Valley area bridge site wind speed prediction method based on valley wind theory and topographic feature decomposition technology

Abstract

The invention provides a valley region bridge site wind speed prediction method based on a valley wind theory and a topographic feature decomposition technology. The method comprises the following steps: acquiring observed wind speed and wind direction data and meteorological element data; acquiring high-resolution terrain elevation data near the bridge site, and calculating terrain feature parameters; determining the canyon trend of the position of the observation station according to the coordinates of the observation station and the terrain elevation data; the observed wind speed and wind direction time sequence is decomposed according to the canyon trend, and axial wind along the canyon trend and normal wind perpendicular to the canyon trend are obtained respectively; respectively predicting an axial wind component and a normal wind component of the target station by using the axial wind prediction model and the normal wind prediction model, and performing projection synthesis on prediction results along the canyon trend to obtain final prediction results of the wind speed and the wind direction; according to the method, the problem that the full wind speed of bridge site wind in a complex mountainous area is difficult to accurately predict due to the influence of valley wind can be relieved, and the accuracy of valley full wind speed prediction is improved.

Description

Technical Field

[0001] This invention relates to the field of disaster monitoring and early warning technology, and more specifically, to a method for predicting wind speed at bridge sites in canyon areas based on valley wind theory and topographic feature decomposition technology. Background Technology

[0002] The coverage of high-speed railways has gradually expanded from the plains of central and eastern China to the complex mountainous regions of central and western China. Mountainous terrain is highly variable, and high-speed railways often need to cross railway bridges over valleys. The steep terrain and diverse surface conditions of complex mountainous areas, driven by strong solar radiation and topographic forcing, can create intense and complex regional and local-scale circulation systems. The strong and unique local wind fields in complex mountainous areas can significantly impact the design, construction, operation, and maintenance of railway bridges, posing a significant threat to train safety. To ensure the safe operation of high-speed railways crossing mountainous railway bridges, accurate wind forecasting near the bridge site is crucial. Currently, accurate ultra-short-term wind speed forecasting remains a challenging problem, primarily due to the complex intermittent, fluctuating, and random characteristics of wind, influenced by numerous factors such as temperature, air pressure, and topography.

[0003] Under complex terrain conditions, background wind fields, local atmospheric circulation, topographic forcing, surface friction, and thermal effects all contribute to wind formation. Commonly observed wind speeds are influenced by both background wind fields and local atmospheric circulation. Mountainous regions exhibit complex circulation structures, which can encompass various systems. Considering only the thermal effects of mountain topography, large-scale mountainous areas exhibit a mountain wind system composed of slope winds, valley winds, and mountain-plain winds, each dominating a different region. Slope winds develop along the slope, exhibiting the smallest temporal and spatial scales, responding immediately to radiation changes, and gradually transferring radiative warming to the valley floor. Due to the smaller environmental capacity within valleys, the rate of temperature rise (or fall) is higher than in plains areas. After a period of time, updraft valley winds (and downdraft valley winds) flow along the valley axis from the plains into the valley floor. The thermal difference between the mountain range and the plains generates mountain-plain wind circulation. In actual observations, these different circulation systems are often confused and difficult to distinguish.

[0004] Local atmospheric circulation in complex terrain is mainly characterized by valley winds. A key feature of valley winds is their diurnal variation in wind speed, with daytime valley winds and nighttime mountain winds alternating. The formation mechanism of valley winds is similar to that of slope winds: air on mountain slopes is closer to the ground, resulting in a faster rate of heating (cooling); while air at the same altitude in valleys is farther from the ground, resulting in a slower rate of heating (cooling). Isotherms slope along the slope, and buoyancy causes airflow to rise (sink) along the slope. Valley winds often refer to the overall three-dimensional structure. During the day-night cycle, temperature differences arise between mountain slopes and valleys, and between mountains and plains. These temperature differences lead to differences in near-surface atmospheric density and pressure. The pressure gradient force propels airflow from high-pressure (low-temperature) areas to low-pressure (high-temperature) areas. Mountain winds and valley winds are caused by the different thermal effects at mountain peaks and valleys throughout the day. During the day, along the valley direction, in higher elevation areas at the valley floor, the temperature is higher due to solar radiation, causing air expansion and a decrease in air pressure. Therefore, air rises along the valley line from lower to higher elevations, creating valley winds. In the morning, as one ascends the slope from the valley, the higher the temperature, the stronger the wind, reaching a maximum of level 3 in the afternoon before gradually weakening. At night, the higher elevation of the slopes causes faster heat dissipation, resulting in a rapid drop in temperature. Cold air descends from the mountaintops and flows into the valley, creating mountain winds. Furthermore, on the slopes perpendicular to the valley line, due to the difference in warming rates between the slopes and the surrounding atmosphere, uphill winds prevail during the day, while downhill winds prevail at night.

[0005] In general: along the valley's orientation, valley winds blow during the day and mountain winds blow at night. On the slopes on both sides of the valley, uphill winds blow during the day and downhill winds blow at night. Therefore, the actual wind vector includes both valley winds along the valley's orientation and uphill and downhill winds perpendicular to the valley's orientation, both exhibiting significant diurnal variations. The presence of valley winds makes the wind field in complex mountainous areas more complex than in plains areas, increasing the difficulty and accuracy of predicting wind speed and direction in bridge site areas, which is detrimental to the safe operation of high-speed railways on mountain bridges.

