Fine prediction method for urban block wind field

Abstract

The invention provides an urban block wind field refined prediction method, and the method comprises the steps: obtaining wind field actual measurement data of a target urban block, and the wind field actual measurement data comprise wind speed data and position data; preprocessing the wind speed data to obtain target wind speed data; inputting the target wind speed data and the position data into a first wind field prediction model to obtain a wind field prediction result of the city block; and inputting the wind field prediction result into the second wind field prediction model to obtain a target wind field prediction result of the city block. Therefore, high-precision and high-efficiency dynamic prediction of the complex wind field of the urban block can be realized through a double-model series prediction framework in combination with the actually measured wind speed and related position data of the wind field.

Description

Technical Field

[0001] This application relates to the field of urban meteorology and environmental engineering technology, and in particular to a method, device, electronic equipment and computer-readable storage medium for refined prediction of wind fields in urban blocks. Background Technology

[0002] Refined wind field prediction at the urban block scale is crucial for assessing the wind environment and preventing wind disasters. Based on this need, related research has developed two core technological approaches: one is wind field reconstruction methods based on CFD (Computational Fluid Dynamics) and AI (Artificial Intelligence), and the other is wind speed prediction methods based on time-series models or numerical weather prediction. However, wind field reconstruction is a static interpolation method lacking temporal prediction capabilities, while wind speed prediction methods either only allow for single-point predictions or have too low spatial resolution to characterize the complex flow fields around buildings. Overall, these methods suffer from functional fragmentation and insufficient physical consistency, making dynamic wind field prediction difficult. Summary of the Invention

[0003] This application aims to at least partially address one of the technical problems in the related art.

[0004] Therefore, the first objective of this application is to propose a refined prediction method for wind fields in urban blocks, which can achieve high-precision and high-efficiency dynamic prediction of complex wind fields in urban blocks by combining measured wind speed and relevant location data through a dual-model cascade prediction framework.

[0005] The second objective of this application is to propose a device for fine-grained prediction of wind fields in urban blocks.

[0006] The third objective of this application is to propose an electronic device.

[0007] The fourth objective of this application is to provide a computer-readable storage medium.

[0008] To achieve the above objectives, the first aspect of this application proposes a method for refined prediction of wind fields in urban blocks, comprising the following steps: acquiring measured wind field data of a target urban block, wherein the measured wind field data includes wind speed data and location data; preprocessing the wind speed data to obtain target wind speed data; inputting the target wind speed data and location data into a first wind field prediction model to obtain a wind field prediction result for the urban block; and inputting the wind field prediction result into a second wind field prediction model to obtain a target wind field prediction result for the urban block.

[0009] According to the urban street wind field refinement prediction method of this application embodiment, the measured wind field data of the target urban street is first obtained, including wind speed data and location data. Then, the wind speed data is preprocessed to obtain target wind speed data. Next, the target wind speed data and location data are input into a first wind field prediction model to obtain the wind field prediction result for the urban street. Finally, the wind field prediction result is input into a second wind field prediction model to obtain the target wind field prediction result for the urban street. Thus, by using a dual-model cascade prediction framework, combining measured wind speed and relevant location data, high-precision and high-efficiency dynamic prediction of complex wind fields in urban streets can be achieved.

[0010] In addition, the urban street wind field refinement prediction method according to the above embodiments of this application may also have the following additional technical features: In one embodiment of this application, obtaining measured wind field data of a target urban block includes: determining the placement location of the target urban block based on a placement strategy, setting the wind speed sensor at the placement location, and pre-setting the location data of the placement location into the corresponding wind speed sensor; and obtaining wind speed data and corresponding location information through the wind speed sensor.

[0011] In one embodiment of this application, the first wind field prediction model includes a regression layer, an embedding layer, and a perception layer, wherein the embedding layer includes a temporal embedding layer and a feature embedding layer, and the perception layer comprises multiple layers.

[0012] In one embodiment of this application, target wind speed data and location data are input into a first wind field prediction model to obtain wind field prediction results for urban blocks. This includes: extracting wind speed time features from the target wind speed data through a time embedding layer; extracting features from the target wind speed data and location data through a feature embedding layer to obtain wind distance features and time-based wind field features; fusing the wind speed time features, wind distance features, and time-based wind field features through a perception layer to obtain target wind field features; and performing wind field prediction based on the target wind field features through a regression layer to obtain wind field prediction results for urban blocks.

