Deep learning-based convective strong wind short-time forecasting method and system

Abstract

The invention relates to the technical field of weather forecasting, in particular to a convective strong wind short-time forecasting method and system based on deep learning, and the method comprises the steps: S1, collecting and preprocessing historical weather data; s2, constructing and training a deep learning model; s3, monitoring and collecting data in real time; s4, outputting a forecast result; s5, real-time data post-processing, extraction and screening; the system comprises a data collection module, a data preprocessing module, a deep learning module, a real-time monitoring module, a forecasting module and a real-time data processing module. The method is used for collecting the topographic feature data of the target region, training the deep learning model in a classified manner, and then performing prediction and forecasting by the deep learning model, thereby improving the accuracy of forecasting.

Description

Technical Field

[0001] The present invention relates to the technical field of meteorological forecasting, and in particular to a method and system for short-term forecasting of convective gale based on deep learning. Background Art

[0002] China's meteorological observation business regulations stipulate that winds with instantaneous wind speeds reaching or exceeding 17 meters per second (or wind speeds reaching or exceeding level 8 by visual estimation) are considered strong winds. Strong winds are one of the main disastrous weather in my country, causing great harm to industry, agriculture, transportation and people's lives. Convective gale is a sudden strong wind weather. Strong winds caused by tornadoes, downbursts and squall lines are all convective gale, which are strong, destructive and have a short defense time. Forecasting the occurrence of convective gale in advance will help residents take protective measures in advance and reduce the harm of convective gale weather.

[0003] In recent years, the application of machine learning methods in meteorology has made great progress. Experts at home and abroad have done a lot of research on convective gale forecasting, making it possible to forecast convective gale by learning the causes of convective gale from historical data. It also improves the defects of numerical forecasting that are subject to initial value errors and model errors, greatly improving the forecast effect. As the research on deep learning methods becomes more and more in-depth, scholars in the field of meteorology have gradually introduced deep learning into the forecast of convective gale, combining artificial intelligence methods with the meteorological industry.

[0004] When applying deep learning models to short-term forecasts of convective gale, the deep learning models are usually trained by only collecting temperature, humidity and air pressure data as key features related to convective gale, and then setting a threshold of less than level 8 gale (wind speed of 17.2m / s). However, since the terrain conditions that affect the wind speed and movement speed of convective gale during displacement are not considered, although the forecast range of convective gale is expanded and the forecast probability is improved, the probability of false alarms is also increased. This method forecasts weather that does not meet the gale standard, which reduces the accuracy of the forecast. Therefore, it is necessary to design a method and system for short-term forecasting of convective gale based on deep learning that can improve the forecast accuracy and reduce the probability of false alarms. Summary of the invention

[0005] To solve the above problems, the present invention provides a method and system for short-term forecasting of convective gale based on deep learning, which is used to collect terrain feature data of the target area, and classify and train the deep learning model, and then the deep learning model makes forecasts separately to improve the accuracy of the forecast.

[0006] In order to achieve the above object, the technical solution of the present invention is as follows: a method for short-term forecasting of convective gale based on deep learning, comprising the following steps:

[0007] S1. Collection and preprocessing of historical meteorological data: First, the target area is divided into several test areas, and several prediction points are set in the test areas. Then, the terrain feature data and the location information of the prediction points in each test area are collected respectively, and each test area is grouped according to the terrain feature data. Then, the wind speed, temperature, humidity and air pressure data corresponding to each test area when encountering convective strong winds are collected respectively, and several temperature, humidity and air pressure data are integrated and summarized for data preprocessing;

[0008] S2. Construction and training of deep learning model: The terrain features and prediction point location information as well as historical temperature, humidity and air pressure are used as input data, and the historical wind speed data collected at each prediction point is used as output data to construct and train the deep learning model;

[0009] S3. Real-time monitoring and data collection: real-time monitoring of temperature, humidity and air pressure data of several test areas, and inputting the data into the trained deep learning model, which analyzes the data and outputs convective gale forecast results at different locations;

[0010] S4. Output of forecast results: When the deep learning model completes the data analysis and determines that there is convective strong wind, it outputs the convective strong wind area and convective strong wind level;

[0011] S5. Post-processing and extraction and screening of real-time data: After completing the forecast of convective strong winds, the collected real-time data will be used as data for updating and training the deep learning model. The deep learning model will be continuously updated and optimized, and the wind speed distribution map of the target area will be extracted to screen out the strong wind areas that need enhanced protection.

[0012] Furthermore, in S1, a number of areas to be measured are classified according to terrain feature data, and the classification groups include wind speed reduction type terrain, wind speed enhancement type terrain and wind speed influence neglected type terrain.