[0006] Generally, based on different principles, wind speed modeling methods can be categorized into four types: physical modeling methods, statistical modeling methods, spatial correlation modeling methods, and data-driven methods. The most common wind speed prediction method is the independent prediction method, which treats wind speed and wind direction as completely independent time series, ignoring their correlation, and using separate models to predict future values. Besides this direct wind speed prediction method, two other common wind speed prediction methods are: orthogonal decomposition method: This method orthogonally decomposes wind speed into east-west and north-south wind speeds according to wind direction. From a physical perspective, forces in the vertical direction do not affect each other; therefore, prediction models can be established for the wind speeds in the two directions separately. Finally, by reconstructing the wind speeds in the two directions using an orthogonal coordinate system, the predicted wind vector value can be obtained. The flowchart is as follows... Figure 2As shown; Dominant Wind Component Prediction Method: In the joint wind speed and direction prediction model, the wind speed time series is decomposed along the dominant wind direction to obtain crosswind and longitudinal wind component series. Assuming no correlation between the two series, the two components are predicted independently, and the prediction results are combined to obtain the final predicted wind speed and direction. The flowchart is shown below. Figure 3 As shown.

[0007] However, research has found that the existing wind speed prediction methods mentioned above still cannot accurately characterize actual atmospheric processes, and the accuracy of wind speed prediction is low. Summary of the Invention

[0008] The purpose of this invention is to provide a wind speed prediction method for bridge sites in canyon areas based on valley wind theory and terrain feature decomposition technology. This method addresses the technical problems in existing technologies where wind speed prediction models for complex mountainous areas lack physical meaning, have poor physical interpretability, cannot accurately represent actual atmospheric processes, and have low wind speed prediction accuracy, thus affecting the effectiveness of the prediction models.

[0009] The objective of this invention can be achieved through the following technical solutions:

[0010] This invention provides a method for predicting wind speed at bridge sites in canyon areas based on valley wind theory and topographic feature decomposition technology, including:

[0011] Obtain observational wind speed and direction data from stations near the bridge site, as well as meteorological element data from standard meteorological stations around the bridge site; the meteorological element data shall include at least temperature, humidity, air pressure, wind speed and direction, total radiation, longwave radiation flux, shortwave radiation flux, sensible heat flux, and latent heat flux.

[0012] High-resolution terrain elevation data near the bridge site is obtained, and terrain feature parameters are calculated. The terrain parameters include the coordinates of the observation station, the terrain slope, slope angle, canyon depth, and canyon width within the canyon area. Based on the coordinates of the observation station and the terrain elevation data, the canyon orientation and canyon normal at the location of the observation station are determined.

[0013] The wind speed and wind direction data obtained from the observation station are projected along the canyon axis and the canyon normal, respectively, to obtain the axial wind component and the normal wind component.

[0014] Based on the sunrise and sunset times at the bridge site, and combined with the canyon topographic features, the switching times of mountain winds and valley winds were obtained, and a time series of valley wind switching was constructed.

[0015] Using the axial wind component as the target variable and the valley wind switching time series and the meteorological element time series of the standard meteorological station as feature factors, an axial wind prediction model is constructed based on the LSTM model; using the normal wind component as the target variable and the valley wind switching time series and the meteorological element time series of the standard meteorological station as feature factors, a normal wind prediction model is constructed based on the LSTM model.

[0016] The axial wind prediction model and the normal wind prediction model are used to predict the axial wind component and normal wind component of the target site, respectively. The prediction results are then projected and synthesized along the canyon direction to obtain the final prediction results of wind speed and wind direction.

[0017] The predicted wind speed results are compared with the actual observed values, and the prediction accuracy is quantitatively evaluated through error indicators.

[0018] Preferably, as one feasible implementation, step S2 determines the canyon orientation and canyon normal based on the coordinates and topographic elevation data of the observation station. Specific methods include:

[0019] Topographic elevation data was cropped with the center point of the bridge site as the center (i.e., topographic DEM data with a side length of 15 kilometers was cropped).

[0020] Extract the canyon area, calculate the terrain slope, and extract the lowest value of the line connecting the terrain heights within the canyon area as the canyon bottom line.

[0021] The canyon orientation is calculated using linear regression and taken as the canyon axis, while the direction perpendicular to the canyon axis is taken as the canyon normal.

[0022] Preferably, as one possible implementation, the calculation method for the axial wind component and the normal wind component is as follows:

[0023] Taking true north as 0 degrees and clockwise rotation as positive, let true north be the positive y-axis, true east be the positive x-axis, wind speed be V, and meteorological wind direction angle be θ. Then, the angle between the wind direction and the positive x-axis... for:

[0024] ;

[0025] Based on orthogonal decomposition:

[0026] ;

[0027] ;

[0028] If the angle between the canyon's axis (direction) and the positive x-axis is... The component along the axial direction

[0029] ;

[0030] Component along the canyon normal (perpendicular to the axis)

[0031] .

