3 into one dimension, and use them together with the actual observation values ​​of the stations as node features, and use the spatial linkage network of meteorological observation stations for prediction.

[0022] S5. Typhoon wind field model training; During the training process, the typhoon wind field model constructed in steps S3 and S4 is optimized using the backpropagation algorithm. By minimizing the L2 loss between the model's predicted noise and the actual noise, the model can learn effectively.

[0023] S6. Typhoon wind field reconstruction: Insert the target grid point as a new node into the site spatial location map and the site feature correlation map, and after constructing the edges according to the process in step S2, complete the calculation of the wind component according to steps S3 and S4.

[0024] Furthermore, the data standardization process in step S1 is as follows:

[0025]

[0026]

[0027] in, Represents standardized operation. This represents the input low-resolution typhoon wind field data. Represents the input high-resolution terrain data, This represents standardized typhoon wind field data. This represents standardized terrain data.

[0028] Further, step S2 specifically involves: based on meteorological observation station data, each station is considered as a node in the graph structure, and the actual observed value of the station, the typhoon wind field value corresponding to the station's location, and the terrain feature value corresponding to the station's location are used as station features. Specifically, a station spatial location map is established: the graph modeling process uses the spatial location of the stations as the basis for constructing edges between nodes. If the Euclidean distance between any two nodes is less than a set threshold R, then an edge is considered to exist between the two nodes, and the edge plays an important role in the subsequent message passing mechanism. A station feature correlation map is also established: the graph modeling process uses the correlation of station features as the basis for constructing edges between nodes. If the cosine similarity of features between any two nodes is greater than a set threshold T, then an edge is considered to exist between the two nodes.

[0029] Furthermore, in step S3, the multi-scale wind-topography spatial correlation extraction module consists of three multi-scale fusion layers (MSF) and three spatial attention layers; standardized wind field and topographic data are input into the wind-topography spatial correlation extraction module, and the multi-scale wind-topography fusion features are output. The calculation formula for the multi-scale wind-topography spatial correlation extraction module is as follows:

[0030]

[0031]

[0032]

[0033]

[0034]

[0035]

[0036] in for Multi-scale feature fusion operation of the module. Features of the first MSF layer fusion The features are after weight adjustment for the first attention module. Features of the second MSF layer fusion The features are after weight adjustment for the second attention module. Features of the third MSF layer fusion The features, after weight adjustment by the third attention module, undergo feature fusion through three MSF layers and three attention modules to obtain the fused features. .

[0037] Furthermore, in step S3, each feature fusion module (MSF) consists of three dilated convolutional layers with different dilation rates, three SiLU activation layers, one convolutional fusion module, and one attention module. The calculation formula for the multi-scale fusion layer of each feature fusion module is as follows:

[0038]

[0039]

[0040]

[0041]

[0042] in A dilated convolution with an inflation rate of 1 A dilated convolution with an inflation rate of 2, This represents a dilated convolution with an inflation rate of 3. This represents the local scale fusion features after processing by a dilated convolutional module with an inflation rate of 1. This is a medium-scale fused feature after being processed by a dilated convolutional module with an inflation rate of 2. This represents the large-scale fused features after processing by a dilated convolutional module with an inflation rate of 3. This represents the multi-scale wind-terrain fusion features output by the MSF layer.

[0043] Furthermore, in step S4, the spatial linkage network of meteorological observation stations includes two different pipeline flows, distributed to process the spatial location map and the correlation map of station features. The network takes the modeling map and node features as input and outputs predictions based on spatial linkage considerations. The formula for the spatial location pipeline flow used to process the spatial location modeling map is as follows:

[0044]

[0045]

[0046] in, Represents network input. represent Message aggregation layer This represents a dense, fully connected layer. Each The message aggregation layer consists of an average aggregator, a max-pooling aggregator, and an adaptive weighted fusion layer, calculated as follows:

[0047]

[0048]

[0049]

[0050]

[0051] in, The average aggregation feature of the i-th node, Let be the number of neighboring nodes of the i-th node. The features of the j-th neighbor node are... Let W be the max-pooling aggregated feature of the i-th node, and let W and b represent the trainable weights and biases, respectively. The weight of the average aggregator. The weight assigned to the maximum pooling aggregator;

[0052] The correlation pipeline used to process feature correlation modeling graphs contains a The message aggregation layer and two graph attention message aggregation layers (Graph Attention, GAT) are defined as follows:

