A high-precision wind speed prediction method and system based on a mesoscale weather model and deep learning
By combining mesoscale meteorological models with deep learning, and employing WRF simulation and singular spectrum analysis for noise reduction, a CNN-GRU model was constructed to correct the wind speed at the bridge site. This approach solved the accuracy and cost issues of wind speed prediction for bridges in complex terrain, achieving high-precision wind speed prediction and meeting the needs of bridge safety and early warning.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HARBIN INST OF TECH
- Filing Date
- 2026-04-29
- Publication Date
- 2026-07-21
AI Technical Summary
Existing technologies struggle to achieve high-precision, low-cost wind speed prediction in complex terrains such as long-span bridges. Mesoscale numerical simulation and deep learning methods each have their limitations and cannot meet engineering requirements.
We employ the WRF mesoscale meteorological model for multi-layer nested grid simulation, combine singular spectrum analysis to denoise the measured data, construct a CNN-GRU deep learning model to correct the bias sequence, and output high-precision wind speed predictions for bridge sites.
It achieves low-cost, high-precision wind speed prediction at bridge sites, improves the stability and generalization ability of prediction, and meets the needs of bridge wind resistance safety and vortex-induced vibration early warning.
Smart Images
Figure CN122432779A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of bridge wind engineering, numerical weather prediction, and deep learning fusion technology, and particularly to a high-precision wind speed prediction method and system based on mesoscale meteorological models and deep learning. Specifically, it relates to a method and system for predicting average wind speed at bridge sites based on numerical weather prediction, singular spectrum analysis, convolutional neural networks, and gated recurrent units. Background Technology
[0002] Long-span bridges are located in complex terrains such as straits and islands, where wind fields exhibit significant spatial heterogeneity and temporal fluctuations. Accurate wind speed prediction at the bridge site is a core prerequisite for ensuring the bridge's wind resistance safety, vortex-induced vibration early warning, and operational management.
[0003] Existing wind speed forecasts are mainly divided into two categories: one is mesoscale numerical weather prediction methods represented by WRF, which can characterize atmospheric physical processes, but are limited by grid resolution and terrain parameterization accuracy, resulting in large local simulation biases and high cost of high-resolution nested computation; the other is pure data-driven machine learning methods, which rely on historical monitoring data, lack physical constraints, have poor generalization ability for extreme weather and small sample scenarios, and are prone to producing prediction results that violate atmospheric laws.
[0004] Currently, there is a lack of high-precision and rapid prediction methods that combine WRF physical simulation with deep learning bias correction and are adapted to the microscale wind field of bridge sites, making it difficult to simultaneously meet the engineering requirements for accuracy and efficiency. Summary of the Invention
[0005] The purpose of this invention is to address the aforementioned shortcomings of existing technologies by providing a high-precision wind speed prediction method and system based on mesoscale meteorological models and deep learning. This method integrates the advantages of numerical simulation physics mechanisms with deep learning temporal modeling, achieving low computational cost, high accuracy, and long-term stable wind speed prediction at bridge sites.
[0006] This invention is achieved through the following technical solution: This invention proposes a high-precision wind speed prediction method based on mesoscale meteorological models and deep learning, the method comprising: Step 1: Initial wind field simulation using WRF; Step 2: Data Acquisition and SSA Noise Reduction; Collect three-dimensional ultrasonic anemometer data at the 1 / 4 span, mid-span, and 3 / 4 span locations of the main span of the bridge, and perform window downsampling; Use Singular Spectrum Analysis (SSA) to construct the trajectory matrix and perform singular value decomposition to separate high-frequency noise and retain the effective wind speed signal; Step 3: Construction of historical deviation sequences; calculation of historical deviations to form a time-series deviation dataset; Step 4: CNN-GRU deep learning model correction; after normalizing the bias sequence, local fluctuation features are extracted by a one-dimensional CNN, and long-term temporal dependencies are captured by GRU; the model outputs single-step / multi-step prediction bias; an early stopping strategy is adopted to prevent overfitting; Step 5: Correct the WRF prediction results with the corrected deviation prediction results to predict the high-precision wind speed time history at the bridge site.
[0007] Furthermore, in step one, a four-layer nested computational domain d01–d04 is adopted. The outer layer provides a large-scale meteorological background, and the innermost layer d04 focuses on the core area of the bridge site, with a horizontal resolution of 100m and 38 vertical layers. The ARW dynamic core is selected, with GFS 0.25°×0.25° data as the initial field and boundary conditions. The simulation output time resolution is 10 minutes to obtain the bridge deck height and wind speed data.