[0013] Furthermore, the classification and grouping are carried out according to the following description: after the collection and processing of the terrain feature data, it basically includes canyons and mountain passes, plains, lakes and oceans, hilly basins, urban industrial areas, and river valleys and rivers. Canyons and mountain passes and river valleys and rivers are wind speed enhanced terrain, plains and lakes and oceans are wind speed neglected terrain, and urban industrial areas and hilly basins are wind speed slowed terrain.

[0014] Furthermore, in S1, the air pressure data of each region are collected mainly at the following levels: the earth's surface, 850 hPa, 700 hPa and 500 hPa.

[0015] Furthermore, in S1, the data preprocessing specifically includes removing duplicate data and invalid data, filling missing values ​​with interpolation or mean substitution, and individually identifying and correcting outliers.

[0016] Furthermore, in S2, the deep learning model is constructed and trained, and the time for convective gales to arrive at any predicted point in a certain tested area from historical data and the intensity of convective gales arriving at each area are collected as test correction data.

[0017] Furthermore, in S3, the short-term forecast of convective gale needs to report the time when the gale arrives in each area, and the arrival time of the convective gale in each area is calculated by the following formula:

[0018]

[0019] Where x is the number of each area, t x To predict the time required for convective winds to reach each area, S x is the distance between each area and the prediction point, n is the type code of wind speed mitigation terrain, wind speed enhancement terrain and wind speed influence neglected terrain, V n is the displacement speed of convective gale of corresponding type in each area;

[0020] The convective wind displacement velocity V predicted by the deep learning model is n , calculated by the following formula:

[0021] V n =y*V

[0022] Where V is the moving speed of convective strong winds predicted by the deep learning model, y is the increase or decrease ratio of the displacement speed of convective strong winds corresponding to different terrain types. In the terrain area with reduced wind speed, y is 0.96-0.98, in the terrain area with enhanced wind speed, y is 1.02-1.04, and in the terrain area with ignored wind speed, y is 1.00.

[0023] Furthermore, the convective wind displacement velocity V in each region is calculated based on the deep learning model. n , V in wind-slowing terrain areas n If the wind speed is greater than 17.5m / s, a convective gale forecast is required. The wind speed has negligible impact on terrain areas V n If the wind speed is greater than 17.0m / s, a convective gale forecast is required. n If the wind speed is greater than 16.5m / s, a convective gale forecast is required.

[0024] Furthermore, in S5, the data is screened specifically to screen out observation data with wind speed greater than or equal to 15 m / s, and in the screened wind speed data, use the threshold segmentation method to extract all connected areas with wind speed greater than the threshold, and mark these areas as potential high wind areas.

[0025] The convective gale short-term forecasting system based on deep learning operates based on the convective gale short-term forecasting method, including:

[0026] A data collection module, used to collect characteristic data of the target area;

[0027] The data preprocessing module is used to identify several areas to be tested as wind speed enhanced terrain, wind speed neglected terrain and wind speed slowed terrain according to the terrain feature data collected by the data collection module, and then search and collect wind speed, temperature, humidity, air pressure and distance from the prediction point of several categories of areas when encountering popular strong winds, integrate the data into the data blocks corresponding to each area, and finally remove or identify and correct the duplicate data, invalid data and erroneous data in the collected data;

[0028] A deep learning module is used to construct and train a deep learning model using terrain features, prediction point location information, and historical temperature, humidity, and air pressure as input data and historical wind data collected at each prediction point as output data;

[0029] The real-time monitoring module is used to monitor and collect temperature, humidity and air pressure data of several test areas in real time, and input the data into the trained deep learning model. The deep learning model analyzes the data and outputs the convective gale prediction results at different locations. The test areas that are about to be subject to convective gale weather are marked with a signal representing "forecast required" and transmitted to the forecast module;

[0030] The forecast module is used to receive the data analysis completed by the deep learning model and determine the presence of convective strong wind weather. When the "forecast required" mark is identified, the convective strong wind intensity and the arrival time t of the strong wind that the area will suffer are forecasted in real time. x ;

[0031] The real-time data processing module is used to use the collected real-time data as data for updating and training the deep learning model after completing the forecast of convective strong winds, continuously update and optimize the deep learning model, extract the wind speed distribution map of the target area, and screen out the strong wind areas that need enhanced protection.

