Cold air and strong wind identification method and system based on multi-source data and deep learning

Abstract

The embodiment of the invention provides a cold air and strong wind identification method and system based on multi-source data and deep learning. The method is applied to the technical field of data analysis and comprises the steps of creating a historical meteorological data set; constructing a meteorological disaster positive sample and a meteorological disaster negative sample; training a cold air and strong wind identification model based on Res-Unet according to the meteorological disaster positive sample and the meteorological disaster negative sample; and acquiring real-time meteorological element data and real-time three-dimensional radar detection data, and inputting the real-time meteorological element data and the real-time three-dimensional radar detection data into the trained cold air and strong wind identification model based on Res-Unet to obtain a strong wind weather prediction result. According to the scheme, the recognition precision of the model in a complex meteorological phenomenon can be remarkably improved, and a strong wind weather prediction result can be obtained in real time. Therefore, early warning can be given out before cold air and strong wind events occur, precautionary measures are taken in advance, and losses caused by disasters are reduced.

Description

Technical Field

[0001] The present disclosure relates to the field of data analysis technology, and in particular to a method and system for identifying cold air and strong winds based on multi-source data and deep learning. Background Art

[0002] In nature, strong winds are a common meteorological phenomenon. Their suddenness, wide impact, and destructive power pose a serious threat to human life and the social economy. To improve our ability to respond to strong winds and reduce the resulting losses, strong wind forecasting is crucial.

[0003] Today, numerical weather forecast models are used to simulate and predict atmospheric conditions, combined with real-time radar monitoring data to capture current weather changes. This process, which includes data preprocessing, model calculations, and result verification and correction, aims to capture the evolving trends of weather systems. In particular, the prediction of specific meteorological events, such as localized cold fronts and strong winds, requires comprehensive consideration of factors such as topography and climate background to enhance forecast accuracy and relevance. Ultimately, forecast results are published through professional platforms, providing timely and effective meteorological information services to decision makers and the public.

[0004] However, although existing strong wind forecasting methods can provide weather information over a wide range, they have limited ability to capture local cold air strong wind events and smaller-scale meteorological changes. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this paper provides a method and system for identifying cold air and high winds based on multi-source data and deep learning. This paper addresses the technical issues that existing high wind forecasting methods, while capable of providing weather information over a wide range, have limited ability to capture localized cold air and high wind events and smaller-scale meteorological changes.

[0006] According to a first aspect of the present disclosure, a cold air and gale identification method based on multi-source data and deep learning is provided, comprising: acquiring historical meteorological element pattern data and historical three-dimensional radar detection data, and creating a historical meteorological dataset based on the historical meteorological element pattern data and the historical three-dimensional radar detection data;

[0007] Determine the high wind single unit characteristic information and the cold wind meteorological characteristic set of the historical meteorological data set, obtain historical disaster-affected area data, and construct meteorological disaster positive samples and meteorological disaster negative samples based on the high wind single unit characteristic information, the cold wind meteorological characteristic set, and the historical disaster-affected area data;

[0008] Constructing a cold air and strong wind recognition model based on Res-Unet, and training the cold air and strong wind recognition model based on Res-Unet according to the meteorological disaster positive samples and the meteorological disaster negative samples;

[0009] Real-time meteorological element data and real-time three-dimensional radar detection data are obtained, and the real-time meteorological element data and real-time three-dimensional radar detection data are input into a trained cold air and gale recognition model based on Res-Unet to obtain a gale weather forecast result.

[0010] According to a second aspect of the present disclosure, a cold air and gale identification system based on multi-source data and deep learning is provided, which is used to execute the method according to the first aspect, including: a dataset creation module, which is used to obtain historical meteorological element pattern data and historical three-dimensional radar detection data, and create a historical meteorological dataset based on the historical meteorological element pattern data and the historical three-dimensional radar detection data;

[0011] A sample creation module is used to determine the high wind single unit characteristic information and the cold wind meteorological characteristic set of the historical meteorological data set, obtain historical disaster-stricken area data, and construct meteorological disaster positive samples and meteorological disaster negative samples based on the high wind single unit characteristic information, the cold wind meteorological characteristic set and the historical disaster-stricken area data;

[0012] A model training module is used to build a cold air and strong wind recognition model based on Res-Unet, and train the cold air and strong wind recognition model based on Res-Unet according to the meteorological disaster positive samples and the meteorological disaster negative samples;

[0013] The prediction module is used to obtain real-time meteorological element data and real-time three-dimensional radar detection data, and input the real-time meteorological element data and real-time three-dimensional radar detection data into the trained Res-Unet-based cold air and gale recognition model to obtain gale weather forecast results.

[0014] According to a third aspect of the present disclosure, an electronic device is provided. The electronic device includes: a memory and a processor. The memory stores a computer program. When the processor executes the program, the method described above is implemented.

[0015] In the cold air and high wind identification method and system based on multi-source data and deep learning, the disclosed embodiments input real-time meteorological element data and three-dimensional radar detection data into the trained Res-Unet model, significantly improving the model's recognition accuracy for complex meteorological phenomena and enabling real-time high wind forecasts. This means that cold air and high wind events can be identified and issued in advance, allowing for proactive preventive measures and reducing losses. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present disclosure or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0017] Figure 1 A schematic flow chart of a cold air and strong wind identification method based on multi-source data and deep learning according to an embodiment of the present disclosure is shown;

[0018] Figure 2 A schematic flow chart of a cold air and strong wind identification method based on multi-source data and deep learning according to an embodiment of the present disclosure is shown;

[0019] Figure 3 A schematic block diagram of a cold air and gale identification system based on multi-source data and deep learning according to an embodiment of the present disclosure is shown;

[0020] Figure 4 A block diagram of an exemplary electronic device according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0021] Various exemplary embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings. It should be noted that unless otherwise specifically stated, the relative arrangement of components and steps, numerical expressions and numerical values set forth in these embodiments do not limit the scope of the present disclosure.

[0022] Those skilled in the art will understand that the terms "first", "second" and the like in the embodiments of the present disclosure are only used to distinguish different steps, devices or modules, etc., and do not represent any specific technical meaning, nor do they represent the necessary logical order between them. It should also be understood that in the embodiments of the present disclosure, "multiple" may refer to two or more, and "at least one" may refer to one, two or more. It should also be understood that any component, data or structure mentioned in the embodiments of the present disclosure can generally be understood as one or more, unless explicitly defined or given a contrary revelation in the context. In addition, the term "and / or" in the present disclosure is merely a description of the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in the present disclosure generally indicates that the associated objects before and after are in an "or" relationship. It should also be understood that the description of each embodiment in the present disclosure emphasizes the differences between the embodiments, and the same or similar aspects thereof can be referenced to each other. For the sake of brevity, they will not be described one by one.

[0023] At the same time, it should be understood that for ease of description, the dimensions of the various parts shown in the drawings are not drawn according to the actual proportional relationship. The following description of at least one exemplary embodiment is actually only illustrative and is in no way intended to limit the present disclosure and its application or use. Technologies, methods and equipment known to those of ordinary skill in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods and equipment should be considered part of the specification. It should be noted that similar numbers and letters represent similar items in the following figures, so once an item is defined in one figure, it does not need to be further discussed in subsequent figures.

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present disclosure more clear, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present disclosure, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present disclosure without making any creative efforts shall fall within the scope of protection of the present disclosure.

[0025] Figure 1 This is a flow chart of a cold air and strong wind identification method based on multi-source data and deep learning provided by an embodiment of the present disclosure. Figure 1 As shown, the method includes:

[0026] S101 , acquiring historical meteorological element pattern data and historical three-dimensional radar detection data, and creating a historical meteorological dataset based on the historical meteorological element pattern data and historical three-dimensional radar detection data.

[0027] Historical meteorological element pattern data refers to datasets that record the changing patterns of meteorological elements and their interrelationships over a period of time. These include parameters such as temperature, air pressure, humidity, wind speed, wind direction, Convective Available Potential Energy (CAPE), and the K index. These data are typically provided by meteorological observation stations, meteorological satellites, and numerical meteorological forecasting systems. They describe atmospheric conditions and meteorological conditions at different altitudes in the atmosphere. CAPE (Convective Available Potential Energy) is a measure of air's convective capacity and is often used to assess the likelihood of severe convective weather. The K index is used to assess the convective potential of weather and is calculated primarily based on the vertical distribution of temperature, humidity, and air pressure. Some of this data can be presented graphically. Temperature distribution maps can display temperature data as heat maps, with temperature variations across different regions represented by color. Pressure contour maps display the air pressure distribution across different regions, with higher and lower pressure areas indicated by different colors. Humidity / wind speed distribution maps display the spatial distribution of humidity and wind speed. Wind speed is typically represented by wind field diagrams or arrows, while humidity is represented by a color gradient. CAPE / K Index Image: Displays the distribution of CAPE or K index in different regions through heat maps or contour maps, highlighting areas of severe convective weather.

[0028] Historical 3D radar detection data can include three-dimensional meteorological information such as reflectivity and radial velocity acquired by weather radar systems. 3D radar can provide atmospheric data from multiple directions and layers, helping to analyze the spatial distribution and dynamics of phenomena such as storms, precipitation, and air currents. Specifically, this data includes reflectivity: the strength of the signal returned by radar-emitted electromagnetic waves after encountering meteorological particles (such as raindrops and hail), reflecting precipitation intensity and cloud moisture content. Radial velocity: the relative speed between radar waves and meteorological targets, reflecting wind speed and direction. Some of this data can be presented graphically. For example, radar reflectivity maps display radar reflectivity data using color coding, with areas of higher reflectivity (such as heavy rain) indicated by red and purple, and areas of lower reflectivity indicated by green and blue. Radial velocity maps display the spatial distribution of the relative velocity between radar waves and meteorological targets, with positive values (storms approaching the radar) indicated by red, and negative values (storms moving away from the radar) indicated by blue.