[0008] Furthermore, in step two, the SSA noise reduction setting window length L=3, the number of principal components r=1, and the noise-reduced sequence retains more than 95% of the original wind speed energy.
[0009] Furthermore, in step three, the historical deviation is calculated using the following formula:
[0010] In the formula, Historical bias reflects the systematic error between WRF simulated values and actual observations. The simulated wind speed output by the WRF model. These are values observed on-site.
[0011] Furthermore, in step four, the CNN-GRU model parameters are input as follows: 12-step historical bias; CNN: Conv1D(64,3), MaxPooling1D(2), Dropout(0.2); GRU: 128 units, Dropout(0.2); optimizer: Adam, learning rate 0.001, and loss function is MSE.
[0012] Furthermore, the loss function MSE is formulated as follows:
[0013] In the formula, N For the sample size, For the first i The predicted value for each sample, For the first i The true value of each sample.
[0014] Furthermore, in step five, the model correction formula is:
[0015] In the formula, The revised forecast results These are the raw future forecast values output by the WRF model. This represents the historical bias after model correction.
[0016] This invention also proposes a high-precision wind speed prediction system based on mesoscale meteorological models and deep learning, the system comprising: The WRF simulation module is used to perform multi-layer nested grid wind field simulations and output initial wind speed results. The data preprocessing module is used for sampling of measured wind speed drop and SSA noise reduction; The deviation calculation module is used to generate a deviation sequence between WRF simulation and measured data; The deep learning correction module is used to run the CNN-GRU model to achieve bias prediction. The results output module is used to output the corrected wind speed prediction value.
[0017] The present invention also proposes an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the high-precision wind speed prediction method based on mesoscale meteorological models and deep learning.
[0018] The present invention also proposes a computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the steps of the high-precision wind speed prediction method based on mesoscale meteorological models and deep learning.
[0019] The beneficial effects of this invention are: This invention proposes a high-precision wind speed prediction method and system based on mesoscale meteorological models and deep learning. First, the invention employs a WRF numerical model to perform a multi-layered nested grid fine-grained wind field simulation of the bridge site area to obtain initial wind speed prediction results. Then, singular spectral analysis (SSA) is used to denoise the measured wind speed data from the field, and the historical deviation sequence between the WRF simulated values and the measured values is calculated. Next, a deep learning model integrating a convolutional neural network (CNN) and a gated recurrent unit (GRU) is constructed to extract local features and model long-term temporal patterns in the deviation sequence, thereby correcting the deviation of the initial WRF prediction results. Finally, a high-precision average wind speed prediction value for the bridge site is output. This invention integrates the physical mechanisms of numerical weather prediction with the data-driven advantages of deep learning, significantly improving the prediction accuracy of wind fields with a resolution of hundreds of meters in complex terrain bridge sites while controlling computational costs. It can meet the engineering requirements of wind resistance safety, vortex-induced vibration early warning, and operational management of long-span bridges, demonstrating excellent prediction stability and generalization ability. Attached Figure Description
[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0021] Figure 1 This is a flowchart illustrating a high-precision wind speed prediction method based on mesoscale meteorological models and deep learning. Figure 2 This is a schematic diagram of the CNN-GRU network structure; Figure 3 This is a schematic diagram of the single-step wind speed prediction results; Figure 4 This is a schematic diagram of the multi-step wind speed prediction results. Detailed Implementation
[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0023] Specifically, in combination Figures 1-4 This invention proposes a high-precision wind speed prediction method based on mesoscale meteorological models and deep learning, the method comprising: Step 1: Initial wind field simulation using WRF; Step 2: Data Acquisition and SSA Noise Reduction; Collect three-dimensional ultrasonic anemometer data at the 1 / 4 span, mid-span, and 3 / 4 span locations of the main span of the bridge, and perform window downsampling (with a 10-minute window for downsampling); Use Singular Spectrum Analysis (SSA) to construct the trajectory matrix and perform singular value decomposition to separate high-frequency noise and retain the effective wind speed signal; Step 3: Construction of historical deviation sequences; calculation of historical deviations to form a time-series deviation dataset; Step 4: CNN-GRU deep learning model correction; after normalizing the bias sequence, local fluctuation features are extracted by a one-dimensional CNN, and long-term temporal dependencies are captured by GRU; the model outputs single-step / multi-step prediction bias; an early stopping strategy is adopted to prevent overfitting; Step 5: Correct the WRF prediction results with the corrected deviation prediction results to predict the high-precision wind speed time history at the bridge site.