[0032] The above scheme has the following beneficial effects:

[0033] 1. This scheme first divides the target area into several areas. After collecting the terrain feature data of several areas to be tested, the areas to be tested are classified according to the terrain feature data. Since various terrain features will affect the speed and direction of airflow, they are classified according to the different effects of terrain on wind speed, which are divided into wind speed reduction terrain, wind speed enhancement terrain and wind speed constant terrain. According to the different effects of terrain on wind speed, the wind speed of convective gales in different terrain areas can be predicted more accurately. In addition, compared with separate predictions for different areas, this classification prediction method can greatly improve the efficiency of prediction. Because we can classify areas according to terrain characteristics, and then make unified predictions and analyses for each category, it avoids repeated and tedious prediction work for each area. This not only saves time and energy, but also improves the accuracy and reliability of predictions.

[0034] 2. This solution builds a deep learning model based on the collected terrain features, temperature, humidity and air pressure data. The model is trained using historical data, and the time when convective gales arrive at each area from the predicted point and the gale intensity upon arrival are collected as test and correction data. Through training, the deep learning model can learn the formation mechanism, propagation law and influencing factors of convective gales, thus having the ability to predict convective gales.

[0035] 3. In this plan, each region will collect multi-level pressure data including 850hPa, 700hPa and 500hPa, not only the surface pressure. Since severe weather such as convective gales mainly occurs in the troposphere, multi-level pressure detection in the troposphere can more accurately capture information on weather changes, thereby improving the accuracy of pressure data. In addition, these accurate data are an important basis for improving the accuracy of weather forecasts.

[0036] 4. In the process of building and training the deep learning model, this scheme mainly uses the historical predicted arrival time and wind intensity of each test area when it is hit by convective strong winds, supplemented by the time and wind intensity of convective strong winds arriving from any predicted point in a test area to any predicted point in the adjacent test area, to test and correct the result output of the deep learning model, thereby improving the prediction accuracy of the deep learning model.

[0037] 5. This solution, the post-processing and extraction and screening design of real-time data, can update and train the deep learning model, making the deep learning model more efficient and accurate. In addition, by screening the windy areas that need to be strengthened, it can strengthen safety protection and improve the ability to respond to convective strong wind weather.

[0038] 6. This scheme performs corresponding forecast calculations and corresponding forecast methods according to different terrains to improve the pertinence of the forecast. Compared with the method of making unified forecasts without considering the terrain area, it can reduce the probability of false alarms and misreports, improve the accuracy of the forecast, reduce the setting of invalid protection, and save social resources.

[0039] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 A schematic diagram of a method embodiment of a method and system for short-term forecasting of convective gale based on deep learning according to the present invention;

[0041] Figure 2 A schematic diagram of regional division of an embodiment of a method and system for short-term forecasting of convective gale based on deep learning of the present invention;

[0042] Figure 3 It is a system operation diagram of a method and system embodiment of a short-term convective gale forecasting method and system based on deep learning of the present invention. DETAILED DESCRIPTION

[0043] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0044] The following is further described in detail through specific implementation methods:

[0045] Embodiment 1:

[0046] As attached Figure 1 and Figure 2 As shown: The short-term forecasting method of convective gale based on deep learning includes the following steps:

[0047] S1. Collection and preprocessing of historical meteorological data: First, the target area is divided into several test areas, and several prediction points are set in the test areas. Then, the terrain feature data of several test areas and the location information of the prediction points are collected respectively, and each test area is grouped according to the terrain feature data. Since various terrain features can affect the speed and direction of airflow, for example, the undulating terrain in hilly areas may cause the airflow to become scattered, the speed to decrease, the direction to change, and the wind speed of convective gales to decrease, the classification group design includes wind speed reduction terrain, wind speed enhancement terrain and wind speed influence neglected terrain. After the terrain feature data are collected and processed, they basically include canyons and mountain passes, plains, lakes and oceans, hilly basins, urban industrial areas and river valleys and rivers. Canyons and mountain passes and river valleys and rivers are wind speed enhancement terrain, plains and lakes and oceans are wind speed influence neglected terrain, and urban industrial areas and hilly basins are wind speed reduction terrain, which is conducive to the subsequent corresponding wind speed forecast calculation for wind speed enhancement terrain, wind speed influence neglected terrain and wind speed reduction terrain. This design classifies the wind speed according to the impact of terrain, and predicts it in three categories, which can improve the efficiency of prediction compared to making separate predictions for different regions. After the classification is completed, the corresponding wind speed, temperature, humidity and air pressure data of several test areas when encountering convective strong winds are collected separately. The collection of air pressure data in each area mainly collects air pressure data at the earth's surface, 850hPa, 700hPa and 500hPa levels. Since convective strong winds are severe weather that mainly occurs in the troposphere, multi-level air pressure detection in the troposphere can improve the accuracy of air pressure data and the accuracy of forecasts. After the collection work is completed, several temperature, humidity and air pressure data are integrated and summarized for data preprocessing. The data preprocessing operation is specifically to remove duplicate data and invalid data. For missing values, interpolation or mean substitution method is used to fill them. For outliers, they are individually identified and corrected. Through data preprocessing, the temperature, humidity and air pressure data are more referenceable, providing more reliable data support for the construction and training of subsequent deep learning models.