[0029] A historical meteorological dataset is a collection of various meteorological elements (such as temperature, air pressure, humidity, and wind speed) as well as historical records related to meteorological disasters. This data is typically collected from weather stations, weather satellites, weather radar, and other equipment, organized and stored in chronological order. It describes weather conditions and historical meteorological events (such as high winds, heavy rains, and heat waves) within a specific time period.

[0030] Historical meteorological element data can be obtained through the following channels: National Meteorological Department / Meteorological Service Agency: For example, the China Meteorological Administration (CMA), the National Oceanic and Atmospheric Administration (NOAA), etc. They will provide public historical meteorological data sets covering meteorological elements (temperature, air pressure, wind speed, etc.) across the country. Meteorological data platform: Some meteorological data platforms provide services to access historical meteorological data through APIs. For example, OpenWeatherMap, Weather Underground, etc. Meteorological satellite data: Global meteorological element data (such as temperature, humidity, wind speed, etc.) are collected through remote sensing satellites. Meteorological data interface: For example, the Global Meteorological Network (GEM), or specific meteorological databases (such as meteorological model data provided by ECMWF).

[0031] Historical 3D radar data can be obtained through the following methods: Meteorological radar data platforms: Meteorological departments in some countries and regions provide historical 3D radar data. This data is typically stored and made available through specialized APIs or file downloads, particularly during the operation of radar systems. Examples include NEXRAD radar data from the United States and Doppler radar data from the China Meteorological Administration. Radar data acquisition and processing systems: This data requires specialized equipment or systems. Radar systems regularly collect and store data such as reflectivity and radial velocity based on meteorological observation requirements.

[0032] Historical meteorological element data and three-dimensional radar data can be collected nationwide. The data can be obtained through the above-mentioned meteorological data platform, meteorological departments, or radar system interfaces. Ensure the integrity of the data set and handle missing data (such as filling or deletion). Standardize the meteorological data so that meteorological data from different time periods and spatial regions can be compared. Store and process all data in a consistent format, such as timestamp, spatial region, etc. Fusion of meteorological element pattern data with radar data to create a comprehensive historical meteorological data set. The data needs to be aligned by time and integrated with information such as wind speed, temperature, precipitation, etc. as needed.

[0033] S102, determining the strong wind single unit characteristic information and the cold wind meteorological characteristic set of the historical meteorological data set, obtaining historical disaster-stricken area data, and constructing meteorological disaster positive samples and meteorological disaster negative samples based on the strong wind single unit characteristic information, the cold wind meteorological characteristic set and the historical disaster-stricken area data.

[0034] A strong wind cell refers to an independent storm system or meteorological phenomenon that can cause strong winds and severe weather. Specifically, the characteristic information of a strong wind cell may include wind speed: the maximum wind speed and sustained wind speed of the strong wind cell, which are key parameters for determining whether it is a strong wind. Wind direction: Changes in the wind direction of strong winds can reveal the direction and range of the storm's movement. Storm intensity: Such as the storm's maximum wind speed, duration, and the core area of the storm (such as the center of the cyclone). Air pressure changes: Strong wind cells are often accompanied by low-pressure areas, and changes in their pressure gradient can reflect the intensity and scale of the storm. CAPE (Convective Potential Energy): This is an indicator of atmospheric instability. A high CAPE value often indicates a stronger storm system. K index: Used to measure instability in the atmosphere. A higher index is usually accompanied by a stronger storm. Updrafts: Strong wind cells are often accompanied by strong updrafts, forming convective activity. Storm development trends: Characteristics of the growth, maturity, and dissipation stages of a storm, including the duration of the storm and the speed at which it strengthens or weakens.

[0035] The cold wind meteorological feature set may refer to meteorological features related to cold air, strong winds, and temperature changes, and may include temperature gradient: the front of cold air (such as a cold front or frontal zone) is usually accompanied by a strong temperature gradient. The greater the temperature difference, the higher the possibility of strong winds. Temperature change: cold winds usually bring about a sharp drop in temperature, and sudden temperature changes may be associated with strong wind events. Humidity: cold winds are accompanied by lower humidity, especially when strong cold air invades, low humidity conditions will occur, which will aggravate the increase in wind speed. Air pressure: cold winds usually occur in low pressure systems, which may bring sustained strong winds. Position and movement of cold fronts: cold fronts or their characteristic areas usually cause strong changes in airflow, bringing about the invasion of cold air and strong winds.

[0036] Historical disaster-affected area data may refer to the spatial and temporal distribution data of high wind or cold wind events that actually occurred in the past. Specifically, it may include the geographic information of the disaster-stricken area: including the coordinates, administrative divisions or specific locations of the disaster-stricken area. Usually it is the location where disasters such as high winds, cold winds, and blizzards occur. Disaster type and intensity: the type of high wind disaster in the historical disaster-stricken area (such as strong winds, hurricanes, etc.) and the intensity of the disaster. Time of disaster occurrence: including the specific date, hour or time period when the high wind occurred. Scope of impact: including the size of the affected area caused by the wind disaster, whether it spans multiple areas, and the affected infrastructure (such as roads, buildings, energy, etc.). Post-disaster losses: if available, post-disaster loss assessment data, including casualties, property losses, infrastructure damage, etc.

[0037] Positive meteorological disaster samples can refer to meteorological images depicting cold air and strong wind events, typically with the cold wind areas annotated. These images mark the areas where strong cold air and strong winds occur and serve as positive samples for model training. Positive sample images must exhibit the following characteristics: they must demonstrate significant characteristics of cold air and strong wind events, such as sudden increases in wind speed, sudden drops in temperature, and dramatic changes in air pressure. Meteorological images of the corresponding areas may contain significant airflow, areas of strong radar reflectivity, and wind field maps. The impact of the cold wind is typically marked in these areas. The annotated cold wind areas are typically distinguished, for example, using polygons, rectangular boxes, or hotspots to clearly indicate the strong wind areas.

[0038] Negative meteorological disaster samples can refer to weather images that do not contain cold air and high wind events. They are typically used as negative samples for model training. These images do not show any cold air and high wind phenomena, and therefore lack the characteristics of high winds. Negative sample images typically exhibit normal weather conditions, with moderate wind speeds, relatively stable temperature and pressure fluctuations, and no severe storms or extreme weather conditions. These images help the model identify images that do not fall within cold air and high wind event areas.

[0039] Based on collected meteorological data, characteristic information related to strong wind cells can be extracted. Specifically, by analyzing historical data, variables with a strong influence on cold air and strong winds can be selected, such as radar echo reflectivity, radial velocity, K index, CAPE index, temperature, and air pressure. Based on these selected variables, computational methods are used to extract features related to strong wind cells. For example, radial velocity shear at different elevation angles, surface pressure gradient, variable pressure gradient, and temperature difference can be calculated. These features can help identify different cold air and strong wind events. To establish a cold wind meteorological feature set, key meteorological parameters closely related to the occurrence and development of cold air and strong winds must first be selected. Specifically, these parameters include temperature, air pressure, humidity, wind speed and direction, surface pressure gradient, temperature difference, and wind shear. Cold air and strong winds are often accompanied by rapid changes in meteorological parameters. Therefore, in addition to the current values of these parameters, their changing trends must also be considered. These changing trends can be extracted using the following methods: Temperature drop: A sharp drop in temperature is often a sign of cold air intrusion and can be quantified by the rate of temperature change. Generally speaking, if the temperature drops by more than a certain threshold (e.g., 5°C) within a short period of time (e.g., within an hour), it may be a sign of cold air intrusion. Temperature change rate = (current temperature - previous temperature) / time interval. Rapid pressure changes: Rapid changes in pressure often indicate the approach of a front or cold air. This can be described by the pressure change rate: Pressure change rate = (current pressure - previous pressure) / time interval. Pressure gradient: The pressure gradient is the rate of change of pressure per unit distance, reflecting the intensity and speed of cold air intrusion. By comparing the pressures of adjacent areas, the pressure gradient is calculated: Pressure gradient = (pressure in area A - pressure in area B) / distance. Wind speed and direction changes: Rapid changes in wind speed and direction often signal a cold air intrusion. Features are extracted by calculating the rate of change of wind speed and the magnitude of the change in wind direction. Temperature difference: Analyze the temperature difference between different layers (e.g., the ground and the upper atmosphere). Large temperature differences often indicate a strong alternation of cold and warm air, possibly accompanied by strong winds. After determining the key meteorological parameters and their changing trends, the next step is to quantify these features and organize them into a feature set. This feature set needs to be able to comprehensively reflect the spatiotemporal changes in meteorological parameters during cold air invasion. It can be constructed through the following steps: Select the feature window: Select an appropriate time window (such as the past 1 hour, 3 hours, etc.) to calculate the rate of change of meteorological parameters, temperature difference, air pressure change, etc. These characteristic values can reflect the spatiotemporal evolution of cold air and strong wind events. Quantify meteorological features: For example, the temperature change rate can be calculated by comparing temperature data at multiple time points, and the set threshold (such as the temperature changes by more than 5°C within 1 hour) is used to calibrate whether it is a cold air invasion event. The pressure gradient and the rate of pressure change can be calculated using the pressure data of adjacent areas to obtain the pressure change rate or gradient value.Wind speed and direction changes: By comparing wind speed and direction data at different time points, we can determine the wind speed change rate and wind direction change amplitude, which serve as indicators of cold winds. Feature combination and fusion: The extracted features are combined and fused to generate the final cold wind meteorological feature set. For example, features such as the temperature change rate, pressure change rate, pressure gradient, and wind speed change can be placed in the same feature set, and each feature can be assigned a specific weight. Based on past meteorological disaster events, historical data on affected areas can be obtained.