[0024] Furthermore, in step one, a four-layer nested computational domain d01–d04 (d01 (12.5km), d02 (2.5km), d03 (500m), d04 (100m)) is adopted with a grid of 50×50. The outer layer provides a large-scale meteorological background, and the innermost layer d04 focuses on the core area of the bridge site with a horizontal resolution of 100m and 38 vertical layers. The ARW dynamic core is selected, and GFS data with a resolution of 0.25°×0.25° is used as the initial field and boundary conditions. The time step meets the CFL stability criterion, with a time step of 0.32s and CFL≤0.4. The simulation output time resolution is 10 minutes to obtain the wind speed data at the bridge deck height (output: 10-minute interval, interpolated wind speed at a height of 54.5m).
[0025] Furthermore, in step two, the SSA noise reduction window length is set to L=3, the number of principal components is r=1, and the denoised sequence retains more than 95% of the original wind speed energy. The noise reduction RMSE is 0.1302m / s.
[0026] Furthermore, in step three, the historical deviation is calculated using the following formula:
[0027] In the formula, Historical bias reflects the systematic error between WRF simulated values and actual observations. The simulated wind speed output by the WRF model. These are values observed on-site.
[0028] Furthermore, in step four, the CNN-GRU model parameters are input as follows: 12-step (2-hour) historical bias; CNN: Conv1D(64,3), MaxPooling1D(2), Dropout(0.2); GRU: 128 units, Dropout(0.2); optimizer: Adam, learning rate 0.001, loss function is MSE.
[0029] Furthermore, the loss function MSE is formulated as follows:
[0030] In the formula, N For the sample size, For the first i The predicted value for each sample, For the first i The true value of each sample.
[0031] Furthermore, in step five, the model correction formula is:
[0032] In the formula, The revised forecast results These are the raw future forecast values output by the WRF model. This represents the historical bias after model correction.
[0033] This invention also proposes a high-precision wind speed prediction system based on mesoscale meteorological models and deep learning, the system comprising: The WRF simulation module is used to perform multi-layer nested grid wind field simulations and output initial wind speed results. The data preprocessing module is used for sampling of measured wind speed drop and SSA noise reduction; The deviation calculation module is used to generate a deviation sequence between WRF simulation and measured data; The deep learning correction module is used to run the CNN-GRU model to achieve bias prediction. The results output module is used to output the corrected wind speed prediction value.
[0034] Example like Figure 1 As shown, this invention proposes a high-precision wind speed prediction method based on mesoscale meteorological models and deep learning, the method comprising: The first step is to use the WRF model to perform a fine simulation of the bridge site area with a four-layer nested grid, and output the initial wind speed field with a resolution of 100 meters.
[0035] The second step involves collecting measured data from a three-dimensional ultrasonic anemometer on the bridge deck, and then using singular spectrum analysis (SSA) to reduce noise and eliminate sensor noise and high-frequency pulsation interference.
[0036] The third step is to calculate the historical deviation sequence between the measured wind speed after noise reduction and the WRF simulated wind speed.
[0037] The fourth step is to input the biased sequence into the CNN. The GRU hybrid model uses CNN to extract local temporal features and GRU to capture long-term temporal dependencies, outputting future bias predictions.
[0038] The fifth step is to superimpose the WRF forecast wind speed with the prediction bias to obtain the corrected average wind speed, thus completing the prediction.
[0039] This invention uses the TensorFlow deep learning framework to build a model that can be used for high-precision wind speed correction and prediction at bridge sites, directly supporting bridge health monitoring, vortex-induced vibration early warning, wind-resistant design, and operational safety management.
[0040] The present invention also proposes an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the high-precision wind speed prediction method based on mesoscale meteorological models and deep learning.
[0041] The present invention also proposes a computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the steps of the high-precision wind speed prediction method based on mesoscale meteorological models and deep learning.
[0042] The memory in this application embodiment can be volatile memory or non-volatile memory, or it can include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct rambus RAM (DR RAM). It should be noted that the memory used in the methods described in this invention is intended to include, but is not limited to, these and any other suitable types of memory.