[0013] In one embodiment of this application, the first wind field prediction model is generated by: obtaining measured wind field training data of the target city block and the first training label corresponding to the measured wind field training data; inputting the measured wind field training data into the first wind field prediction model to generate a predicted first wind field result; generating a first loss value based on the predicted first wind field result and the first training label, and training the first wind field prediction model based on the first loss value.

[0014] In one embodiment of this application, the second wind field prediction model is generated by: obtaining wind field simulation data of the target city block and the second training label corresponding to the wind field simulation data; inputting the wind field simulation data into the second wind field prediction model to generate the predicted second wind field result; generating a second loss value based on the predicted second wind field result and the second training label, and training the second wind field prediction model based on the second loss value.

[0015] In one embodiment of this application, the layout strategy is generated by: acquiring a three-dimensional map of the target city block; and generating the layout strategy based on the three-dimensional map and / or wind field simulation data.

[0016] To achieve the above objectives, a second aspect of this application proposes a device for refined prediction of wind fields in urban blocks, comprising: an acquisition module for acquiring measured wind field data of a target urban block, wherein the measured wind field data includes wind speed data and location data; a preprocessing module for preprocessing the wind speed data to obtain target wind speed data; a first prediction module for inputting the target wind speed data and location data into a first wind field prediction model to obtain a wind field prediction result for the urban block; and a second prediction module for inputting the wind field prediction result into a second wind field prediction model to obtain a target wind field prediction result for the urban block.

[0017] According to the urban street wind field refinement prediction device of this application embodiment, the device first acquires measured wind field data of the target urban street through an acquisition module, wherein the measured wind field data includes wind speed data and location data. Then, the wind speed data is preprocessed by a preprocessing module to obtain target wind speed data. Next, the target wind speed data and location data are input into a first wind field prediction model through a first prediction module to obtain the wind field prediction result of the urban street. Finally, the wind field prediction result is input into a second wind field prediction model through a second prediction module to obtain the target wind field prediction result of the urban street. Thus, by using a dual-model cascade prediction framework, combining measured wind speed and relevant location data, high-precision and high-efficiency dynamic prediction of complex wind fields in urban streets can be achieved.

[0018] To achieve the above objectives, a third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements any of the above-mentioned methods for refined prediction of urban street wind fields.

[0019] The electronic device according to the embodiments of this application implements any of the above-mentioned methods for refined prediction of urban street wind fields when the processor executes a computer program. It realizes high-precision and high-efficiency dynamic prediction of complex wind fields in urban streets by combining measured wind speed and related location data through a dual-model cascade prediction framework.

[0020] To achieve the above objectives, a fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement any of the above-described methods for fine-grained prediction of urban street wind fields.

[0021] According to the embodiments of this application, a computer-readable storage medium storing a computer program thereon implements any of the above-described methods for refined prediction of urban street wind fields when executed by a processor. By using a dual-model cascade prediction framework and combining measured wind speeds and related location data, it achieves high-precision and high-efficiency dynamic prediction of complex wind fields in urban streets.

[0022] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0023] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart illustrating a method for refined prediction of urban street wind fields according to some embodiments of this application; Figure 2 This is a framework diagram of a first wind field prediction model according to some embodiments of this application; Figure 3 This is a flowchart illustrating a method for refined prediction of urban street wind fields according to other embodiments of this application; Figure 4 This is a flowchart illustrating a method for refined prediction of urban street wind fields according to a specific embodiment of this application; Figure 5 A block diagram of a refined urban street wind field prediction device according to some embodiments of this application; and Figure 6 This is a schematic diagram of the structure of an electronic device according to some embodiments of this application. Detailed Implementation

[0024] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0025] The following describes, with reference to the accompanying drawings, a method, apparatus, electronic device, and computer-readable storage medium for refined prediction of urban street wind fields according to embodiments of this application.

[0026] The urban street wind field refinement prediction method provided in this application embodiment can be executed by an electronic device, such as a mobile phone, tablet computer, handheld computer or server, etc., without any limitation.

[0027] In this embodiment, the electronic device may include a processing component, a storage component, and a driving component. Optionally, the driving component and the processing component may be integrated, and the storage component may store an operating system, application programs, or other program modules. The processing component implements the urban street wind field refinement prediction method provided in this embodiment by executing the application programs stored in the storage component.

[0028] like Figure 1 As shown, the urban street wind field refinement prediction method of this application embodiment may include the following steps: Step S1: Obtain measured wind field data of the target city block, including wind speed data and location data.