[0032] Preferably, as one feasible implementation; the step of modifying the sunrise and sunset times based on the bridge site location and combining them with the canyon topographic features to obtain the switching time between mountain winds and valley winds, and constructing a valley wind switching time series, includes the following modification steps:

[0033] The sunrise and sunset times at the bridge site are calculated based on latitude and longitude as a benchmark;

[0034] Based on the canyon width and depth, the slope, aspect and altitude of the terrain on both sides, the sunrise and sunset times are corrected to obtain the switching time of mountain wind and valley wind, and construct the valley wind switching time series.

[0035] Preferably, as an feasible implementation, the sunrise and sunset times at the bridge site are calculated based on latitude and longitude as a benchmark; and the sunrise and sunset times are corrected according to the canyon width and depth, the slope, aspect, and altitude of the terrain on both sides to obtain the switching time of mountain winds and valley winds, and to construct a time series of valley wind switching, including the following steps:

[0036] Calculate the daily sunrise and sunset times, with an initial correction time of 0;

[0037] The correction is made based on the canyon width and depth. If the canyon width is less than 800 meters or the canyon depth is greater than 500 meters, the correction time for the occurrence of valley winds will be delayed by 60 minutes.

[0038] The correction is made according to the terrain slope. If the terrain slope is greater than 30 degrees, the correction time for the occurrence of valley winds is advanced by 30 minutes.

[0039] The correction is made according to the slope aspect. If the slope aspect is north, it is a sunny slope, and the occurrence time of the valley wind is adjusted to be postponed by 60 minutes.

[0040] The timing of valley winds is adjusted based on altitude, with a 30-minute delay for every 1000-meter increase in altitude. The state of mountain and valley winds is marked by cosine and sine function encoding, and the sunrise and sunset times and the adjusted times are converted into features that can be recognized by the LSTM model.

[0041] Preferably, as an implementation scheme, when constructing the axial wind prediction model and the normal wind prediction model, the optimal model parameters and the optimal feature combination are selected by grid search and optimal feature selection methods.

[0042] Preferably, as one possible implementation, the construction of the axial wind prediction model and the construction of the normal wind prediction model include:

[0043] The axial wind of the station to be predicted is taken as the target variable, and the time series of valley wind switching and the time series of standard meteorological station elements are taken as feature factors. The training set and the test set are divided. The optimal model parameters and the optimal feature combination are selected by grid search and optimal feature selection method to minimize the root mean square error between the actual value and the predicted value. Finally, the model converges. Based on the LSTM model, an axial wind prediction model is constructed.

[0044] The normal wind at the station to be predicted is taken as the target variable, and the time series of valley wind switching and the time series of standard meteorological station elements are taken as feature factors. The training set and the test set are divided. The optimal model parameters and the optimal feature combination are selected by grid search and optimal feature selection methods to minimize the root mean square error between the actual value and the predicted value. Finally, the model converges. Based on the LSTM model, a normal wind prediction model is constructed.

[0045] Preferably, as one possible implementation, the projection synthesis method is as follows: based on the predicted axial wind component and normal wind component, as well as the canyon orientation angle, the wind speed and wind direction angle are calculated by vector synthesis.

[0046] Preferably, as one possible implementation, the error index includes at least one of root mean square error, absolute deviation, and relative error percentage.

[0047] Compared with the prior art, the embodiments of this application have at least the following technical effects:

[0048] This invention provides a wind speed prediction method for bridge sites in canyons based on valley wind theory and topographic feature decomposition technology. Wind speed is decomposed into an axial component (mainly constrained by topography, with relatively stable wind direction) and a normal component (mainly driven by diurnal variations of valley winds, with wind direction reversing between day and night). This allows the LSTM model to learn the two physical mechanisms separately, avoiding interference from mixed signals. Simultaneously, standard meteorological station data (such as radiation and sensible heat flux) are introduced as feature factors to provide indirect observational information on the thermal driving process, compensating for the scarcity of stations. Finally, the full wind speed is recovered through projection synthesis, and error index evaluation ensures that the prediction results meet engineering accuracy requirements.

[0049] Therefore, this invention provides a method for predicting wind speed at bridge sites in canyons based on valley wind theory and topographic feature decomposition technology. The core of this method lies in decoupling the complex actual wind field at the bridge site into two physically distinct independent components based on valley wind theory: one is the wind component along the canyon direction (axial wind), mainly influenced by topographic constraints and large-scale airflow within the canyon; the other is the wind component perpendicular to the canyon direction (normal wind), mainly dominated by the circulation of valley winds (uphill and downhill winds). These two wind components have different formation mechanisms and are influenced by different physical processes, which can solve the problem of unpredictable overall wind speed at bridge sites due to the influence of valley winds, improve the accuracy of overall wind speed prediction in canyons, and benefit the safe operation of high-speed railways in canyon areas. By independently constructing LSTM prediction models for these two components, the evolution patterns influenced by different physical processes can be precisely characterized, thus effectively solving the problem of traditional methods' difficulty in accurately predicting overall wind speed in canyon areas due to hybrid modeling. Attached Figure Description

[0050] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1 A schematic diagram showing the decomposition of axial wind and normal wind;

[0052] Figure 2 A flowchart illustrating the prediction process of an existing wind speed prediction model based on the orthogonal method.