[0040] Positive samples are constructed using the following method: By analyzing historical meteorological data for characteristics of strong winds (such as a sharp drop in temperature, a sharp increase in wind speed, and changes in air pressure) and cold winds (such as the passage of a cold front and increased wind speed), time periods and regions where cold air and strong wind events occurred are identified. Within these time periods and regions, meteorological images are extracted from multiple sources, including satellite and radar imagery. These images should demonstrate distinct characteristics of cold air and strong wind events, such as precipitation intensity, storm cloud configuration, and areas of sudden temperature drops. Based on these acquired meteorological images, historical data on affected areas is combined to further determine the impact of the cold wind event. This data can be sourced from meteorological disaster reports, emergency management data, and post-disaster assessment reports. The impact area typically includes the affected geographic area, the extent of damage to buildings and infrastructure, and the extent of the impact. Using this data, geographical annotations can be added to cold wind areas in the imagery, enabling the model to identify not only the presence of cold winds but also their specific impact area. For example, on satellite or radar imagery, affected areas can be annotated with rectangular boxes, circular labels, or colored areas, along with the impact area of the cold wind. This positive sample not only reflects the spatiotemporal characteristics of the cold wind event but also correlates it with the disaster-affected area. By combining meteorological images and data on the affected areas, the cold wind areas are labeled. Labeling methods can include using rectangular or polygonal boxes to mark the cold wind-affected areas. Labeling cold wind events and affected areas with different colors allows the model to learn that the impact of cold winds is not limited to the meteorological phenomenon itself, but also includes its impact on geographic space.

[0041] Negative samples are constructed as follows: Historical meteorological data is filtered to identify time periods and regions that do not contain cold air and strong wind events. These areas typically lack cold wind characteristics, such as a sharp increase in wind speed or a sharp drop in temperature. Meteorological imagery, such as satellite and radar images, is extracted that does not contain cold wind characteristics. These images typically do not show heavy precipitation, sudden temperature changes, or changes in air pressure, nor do they show a sharp increase in wind speed. In the process of constructing negative samples, data on affected areas helps exclude areas that may not be associated with the disaster. For example, if a region has not experienced a cold wind event or suffered damage, images from that region can be identified as negative samples. For these images, there is no need to label the cold wind area, as they are already free of cold wind disasters.

[0042] Based on the above technical solution, optionally, before determining the strong wind single body characteristic information and the cold wind meteorological characteristic set of the historical meteorological data set, the method further includes:

[0043] Abnormal data in the historical meteorological data set is identified, the abnormal data is deleted, and the historical meteorological data set is updated.

[0044] In this solution, abnormal data can be those abnormal data that deviate from the normal pattern or expectation and may be caused by measurement errors, equipment failure, data loss and other factors. Specifically, the types of abnormal data may include measurement errors: including inaccurate data due to instrument failure or calibration problems, such as extreme temperature and air pressure values. Missing values: refers to the absence of data records at certain times or areas, which may be due to sensor failure or network problems. Noisy data: Sometimes there is random noise in the data, which may be due to instantaneous changes in environmental conditions or interference factors (such as radar or satellite signal reflections, etc.). Data that deviates from the pattern: Although these data do not have obvious errors, their values are significantly different from most other data points in the data set, which may be caused by temporary abnormal meteorological phenomena or other special reasons. Extreme values: Certain data values deviate greatly from the normal range (such as very high or very low temperature, air pressure, etc.), which are usually erroneous or abnormal records.

[0045] Methods for detecting abnormal data may include anomaly detection based on statistical methods: Standard deviation method: Calculate the mean and standard deviation of each meteorological element (such as temperature, humidity, wind speed, etc.). If a data point deviates from the mean by more than a certain multiple (usually 3 times the standard deviation), it can be considered as abnormal data. For example: Assuming that the mean temperature is 30°C and the standard deviation is 2°C, if a data point is 10°C, it can be considered abnormal. Box plot method: Use a box plot to identify outliers in a data set. The upper and lower edges of the box plot usually represent the upper quartile (Q3) and lower quartile (Q1) of the data. Outliers are usually data points outside the upper limit (Q3+1.5IQR) or lower limit (Q1-1.5IQR).

[0046] Anomaly detection based on time series: Seasonality and trend detection: If the meteorological data shows seasonal or cyclical changes, seasonal decomposition (such as STL: Seasonal and Trend decomposition using Loess) can be used to detect abnormal data that does not conform to the long-term trend or seasonal changes. For example, temperature data should be lower than a certain threshold in winter. If the temperature data for a certain time period is higher than the normal range, it is an anomaly. Time window detection: In historical meteorological data sets, the moving average method or sliding window method can be used to calculate the predicted value within a certain time window and compare it with the actual data. If the error exceeds the preset threshold, it is marked as abnormal data.

[0047] Anomaly detection based on machine learning: Clustering algorithm: Use clustering algorithms (such as K-means, DBSCAN, etc.) to cluster meteorological data and divide the data into multiple clusters. If some data points do not belong to any cluster (that is, they are far away from other data points), they may be anomalies. Isolation Forest Algorithm: Isolation Forest is a commonly used anomaly detection algorithm that divides the data by randomly selecting the features of the data points and randomly selecting the split points to identify anomalies that are different from other data points. Autoencoder: Autoencoder is a deep learning model that can be used for anomaly detection. By training the autoencoder to reconstruct the input data, the model cannot reconstruct the abnormal data well during reconstruction, and thus can identify outliers.

[0048] Missing data handling: Interpolation: For missing data, you can use interpolation methods (such as linear interpolation, spline interpolation, etc.) to fill in the missing values. This method is suitable for cases where there are not many missing values. Deleting missing values: If the amount of missing data is large and it is difficult to effectively infer reasonable values, you can consider deleting these missing data points.

[0049] You can collect and organize historical meteorological data sets, ensure that the data is arranged in time series, and mark all outliers and missing values. Identify outliers in the data based on the various methods mentioned above (statistical methods, machine learning algorithms, manual inspection, etc.). For example, use the standard deviation method to check the value of each meteorological element, use the box plot method to detect extreme values, and use time series analysis methods to check data that does not conform to the trend. For data points marked as abnormal, you can choose to delete these data directly or interpolate and fill in missing values. After deleting the abnormal data, update the historical meteorological data set to ensure the accuracy and completeness of the data set. The cleaned data can be saved in a new file for subsequent analysis and modeling.

[0050] S103: Constructing a cold air and strong wind recognition model based on Res-Unet, and training the cold air and strong wind recognition model based on Res-Unet according to the meteorological disaster positive samples and the meteorological disaster negative samples.

[0051] Res-Unet is a deep learning model that combines the Residual Network (ResNet) and the U-Net architecture, and is commonly used in image segmentation tasks. It adds residual connections (ResidualConnections) to the traditional U-Net architecture, which can effectively alleviate the gradient vanishing problem and enhance the model's ability to capture low-frequency information. Res-Unet is particularly suitable for tasks that require fine image segmentation, such as applications in remote sensing images, medical imaging, and other fields. In the problem of cold air and strong wind identification, the Res-Unet-based model can be used to segment areas related to cold air and strong winds (such as the impact range of strong winds) from meteorological images. The specific task is to train the model based on positive samples of meteorological disasters (such as images containing cold air and strong winds) and negative samples of meteorological disasters (such as images that do not contain cold air and strong winds) to achieve accurate identification of cold air and strong winds.

[0052] Res-Unet consists of three parts: Downsampling: This part uses convolution and pooling operations to gradually reduce the size of the image while extracting low-level features. Each downsampling step increases the number of features, which helps capture deeper feature information.

[0053] Upsampling: Through transposed convolution or upsampling operations, the image size is gradually restored to generate high-level features. This part not only enlarges the image but also performs feature fusion to restore the details lost during the downsampling process.

[0054] Skip connections: Connect features of the same resolution during downsampling and upsampling. In this way, the network can better capture multi-scale information, thereby maintaining image details while avoiding the loss of feature information.

[0055] Residual Connections: Residual connections are used in each layer of Res-Unet, summing the input information with the output of the convolution. This approach improves training efficiency, prevents gradient vanishing, and enhances the model's expressiveness.

[0056] Prepare a dataset: Positive meteorological disaster samples: These include meteorological images showing cold air and strong winds, with the areas where the cold winds occurred marked. Negative meteorological disaster samples: These include meteorological images not showing cold air and strong winds, used as negative samples. Datasets typically include meteorological images (such as satellite images and radar images), each with a corresponding label, typically a binary image, identifying the area where the strong winds occurred. Standardize the image data by scaling pixel values to the [0, 1] range or normalizing the mean and variance. Enhance the images using rotation, scaling, cropping, flipping, and other methods to increase the robustness of the model. Ensure that the label map (mask map) for each image has the same size and format as the original image. When building the Res-Unet model, the main focus is on building an architecture that incorporates downsampling, upsampling, and skip connections. Deep learning frameworks such as TensorFlow, Keras, and PyTorch can be used to implement the model. Downsampling: Features are extracted through convolution operations (Conv2D), and image size is reduced through pooling operations (MaxPooling). After each convolution, the number of feature maps gradually increases. Skip connections: By connecting the output features of the downsampled layer with the input features of the upsampled layer, more low-level feature information is retained. Upsampling: Transposed convolutions (Conv2DTranspose) are used to gradually restore the image size and fuse the downsampled feature maps to achieve fine image segmentation. Output layer: Convolutional layers (Conv2D) are used to generate an output image of the same size as the input image, which is the predicted segmentation map (a binary image of high wind areas and non-high wind areas). The Res-Unet model is trained using positive and negative meteorological disaster samples. Loss function: Common loss functions include cross-entropy loss (Cross-Entropy Loss) and Dice coefficient loss (Dice Loss), which measure the similarity between the prediction and the true label. Optimizer: Select an optimizer such as Adam for model training to improve convergence speed and accuracy. Training parameters: Set hyperparameters such as the number of training epochs and batch size to ensure effective model training and avoid overfitting. The model's classification performance on positive and negative samples is then evaluated. Calculate the degree of overlap between the segmented regions output by the model and the actual high wind areas. This is used to evaluate the performance of binary segmentation tasks, especially in cases of class imbalance. After training, the Res-Unet model can be used to predict and identify actual cold air and high wind events. Applying the trained model to real-time meteorological image data automatically detects areas of cold wind and high winds and issues warnings.