[0053]

[0054]

[0055]

[0056] in, Represents network input. represent Message aggregation layer This represents the graph attention message aggregation layer. Each GAT layer contains a multi-head attention mechanism that dynamically weights the messages passed by neighboring nodes based on their importance. Each attention head has independent parameters, and the outputs of all attention heads are concatenated. The calculation method is as follows:

[0057]

[0058] Where K is the number of heads receiving multi-head attention. Represents the i-th node's... Layer embedding of latent variables, For activation, Let be the number of neighboring nodes of the i-th node. The standardized attention matrix, Let be the linear transformation weight matrix for the k-th attention head;

[0059] The fully connected prediction module outputs two different pipeline flows. and As input, the features are projected into a high-dimensional space through four fully connected-activation layers and one fully connected layer, generating an output based on site spatial linkage considerations. The network formula is as follows:

[0060]

[0061]

[0062]

[0063]

[0064]

[0065] in Features extracted from the first fully connected-activation layer of the fully connected prediction module. The features extracted from the second fully connected-activation layer of the second fully connected prediction module. Features extracted from the third fully connected-activation layer of the fully connected prediction module. Features extracted from the fourth fully connected-activation layer of the fully connected prediction module. The predicted wind component.

[0066] Furthermore, the calculation process of the loss function in step S5 is as follows:

[0067]

[0068] in, Represents the loss function. and The wind component observed at the representative station.

[0069] This adaptive typhoon wind field simulation system, which enhances the interaction between complex terrain effects and site monitoring, includes:

[0070] The data acquisition and standardization preprocessing module acquires data from meteorological observation stations and low-resolution typhoon wind field data. and high-resolution terrain data Furthermore, standardized preprocessing was performed on wind field and topographic data to generate standardized data;

[0071] Geographic location modeling module: Based on meteorological observation station data, establish station spatial location maps and station feature correlation maps;

[0072] Multi-scale wind-terrain feature fusion module: Taking each node as the center, extract the wind field and terrain spatial information of the neighborhood, and input the standardized wind field and terrain data into the multi-scale wind-terrain spatial correlation extraction module;

[0073] Spatial Linkage Network Prediction Module: The multi-scale wind-topography fusion features of each graph node are flattened into one dimension and used together with the actual observation values ​​of the stations as node features. The spatial linkage network of meteorological observation stations is used for prediction.

[0074] Typhoon wind field model training module: In the training of the constructed typhoon wind field model, the backpropagation algorithm is used to optimize the model parameters. By minimizing the L2 loss between the model's predicted noise and the actual noise, the model can learn effectively.

[0075] Typhoon wind field reconstruction module: Insert the target grid points as new nodes into the site spatial location map and the site feature correlation map, construct edges according to the geographical location modeling module, and complete the calculation of wind components based on the spatial linkage network prediction module and the typhoon wind field model training module.

[0076] An electronic device includes: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of the above-described adaptive method for simulating the influence of complex terrain and the linkage effect of sites in typhoon wind fields.

[0077] A computer-readable storage medium storing a computer program, which, when executed by a processor, is used to implement the steps of the above-described adaptive method for simulating the influence of complex terrain and the linkage effect of sites in typhoon wind fields.

[0078] The beneficial effects of this invention are as follows:

[0079] 1. The typhoon wind field simulation method disclosed in this invention, which adaptively enhances the influence of complex terrain and the linkage effect between observation stations, is a typhoon wind field simulation model that can be constructed without relying on preset parameter structures. This model simultaneously enhances the local modulation effect of complex terrain on the wind field and the spatial synergy between wind speed observations from multiple stations. To this end, this method introduces a deep learning framework, fusing high-resolution terrain features, typhoon path and intensity elements, and time-series wind speed data from multiple stations to establish an end-to-end wind field prediction model. This model can automatically capture the spatial heterogeneity between terrain and wind field and mine the dynamic linkage characteristics between observation stations, thereby improving prediction accuracy and regional adaptability while maintaining physical consistency. Compared with existing techniques that rely on static physical modeling or empirical correction, the method of this invention has good scalability and generalization ability, and is applicable to wind field simulation and rapid reconstruction tasks under different typhoon paths, terrain complexity, and observation data integrity conditions.