[0053] Figure 3 Here is a flowchart of the prediction process based on the prevailing wind direction method in existing technologies;

[0054] Figure 4 A simplified full-process diagram of the canyon area bridge site wind speed prediction method based on valley wind theory and terrain feature decomposition technology provided by the present invention;

[0055] Figure 5 A schematic diagram of the main process of the canyon area bridge site wind speed prediction method based on valley wind theory and terrain feature decomposition technology provided by the present invention;

[0056] Figure 6 This is a schematic diagram of the canyon's orientation during a verification test in a specific case. Detailed Implementation

[0057] 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.

[0058] Example 1

[0059] See Figure 5 Embodiment 1 of this invention provides a method for predicting wind speed at bridge sites in canyons based on valley wind theory and topographic feature decomposition technology. It fully considers the important role of valley winds in local wind fields in mountainous areas and the impact of their daily periodic variations on wind speed and direction. Specific implementation steps include:

[0060] S1: Obtain observational wind speed and direction data from stations near the bridge site, as well as meteorological element data from standard meteorological stations around the bridge site; the meteorological element data shall include at least temperature, humidity, air pressure, wind speed and direction, total radiation, longwave radiation flux, shortwave radiation flux, sensible heat flux, and latent heat flux.

[0061] S2: Obtain high-resolution terrain elevation data near the bridge site and calculate terrain feature parameters; the terrain parameters include the coordinates of the observation station, the terrain slope, slope angle, canyon depth and canyon width within the canyon area; based on the coordinates of the observation station and the terrain elevation data, determine the canyon orientation (i.e., canyon axis) and canyon normal at the location of the observation station.

[0062] S3: Project the wind speed and wind direction data obtained from the observation station along the canyon axis and the canyon normal, respectively, to obtain the axial wind component and the normal wind component;

[0063] S4: Based on the sunrise and sunset times at the bridge site, and combined with the canyon topographic features, the switching times of mountain winds and valley winds are obtained, and a time series of valley wind switching is constructed.

[0064] S5: Using the axial wind component as the target variable, and the valley wind switching time series and the meteorological element time series of the standard meteorological station as feature factors, an axial wind prediction model is constructed based on the LSTM model; using the normal wind component as the target variable, and the valley wind switching time series and the meteorological element time series of the standard meteorological station as feature factors, a normal wind prediction model is constructed based on the LSTM model.

[0065] S6: The axial wind prediction model and the normal wind prediction model are used to predict the axial wind component and normal wind component of the target station, respectively, and the prediction results are projected and synthesized along the canyon direction to obtain the final prediction results of wind speed and wind direction.

[0066] S7: Compare the predicted wind speed with the actual observed values, and quantitatively evaluate the prediction accuracy through error indicators.

[0067] The present invention provides a method for predicting wind speed at bridge sites in canyons based on valley wind theory and topographic feature decomposition technology. The main steps include: S1: Obtaining observed wind speed and direction data from stations near the bridge site, as well as meteorological element data from standard meteorological stations around the bridge site; S2: Obtaining high-resolution topographic elevation data near the bridge site and calculating topographic feature parameters; the topographic parameters include the coordinates of the observation stations, the topographic slope within the canyon area, the slope angle, the canyon depth, and the canyon width; based on the coordinates and topographic elevation data of the observation stations, determining the canyon orientation and normal direction at the location of the observation stations; S3: Projecting the wind speed and direction data obtained from the observation stations along the canyon axis and the canyon normal direction, respectively, to obtain the axial wind component and the normal wind component; S4: Based on the sunrise and sunset times at the bridge site location, combined with the canyon topographic feature decomposition technology... S5: Using the axial wind component as the target variable, and the valley wind switching time series and the meteorological element time series of the standard meteorological station as feature factors, an axial wind prediction model is constructed based on the LSTM model; using the normal wind component as the target variable, and the valley wind switching time series and the meteorological element time series of the standard meteorological station as feature factors, a normal wind prediction model is constructed based on the LSTM model; S6: The axial wind component and the normal wind component are predicted by the axial wind prediction model and the normal wind prediction model, respectively, and the prediction results are projected and synthesized along the valley direction to obtain the final prediction results of wind speed and wind direction; S7: The predicted wind speed results are compared with the actual observed values, and the prediction accuracy is quantitatively evaluated by error index.

[0068] This invention provides a method for predicting wind speed at bridge sites in canyons based on valley wind theory and topographic feature decomposition technology. Wind speed is decomposed into an axial component (mainly constrained by topography, with relatively stable wind direction) and a normal component (mainly driven by diurnal variations of valley winds, with wind direction reversing between day and night). This allows the LSTM model to learn both types of physical mechanisms separately, avoiding interference from mixed signals. Simultaneously, standard meteorological station data (such as radiation and sensible heat flux) are introduced as feature factors to provide indirect observational information on the thermal driving process, compensating for the scarcity of stations. Finally, the full wind speed is recovered through projection synthesis, and the prediction results are evaluated using error indices to ensure they meet engineering accuracy requirements. This invention provides a method for predicting wind speed at bridge sites in canyons based on valley wind theory and topographic feature decomposition technology, realizing a directional prediction-synthesis framework for wind speed at canyon bridge sites. The complex wind field is decomposed into an axial component controlled by topography and a normal component influenced by thermal cycles, which are modeled separately and then synthesized, significantly reducing model complexity and improving prediction accuracy.