[0057] S104, obtaining real-time meteorological element data and real-time three-dimensional radar detection data, inputting the real-time meteorological element data and real-time three-dimensional radar detection data into the trained Res-Unet-based cold air and gale recognition model to obtain a gale weather forecast result.

[0058] Real-time meteorological element pattern data refers to datasets that record the changing patterns of meteorological elements and their interrelationships over a period of time. These include parameters such as temperature, air pressure, humidity, wind speed, wind direction, Convective Available Potential Energy (CAPE), and the K index. These data are typically provided by meteorological observation stations, meteorological satellites, and numerical meteorological forecasting systems. They describe atmospheric conditions and meteorological conditions at different altitudes in the atmosphere. CAPE (Convective Available Potential Energy) is a measure of air's convective capacity and is often used to assess the likelihood of severe convective weather. The K index is used to assess the convective potential of weather and is calculated primarily based on the vertical distribution of temperature, humidity, and air pressure. Some of this data can be presented graphically. Temperature distribution maps can display temperature data as heat maps, with temperature variations across different regions represented by color. Pressure contour maps display the air pressure distribution across different regions, with higher and lower pressure areas indicated by different colors. Humidity / wind speed distribution maps display the spatial distribution of humidity and wind speed. Wind speed is typically represented by wind field diagrams or arrows, while humidity is represented by a color gradient. CAPE / K Index Image: Displays the distribution of CAPE or K index in different regions through heat maps or contour maps, highlighting areas of severe convective weather.

[0059] Real-time 3D radar detection data can include three-dimensional meteorological information such as reflectivity and radial velocity acquired by weather radar systems. 3D radar can provide atmospheric data from multiple directions and layers, helping to analyze the spatial distribution and dynamics of phenomena such as storms, precipitation, and air currents. Specifically, this data includes reflectivity: the strength of the signal returned by radar-emitted electromagnetic waves after encountering meteorological particles (such as raindrops and hail), reflecting precipitation intensity and cloud moisture content. Radial velocity: the relative speed between radar waves and meteorological targets, reflecting wind speed and direction. Some of this data can be presented graphically. For example, radar reflectivity maps display radar reflectivity data using color coding, with areas of higher reflectivity (such as heavy rain) indicated by red and purple, and areas of lower reflectivity by green and blue. Radial velocity maps display the spatial distribution of the relative velocity between radar waves and meteorological targets, with positive values (storms approaching the radar) indicated by red, and negative values (storms moving away from the radar) indicated by blue.

[0060] The high wind weather forecast result can be the result of predicting cold air high wind events based on the input real-time meteorological element data and three-dimensional radar detection data through the trained Res-Unet model. Specifically, it can include the time period when high wind weather occurs: the model can predict the possible occurrence time of high wind weather and provide a predicted time window. The affected area of high wind weather: based on the input meteorological data, the Res-Unet model will identify which areas are affected by cold air high winds. This result can be presented in the form of a spatial distribution (such as a heat map, annotated areas, etc.) to show the impact range of high wind weather. Wind speed intensity: Through the model's prediction of wind speed, the wind speed intensity can be obtained to determine whether it meets the high wind standard (for example, the wind speed reaches a certain value). The path and changes of the storm or cold front: predict the movement path of cold winds and cold fronts, as well as the areas and duration that may be affected.

[0061] Real-time meteorological data can be collected through sensors such as weather stations, satellites, and radar. Real-time 3D radar data is collected through a meteorological radar data platform and radar data collection and processing system. Real-time meteorological data is standardized and denoised to ensure accuracy. Radar data is converted to a suitable format to ensure compatibility with meteorological data. Radar data can be spatially interpolated to address issues such as radar blind spots. Real-time meteorological data and 3D radar data are input into a trained Res-Unet model in temporal and spatial sequences. Res-Unet makes predictions based on features learned during historical training. Based on the input meteorological data, the Res-Unet model outputs predictions, including the time of occurrence of high winds, the affected area, and wind speed and intensity. The model can generate images or heat maps based on these predictions, highlighting the areas affected by high winds and providing the predicted range for strong winds. The prediction results can be visualized to generate real-time high wind forecast maps, helping meteorological and disaster management agencies to formulate appropriate response measures.

[0062] After completing the high wind weather forecast, you can issue an early warning through the following steps: Based on the high wind weather forecast results, you first need to set the early warning standards. The standards may include the following items: Wind speed threshold: Set the early warning to be triggered when the predicted wind speed exceeds a certain threshold (such as 50km / h or 60km / h). Wind speed duration: If the wind speed duration exceeds a certain period of time (such as more than 1 hour), the warning intensity can be increased. Affected area: If the predicted wind speed intensity reaches a high level in an important area (such as a densely populated area, transportation hub or building complex), an early warning should also be triggered. Wind speed change rate: If the wind speed increases sharply in a short period of time, such as an increase of more than 20km / h within 30 minutes, it may indicate the beginning of a storm and trigger an early warning.

[0063] Wind speed and disaster risk correlation: Predicted wind speed: Combined with the model's predicted wind speed intensity, special attention is paid to wind speeds in strong wind areas. If the wind speed intensity exceeds or approaches the predetermined risk wind speed threshold, an early warning state can be entered. Disaster risk assessment: Based on the model-generated heat maps or impact areas, disaster risk in different regions is assessed. For example, coastal areas, mountainous areas, or densely built-up areas require special attention if wind speed has a significant impact.

[0064] Generating Warning Levels: Based on different forecast results, high wind weather is divided into different warning levels. Generally, they are divided into: Blue Warning (Low Risk): Low wind speeds, short duration, and a small impact area; Yellow Warning (Medium Risk): High wind speeds, long duration, and a wider impact area; Orange Warning (High Risk): Extreme wind speeds, long duration, and impacting critical areas; Red Warning (Extreme Risk): Extreme wind speeds, long duration, and wide impact areas, potentially leading to serious disasters.

[0065] Warning triggering and dissemination: Automatic generation of warning information: Once the forecast results meet certain warning criteria, the system automatically generates the corresponding warning information. This information should include detailed information such as wind speed, warning level, affected area, and duration. Real-time notification of relevant departments: Warning information is transmitted in real time through the system to relevant departments such as meteorological departments, emergency management departments, local governments, and public security departments. Warning information can be released through multiple channels such as text messages, emails, app push notifications, and alarm systems. Public warning notification: Warning information is released to the public through channels such as media broadcasts, public broadcasting systems, and government portals to guide residents and businesses to take preventive measures.

[0066] In an embodiment of the present application, historical meteorological element pattern data and historical three-dimensional radar detection data are obtained, and a historical meteorological data set is created based on the historical meteorological element pattern data and the historical three-dimensional radar detection data; the gale unit characteristic information and the cold wind meteorological feature set of the historical meteorological data set are determined, and historical disaster-stricken area data are obtained. According to the gale unit characteristic information, the cold wind meteorological feature set and the historical disaster-stricken area data, meteorological disaster positive samples and meteorological disaster negative samples are constructed; a cold air gale recognition model based on Res-Unet is constructed, and the cold air gale recognition model based on Res-Unet is trained based on the meteorological disaster positive samples and the meteorological disaster negative samples; real-time meteorological element data and real-time three-dimensional radar detection data are obtained, and the real-time meteorological element data and real-time three-dimensional radar detection data are input into the trained cold air gale recognition model based on Res-Unet to obtain gale weather forecast results. Through the above-mentioned cold air gale recognition method based on multi-source data and deep learning, the real-time meteorological element data and the three-dimensional radar detection data are input into the trained Res-Unet model, which can significantly improve the recognition accuracy of the model in complex meteorological phenomena and can obtain gale weather forecast results in real time. This means that cold air and strong wind events can be identified and warned in advance before they occur, so that preventive measures can be taken in advance to reduce the losses caused by the disaster.

[0067] Based on the above technical solution, optionally, after obtaining the strong wind weather forecast result, the method further includes:

[0068] Obtaining the current system time, sending a traffic flow acquisition request to the control center according to the current system time, and receiving normal traffic flow data sent by the control center according to the traffic flow acquisition request;

[0069] Determining a predicted wind speed based on a high wind weather forecast result, and determining a wind speed-traffic response coefficient based on the predicted wind speed, current traffic flow data, and a preset response coefficient standard;

[0070] Calculate the predicted traffic flow based on normal traffic flow data, predicted wind speed, wind speed-traffic response coefficient and preset traffic flow calculation formula;

[0071] Target traffic control measures are determined based on the predicted traffic flow and preset traffic control measure standards, and the target traffic control measures are sent to the control center.

[0072] In this scenario, the current system time is the date and time of a computer or server at a given moment. It is used to determine the time window for operations and to help synchronize operations and schedule them. It is typically expressed in the format of "year-month-day hour:minute:second".

[0073] The traffic flow acquisition request is a request sent to the control center to obtain traffic flow data at the current time point in the historical records without special circumstances.

[0074] Normal traffic flow data refers to traffic flow data at the current time under normal circumstances, without the influence of special meteorological conditions (such as strong winds, rain or snow). These data reflect the normal level of traffic flow and serve as a benchmark for prediction and control measures.

[0075] The pre-defined response factor standard is a mathematical criterion or formula used to determine the impact of wind speed on traffic flow. Based on historical data and the relationship between wind speed and traffic flow, it determines the degree of traffic flow response at different wind speeds (for example, the degree to which greater wind speeds may reduce traffic flow). This standard serves as a reference for predicting and adjusting traffic flow.

[0076] The wind speed-traffic response factor can be a calculated value used to represent the impact of high winds on traffic flow. It is calculated based on predicted wind speeds, historical traffic flow data, and preset response factor standards. When wind speeds are higher, traffic flow may change more significantly, and the wind speed-traffic response factor is a quantitative expression of this relationship. The wind speed-traffic response factor can range from 0.0001 to 0.01. For example, for light wind speeds (e.g., less than 30 km / h), the wind speed-traffic response factor may be 0.0001 to 0.001; for moderate wind speeds (e.g., 30 km / h to 50 km / h), the wind speed-traffic response factor may be 0.001 to 0.005; and for strong wind speeds (e.g., greater than 50 km / h), the wind speed-traffic response factor may be 0.005 to 0.01.