[0029] Specifically, this step aims to provide accurate and reliable initial input data for subsequent model predictions. The measured wind field data mainly includes wind speed data and location data. Wind speed data reflects the velocity information of airflow at the measurement point, while location data precisely records the spatial coordinates of the marked data collection points.

[0030] In one embodiment of this application, obtaining measured wind field data of a target urban block includes: determining the placement location of the target urban block based on a placement strategy, setting the wind speed sensor at the placement location, and pre-setting the location data of the placement location into the corresponding wind speed sensor; and acquiring wind speed data and corresponding location information through the wind speed sensor. The placement strategy can be calibrated according to actual conditions.

[0031] Specifically, firstly, based on the deployment strategy, wind speed sensors are installed and fixed at corresponding locations in the target city blocks. During installation, it must be ensured that their orientation is consistent with the measurement axis. During the sensor initialization phase, the three-dimensional coordinates and ground elevation of the point, obtained beforehand through precise measurements (such as total station or RTK (Real-Time Kinematic)), are preset as location data into the sensor's internal memory or associated data acquisition unit via its accompanying configuration software or hardware interface. Then, the sensor is activated for long-term or continuous monitoring over a specific period, collecting wind speed and direction data at a set frequency, and binding the measured values ​​with the corresponding preset location information in real time, forming a complete record containing timestamps, wind speed, wind direction, and spatial coordinates.

[0032] It should be noted that there may be multiple wind speed sensors described in this embodiment, and the aforementioned position data can be used to represent the relative positions between multiple wind speed sensors.

[0033] In one embodiment of this application, the layout strategy is generated by: acquiring a three-dimensional map of the target city block; and generating the layout strategy based on the three-dimensional map and / or wind field simulation data.

[0034] Specifically, a high-precision 3D map of the target street can be acquired, and a deployment strategy (e.g., equidistant placement) can be generated based on the spatial geometry of this 3D map. Alternatively, existing wind field simulation data (such as from historical projects or standard databases) can be used for feature extraction to generate a deployment strategy. A combination of both methods can be used for more precise decision-making. Using the 3D map, a simulation database covering different inflow conditions can be constructed using computational fluid dynamics tools (such as OpenFOAM, ANSYS Fluent, etc.). Flow field analysis based on this database can employ either a uniform point selection method (ensuring uniform spatial coverage) or a feature-based point selection method (focusing on key flow field regions, such as acceleration zones and vortex zones) to determine a set of sparse feature points. These points will serve as the actual deployment locations for subsequent wind speed sensors and also as input points for the physical information neural network model. This ensures that subsequent sensor deployment comprehensively covers key areas while maintaining a reasonable point density to guarantee the economy and efficiency of data acquisition.

[0035] Step S2: Preprocess the wind speed data to obtain the target wind speed data.

[0036] Specifically, this step aims to preprocess the raw measured wind speed data directly acquired from the sensor, eliminating interference and invalid information in the raw data. First, the raw wind speed data collected by the wind speed sensor in step S1 is extracted and matched with preset location data to remove wind speed data without location association caused by sensor misplacement or signal transmission interruption. Statistical analysis methods (such as...) can also be used subsequently. Outlier identification is performed on valid raw wind speed data using either statistical analysis (based on criteria) or physical consistency testing methods. Specifically, statistical analysis calculates the mean and standard deviation of the wind speed data; preferably, data deviating more than three times the standard deviation from the mean are considered outliers. Physical consistency testing, combined with wind field simulation data from the aforementioned simulation database, identifies measured data exceeding the reasonable range of simulated wind speeds at the corresponding location and under the corresponding inflow conditions as outliers. All identified outliers are then removed. Finally, considering the convergence efficiency and accuracy requirements of subsequent model training, the wind speed data can be standardized, such as by normalizing the data to a certain degree. The interval is used to eliminate the dimensional differences in wind speed data at different locations and under different inflow conditions, and finally obtain target wind speed data that meets the model input requirements and is both continuous and reasonable.

[0037] Step S3: Input the target wind speed data and location data into the first wind field prediction model to obtain the wind field prediction results for the urban block.