[0080] 2. The adaptive typhoon wind field simulation method for enhancing the influence of complex terrain and site linkage effects disclosed in this invention effectively solves the problem of insufficient representation of complex terrain and site linkage effects in existing typhoon wind field simulation methods by introducing deep learning methods. Compared with traditional methods, this invention has the following significant advantages:

[0081] 1) Strong adaptability: Through adaptive training of deep learning models, it can flexibly capture wind field characteristics under different typhoon paths, intensities and terrain changes, avoiding the excessive dependence on physical models or parameters in traditional methods, and has stronger generalization ability.

[0082] 2) Improve simulation accuracy: By integrating wind speed data from multiple stations with complex terrain information, the simulation of wind fields is refined. In particular, under complex terrain conditions, the simulation accuracy of local wind field changes is effectively enhanced, and the spatial resolution and physical consistency of the wind field are significantly improved.

[0083] 3) Enhance the linkage effect of stations: Through multi-station data collaborative modeling, the linkage relationship between observation points can be accurately reflected, which improves the accuracy and reliability of typhoon wind speed time series in the region and helps to more accurately assess the wind disaster risk brought by typhoons.

[0084] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0085] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein:

[0086] Figure 1 This is a flowchart of the typhoon wind field simulation method of the present invention, which adaptively enhances the influence of complex terrain and the site linkage effect;

[0087] Figure 2 This is a flowchart of step S3, multi-scale wind-terrain feature fusion, of the present invention.

[0088] Figure 3 This is a flowchart of the multi-scale fusion layer calculation for each feature fusion module in step S3 of the present invention, which is a multi-scale wind-terrain feature fusion. Detailed Implementation

[0089] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention.

[0090] like Figure 1 The adaptive method for simulating typhoon wind fields by enhancing the influence of complex terrain and site linkage effects, as shown, includes the following steps:

[0091] S1. Data Acquisition and Standardization Preprocessing; Acquiring meteorological observation station data and low-resolution typhoon wind field data. and high-resolution terrain data The acquired wind field data and topographic data from the engineering typhoon model are preprocessed, and the different data are standardized to generate standardized data. The standardized wind field data and standardized topographic data are represented as follows: and ;

[0092] The data standardization process is as follows:

[0093]

[0094]

[0095] in, Represents standardized operation. This represents the input low-resolution typhoon wind field data. Represents the input high-resolution terrain data, This represents standardized typhoon wind field data. This represents standardized terrain data.

[0096] S2. Geographic location modeling: Based on the meteorological observation station data from step S1, establish a spatial location map of the stations and a correlation map of station characteristics.

[0097] Specifically, a spatial location map of the stations is established: based on meteorological observation station data, each station is regarded as a node in the graph structure, and the actual observed value of the station, the typhoon wind field value corresponding to the station's location, and the terrain feature value corresponding to the station's location are used as station features. In the graph modeling process, the spatial location of the stations is used as the basis for constructing edges between nodes. If the Euclidean distance of latitude and longitude between any two nodes is less than a set threshold R, it is considered that there is an edge between the two nodes. The edge plays an important role in the subsequent message passing mechanism.

[0098] Establish a site feature correlation graph; based on meteorological observation station data, each station is regarded as a node in the graph structure, and the actual observed value of the station, the typhoon wind field value corresponding to the station location, and the terrain feature value corresponding to the station location are used as station features. In the graph modeling process, the correlation between station features is used as the basis for constructing edges between nodes. If the feature cosine similarity between any two nodes is greater than a set threshold T, then an edge is considered to exist between the two nodes.

[0099] S3, Multi-scale wind-topography feature fusion; such as Figure 2 As shown, taking each node in step S1 as the center, the neighborhood wind field and terrain spatial information are extracted, and the standardized wind field and terrain data are input into the multi-scale wind-terrain spatial correlation extraction module.

[0100] The multi-scale wind-topography spatial correlation extraction module consists of three multi-scale fusion (MSF) layers and three spatial attention layers. Standardized wind field and topographic data are input into the module, which outputs multi-scale wind-topography fusion features. The calculation formula for the multi-scale wind-topography spatial correlation extraction module is as follows:

[0101]

[0102]

[0103]

[0104]

[0105]

[0106]

[0107] in for Multi-scale feature fusion operation of the module. Features of the first MSF layer fusion The features are after weight adjustment for the first attention module. Features of the second MSF layer fusion The features are after weight adjustment for the second attention module. Features of the third MSF layer fusion The features, after weight adjustment by the third attention module, undergo feature fusion through three MSF layers and three attention modules to obtain the fused features. .