[0077] Traffic flow forecasts can be calculated using models based on current traffic flow data, predicted wind speed, the wind speed-traffic response coefficient, and other possible factors (such as holidays and accidents). Traffic flow forecasts help traffic management departments prepare for emergency responses and resource allocation in advance.

[0078] Pre-set traffic control measures can be pre-set standards for traffic management measures to be taken under different traffic flow levels and wind speed conditions. These standards may include adjusting traffic lights, limiting speeds, closing certain roads, and initiating traffic control measures to address the impact of high winds on traffic flow.

[0079] Targeted traffic control measures can be specific traffic management measures based on predicted traffic flow and wind speed-traffic response factors, combined with pre-set control standards. These measures can be tailored to different wind speeds or traffic flow levels, such as adjusting lane settings, issuing speed limits, or restricting traffic flow in certain areas.

[0080] The current system time can be obtained using the computer system or server's built-in clock. Based on the current system time, the system sends a request to the traffic control center for normal traffic flow data at that time. Upon receiving the request, the control center returns the traffic flow data for that time. For example, the data returned by the control center indicates that the traffic flow for that period is 2,000 vehicles per hour. Next, the system determines the wind speed based on the high wind forecast. For example, the wind speed is predicted to reach 50 km / h within the next hour. The predicted wind speed and current traffic flow data are then matched against a preset response coefficient. For example, if the current traffic flow is 2,000 vehicles per hour and the predicted wind speed is between 30 and 50 km / h, the response coefficient is 0.005. The normal traffic flow data, predicted wind speed, and wind speed-traffic response coefficient are then substituted into the preset traffic flow calculation formula to calculate the predicted traffic flow. Based on the predicted traffic flow, the system then determines whether traffic control measures are necessary, referring to the preset traffic control measures criteria. For example, the preset traffic control measures are as follows: When traffic flow is below 1,200 vehicles per hour, normal traffic flow is allowed, with no control required. When traffic flow is between 1,200 and 1,500 vehicles per hour, light control measures, such as lane restrictions or speed adjustments, are implemented. When traffic flow exceeds 1,500 vehicles per hour, area-wide control measures are activated. Once the target traffic control measures are determined, they are transmitted to the control center via wireless communication technology.

[0081] In this solution, by combining weather forecasts and current traffic flow data, the impact of strong winds on traffic flow can be accurately predicted, avoiding traffic jams or accidents caused by emergencies.

[0082] Based on the above technical solution, an optional, preset traffic flow calculation formula is:

[0083] F new =F base ×(1-K×V);

[0084] Among them, F new To predict traffic flow; F base is the normal traffic flow data; K is the wind speed-traffic response coefficient; V is the predicted wind speed.

[0085] In this scenario, for example, the normal traffic flow F base The number of vehicles per hour is 2000, the predicted wind speed V is 50 km / h, and the wind speed-traffic response coefficient K is 0.005, then F new =2000×(1-0.005×50)=1500. The corresponding target traffic control measures can be to implement mild control, such as restricting lanes or adjusting vehicle speeds.

[0086] Based on the above technical solution, optionally, after sending the target traffic control measures to the control center, the method further includes:

[0087] If the preset update interval is reached, updating the strong wind weather forecast result, and updating the predicted wind speed according to the strong wind weather forecast result;

[0088] updating the building wind pressure according to the updated predicted wind speed; if the updated building wind pressure still exceeds a preset building wind pressure threshold, regenerating an alarm message according to the updated building wind pressure; and updating a disaster prevention and emergency response plan according to the updated building wind pressure, and transmitting the regenerated alarm message and the updated disaster prevention and emergency response plan to a control center;

[0089] Accordingly, after updating the predicted wind speed according to the high wind weather forecast result, the method further includes:

[0090] Re-acquire the current system time, re-acquire normal traffic flow data according to the re-acquired current system time, and recalculate the predicted traffic flow according to the re-acquired normal traffic flow data and the updated predicted wind speed;

[0091] The target traffic control measures are re-determined according to the recalculated predicted traffic flow, and the re-determined target traffic control measures are sent to the control center.

[0092] In this scenario, the preset update interval refers to a system-defined time interval that controls the frequency of data updates. This interval is typically measured in minutes, hours, or days, depending on business needs and the timeliness of the forecast. For example, weather forecasts and related data might be updated every 30 minutes to ensure that traffic control measures, wind pressure on buildings, and other data reflect the latest meteorological conditions.

[0093] When the preset update interval is reached, the system automatically updates the high wind forecast and adjusts the predicted wind speed based on the updated weather data. The system then uses the updated wind speed to recalculate the wind pressure on the building and checks whether it exceeds the set wind pressure threshold. If the threshold is exceeded, the system generates a new alarm message, updates the disaster prevention and emergency response plan, and transmits this information to the control center. At the same time, the system re-acquires the current time and the latest normal traffic flow data, and recalculates the predicted traffic flow based on the updated wind speed. Based on the newly calculated traffic flow, the system adjusts traffic control measures and sends the new control strategy to the control center.

[0094] This solution can promptly adjust building wind pressure assessments based on the latest high wind forecasts, ensuring a timely response to wind pressure changes and preventing damage to buildings due to excessive wind pressure. As wind speeds change, the system dynamically adjusts traffic flow forecasts and, in conjunction with traffic flow calculation formulas, provides real-time predictions of future traffic flows, ensuring the effectiveness of traffic control measures and mitigating the negative impact of high winds on traffic.

[0095] Figure 2 This is a flow chart of a cold air and strong wind identification method based on multi-source data and deep learning provided by an embodiment of the present disclosure. Figure 2 As shown, the method includes:

[0096] S201 , acquiring historical meteorological element pattern data and historical three-dimensional radar detection data, and creating a historical meteorological dataset based on the historical meteorological element pattern data and historical three-dimensional radar detection data.

[0097] S202, determining the strong wind single unit characteristic information and the cold wind meteorological characteristic set of the historical meteorological data set, obtaining historical disaster-stricken area data, and constructing meteorological disaster positive samples and meteorological disaster negative samples based on the strong wind single unit characteristic information, the cold wind meteorological characteristic set and the historical disaster-stricken area data.

[0098] S203: Constructing a cold air and strong wind recognition model based on Res-Unet, and training the cold air and strong wind recognition model based on Res-Unet according to the meteorological disaster positive samples and the meteorological disaster negative samples.

[0099] S204, obtaining real-time meteorological element data and real-time three-dimensional radar detection data, inputting the real-time meteorological element data and real-time three-dimensional radar detection data into the trained Res-Unet-based cold air and gale recognition model to obtain a gale weather forecast result.

[0100] S205 , determining a predicted wind speed according to the strong wind weather prediction result; if the predicted wind speed is greater than a preset wind speed threshold, obtaining building shape information; and determining a building shape coefficient according to the building shape information and a preset shape coefficient standard.

[0101] Predicted wind speeds are predicted using a Res-Unet-based cold wind identification model trained on historical meteorological data and real-time radar detection data. This model uses real-time meteorological data (such as wind speed, wind direction, and temperature) and relevant characteristics of windy weather to predict wind speed changes in a specific area over a period of time.

[0102] The preset wind speed threshold can be a pre-set wind speed value used to determine whether the wind speed exceeds an acceptable safety range. Typically, this threshold is determined based on regional building codes, safety standards, or meteorological disaster protection requirements. For example, if the predicted wind speed exceeds the threshold, a specific safety response or warning may be triggered.

[0103] Building shape information refers to geometric features of a building, including its shape, dimensions (such as height, width, and length), and structural type. This information is crucial for a building's wind resistance. For example, the wind resistance and pressure distribution of a high-rise building differ from those of a low-rise building. Different building shapes also affect the distribution of wind speed and the impact of wind pressure.

[0104] A pre-set shape factor standard can be a standard factor based on different building shapes, used to estimate the impact of wind speed on a building. This factor is typically determined based on the building's shape and structure, as well as the wind conditions in the area. Shape factor standards are often specified by building wind design codes, meteorological standards, or building wind load calculation standards to ensure the safety of buildings under wind loads.

[0105] The building shape factor is a numerical value that describes the effect of a building's shape on wind pressure. It is calculated based on the building's shape, dimensions, materials, and relative wind direction, and is typically used to assess wind loads on a building at specific wind speeds. A larger factor indicates a building's ability to withstand stronger wind forces and a more significant wind impact. It is typically determined through experiments, simulations, or code calculations, reflecting the wind resistance of a building's design.

[0106] The predicted wind speed can be extracted from the high wind forecast results. If the predicted wind speed exceeds the preset wind speed threshold, the next step is to obtain building shape information and calculate the shape coefficient. If the predicted wind speed is less than or equal to the wind speed threshold, no further action may be required, or routine monitoring may be performed. When the predicted wind speed exceeds the preset wind speed threshold, detailed building shape data is obtained. This data is typically collected through architectural design drawings, building information models (BIM), geographic information systems (GIS), or sensors. Building geometric parameters (e.g., height, width, length, surface material, window distribution, etc.) can be obtained. If necessary, laser scanning (LIDAR) or drones can be used to obtain accurate 3D building data. Based on the building shape information (e.g., height, surface characteristics, etc.), a preset shape coefficient standard is searched. For example, for a rectangular building, the shape coefficient is typically related to the building's aspect ratio, facade characteristics (whether there are openings such as windows and doors), and wind direction. For low-rise rectangular buildings, the shape coefficient is typically fixed, such as 1.2 to 1.5. Circular buildings: For cylindrical buildings, the calculation of the shape coefficient may depend on the height and diameter of the building, or existing standards may be used directly. The shape coefficient of circular buildings is generally small (for example, 0.6-0.8), and the wind speed is closely related to the diameter and height of the building. Buildings with pointed or sloping roofs: For buildings with pointed roofs, the shape coefficient is usually closely related to the slope of the roof, the angle of the pointed roof, and the height and width of the roof. Some standard documents or software provide specific formulas or charts to help calculate the shape coefficient. Buildings with complex shapes: For buildings with irregular shapes (such as polygonal facades, large windows or cantilevered parts), the calculation of the shape coefficient is more complicated and may require wind tunnel experiments, numerical simulations or reference to correction factors in specific literature.