[0038] Specifically, the target wind speed data and its corresponding location data are input into a specially designed neural network model. This model, by learning the spatiotemporal evolution and physical constraints of complex wind fields, can accurately predict the complete wind field within a continuous spatial range of the target urban block. This result can include wind speed and direction information at any location within the block at a specific time, achieving an intelligent mapping from sparse point measurements to full-field information. In this embodiment, the model can be a multivariate long-sequence wind speed prediction model with spatiotemporal feature embedding.

[0039] In one embodiment of this application, the first wind field prediction model includes a regression layer, an embedding layer, and a perception layer, wherein the embedding layer includes a temporal embedding layer and a feature embedding layer, and the perception layer comprises multiple layers.

[0040] Specifically, the regression layer, located at the end of the model, typically consists of one or more fully connected layers. Its function is to decode the high-dimensional target wind field features encoded by the perception layer and map them into specific physical prediction values, i.e., outputting the three-dimensional wind speed components or synthetic wind speed and direction for each predicted location. The embedding layer, located at the beginning of the model, is the key part for information extraction and representation, containing two parallel branches: the temporal embedding layer, specifically designed to extract temporal dependencies and periodic patterns from the target wind speed data. It receives wind speed sequences with timestamps and learns the evolution of wind speed over time, such as inertial changes and periodic fluctuations, through temporal models such as Recurrent Neural Networks (RNNs), Long Short-Term Memory Networks (LSTMs), or Transformer encoders, outputting temporal features of wind speed containing temporal dynamics. The feature embedding layer is mainly responsible for deep feature fusion and extraction from the input target wind speed and location data. It combines wind speed values ​​with their spatial coordinates (x, y, z), and through structures such as Multilayer Perceptrons (MLPs), learns the correlation between spatial points and the statistical laws governing the spatial distribution of wind speed. It extracts wind distance features containing spatial correlations (such as features describing the wind speed attenuation relationship between two points) and temporal wind field features reflecting the spatial wind speed distribution pattern at a specific moment. The perception layer is the intermediate module connecting the embedding layer and the regression layer, composed of multiple cascaded neural network layers (such as MLP blocks and attention blocks). Its main function is to receive and deeply fuse wind speed temporal features, wind distance features, and temporal wind field features from different embedding paths. Through multi-level nonlinear transformations and feature interactions, it captures the complex spatiotemporal coupling relationship of wind speed, such as how wind speed changes at a certain location are influenced by upstream historical wind conditions and the current surrounding wind field. It gradually abstracts and refines target wind field features that can comprehensively and completely characterize the flow field state and evolution trend of the entire block at the current moment.

[0041] In one embodiment of this application, target wind speed data and location data are input into a first wind field prediction model to obtain wind field prediction results for urban blocks. This includes: extracting wind speed time features from the target wind speed data through a time embedding layer; extracting features from the target wind speed data and location data through a feature embedding layer to obtain wind distance features and time-based wind field features; fusing the wind speed time features, wind distance features, and time-based wind field features through a perception layer to obtain target wind field features; and performing wind field prediction based on the target wind field features through a regression layer to obtain wind field prediction results for urban blocks.

[0042] Specifically, the entire model's inference process is a hierarchical, step-by-step feature extraction, fusion, and mapping process. Wind distance, in particular, describes the functional distance between any two sensor locations, taking into account the ease or difficulty of actual airflow. This distance calculation depends not only on the geometric path length between the two points but, more importantly, on the velocity fluctuation information obtained from the simulation database, reflecting the degree of building disturbance to airflow along each segment of the path. Unobstructed paths with less disturbance contribute a smaller effective distance, while obstructed paths with greater disturbance contribute a larger effective distance. Therefore, wind distance can more realistically reflect the correlation between two points in an urban wind field than simple straight-line distance or Manhattan distance. Furthermore, in this embodiment, key temporal context features, namely hour and month labels, are constructed. The hour (0-23) and month (1-12) information corresponding to the prediction time are transformed into learnable embedding vectors for the model, providing the model with prior knowledge of the inherent daily and annual cycles in wind speed changes.

[0043] See Figure 2 The temporal embedding layer extracts time-series data from the target wind speed data and converts it into wind speed temporal features via a corresponding encoder. These features may include short-term fluctuations. The feature embedding layer extracts features from the target wind speed and location data to obtain corresponding spatial and temporal features, which are then encoded to obtain corresponding wind distance and temporal wind field features. The perception layer (MLP) then fuses the wind speed temporal features, wind distance features, and temporal wind field features (e.g., by stitching fusion) to obtain the target wind field features. Finally, the regression layer performs wind field prediction (regression prediction) based on these target wind field features to obtain the wind field prediction results for urban blocks.