[0069] Preferably, as one feasible implementation, step S2 determines the canyon orientation and canyon normal based on the coordinates and topographic elevation data of the observation station. Specific methods include:

[0070] Topographic elevation data was cropped with the center point of the bridge site as the center (i.e., topographic DEM data with a side length of 15 kilometers was cropped).

[0071] Extract the canyon area, calculate the terrain slope, and extract the lowest value of the line connecting the terrain heights within the canyon area as the canyon bottom line.

[0072] The canyon orientation is calculated using linear regression and taken as the canyon axis, while the direction perpendicular to the canyon axis is taken as the canyon normal.

[0073] Preferably, as one possible implementation, the calculation method for the axial wind component and normal wind component in step S3 is as follows:

[0074] Taking true north as 0 degrees and clockwise rotation as positive, let true north be the positive y-axis, true east be the positive x-axis, wind speed be V, and meteorological wind direction angle be θ. Then, the angle between the wind direction and the positive x-axis... for: Based on orthogonal decomposition: ; If the angle between the canyon's axial direction (or orientation) and the positive x-axis is... Then the component along the axial direction:

[0075] ;

[0076] Component along the canyon normal (perpendicular to the axis):

[0077] .

[0078] In the above technical solution, wind speed and direction data obtained from all observation points near the bridge site are projected axially and normally to obtain axial and normal wind components. Axial wind represents valley wind processes, while normal wind represents uphill and downhill wind processes. In meteorology, true north is usually considered 0°, and clockwise rotation is considered positive. Figure 1 As shown, if due north is taken as the positive y-axis and due east as the positive x-axis, and the wind speed is V, then the meteorological wind direction angle is... The angle between the wind direction and the positive x-axis is... for: Then perform orthogonal decomposition: ; ;

[0079] Further processing is then performed. If the angle between the canyon axis (direction) and the positive x-axis is... The component along the axial direction Determine the component along the canyon normal (perpendicular to the axis). ;

[0080] Preferably, as one feasible implementation; step S4, based on the sunrise and sunset times of the bridge site location, and combined with the canyon topographic features, is corrected to obtain the switching time between mountain winds and valley winds, and a valley wind switching time series is constructed, including the following correction steps:

[0081] The sunrise and sunset times at the bridge site are calculated based on latitude and longitude as a benchmark;

[0082] Based on the canyon width and depth, the slope, aspect and altitude of the terrain on both sides, the sunrise and sunset times are corrected to obtain the switching time of mountain wind and valley wind, and construct the valley wind switching time series.

[0083] Preferably, as an feasible implementation, the sunrise and sunset times at the bridge site are calculated based on latitude and longitude as a benchmark; and the sunrise and sunset times are corrected according to the canyon width and depth, the slope, aspect, and altitude of the terrain on both sides to obtain the switching time of mountain winds and valley winds, and to construct a time series of valley wind switching, including the following steps:

[0084] Calculate the daily sunrise and sunset times, with an initial correction time of 0;

[0085] The correction is made based on the canyon width and depth. If the canyon width is less than 800 meters or the canyon depth is greater than 500 meters, the correction time for the occurrence of valley winds will be delayed by 60 minutes.

[0086] The correction is made according to the terrain slope. If the terrain slope is greater than 30 degrees, the correction time for the occurrence of valley winds is advanced by 30 minutes.

[0087] The correction is made according to the slope aspect. If the slope aspect is north, it is a sunny slope, and the occurrence time of the valley wind is adjusted to be postponed by 60 minutes.

[0088] The timing of valley winds is adjusted based on altitude, with a 30-minute delay for every 1000-meter increase in altitude. The state of mountain and valley winds is marked by cosine and sine function encoding, and the sunrise and sunset times and the adjusted times are converted into features that can be recognized by the LSTM model.

[0089] It should be noted that the switching time of mountain winds and valley winds near the bridge site was obtained. The method for obtaining the switching time of mountain winds and valley winds is as follows: the sunrise and sunset times of the bridge site are calculated based on latitude and longitude, and the sunrise and sunset times are corrected according to the width and depth of the canyon, the slope, aspect and altitude of the terrain on both sides, so as to obtain the switching time of mountain winds and valley winds and construct the valley wind switching time series.

[0090] The specific steps are as follows:

[0091] S41: Calculate the sunrise and sunset times for each day, with a time correction of 0;

[0092] S42: Execution is based on valley width and valley depth correction: If the valley width at the observation point is less than 800 meters or the valley depth is greater than 500 meters, the correction time for the occurrence of valley winds is delayed by 60 minutes.

[0093] S43: Implement slope correction: If the slope of the observation point is greater than 30 degrees, the correction time for the occurrence of valley winds will be advanced by 30 minutes.

[0094] S44: Perform slope correction: If the slope is north-facing, it is a sunny slope, and the correction time for the occurrence of valley winds is postponed by 60 minutes.

[0095] S45: Implement altitude correction: For every 1000 meters increase in altitude, the correction time for the occurrence of valley winds is delayed by 30 minutes;

[0096] S46: By encoding with cosine and sine functions, the mountain wind / valley wind state is marked, and the sunrise and sunset times and correction times are converted into features that can be recognized by the LSTM model.