[0107] S206: Acquire building type information, and determine the building dynamic pressure coefficient according to the building type information and a preset dynamic pressure coefficient standard.

[0108] Building type information refers to attribute data describing a building's basic characteristics and functions. This information typically includes the building's purpose, structural type, and facade features. This information directly affects the magnitude and distribution of wind loads on the building. Therefore, when calculating wind loads, it's important to select an appropriate dynamic pressure coefficient based on the building type.

[0109] The dynamic pressure coefficient is a coefficient used to describe the pressure changes caused by wind on a building's surface. Different types of buildings exhibit varying wind pressure distributions under wind loads, so the appropriate dynamic pressure coefficient standard must be selected based on the building's shape, structure, and purpose.

[0110] Preset dynamic pressure coefficient standards can be given based on factors such as building type, wind speed, and building surface characteristics. Dynamic pressure coefficient standards are generally provided by building wind load specifications, design standards, or wind engineering studies. Specific standards include: Building shape and size: Standard dynamic pressure coefficients for rectangular, circular, and pointed-roof buildings. Building height and structure: High-rise buildings have different dynamic pressure coefficients than low-rise buildings, and taller buildings generally experience greater wind loads. Roof shape: The dynamic pressure coefficients of flat roofs, pitched roofs, and pointed roofs will vary, with roofs with larger slopes generally having higher dynamic pressure coefficients.

[0111] The building dynamic pressure coefficient is a dimensionless coefficient that describes the dynamic pressure effect of wind on a building's surface. When wind passes over a building's surface, it exerts varying pressures on different parts of the building. The magnitude of these pressures is typically related to factors such as wind speed, building shape, and wind direction.

[0112] The dynamic pressure coefficient of a building can be determined by looking up the standard dynamic pressure coefficient table by building type. Specifically, the search method is as follows: Building type classification: Determine the category of the building based on the building's purpose, structure, roof type and other information. For example, a rectangular high-rise building, a building with a pointed roof, a circular tower, etc. Consult the standard: Refer to the dynamic pressure coefficients of different types of buildings listed in the specification. For example, a flat-roofed building may have a fixed dynamic pressure coefficient (such as 0.8-1.2), while a building with a pointed roof may have a higher dynamic pressure coefficient (such as 1.5-2.0). Determine the coefficient value: Select the corresponding dynamic pressure coefficient based on the building category.

[0113] S207, obtaining building roughness, building surface material, and building additional structure, and inputting the building roughness, building surface material, building additional structure, and building shape information into a preset drag coefficient calculation model to obtain the building drag coefficient.

[0114] Building roughness can refer to the degree of influence of the building surface or surrounding environment on wind flow. Roughness is mainly related to the undulations of the building surface, the distribution and shape of objects around the building (such as trees, roads, other buildings, etc.). Building roughness determines the friction of wind flow when passing through the surface of the building, which in turn affects the change in wind speed and the direction of wind flow. High roughness: If there are many protrusions on the surface of the building or there are many obstacles around it (such as trees, fences, etc.), it will cause strong wind flow disturbance and reduce wind speed. Low roughness: If the building surface is smooth and the surrounding area is open, the wind flow is almost undisturbed and the wind speed changes little.

[0115] Building surface material refers to the type of material covering a building's exterior facade, such as concrete, steel, glass, masonry, and so on. Different materials have varying wind resistance properties and the pressure exerted by wind on their surfaces. Smooth materials, such as glass and metal panels, generally have lower wind resistance, resulting in minimal wind disturbance. Rough materials, such as stone, wood, and concrete, create greater resistance to wind flow, reducing wind speed and increasing wind loads.

[0116] Building additional structures can refer to additional devices installed on the surface or top of a building, such as billboards, air-conditioning units, balconies, awnings, etc. Additional structures may change the aerodynamic characteristics of the building and cause different pressure distribution when the wind flows around these additional structures.

[0117] The preset drag coefficient calculation model may be a mathematical model that estimates the drag coefficient based on the characteristics of the building, taking into account factors such as the building's roughness, surface material, and additional structures.

[0118] The drag coefficient of a building is a measure of the building's resistance to wind and is commonly used in wind pressure calculations. The drag coefficient reflects the resistance created by wind flowing over the building's surface, which in turn affects wind pressure calculations.

[0119] Satellite or aerial imagery can be used to analyze the complexity of the surrounding environment and infer the roughness level of the area where a building is located. Building surface material information can typically be obtained from architectural design drawings or remote sensing imagery, or confirmed through on-site surveys and the building's construction records. Design information for additional structures, such as billboards, balconies, and external ducting, can be obtained from the building's design records. Once the building's roughness, surface material, additional structures, and shape information are obtained, they are then input into a pre-defined drag coefficient calculation model. Based on this building's characteristics, the pre-defined drag coefficient calculation model estimates the building's drag coefficient using numerical methods or experimental formulas. These models are typically based on wind tunnel test results, CFD (computational fluid dynamics) simulations, or empirical formulas. The model considers the following factors: Roughness: Different roughness levels have different effects on wind resistance. Generally, urban or industrial areas have higher roughness and higher drag coefficients. Surface material: Different surface materials alter the viscosity and adhesion properties of the wind flow, resulting in different drag effects. Additional structures: Additional structures can alter the complexity of a building's exterior, affecting wind separation and vortex formation, thereby increasing or decreasing wind resistance. Shape: A building's shape influences airflow patterns, particularly tall buildings and those with unusual shapes, which typically have higher drag coefficients. The output of the pre-set drag coefficient calculation model is the building's drag coefficient, which represents the amount of resistance generated by wind on the building's surface.

[0120] The training process of the preset drag coefficient calculation model is as follows:

[0121] A dataset containing building characteristic information can be prepared. This information typically comes from historical building data, wind tunnel test data, CFD simulation results, and so on. The dataset should include the following: Building characteristic data (input features): This includes information such as the building's roughness, surface material, additional structures, and shape. This data can be numerical or categorical variables. For example, building roughness can have different levels (such as urban areas and rural areas), and the surface material can be steel, wood, etc. Drag coefficient data (labels): This data typically comes from wind tunnel tests or CFD simulations and represents the drag coefficient of different buildings at different wind speeds. The drag coefficient is the label value, usually a numerical value. The drag coefficient, as a label, is the target of model training. The label corresponding to each data point is the drag coefficient of the building, and these labels are obtained from experimental or simulation data. The drag coefficient typically varies with characteristics such as the building's shape and material. To train a drag coefficient calculation model, the following types of models are typically selected: Linear regression: This is suitable for simple cases where there is a linear relationship between the features and the target variable. For complex building drag coefficient predictions, more complex models may be required. Decision trees / random forests: These are suitable for nonlinear relationships between features and labels and can handle the complex effects of various building characteristics on the drag coefficient. Neural networks: Deep learning models are suitable for handling complex feature relationships and can capture highly nonlinear relationships between features. Common neural network models include multi-layer perceptrons (MLPs). Building characteristic data (such as roughness and surface material) are then standardized or normalized to ensure consistent scales across features. If the dataset contains missing values, interpolation or deletion is necessary. Feature selection methods (such as principal component analysis (PCA)) can be used to reduce feature dimensionality, retaining features that have a significant impact on drag coefficient prediction. The dataset is divided into training and test sets, typically with a 70% / 30% or 80% / 20% split. The model is trained using the training set data, and the model parameters are adjusted using optimization algorithms (such as gradient descent) to ensure that the predicted drag coefficient is as close as possible to the actual label value. During training, the validation set is used to evaluate model performance to ensure that the model does not overfit. Hyperparameters (such as the learning rate and tree depth) can be adjusted to optimize the model. After training, the model is evaluated using the test set. Common evaluation indicators include: Mean Square Error (MSE): Calculates the sum of squared errors between the predicted value and the actual value, which is used to measure the degree of fit of the model. 2 ): Used to evaluate the model's ability to explain changes in the drag coefficient. A trained and qualified drag coefficient calculation model can be applied to real-world scenarios, inputting building characteristic data and outputting the corresponding drag coefficient.

[0122] S208, obtaining the current air density and the building's windward area, and calculating the building's wind pressure based on the current air density, the building's windward area, the building's shape coefficient, the building's dynamic pressure coefficient, the building's wind resistance coefficient, the predicted wind speed, and a preset wind pressure calculation formula.

[0123] Air density refers to the mass of air per unit volume, usually expressed in kilograms per cubic meter (kg / m 3 Air density is affected by meteorological factors such as temperature, air pressure, and humidity, and varies particularly at different altitudes and in different weather conditions. Air density is crucial for calculating wind pressure because it directly affects the force of wind on a building.

[0124] The windward area of a building refers to the surface area of the building facing the wind, and the unit is usually square meters (m 2 The windward area of a building affects the magnitude of wind pressure, because a larger windward area means that the wind will exert a greater force on the building.

[0125] Building wind pressure can be the pressure exerted by wind on the surface of the building, usually measured in Pascals.

[0126] The current air density can be obtained in the following ways: Meteorological database: Many meteorological services provide historical and real-time data, which usually include direct measurements of air density. Meteorological API: For example, APIs such as OpenWeatherMap and WeatherStack provide weather information, which may also include parameters such as air density, air pressure, and temperature. By calling these APIs, you can directly obtain the required meteorological data and then calculate the air density. The windward area of the building can be estimated by the projected area of the building surface facing the wind direction. Then substitute the current air density, the windward area of the building, the building shape coefficient, the building dynamic pressure coefficient, the building drag coefficient, and the predicted wind speed into the preset wind pressure calculation formula to calculate the building wind pressure.