[0044] Furthermore, the received input consists of target wind speed data from N sensor locations over the past T time steps, forming a tensor. The preprocessed target wind speed data and their location data, distributed across N sparse feature points, are transformed into wind speed predictions for the entire city block over S time steps at N points, as shown in the following formula:

[0045] Where i represents the position variable, t represents the time variable, and S is the number of time steps.

[0046] The time embedding layer will embed the original historical time series Embedded into the hidden space In, as described in the following formula:

[0047] in, This represents the fully connected layer used for time series embedding, and represents the dimension of the embedding.

[0048] Simultaneously, the system constructs and encodes a set of heterogeneous auxiliary features. The feature embedding layer constructs features to assist in wind speed prediction. These features include: spatial features, wind distance features reflecting the physical correlation of urban wind fields calculated based on sensor location data; and temporal context features, hourly and monthly labels corresponding to the prediction time, used to provide prior knowledge of daily and annual cycles. These heterogeneous auxiliary features are input together into a feature embedding layer, which is encoded and fused through another fully connected network, outputting unified auxiliary features, as shown in the following formula:

[0049] in, This represents the feature embedding layer.

[0050] The first wind field prediction model aggregates the embedded features into a single feature. As shown in the following formula:

[0051] in, .

[0052] The first wind field prediction model concatenates the equal-dimensional features output from the two paths and inputs the concatenated fused features into a deep encoding network consisting of n perceptual layers. Residual connections and Dropout techniques are introduced to prevent overfitting and ensure training stability, as shown in the following formula:

[0053] in, This indicates the Drop-out layer.

[0054] Finally, the regression layer receives the target wind field features after deep encoding and maps them back to physical space to calculate the wind field prediction results at N target spatial points over the next S time steps:

[0055] Step S4: Input the wind field prediction results into the second wind field prediction model to obtain the target wind field prediction results for the urban block.

[0056] Specifically, this step aims to achieve a key leap from sequential prediction of future wind speeds at sparse locations to high-resolution, high-fidelity 3D flow field reconstruction and prediction for the entire urban block. The cascading of these two models forms a complete technical chain of "sequence prediction - physical reconstruction," ultimately enabling dynamic, high-resolution prediction of wind fields in urban blocks. In this embodiment, the model can be a physical information neural network (SA-PINN) model based on an adaptive soft attention mechanism. Its design concept combines the physical principles of computational fluid dynamics (CFD) with the powerful fitting capabilities of deep learning.

[0057] The wind field prediction results at sparse feature points at a specific future time, output by the first wind field prediction model obtained in step S3, are used as part of the supervision data and boundary conditions for the SA-PINN model. At the same time, the coordinate information of the three-dimensional geometric domain, i.e. the computational domain, of the target block is used as the spatial input of the SA-PINN model. Finally, the SA-PINN model performs forward calculation and physical constraint optimization within the computational domain, outputting a series of three-dimensional velocity and pressure fields covering the continuous space of the entire block, thus obtaining the target wind field prediction results for the urban block.

[0058] This embodiment first acquires measured wind field data for the target urban area, including wind speed and location data. The wind speed data is then preprocessed to obtain target wind speed data. This target wind speed and location data are then input into a first wind field prediction model to obtain the wind field prediction result for the urban area. Finally, the wind field prediction result is input into a second wind field prediction model to obtain the target wind field prediction result for the urban area. Thus, a dual-model cascaded prediction framework, combining measured wind speed and relevant location data, enables high-precision and high-efficiency dynamic prediction of complex wind fields in urban areas.

[0059] In one embodiment of this application, the first wind field prediction model is generated by: obtaining measured wind field training data of the target city block and the first training label corresponding to the measured wind field training data; inputting the measured wind field training data into the first wind field prediction model to generate a predicted first wind field result; generating a first loss value based on the predicted first wind field result and the first training label, and training the first wind field prediction model based on the first loss value.

[0060] Specifically, the wind field training data can be the target wind speed data collected over a long period by sensors deployed in actual urban areas in step S1 and preprocessed in step S2. This data is constructed into multiple training samples, each containing wind speed observation sequences from N sensors over T consecutive historical time steps, along with corresponding auxiliary features. The first training label corresponds to the future ground truth value for each training sample. Specifically, the label is the actual observed wind speed of the same N sensors over the subsequent S future time steps. The label data also comes from actual measurements, ensuring that the model learns the true physical evolution laws.