[0097] Preferably, as one feasible implementation, when constructing the axial wind prediction model and the normal wind prediction model in step S5, the optimal model parameters and the optimal feature combination are selected by grid search and optimal feature selection methods.

[0098] Preferably, as one possible implementation, the construction of the axial wind prediction model and the construction of the normal wind prediction model include:

[0099] S51: The axial wind of the station to be predicted is taken as the target variable, and the valley wind switching time series and the standard meteorological station element time series are taken as feature factors. The training set and the test set are divided. The optimal model parameters and the optimal feature combination are selected by grid search and optimal feature selection method to minimize the root mean square error between the actual value and the predicted value. Finally, the model converges. Based on the LSTM model, the axial wind prediction model is constructed.

[0100] S52: The normal wind at the station to be predicted is taken as the target variable, and the valley wind switching time series and the standard meteorological station element time series are taken as feature factors. The training set and test set are divided. The optimal model parameters and optimal feature combination are selected by grid search and optimal feature selection method to minimize the root mean square error between the actual value and the predicted value. Finally, the model converges. Based on the LSTM model, a normal wind prediction model is constructed.

[0101] Preferably, as one possible implementation, the projection synthesis method in step S6 is: based on the predicted axial wind component and normal wind component, as well as the canyon orientation angle, the wind speed and wind direction angle are calculated by vector synthesis.

[0102] Preferably, as one possible implementation, the error index in step S7 includes at least one of root mean square error, absolute deviation, and relative error percentage.

[0103] This invention significantly improves the accuracy of wind speed prediction at canyon bridge sites. Its technical advantages are reflected in: objectively determining the canyon axis through high-resolution terrain analysis, providing a precise geometric benchmark for wind speed decomposition; physically correcting the diurnal wind transition time of valleys using terrain parameters, enabling the model to more accurately learn key diurnal variation nodes; and employing a component-based modeling and vector synthesis framework, reducing the complexity of a single model and enhancing its generalization ability. This method provides a more reliable meteorological basis for the wind-resistant design and safe operation of bridges in canyon areas, especially high-speed railway bridges, and helps improve the risk prevention and control level of infrastructure under complex terrain conditions.

[0104] The following is a specific implementation case of a wind speed prediction method for bridge sites in canyons based on valley wind theory and topographic feature decomposition technology.

[0105] like Figure 4 As shown, this invention provides a method for wind speed decomposition and short-term wind speed prediction at complex mountain bridge sites based on valley wind theory, specifically including the following steps:

[0106] I. Data Acquisition. Acquire latitude and longitude information from wind speed observation stations near the bridge site, providing minute-level wind speed and direction data. Acquire hourly data from standard meteorological stations near the bridge site, including wind speed, temperature, humidity, air pressure, wind direction, wind speed, total radiation, longwave radiation flux, shortwave radiation flux, sensible heat flux, and latent heat flux, converting these to minute-level data through spline interpolation. Acquire high-resolution terrain elevation data of the canyon, such as 90-meter resolution SRTM data.

[0107] 2. Based on the topographic elevation data, extract the topographic elevation of the canyon area (centered on the bridge site, with a side length of 15 kilometers), and calculate topographic parameters such as slope, hillside inclination, canyon depth, and width within the canyon area. Extract the valley floor and obtain its coordinates. Fit the valley floor using linear regression and calculate its orientation angle (the angle of clockwise rotation relative to true north), which is used as the canyon's axis. The direction perpendicular to the axis at 90° is used as the canyon's normal.

[0108] III. Based on the location, longitude, and latitude of the bridge site, a Python program was written to calculate the sunrise and sunset times at the bridge site using the Astral library. The specific steps are as follows: after installing the Astral library in the Python environment, input the longitude and latitude parameters of the surrounding stations, and calculate the sunrise and sunset times using the Astral library. Generally, when the slope is steep, valley winds appear 1-2 hours after sunrise, while on gentler slopes, they form 3-4 hours after sunrise. When the slope is steep, mountain winds begin 1-2 hours after sunset, while on gentler slopes, they form 2-3 hours after sunset. The time it takes for the upslope and downslope winds to change direction is basically consistent with the switching time of valley winds. In this invention, the mountain wind begins at sunset time + 2 hours, and the valley wind begins at sunrise time + 1 hour; the upslope wind begins at sunrise time + 1 hour, and the downslope wind begins at sunset time + 2 hours.

[0109] IV. Projection of Wind Speed ​​Along the Axial and Normal Directions of the Canyon. A station near the bridge site was selected as the target station. Wind speed and direction data at 1-minute intervals were projected along the canyon's axis and normal direction to obtain the axial wind component Vaxis and the normal wind component Vnormal. Within the corresponding time intervals of the wind speeds, a temporal feature sequence TT(t1, t2, t3…) of valley winds was constructed. The sequence value was 1 for mountain winds and downhill winds, and 0 for valley winds and uphill winds. The hourly interval data of meteorological stations near the bridge site were upscaled using linear interpolation to become minute-interval data. These elements were then used as feature values, such as T(t1, t2, t3…), P(t1, t2, t3…), and RH(t1, t2, t3…).