[0043] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., high-density digital video discs (DVDs)), or semiconductor media (e.g., solid-state disks (SSDs)).
[0044] In implementation, each step of the above method can be completed by integrated logic circuits in the processor's hardware or by instructions in software. The steps of the method disclosed in the embodiments of this application can be directly implemented by a hardware processor, or by a combination of hardware and software modules in the processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, detailed descriptions are omitted here.
[0045] It should be noted that the processor in the embodiments of this application can be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method embodiments can be completed by the integrated logic circuitry in the processor's hardware or by instructions in software form. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied as execution by a hardware decoding processor, or as a combination of hardware and software modules in the decoding processor. The software modules can be located in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory, and the processor reads the information in the memory and, in conjunction with its hardware, completes the steps of the above methods.
[0046] The above provides a detailed description of a high-precision wind speed prediction method and system based on mesoscale meteorological models and deep learning proposed in this invention. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.
Claims
1. A high-precision wind speed prediction method based on mesoscale meteorological models and deep learning, characterized in that, The method includes: Step 1: Initial wind field simulation using WRF; Step 2: Data Acquisition and SSA Noise Reduction; Collect three-dimensional ultrasonic anemometer data at the 1 / 4 span, mid-span, and 3 / 4 span locations of the main span of the bridge, and perform window downsampling; Use Singular Spectrum Analysis (SSA) to construct the trajectory matrix and perform singular value decomposition to separate high-frequency noise and retain the effective wind speed signal; Step 3: Construction of historical deviation sequences; calculation of historical deviations to form a time-series deviation dataset; Step 4: CNN-GRU deep learning model correction; after normalizing the bias sequence, local fluctuation features are extracted by a one-dimensional CNN, and long-term temporal dependencies are captured by GRU; the model outputs single-step / multi-step prediction bias; an early stopping strategy is adopted to prevent overfitting; Step 5: Correct the WRF prediction results with the corrected deviation prediction results to predict the high-precision wind speed time history at the bridge site.
2. The method according to claim 1, characterized in that, In step one, a four-layer nested computational domain d01–d04 is used. The outer layer provides a large-scale meteorological background, and the innermost layer d04 focuses on the core area of the bridge site. The horizontal resolution is 100m, and the vertical layer is 38 layers. The ARW dynamic core is selected, and the GFS 0.25°×0.25° data is used as the initial field and boundary conditions. The simulation output time resolution is 10 minutes to obtain bridge deck height and wind speed data.
3. The method according to claim 1, characterized in that, In step two, the SSA noise reduction setting window length L=3 and the number of principal components r=1, and the noise-reduced sequence retains more than 95% of the original wind speed energy.
4. The method according to claim 1, characterized in that, In step three, the historical deviation is calculated using the following formula: In the formula, Historical bias reflects the systematic error between WRF simulated values and actual observations. The simulated wind speed output by the WRF model. These are values observed on-site.
5. The method according to claim 1, characterized in that, In step four, the CNN-GRU model parameters are input as follows: 12-step historical bias; CNN: Conv1D(64,3), MaxPooling1D(2), Dropout(0.2); GRU: 128 units, Dropout(0.2); optimizer: Adam, learning rate 0.001, loss function is MSE.
6. The method according to claim 5, characterized in that, The loss function MSE is formulated as follows: In the formula, N For the sample size, For the first i The predicted value for each sample, For the first i The true value of each sample.
7. The method according to claim 1, characterized in that, In step five, the model correction formula is: In the formula, The revised forecast results These are the raw future forecast values output by the WRF model. This represents the historical bias after model correction.
8. A high-precision wind speed prediction system based on mesoscale meteorological models and deep learning, characterized in that, The system includes: The WRF simulation module is used to perform multi-layer nested grid wind field simulations and output initial wind speed results. The data preprocessing module is used for sampling of measured wind speed drop and SSA noise reduction; The deviation calculation module is used to generate a deviation sequence between WRF simulation and measured data; The deep learning correction module is used to run the CNN-GRU model to achieve bias prediction. The results output module is used to output the corrected wind speed prediction value.
9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1-7.
10. A computer-readable storage medium for storing computer instructions, characterized in that, When the computer instructions are executed by the processor, they implement the steps of the method according to any one of claims 1-7.