[0061] In this embodiment, see Figure 3 The system inputs the constructed measured wind field training data into the first wind field prediction model to be trained, and obtains the corresponding prediction results. The difference between the predicted results and the true labels is calculated to form the first loss value. Mean squared error is typically used as the loss function to minimize the overall deviation between the predicted and measured wind speeds. Based on the first loss value, the gradients of the model parameters (including the weights and biases of each embedding layer, perception layer, and regression layer) are calculated using the backpropagation algorithm. Subsequently, an optimization algorithm (such as Adam) is used to update the model parameters based on the gradients. This process is iterated until the model's prediction accuracy on the validation set converges, ultimately resulting in a trained first wind field prediction model capable of reliably predicting future wind speeds based on historical sparse measured data.

[0062] In one embodiment of this application, the second wind field prediction model is generated by: obtaining wind field simulation data of the target city block and the second training label corresponding to the wind field simulation data; inputting the wind field simulation data into the second wind field prediction model to generate the predicted second wind field result; generating a second loss value based on the predicted second wind field result and the second training label, and training the second wind field prediction model based on the second loss value.

[0063] Specifically, the wind field simulation data was constructed using computational fluid dynamics (CFD) tools, based on a three-dimensional geometric model of the target block, through extensive pre-simulation. The data covers various inflow conditions such as wind speed and direction. A large number of simulation cases were extracted from this database, each providing complete flow field data of the target block under high-resolution CFD simulation. The second training label is the global high-resolution flow field ground truth for each simulation case, i.e., the velocity components (u, v, w) and pressure p at each grid point obtained from the CFD solution. These labeled data are physically rigorously self-consistent and highly accurate.

[0064] In this embodiment, see Figure 3For each simulation case, a set of wind speed data is sampled from the high-resolution CFD flow field at the same spatial locations as the sparse feature points actually deployed in step S1, simulating the observed values ​​at the measured points. This set of sparse point wind speed data, along with the spatial coordinates of these points and the coordinate set of the global computational domain, constitutes the wind field simulation data input for the SA-PINN model. The corresponding global CFD flow field serves as the second training label for supervised learning. The constructed input (sparse point wind speed + global coordinates) is input into the SA-PINN model to be trained, and the model outputs a prediction of the global flow field. The second loss value consists of two parts: data loss, which calculates the mean square error between the model's predicted wind speed at the sparse sampling point locations and the simulated observations extracted from the CFD. This ensures that the model's output at known points is consistent with the true value; and physical loss, which calculates the NS equation residual corresponding to the flow field output by the model at the global coordinate points. By minimizing this residual, the flow field output by the hard-constrained model satisfies the laws of mass conservation and momentum conservation. The adaptive soft attention mechanism dynamically adjusts the weights of the physical loss in different spatial regions (especially high-gradient regions such as the boundary layer and turbulent areas) to enhance the fitting of these physically complex regions. The total loss, obtained by summing the data loss and the weighted physical loss, is then used to update the SA-PINN network parameters via backpropagation. Through iterative training with numerous simulation samples of different inflow conditions, the model eventually learns to reconstruct a physically plausible flow field with high accuracy across the entire domain, based solely on sparse point wind speed information. This pre-trained model lays a solid foundation for subsequently receiving predictions from the first model and performing real-time, high-resolution wind field reconstruction.

[0065] As a specific embodiment of this application, such as Figure 4 As shown, the method for refined prediction of wind fields in urban blocks may include the following steps: S110 generates sensor deployment strategies based on 3D maps and wind field simulation data.

[0066] S120, based on the deployment strategy, sets up wind speed sensors at designated locations to collect wind speed data and corresponding location information.

[0067] S130 preprocesses the measured wind speed data to obtain the target wind speed data.

[0068] S140: Input the target wind speed data and location data into the first wind field prediction model to obtain the wind field prediction results for the urban block.

[0069] S150: Input the wind field prediction results into the second wind field prediction model to obtain the target wind field prediction results for the urban block.

[0070] In summary, the refined wind field prediction method for urban blocks according to the embodiments of this application first obtains measured wind field data of the target urban block, including wind speed data and location data. Then, the wind speed data is preprocessed to obtain target wind speed data. Next, the target wind speed data and location data are input into a first wind field prediction model to obtain the wind field prediction result for the urban block. Finally, the wind field prediction result is input into a second wind field prediction model to obtain the target wind field prediction result for the urban block. Thus, by using a dual-model cascade prediction framework that combines measured and simulated data, both spatiotemporal continuity and physical consistency can be achieved, enabling high-precision and high-efficiency prediction of complex wind fields in urban blocks.