[0108] like Figure 3 Each feature fusion module (MSF) shown consists of three dilated convolutional layers with different dilation rates, three SiLU activation layers, one convolutional fusion module, and one attention module. The calculation formula for the multi-scale fusion layer of each feature fusion module is as follows:

[0109]

[0110]

[0111]

[0112]

[0113] in A dilated convolution with an inflation rate of 1 A dilated convolution with an inflation rate of 2, This represents a dilated convolution with an inflation rate of 3. This represents the local scale fusion features after processing by a dilated convolutional module with an inflation rate of 1. This is a medium-scale fused feature after being processed by a dilated convolutional module with an inflation rate of 2. This represents the large-scale fused features after processing by a dilated convolutional module with an inflation rate of 3. This represents the multi-scale wind-terrain fusion features output by the MSF layer.

[0114] S4. Spatial Linkage Network Prediction: Flatten the multi-scale wind-topography fusion features of each graph node in step S3 into one dimension, and use them together with the actual observation values ​​of the stations as node features, and use the spatial linkage network of meteorological observation stations for prediction.

[0115] The meteorological observation station spatial linkage network includes two different pipeline flows, distributed to process station spatial location maps and station feature correlation maps. The network takes the modeling map and node features as input and outputs predictions based on station spatial linkage considerations. The formula for the spatial location pipeline flow used to process the spatial location modeling map is as follows:

[0116]

[0117]

[0118] in, Represents network input. represent Message aggregation layer This represents a dense, fully connected layer. Each The message aggregation layer consists of an average aggregator, a max-pooling aggregator, and an adaptive weighted fusion layer, calculated as follows:

[0119]

[0120]

[0121]

[0122]

[0123] in, The average aggregation feature of the i-th node, Let be the number of neighboring nodes of the i-th node. The features of the j-th neighbor node are... Let W be the max-pooling aggregated feature of the i-th node, and let W and b represent the trainable weights and biases, respectively. The weight of the average aggregator. The weight assigned to the maximum pooling aggregator.

[0124] This message aggregation layer aggregates the average and maximum features from neighboring nodes and can dynamically allocate the weights of the two features based on different nodes and different graph characteristics, effectively leveraging the spatial linkage between sites and improving the network's predictive performance.

[0125] The correlation pipeline used to process feature correlation modeling graphs contains a The message aggregation layer and two graph attention message aggregation layers (Graph Attention, GAT) are defined as follows:

[0126]

[0127]

[0128]

[0129] in, Represents network input. represent Message aggregation layer This represents the graph attention message aggregation layer. Each GAT layer contains a multi-head attention mechanism that dynamically weights the messages passed by neighboring nodes based on their importance. Each attention head has independent parameters, and the outputs of all attention heads are concatenated. The calculation method is as follows:

[0130]

[0131] Where K is the number of heads receiving multi-head attention. Represents the i-th node's... Layer embedding of latent variables, For activation, Let be the number of neighboring nodes of the i-th node. The standardized attention matrix, Let be the linear transformation weight matrix of the k-th attention head.

[0132] Through the multi-head attention mechanism, the network can discover the potential connections between nodes and their neighboring nodes, thereby fully capturing the spatial linkages between target grid points and sites.

[0133] The fully connected prediction module outputs two different pipeline flows. and As input, the features are projected into a high-dimensional space through four fully connected-activation layers and one fully connected layer, generating an output based on site spatial linkage considerations. The network formula is as follows:

[0134]

[0135]

[0136]

[0137]

[0138]

[0139] in Features extracted from the first fully connected-activation layer of the fully connected prediction module. The features extracted from the second fully connected-activation layer of the second fully connected prediction module. Features extracted from the third fully connected-activation layer of the fully connected prediction module. Features extracted from the fourth fully connected-activation layer of the fully connected prediction module. The predicted wind component.

[0140] S5. Typhoon Wind Field Model Training; During the training process, the typhoon wind field model constructed in steps S3 and S4 utilizes the backpropagation algorithm to optimize model parameters. Effective model learning is achieved by minimizing the L2 loss between predicted and actual noise. Since the grid points lack site observation data, the observation values ​​for each batch are set to zero during the training phase to simulate the generalization inference process. The model learns in this way how to infer wind components based on the multi-scale features of the wind-topography domain and observation data from neighboring sites. The calculation process of its loss function is shown below:

[0141]

[0142] in, Represents the loss function. and The wind component observed at the representative station.