[0110] 5. Using Vaxis as the predictor variable and T, P, RH, etc. as feature parameters, the dataset is divided into training and test sets in a 7:3 ratio. Based on the LSTM model, grid search and feature engineering are used to select the optimal parameter and feature combination, minimizing the root mean square error between the actual and predicted values, and finally constructing the prediction model model_axis. Using Vnormal as the predictor variable and T, P, RH, etc. as feature parameters, the dataset is divided into training and test sets in a 7:3 ratio. Based on the LSTM model, grid search and feature engineering are used to select the optimal parameter and feature combination, minimizing the root mean square error between the actual and predicted values, and finally constructing the prediction model model_normal.

[0111] 6. Use `model_axsi` and `model_normal` to predict the test set, then perform a pseudo-projection based on the canyon orientation to obtain the total wind speed and direction. Statistically calculate the root mean square error (RMSE) of the actual wind speed and the predicted total wind speed, the percentage of relative error, etc., to determine the prediction accuracy of the faulty model.

[0112] To evaluate the effectiveness of this invention, experiments were conducted, the details of which are as follows.

[0113] 1. Taking a bridge site in a certain canyon as an example, there is a wind speed and direction observation station near the bridge site. The station information is shown in Table 1.

[0114] Table 1 shows information on observation stations near a major bridge in a canyon.

[0115] Site longitude latitude Altitude (meters) Chengdu Shore Tower Top (CDATD) 102.2354°E 29.9266°N 1827.9

[0116] 2. Canyon topography, elevation, and station locations, as follows: Figure 6 As shown: Figure 6 The red dots represent the actual locations of the stations listed in Table 1. The red line represents the fitted canyon orientation, and the green and yellow arrows point to the canyon's axis and normal direction (counterclockwise is positive relative to due east). As shown in the figure, the canyon's orientation is southwest-southeast-northeast, with an orientation (axis) of 73.3° and a normal of 163.3°.

[0117] 3. Components of wind speed and direction projected along the axial and normal axes of the canyon. As described in step S3, taking the top of the Chengdu shore tower as an example, the wind speed is divided into axial wind and normal wind.

[0118] 4. Comparison of the prediction method used in the embodiments of the present invention with the results of other wind speed prediction methods.

[0119] Table 2 is a comparison table of indicators between the present invention and existing methods.

[0120] Mean Absolute Error (MAE) Mean Squared Error (MSE) Root Mean Square Error (RMSE) Mean Absolute Error Percentage (MAPE) (%) Coefficient of determination (R²) Orthogonal method 0.281437 0.164205 0.405223 37.539391 0.818294 Dominant wind direction decomposition method 0.280199 0.163777 0.404693 36.192661 0.818769 Prediction method of the present invention 0.277412 0.162606 0.403244 36.144227 0.820064

[0121] The prediction method used in this embodiment of the invention is compared with the traditional orthogonal method (zonal wind and meridional wind) and the dominant wind direction decomposition method. The models all use the same structure and hyperparameters and are trained for the same number of rounds. The wind speed of the test set is predicted, and the results are shown in Table 2.

[0122] The results show that, compared with existing orthogonal methods (zonal and meridional winds) and dominant wind direction decomposition methods, the method provided by this invention performs best on multiple evaluation indicators. Specifically, the prediction method used in this embodiment of the invention is optimal in terms of root mean square error, mean square error, and mean absolute error compared to the orthogonal method and the dominant wind direction decomposition method. The percentage of mean absolute error is also lower than that of the orthogonal method and the dominant wind direction decomposition method. Regarding the coefficient of determination, the prediction method used in this embodiment of the invention has the highest coefficient of determination, indicating that the prediction method used in this embodiment of the invention has a significant advantage over the orthogonal method and the dominant wind direction decomposition method in predicting wind speed. This is crucial for the accuracy of wind speed prediction near bridge sites in complex mountainous areas.

[0123] In summary, this invention provides a method for predicting wind speed at bridge sites in canyons based on valley wind theory and topographic feature decomposition technology. It realizes a directional prediction-synthesis framework for wind speed at canyon bridge sites, decomposes complex wind fields into axial and normal components, models them separately, and then synthesizes them, which significantly reduces model complexity and improves prediction accuracy.

[0124] The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, and all such modifications and additions should fall within the protection scope of the present invention.