[0071] Corresponding to the above embodiments, this application also proposes a method for refined prediction of wind fields in urban blocks.

[0072] like Figure 5 As shown, the urban street wind field fine prediction device 500 of this application embodiment includes: acquisition module 510, preprocessing module 520, first prediction module 530 and second prediction module 540.

[0073] The system includes: an acquisition module 510 for acquiring measured wind field data of the target city block, including wind speed data and location data; a preprocessing module 520 for preprocessing the wind speed data to obtain target wind speed data; a first prediction module 530 for inputting the target wind speed data and location data into a first wind field prediction model to obtain wind field prediction results for the city block; and a second prediction module 540 for inputting the wind field prediction results into a second wind field prediction model to obtain target wind field prediction results for the city block.

[0074] According to one embodiment of this application, the acquisition module 510 is specifically used to acquire measured wind field data of a target urban block, including: determining the placement location of the target urban block based on the placement strategy, setting the wind speed sensor at the placement location, and pre-setting the location data of the placement location into the corresponding wind speed sensor; and acquiring wind speed data and corresponding location information through the wind speed sensor.

[0075] According to one embodiment of this application, the first wind field prediction model includes a regression layer, an embedding layer, and a perception layer, wherein the embedding layer includes a temporal embedding layer and a feature embedding layer, and the perception layer comprises multiple layers.

[0076] According to one embodiment of this application, the preprocessing module 520 is specifically used to input target wind speed data and location data into a first wind field prediction model to obtain wind field prediction results for urban blocks, including: extracting wind speed time features from the target wind speed data through a time embedding layer; extracting features from the target wind speed data and location data through a feature embedding layer to obtain wind distance features and time-based wind field features; fusing the wind speed time features, wind distance features, and time-based wind field features through a perception layer to obtain target wind field features; and performing wind field prediction based on the target wind field features through a regression layer to obtain wind field prediction results for urban blocks.

[0077] According to one embodiment of this application, the first wind field prediction model is generated in the following manner: obtaining the measured wind field training data of the target city block and the first training label corresponding to the measured wind field training data; inputting the measured wind field training data into the first wind field prediction model to generate the predicted first wind field result; generating a first loss value based on the predicted first wind field result and the first training label, and training the first wind field prediction model based on the first loss value.

[0078] According to one embodiment of this application, the second wind field prediction model is generated by: obtaining wind field simulation data of the target city block and the second training label corresponding to the wind field simulation data; inputting the wind field simulation data into the second wind field prediction model to generate the predicted second wind field result; generating a second loss value based on the predicted second wind field result and the second training label, and training the second wind field prediction model based on the second loss value.

[0079] According to one embodiment of this application, the layout strategy is generated by: acquiring a three-dimensional map of the target city block; and generating the layout strategy based on the three-dimensional map and / or wind field simulation data.

[0080] It should be noted that the above explanation of the embodiments and beneficial effects of the method for refined prediction of wind fields in urban blocks also applies to the refined prediction device for wind fields in urban blocks in the embodiments of this application. To avoid redundancy, it will not be elaborated in detail here.

[0081] In summary, the urban street wind field refinement prediction device according to the embodiments of this application first acquires measured wind field data of the target urban street through an acquisition module. This measured wind field data includes wind speed data and location data. Then, a preprocessing module preprocesses the wind speed data to obtain target wind speed data. Next, a first prediction module inputs the target wind speed data and location data into a first wind field prediction model to obtain the wind field prediction result for the urban street. Finally, a second prediction module inputs the wind field prediction result into a second wind field prediction model to obtain the target wind field prediction result for the urban street. Thus, by using a dual-model cascade prediction framework, combining measured wind speed and relevant location data, high-precision and high-efficiency dynamic prediction of complex wind fields in urban streets is achieved.

[0082] Corresponding to the above embodiments, this application also proposes an electronic device.

[0083] like Figure 6 As shown, the electronic device 600 of this application embodiment includes a memory 610, a processor 620, and a computer program stored in the memory and executable on the processor. The processor executes the program to implement any of the above-mentioned methods for refined prediction of urban street wind fields.