[0143] S6. Typhoon Wind Field Reconstruction: This step first inserts the target grid points as new nodes into the site spatial location map and the site feature correlation map, and constructs edges according to the process in step S2. Subsequently, the wind components are calculated according to steps S3 and S4. Because the model has learned how to infer wind components based on the multi-scale features of the domain wind-topography and the observation data of neighboring sites, it can generalize the inference process to all grid points, thereby realizing the reconstruction of the typhoon wind field.

[0144] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for adaptively enhancing the simulation of typhoon wind fields by incorporating the effects of complex terrain and site linkages, characterized in that, Includes the following steps: S1. Data Acquisition and Standardization Preprocessing; Acquiring meteorological observation station data and low-resolution typhoon wind field data. and high-resolution terrain data The acquired wind field data and terrain data from the engineering typhoon model are preprocessed to generate standardized data. S2. Geographic Location Modeling: Based on the meteorological observation station data from step S1, establish a spatial location map and a station feature correlation map. Specifically, using the meteorological observation station data as a foundation, each station is considered a node in the graph structure. The actual observed value of the station, the typhoon wind field value corresponding to the station's location, and the terrain feature value corresponding to the station's location are used as station features. Establishing the spatial location map: The graph modeling process uses the spatial location of the stations as the basis for constructing edges between nodes. If the Euclidean distance between any two nodes is less than a set threshold R, then an edge is considered to exist between the two nodes. Edges play an important role in the subsequent message passing mechanism. Establishing the station feature correlation map: The graph modeling process uses the correlation of station features as the basis for constructing edges between nodes. If the cosine similarity of features between any two nodes is greater than a set threshold T, then an edge is considered to exist between the two nodes. S3, Multi-scale wind-topography feature fusion; Centered on each node in step S1, extract neighborhood wind field and terrain spatial information. Input the standardized wind field and terrain data into the multi-scale wind-terrain spatial correlation extraction module, which consists of three multi-scale fusion layers (MSF) and three spatial attention layers. The standardized wind field and terrain data are input into the wind-terrain spatial correlation extraction module, which outputs multi-scale wind-terrain fusion features. The calculation formula for the multi-scale wind-topography spatial correlation extraction module is as follows: in for Multi-scale feature fusion operation of the module. Features of the first MSF layer fusion The features are after weight adjustment for the first attention module. Features of the second MSF layer fusion The features are after weight adjustment for the second attention module. Features of the third MSF layer fusion The features, after weight adjustment by the third attention module, undergo feature fusion through three MSF layers and three attention modules to obtain the fused features. ; S4, Spatial Linkage Network Prediction; In step S3, the multi-scale wind-topography fusion features of each graph node are flattened into one dimension and used together with the actual observations of the stations as node features. The meteorological observation station spatial linkage network is then used for prediction. S5. Typhoon wind field model training; During the training process, the typhoon wind field model constructed in steps S3 and S4 is optimized using the backpropagation algorithm. By minimizing the L2 loss between the model's predicted noise and the actual noise, the model can learn effectively. S6. Typhoon wind field reconstruction: Insert the target grid point as a new node into the site spatial location map and the site feature correlation map, and after constructing the edges according to the process in step S2, complete the calculation of the wind component according to steps S3 and S4.

2. The typhoon wind field simulation method as described in claim 1, characterized in that, The data standardization process in step S1 is as follows: in, Represents standardized operation. This represents the input low-resolution typhoon wind field data. Represents the input high-resolution terrain data, This represents standardized typhoon wind field data. This represents standardized terrain data.

3. The typhoon wind field simulation method as described in claim 2, characterized in that, In step S3, each feature fusion module (MSF) consists of three dilated convolutional layers with different dilation rates, three SiLU activation layers, one convolutional fusion module, and one attention module. The calculation formula for the multi-scale fusion layer of each feature fusion module is as follows: in A dilated convolution with an inflation rate of 1 A dilated convolution with an inflation rate of 2, This represents a dilated convolution with an inflation rate of 3. This represents the local scale fusion features after processing by a dilated convolutional module with an inflation rate of 1. This is a medium-scale fused feature after being processed by a dilated convolutional module with an inflation rate of 2. This represents the large-scale fused features after processing by a dilated convolutional module with an inflation rate of 3. This represents the multi-scale wind-terrain fusion features output by the MSF layer.