Claims

1. A method for predicting wind speed at bridge sites in canyon areas based on valley wind theory and topographic feature decomposition technology, characterized in that, The method includes the following steps: Obtain observational wind speed and direction data from stations near the bridge site, as well as meteorological element data from standard meteorological stations around the bridge site; the meteorological element data shall include at least temperature, humidity, air pressure, wind speed and direction, total radiation, longwave radiation flux, shortwave radiation flux, sensible heat flux, and latent heat flux. High-resolution topographic elevation data near the bridge site is acquired, and topographic feature parameters are calculated. The topographic parameters include the coordinates of the observation station, the topographic slope within the canyon area, the slope angle, the canyon depth, and the canyon width. Based on the coordinates of the observation station and the topographic elevation data, the canyon axis and canyon normal at the location of the observation station are determined. The canyon axis is the canyon direction. The wind speed and wind direction data obtained from the observation station are projected along the canyon axis and the canyon normal, respectively, to obtain the axial wind component and the normal wind component. Based on the sunrise and sunset times at the bridge site, and combined with the canyon topographic features, the switching times of mountain winds and valley winds were obtained, and a time series of valley wind switching was constructed. Using the axial wind component as the target variable and the valley wind switching time series and the meteorological element time series of the standard meteorological station as feature factors, an axial wind prediction model is constructed based on the LSTM model; using the normal wind component as the target variable and the valley wind switching time series and the meteorological element time series of the standard meteorological station as feature factors, a normal wind prediction model is constructed based on the LSTM model. The axial wind prediction model and the normal wind prediction model are used to predict the axial wind component and normal wind component of the target site, respectively. The prediction results are then projected and synthesized along the canyon direction to obtain the final prediction results of wind speed and wind direction. The predicted wind speed results are compared with the actual observed values, and the prediction accuracy is quantitatively evaluated through error indicators.

2. The method according to claim 1, characterized in that, Based on the coordinates and topographic elevation data of the observation station, the canyon orientation and normal direction at the location of the observation station are determined. Specific methods include: Topographic elevation data was cropped with the center point of the bridge site as the center (i.e., topographic DEM data with a side length of 15 kilometers was cropped). Extract the canyon area, calculate the terrain slope, and extract the lowest value of the line connecting the terrain heights within the canyon area as the canyon bottom line. The canyon orientation is calculated using linear regression and taken as the canyon axis, while the direction perpendicular to the canyon axis is taken as the canyon normal.

3. The method according to claim 1, characterized in that, The calculation methods for the axial wind component and the normal wind component are as follows: Taking true north as 0 degrees and clockwise rotation as positive, let true north be the positive y-axis, true east be the positive x-axis, wind speed be V, and meteorological wind direction angle be θ. Then, the angle between the wind direction and the positive x-axis... for: ； Based on orthogonal decomposition: ； ； If the angle between the canyon's axis (direction) and the positive x-axis is... The component along the axial direction ； Component along the canyon normal (perpendicular to the axis) 。 4. The method according to claim 1, characterized in that, The method involves adjusting the sunrise and sunset times based on the bridge site location, combined with the canyon topographic features, to obtain the switching time between mountain winds and valley winds, and constructing a valley wind switching time series. This includes the following adjustment steps: The sunrise and sunset times at the bridge site are calculated based on latitude and longitude as a benchmark; Based on the canyon width and depth, the slope, aspect and altitude of the terrain on both sides, the sunrise and sunset times are corrected to obtain the switching time of mountain wind and valley wind, and construct the valley wind switching time series.

5. The method according to claim 4, characterized in that, The sunrise and sunset times at the bridge site are calculated based on latitude and longitude as a baseline. Then, the sunrise and sunset times are corrected according to the canyon width and depth, as well as the slope, aspect, and altitude of the terrain on both sides. This yields the switching times of mountain and valley winds, and a time series of these wind transitions is constructed, including the following steps: Calculate the daily sunrise and sunset times, with an initial correction time of 0; The correction is made based on the canyon width and depth. If the canyon width is less than 800 meters or the canyon depth is greater than 500 meters, the correction time for the occurrence of valley winds will be delayed by 60 minutes. The correction is made according to the terrain slope. If the terrain slope is greater than 30 degrees, the correction time for the occurrence of valley winds is advanced by 30 minutes. The correction is made according to the slope aspect. If the slope aspect is north, it is a sunny slope, and the occurrence time of the valley wind is adjusted to be postponed by 60 minutes. The timing of valley winds is adjusted based on altitude, with a 30-minute delay for every 1000-meter increase in altitude. The state of mountain and valley winds is marked by cosine and sine function encoding, and the sunrise and sunset times and the adjusted times are converted into features that can be recognized by the LSTM model.

6. The method according to claim 1, characterized in that, When constructing axial wind prediction models and normal wind prediction models, grid search and optimal feature selection methods are used to select the optimal model parameters and optimal feature combinations.

7. The method as described in claim 6, characterized in that, The construction of axial wind prediction models and normal wind prediction models also includes: The axial wind of the station to be predicted is taken as the target variable, and the time series of valley wind switching and the time series of standard meteorological station elements are taken as feature factors. The training set and the test set are divided. The optimal model parameters and the optimal feature combination are selected by grid search and optimal feature selection method to minimize the root mean square error between the actual value and the predicted value. Finally, the model converges. Based on the LSTM model, an axial wind prediction model is constructed. The normal wind at the station to be predicted is taken as the target variable, and the time series of valley wind switching and the time series of standard meteorological station elements are taken as feature factors. The training set and the test set are divided. The optimal model parameters and the optimal feature combination are selected by grid search and optimal feature selection methods to minimize the root mean square error between the actual value and the predicted value. Finally, the model converges. Based on the LSTM model, a normal wind prediction model is constructed.

8. The method according to claim 1, characterized in that, The projection synthesis method is as follows: based on the predicted axial wind component and normal wind component, as well as the canyon orientation angle, the wind speed and wind direction angle are calculated through vector synthesis.

9. The method according to claim 1, characterized in that, The error metrics include at least one of root mean square error, absolute deviation, and relative error percentage.

Images