[0084] The electronic device according to the embodiments of this application implements any of the above-mentioned methods for refined prediction of urban street wind fields when the processor executes a computer program. It realizes high-precision and high-efficiency dynamic prediction of complex wind fields in urban streets by combining measured wind speed and related location data through a dual-model cascade prediction framework.

[0085] Corresponding to the above embodiments, this application also proposes a computer-readable storage medium.

[0086] The computer-readable storage medium of this application embodiment stores a computer program that is executed by a processor to implement any of the above-described methods for refined prediction of urban street wind fields.

[0087] The computer-readable storage medium according to the embodiments of this application implements any of the above-mentioned methods for refined prediction of urban street wind fields when the computer program stored thereon is executed by a processor. It realizes high-precision and high-efficiency dynamic prediction of complex wind fields in urban streets by combining measured wind speed and related location data through a dual-model cascade prediction framework.

[0088] Specifically, in the embodiments of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0089] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0090] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for fine prediction of urban street wind field, characterized in that, include: Obtain measured wind field data of the target city blocks, wherein the measured wind field data includes wind speed data and location data; The wind speed data is preprocessed to obtain the target wind speed data; The target wind speed data and the location data are input into the first wind field prediction model to obtain the wind field prediction results of the urban block; The wind field prediction results are input into the second wind field prediction model to obtain the target wind field prediction results for the urban block.

2. The urban street canyon wind field refinement prediction method of claim 1, wherein, The acquisition of measured wind field data for the target city blocks includes: Based on the deployment strategy, the deployment location of the target city block is determined, and the wind speed sensor is set at the deployment location, and the location data of the deployment location is preset into the corresponding wind speed sensor. The wind speed data and corresponding location information are obtained through the wind speed sensor.

3. The method for refined prediction of urban street wind fields according to claim 1, characterized in that, The first wind field prediction model includes a regression layer, an embedding layer, and a perception layer. The embedding layer includes a temporal embedding layer and a feature embedding layer, and the perception layer consists of multiple layers.

4. The method for refined prediction of urban street wind fields according to claim 3, characterized in that, The step of inputting the target wind speed data and the location data into the first wind field prediction model to obtain the wind field prediction result of the urban block includes: The wind speed time feature is extracted from the target wind speed data through the time embedding layer. The feature embedding layer is used to extract features from the target wind speed data and the location data to obtain wind distance features and time-varying wind field features. The wind speed time features, wind distance features, and time-varying wind field features are fused by the perception layer to obtain the target wind field features. Wind field prediction results for the urban blocks are obtained by performing wind field prediction based on the target wind field characteristics through a regression layer.

5. The method for refined prediction of urban street wind fields according to claim 1, characterized in that, The first wind field prediction model is generated in the following way: Obtain the measured wind field training data of the target city blocks and the first training label corresponding to the measured wind field training data; The measured wind field training data is input into the first wind field prediction model to generate the predicted first wind field result; A first loss value is generated based on the predicted first wind field result and the first training label, and the first wind field prediction model is trained based on the first loss value.

6. The method for refined prediction of urban street wind fields according to claim 1, characterized in that, The second wind field prediction model is generated in the following way: Obtain wind field simulation data of the target city block and the second training label corresponding to the wind field simulation data; The wind field simulation data is input into the second wind field prediction model to generate the predicted second wind field result; A second loss value is generated based on the predicted second wind field result and the second training label, and the second wind field prediction model is trained based on the second loss value.

7. The method for refined prediction of urban street wind fields according to claim 2, characterized in that, The layout strategy is generated in the following way: Obtain a 3D map of the target city blocks; The layout strategy is generated based on the 3D map and / or the wind field simulation data.

8. A device for refined prediction of wind fields in urban blocks, characterized in that, include: The acquisition module is used to acquire measured wind field data of the target city blocks, wherein the measured wind field data includes wind speed data and location data; A preprocessing module is used to preprocess the wind speed data to obtain target wind speed data; The first prediction module is used to input the target wind speed data and the location data into the first wind field prediction model to obtain the wind field prediction result of the urban block. The second prediction module is used to input the wind field prediction results into the second wind field prediction model to obtain the target wind field prediction results of the urban block.

9. An electronic device, characterized in that, include: The memory, the processor, and the computer program stored in the memory and executable on the processor, the processor executing the program to implement the urban block wind field refinement prediction method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by the processor to implement the urban block wind field refinement prediction method as described in any one of claims 1-7.

Images