[0127] On the basis of the above technical solution, an optional, preset wind pressure calculation formula is:

[0128] P wind =0.5·ρ·C d ·V 2 ·A·C s ·C f ;

[0129] Among them, P wind is the wind pressure on the building; ρ is the current air density; C d is the building dynamic pressure coefficient; V is the predicted wind speed; A is the building's windward area; C s is the building shape coefficient; C fis the building's drag coefficient.

[0130] S209: If the building wind pressure exceeds a preset building wind pressure threshold, generate alarm information according to the building wind pressure, and transmit the alarm information to a control center.

[0131] The preset building wind pressure threshold can be a predefined wind pressure value that represents the maximum wind pressure that the building can withstand. When the wind pressure on the building exceeds this threshold, it indicates that the building may be at risk of wind disaster and an alarm needs to be issued.

[0132] An alarm message can be a notification or alarm generated by the system when the wind pressure on a building exceeds a preset wind pressure threshold. It contains the necessary information to alert operators or the control center for an emergency response.

[0133] A control center is a centralized monitoring and command management system, typically used to monitor a building's operational status, environmental data, and emergency response. The control center receives data from the building and other monitoring systems (such as wind pressure, wind speed, and temperature) and issues alerts or takes necessary response measures based on preset thresholds or rules.

[0134] Once wind pressure exceeds a threshold, the system generates an alarm. This information may include the building ID, current wind pressure value, threshold comparison, alarm level, and location and time. The alarm is transmitted to the control center via a network or communication channel. This can be done via various channels (such as text messages, emails, and system notifications). Upon receiving the alarm, the control center assesses the risk based on the alarm level and wind pressure data and takes appropriate emergency response measures. Control center staff may inspect the building's condition, notify maintenance personnel, activate the wind disaster warning system, or take other actions.

[0135] This embodiment enables dynamic wind pressure calculation and real-time monitoring of buildings exposed to high winds. This allows foreseeing and mitigating potential risks before they occur, providing timely warnings and emergency responses. This not only ensures building safety but also optimizes management and decision-making processes, improving wind resistance and emergency response efficiency.

[0136] Based on the above technical solution, optionally, after transmitting the alarm information to the control center, the method further includes:

[0137] Obtain building usage status data, building structure data and building surrounding environment data, input the building wind pressure, building usage status data, building structure data, building surrounding environment data, building type information, building shape information, building surface material and building wind-exposed area into a preset disaster prevention and emergency plan formulation model to obtain a disaster prevention and emergency plan, and transmit the disaster prevention and emergency plan to a control center.

[0138] In this solution, building usage data refers to various information related to the current use of a building. For example, whether the building is occupied or occupied, whether important equipment is in operation, or whether renovations or maintenance are being performed. This information can reflect whether the building is experiencing special loads or structural conditions. Specifically, it may include building occupancy, internal equipment operating status, occupancy distribution, and the building's function (e.g., commercial, residential, or industrial).

[0139] Building structural data can include structural characteristics and design parameters, describing a building's load-bearing capacity, wind resistance, and earthquake resistance. This data helps assess a building's resilience to extreme weather conditions (such as strong winds, heavy snow, and earthquakes). Specifically, it can include the number of floors, structural type (such as reinforced concrete or steel), wall materials, frame design, support structure distribution, and wind resistance.

[0140] Building environmental data refers to the environmental conditions surrounding a building, including factors such as its geographic location, neighboring buildings, topography, vegetation cover, road infrastructure, and urban planning, which can affect wind distribution and pressure at a building under extreme weather conditions. Specifically, this data includes wind direction and speed, topographical features (such as the presence of mountains, canyons, and open areas), the height and layout of surrounding buildings, natural obstacles (such as trees and rivers), and man-made environments (such as roads and bridges).

[0141] The pre-defined disaster prevention and emergency response planning model can be a mathematical model or intelligent system that develops specific emergency response strategies based on various building data (such as wind pressure, structure, surrounding environment, and usage status). This model will comprehensively consider the building's safety, external environment, and the type and intensity of the disaster to generate a detailed emergency response plan.

[0142] A disaster emergency plan is a set of emergency response measures for extreme disasters (such as high winds, earthquakes, and fires) developed based on building safety assessments and environmental data analysis. The plan aims to minimize the impact of disasters on buildings and personnel, ensure personal safety, and implement timely and effective measures to reduce property losses.

[0143] Building occupancy data can be collected through building management systems, smart sensors, and IoT devices. For example, smart sensors can monitor building occupancy in real time (whether it's occupied, whether equipment is running, whether renovations have been made, etc.). Building structural characteristics can be obtained through architectural design drawings and structural health monitoring systems (such as structural sensors and stress sensors). Environmental information about the surrounding environment can be obtained through environmental sensors, weather station data, and geographic information systems (GIS). Building wind pressure, building occupancy data, building structure data, building surrounding environment data, building type information, building shape information, building surface material, and building windward area are then input into a pre-defined disaster prevention and emergency response plan development model. The model may first normalize the input data to ensure consistent dimensionality across all features and prevent certain features from overly influencing the results. The model analyzes the importance of each feature and selects the data features most influential in generating the disaster prevention and emergency response plan (e.g., wind pressure, building structure, occupancy status, etc.). The model conducts a risk assessment based on factors such as building type, shape, material, and windward area, combined with factors such as wind pressure. This process may employ a multi-objective optimization algorithm to assess the building's safety under different disaster levels and calculate the types of risks it may face (e.g., structural damage, collapse due to excessive wind pressure, etc.). Based on the input data and risk assessment, the model generates specific emergency response strategies. These strategies may include: whether immediate building reinforcement is necessary; whether evacuation is necessary; which emergency facilities (e.g., generators, fire protection systems, etc.) should be activated; and recommendations for various resource deployments (e.g., the deployment of rescue personnel and equipment). The model may combine machine learning and a decision support system (DSS) to optimize the emergency response strategy, continuously optimizing and adjusting the response plan based on historical data. After completing the analysis, the model generates a disaster prevention and emergency response plan and transmits it to the control center. The specific output and transmission process is as follows: The disaster prevention and emergency response plan includes various emergency response recommendations and instructions, such as measures to be initiated when wind pressure exceeds a certain threshold; customized emergency responses for different building types and structural conditions; and necessary resource deployment and evacuation plans. The generated emergency plan is usually formatted as an electronic document (such as PDF or XML format) and contains a detailed action plan and decision-making basis. After the plan is generated, the system will send it to the control center through a secure communication channel. The control center is usually responsible for handling emergencies, coordinating resources and making decisions. The transmission method can be: Real-time transmission: Real-time transmission to the control center using the Internet, dedicated lines or wireless communication technology. Automatic triggering: When the wind pressure reaches a certain threshold, the system will automatically trigger the plan and send it to the control center.

[0144] The training process of the preset disaster prevention and emergency plan formulation model is as follows:

[0145] Create a large dataset containing historical data. The dataset typically includes the following: Building wind pressure: Wind pressure on buildings during historical wind disasters. Building occupancy data: Historical occupancy of buildings (whether they were occupied, whether equipment was operating, etc.). Building structural data: Data on the building's design, wind resistance, and structural materials. Building surrounding environment data: This includes topography, distribution of surrounding buildings, and weather conditions. Building type: Building type (residential, commercial, industrial, etc.). Building shape: External features of the building, such as height, width, window type, and roof form. Building surface material: Exterior wall and window materials. Building windward area: The windward area of the building, used to calculate wind pressure. The dataset label is "Emergency Plan." Before training the model, you can choose to fill in missing values (e.g., using the mean or median) or delete rows with significant missing values. By setting reasonable ranges (e.g., wind speeds outside the normal range), you can remove unreasonable data. Ensure that numerical features (such as wind pressure and building height) are on the same scale to avoid certain features dominating during training. Non-numeric data (such as building type and material) needs to be converted to numerical form through one-hot encoding or label encoding so that the model can understand it. Divide the dataset into training, validation, and test sets. Typically, 80% is used for training, 10% for validation, and 10% for testing. Depending on the characteristics of the problem, you can choose from a variety of machine learning algorithms to train the model. Common algorithms include: Decision Trees: These can handle both categorical and numerical data and are suitable for classification problems. They generate a series of decision rules to generate emergency response plans based on input features. Random Forests: Ensembles multiple decision trees to improve prediction accuracy and reduce overfitting. Gradient Boosted Decision Trees (GBDTs): These combine multiple weak classifiers (trees) into a strong classifier, capable of handling complex relationships between features. Support Vector Machines (SVMs): These are suitable for classification problems and can find optimal decision boundaries. Neural Networks: Deep learning methods, such as Multilayer Perceptrons (MLPs), are suitable for large datasets and complex nonlinear relationships. XGBoost or LightGBM: These are currently very popular and efficient gradient boosting algorithms suitable for large datasets. Input: The model's input features include all information about each building in the dataset, including wind pressure, building occupancy data, building structural data, building surrounding environment data, building type information, building shape information, building surface material, and the building's wind-exposed area. Output (label): Emergency plan label. This label is the model's target output and indicates the emergency measures that should be taken by the building under specific wind pressure conditions. During training, the model is trained using the input data and its parameters are optimized to accurately predict emergency plans based on the input features.Since the dataset contains labels (emergency plans), the model will be trained through supervised learning, adjusting the weights so that the predicted value is closest to the actual label. Then, an appropriate loss function is selected according to the task. For example, the cross-entropy loss function can be used for classification problems. Commonly used optimization algorithms include gradient descent, Adam, etc., the purpose of which is to minimize the prediction error by optimizing model parameters. Cross-validation is then performed on the validation set to adjust model hyperparameters (such as learning rate, tree depth, etc.). If the model performs poorly on the validation set, different algorithms or optimized features can be tried. The test set is then used to evaluate the generalization ability of the model to ensure that the model not only performs well on the training data, but can also make effective emergency plan predictions on new data. After the model is trained, it is deployed in a real environment. When new building data (such as wind pressure, building shape, etc.) is input, the model can automatically predict the emergency plan and send it to the control center.