4. The typhoon wind field simulation method as described in claim 3, characterized in that, In step S4, the meteorological observation station spatial linkage network includes two different pipeline flows, distributed to process the station spatial location map and the station feature correlation map. The network takes the modeling map and node features as input and outputs a prediction based on station spatial linkage considerations. The formula for the spatial location pipeline flow used to process the spatial location modeling map is as follows: in, Represents network input. represent Message aggregation layer Represents a dense fully connected layer; each The message aggregation layer consists of an average aggregator, a max-pooling aggregator, and an adaptive weighted fusion layer, calculated as follows: in, The average aggregation feature of the i-th node, Let be the number of neighboring nodes of the i-th node. The features of the j-th neighbor node are... Let W be the max-pooling aggregated feature of the i-th node, and let W and b represent the trainable weights and biases, respectively. The weight of the average aggregator. The weight assigned to the maximum pooling aggregator; The correlation pipeline used to process feature correlation modeling graphs contains a The message aggregation layer and two graph attention message aggregation layers are defined by the following formula: in, Represents network input. represent Message aggregation layer The GAT (Graph Attention Message Aggregation) layer represents the graph attention message aggregation layer. Each GAT layer contains a multi-head attention mechanism that dynamically weights the messages passed by neighboring nodes based on their importance. Each attention head has independent parameters, and the outputs of all attention heads are concatenated. The calculation method is as follows: Where K is the number of heads receiving multi-head attention. Represents the i-th node's... Layer embedding of latent variables, For activation, Let be the number of neighboring nodes of the i-th node. The standardized attention matrix, Let be the linear transformation weight matrix for the k-th attention head; The fully connected prediction module outputs two different pipeline flows. and The input is processed through four fully connected-activation layers and one fully connected layer to project the features into a high-dimensional space, generating an output based on site spatial linkage considerations; the network formula is as follows: in Features extracted from the first fully connected-activation layer of the fully connected prediction module. The features extracted from the second fully connected-activation layer of the second fully connected prediction module. Features extracted from the third fully connected-activation layer of the fully connected prediction module. Features extracted from the fourth fully connected-activation layer of the fully connected prediction module. The predicted wind component.

5. The typhoon wind field simulation method as described in claim 4, characterized in that, The calculation process of the loss function in step S5 is as follows: in, Represents the loss function. and The wind component observed at the representative station.

6. A system based on the adaptive enhancement method for typhoon wind field simulation of complex terrain influence and site linkage effect as described in claim 1, characterized in that, include: The data acquisition and standardization preprocessing module acquires data from meteorological observation stations and low-resolution typhoon wind field data. and high-resolution terrain data Furthermore, standardized preprocessing was performed on wind field and topographic data to generate standardized data; Geographic location modeling module: Based on meteorological observation station data, establish station spatial location maps and station feature correlation maps; Multi-scale wind-terrain feature fusion module: Taking each node as the center, extract the wind field and terrain spatial information of the neighborhood, and input the standardized wind field and terrain data into the multi-scale wind-terrain spatial correlation extraction module; Spatial Linkage Network Prediction Module: The multi-scale wind-topography fusion features of each graph node are flattened into one dimension and used together with the actual observation values ​​of the stations as node features. The spatial linkage network of meteorological observation stations is used for prediction. Typhoon wind field model training module: In the training of the constructed typhoon wind field model, the backpropagation algorithm is used to optimize the model parameters. By minimizing the L2 loss between the model's predicted noise and the actual noise, the model can learn effectively. Typhoon wind field reconstruction module: Insert the target grid points as new nodes into the site spatial location map and the site feature correlation map, construct edges according to the geographical location modeling module, and complete the calculation of wind components based on the spatial linkage network prediction module and the typhoon wind field model training module.

7. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor is used to execute a computer program to implement the steps of the typhoon wind field simulation method according to any one of claims 1 to 5, which adaptively enhances the influence of complex terrain and the linkage effect of sites.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, is used to implement the steps of the typhoon wind field simulation method for adaptively enhancing the influence of complex terrain and the linkage effect of sites as described in any one of claims 1 to 5.

Images