[0146] In this solution, real-time assessment of building wind pressure, structural conditions, and surrounding environmental data enables the rapid and accurate generation of emergency response plans. These plans can help control centers take timely action, reduce manual intervention, and improve emergency response efficiency.

[0147] Figure 3 A schematic block diagram of a cold air and strong wind identification system based on multi-source data and deep learning provided in an embodiment of the present disclosure. The system includes:

[0148] A data set creation module 301 is configured to obtain historical meteorological element pattern data and historical three-dimensional radar detection data, and create a historical meteorological data set based on the historical meteorological element pattern data and the historical three-dimensional radar detection data;

[0149] The sample creation module 302 is configured to determine the high wind single element characteristic information and the cold wind meteorological characteristic set of the historical meteorological data set, obtain historical disaster-affected area data, and construct meteorological disaster positive samples and meteorological disaster negative samples based on the high wind single element characteristic information, the cold wind meteorological characteristic set, and the historical disaster-affected area data;

[0150] A model training module 303 is used to build a cold air and strong wind recognition model based on Res-Unet, and train the cold air and strong wind recognition model based on Res-Unet according to the meteorological disaster positive samples and the meteorological disaster negative samples;

[0151] The prediction module 304 is used to obtain real-time meteorological element data and real-time three-dimensional radar detection data, and input the real-time meteorological element data and real-time three-dimensional radar detection data into the trained Res-Unet-based cold air and strong wind recognition model to obtain strong wind weather prediction results.

[0152] Figure 4A schematic block diagram of an electronic device 400 that can be used to implement an embodiment of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0153] The electronic device 400 includes a computing unit 401, which can perform various appropriate actions and processes according to a computer program stored in a ROM 402 or a computer program loaded from a storage unit 408 into a RAM 404. The RAM 404 can also store various programs and data required for the operation of the electronic device 400. The computing unit 401, the ROM 402, and the RAM 404 are connected to each other via a bus 404. An I / O interface 405 is also connected to the bus 404.

[0154] Multiple components in the electronic device 400 are connected to the I / O interface 405, including an input unit 406, such as a keyboard, a mouse, etc.; an output unit 407, such as various types of displays, speakers, etc.; a storage unit 408, such as a magnetic disk, an optical disk, etc.; and a communication unit 409, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 409 allows the electronic device 400 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0155] The computing unit 401 can be various general-purpose and / or specialized processing components with processing and computing capabilities. Some examples of the computing unit 401 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units that run machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The computing unit 401 performs the various methods and processes described above, such as the cold air and wind identification method based on multi-source data and deep learning. For example, in some embodiments, the cold air and wind identification method based on multi-source data and deep learning can be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as the storage unit 408. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 400 via the ROM 402 and / or the communication unit 409. When the computer program is loaded into the RAM 404 and executed by the computing unit 401, one or more steps of the cold air and wind identification method based on multi-source data and deep learning described above can be performed. Alternatively, in other embodiments, the computing unit 401 may be configured to execute the cold air and strong wind identification method based on multi-source data and deep learning in any other appropriate manner (for example, by means of firmware).

[0156] Various embodiments of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system that includes at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0157] The program code for implementing the method of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device so that when the program code is executed by the processor or controller, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0158] In the context of the present disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium can include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0159] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0160] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.

[0161] A computer system may include a client and a server. The client and server are generally remote from each other and typically interact through a communication network. The client-server relationship arises through computer programs running on the respective computers and having a client-server relationship with each other. The server may be a cloud server, a server in a distributed system, or a server integrated with a blockchain.

[0162] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this disclosure can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved. This is not limited herein.

Claims

1. A cold air and strong wind identification method based on multi-source data and deep learning is characterized by: The method comprises: Acquire historical meteorological element pattern data and historical three-dimensional radar detection data, and create a historical meteorological dataset based on the historical meteorological element pattern data and historical three-dimensional radar detection data; Determine the high wind single unit characteristic information and the cold wind meteorological characteristic set of the historical meteorological data set, obtain historical disaster-affected area data, and construct meteorological disaster positive samples and meteorological disaster negative samples based on the high wind single unit characteristic information, the cold wind meteorological characteristic set, and the historical disaster-affected area data; Constructing a cold air and strong wind recognition model based on Res-Unet, and training the cold air and strong wind recognition model based on Res-Unet according to the meteorological disaster positive samples and the meteorological disaster negative samples; Real-time meteorological element data and real-time three-dimensional radar detection data are obtained, and the real-time meteorological element data and real-time three-dimensional radar detection data are input into a trained cold air and gale recognition model based on Res-Unet to obtain a gale weather forecast result.

2. The method according to claim 1, characterized in that in, Before determining the strong wind single body characteristic information and the cold wind meteorological characteristic set of the historical meteorological data set, the method further includes: Abnormal data in the historical meteorological data set is identified, the abnormal data is deleted, and the historical meteorological data set is updated.

3. The method according to claim 1, characterized in that in, After obtaining the high wind weather forecast result, the method further includes: determining a predicted wind speed according to the high wind weather forecast result, and if the predicted wind speed is greater than a preset wind speed threshold, obtaining building shape information, and determining a building shape coefficient according to the building shape information and a preset shape coefficient standard; Obtaining building type information, and determining the building dynamic pressure coefficient according to the building type information and a preset dynamic pressure coefficient standard; Obtaining building roughness, building surface material, and building additional structure, and inputting the building roughness, building surface material, building additional structure, and building shape information into a preset drag coefficient calculation model to obtain the building drag coefficient; Obtaining the current air density and the building's windward area, and calculating the building's wind pressure based on the current air density, the building's windward area, the building's shape coefficient, the building's dynamic pressure coefficient, the building's wind resistance coefficient, the predicted wind speed, and a preset wind pressure calculation formula; If the building wind pressure exceeds a preset building wind pressure threshold, an alarm message is generated according to the building wind pressure, and the alarm message is transmitted to a control center.

4. The method according to claim 3, characterized in that in, The preset wind pressure calculation formula is: P.S wind 0.5·ρ·C d ·V 2 ·A·C s ·C f 100. Among them, P wind is the wind pressure on the building; ρ is the current air density; C d is the building dynamic pressure coefficient; V is the predicted wind speed; A is the building's windward area; C s is the building shape coefficient; C f is the building's drag coefficient.

5. The method according to claim 3, characterized in that in, After transmitting the alarm information to the control center, the method further includes: Obtain building usage status data, building structure data and building surrounding environment data, input the building wind pressure, building usage status data, building structure data, building surrounding environment data, building type information, building shape information, building surface material and building wind-exposed area into a preset disaster prevention and emergency plan formulation model to obtain a disaster prevention and emergency plan, and transmit the disaster prevention and emergency plan to a control center.

6. The method according to claim 1, characterized in that in, After obtaining the high wind weather forecast result, the method further includes: Obtaining the current system time, sending a traffic flow acquisition request to the control center according to the current system time, and receiving normal traffic flow data sent by the control center according to the traffic flow acquisition request; Determining a predicted wind speed based on a high wind weather forecast result, and determining a wind speed-traffic response coefficient based on the predicted wind speed, current traffic flow data, and a preset response coefficient standard; Calculate the predicted traffic flow based on normal traffic flow data, predicted wind speed, wind speed-traffic response coefficient and preset traffic flow calculation formula; Target traffic control measures are determined based on the predicted traffic flow and preset traffic control measure standards, and the target traffic control measures are sent to the control center.

7. The method according to claim 6, characterized in that in, The default traffic flow calculation formula is: F new =F base ×(1-K×V); Among them, F new To predict traffic flow; F base is the normal traffic flow data; K is the wind speed-traffic response coefficient; V is the predicted wind speed.

8. The method according to claim 6, characterized in that in, After sending the target traffic control measure to the control center, the method further includes: If the preset update interval is reached, updating the strong wind weather forecast result, and updating the predicted wind speed according to the strong wind weather forecast result; updating the building wind pressure according to the updated predicted wind speed; if the updated building wind pressure still exceeds a preset building wind pressure threshold, regenerating an alarm message according to the updated building wind pressure; and updating a disaster prevention and emergency response plan according to the updated building wind pressure, and transmitting the regenerated alarm message and the updated disaster prevention and emergency response plan to a control center; Accordingly, after updating the predicted wind speed according to the high wind weather forecast result, the method further includes: Re-acquire the current system time, re-acquire normal traffic flow data according to the re-acquired current system time, and recalculate the predicted traffic flow according to the re-acquired normal traffic flow data and the updated predicted wind speed; The target traffic control measures are re-determined according to the recalculated predicted traffic flow, and the re-determined target traffic control measures are sent to the control center.

9. A cold air and strong wind identification system based on multi-source data and deep learning, used to execute the method according to any one of claims 1 to 8, characterized in that: The system comprises: A data set creation module is used to obtain historical meteorological element pattern data and historical three-dimensional radar detection data, and create a historical meteorological data set based on the historical meteorological element pattern data and historical three-dimensional radar detection data; A sample creation module is used to determine the high wind single unit characteristic information and the cold wind meteorological characteristic set of the historical meteorological data set, obtain historical disaster-stricken area data, and construct meteorological disaster positive samples and meteorological disaster negative samples based on the high wind single unit characteristic information, the cold wind meteorological characteristic set and the historical disaster-stricken area data; A model training module is used to build a cold air and strong wind recognition model based on Res-Unet, and train the cold air and strong wind recognition model based on Res-Unet according to the meteorological disaster positive samples and the meteorological disaster negative samples; The prediction module is used to obtain real-time meteorological element data and real-time three-dimensional radar detection data, and input the real-time meteorological element data and real-time three-dimensional radar detection data into the trained Res-Unet-based cold air and gale recognition model to obtain gale weather forecast results.

10. An electronic device comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 8.

Images