[0048] S2. Construction and training of deep learning model: The deep learning model is constructed with terrain features as the algorithm differentiation standard, so that the deep learning model has the ability to predict multiple types of terrain, such as wind speed slowing terrain, wind speed enhancing terrain and wind speed neglected terrain. Then, the historical temperature, humidity and air pressure are used as input data, and the historical wind speed data collected at each prediction point is used as output data to train the model. During the training process of the deep learning model, the time when convective gales arrive at any prediction point in a certain test area from any prediction point and the gale intensity in the adjacent test area are collected in the historical data as test and correction data, so that the deep learning model has the function of predicting the intensity and arrival time of convective gales at the same time, which is beneficial for various regions to take safety precautions against convective gales.

[0049] S3. Real-time monitoring and data collection: Real-time monitoring of the temperature, humidity and air pressure data of several test areas, and inputting the data into the trained deep learning model. The deep learning module predicts the convective gale weather. The specific prediction process is to calculate the arrival time of the convective gale in each area through the following formula:

[0050]

[0051] Where x is the number of each area, t x To predict the time required for convective winds to reach each area, S x is the distance between each area and the prediction point, n is the type code of wind speed mitigation terrain, wind speed enhancement terrain and wind speed influence neglected terrain, V n is the displacement speed of convective gale of corresponding type in each area;

[0052] The convective wind displacement velocity V predicted by the deep learning model is n , calculated by the following formula:

[0053] V n =y*V

[0054] Where V is the convective gale moving speed predicted by the deep learning model, y is the increase or decrease ratio of the convective gale displacement speed corresponding to different terrain types, y is 0.96-0.98 in the wind speed reduction terrain area, y is 1.02-1.04 in the wind speed enhancement terrain area, and y is 1.00 in the wind speed neglected terrain area. After calculating the wind speed data corresponding to each area, combined with the analysis of air pressure, the convective gale displacement speed V of each area is calculated based on the prediction module. n , when V in the wind speed slowing terrain area n If the wind speed is greater than 17.5m / s, a convective gale forecast is required. n If the wind speed is greater than 17.0m / s, a convective gale forecast is required. n When the wind speed is greater than 16.5m / s, a convective gale forecast is required. This design is conducive to reducing the probability of false alarms of weather with convective gale intensity, while increasing the success rate of forecasting situations where gale accelerates to form during the displacement process, thereby improving safety.

[0055] S4. Output of forecast results: After the deep learning model predicts the wind speed, it determines whether the intensity corresponding to the wind speed exceeds the set threshold, and combines the data analysis of air pressure to determine whether the strong wind is a convective strong wind. If it is determined to be a convective strong wind, the forecast result of the corresponding area is output;

[0056] S5. Post-processing and extraction and screening of real-time data: After completing the forecast of convective gale, the collected real-time data is used as the data for updating and training the deep learning model. The deep learning model is continuously updated and trained, and the wind speed distribution map of the target area is extracted to screen out the gale areas that need to be strengthened. The specific screening method is to screen out the observation data with wind speed greater than or equal to 15m / s, and use the threshold segmentation method to extract all connected areas with wind speed greater than the threshold in the screened wind speed data, and mark these areas as potential gale areas. By marking the gale areas that need to be strengthened, it is beneficial to strengthen the safety protection of these gale areas when they experience convective gale in the future, thereby reducing the damage rate of convective gale to the target area.

[0057] Embodiment 2:

[0058] As attached Figure 3 As shown: A convective gale short-term forecasting system based on deep learning is operated based on the convective gale short-term forecasting method described in Example 1, and mainly includes a data collection module, a data preprocessing module, a deep learning module, a real-time monitoring module, a forecasting module and a real-time data processing module.

[0059] The data collection module collects the characteristic data of the target area and integrates and passes the data to the data preprocessing module.

[0060] In the data preprocessing module, first, the user divides the target area into several test areas and sets several test points in the test areas. Then, according to the terrain feature data collected by the receiving data collection module, the features of several test areas are marked as "1", "0" and "-1" respectively according to the wind speed enhancement terrain, wind speed neglect terrain and wind speed reduction terrain. Then, the wind speed, temperature, humidity, air pressure and distance from the forecast point of several categories of areas when encountering strong winds are searched in the meteorological history database, and each data is integrated into the corresponding data block of each area. Finally, the operation of removing or identifying and correcting the duplicate data, invalid data and erroneous data in the collected data is performed to improve the reference value and reliability of the data. After data preprocessing, the data is passed to the deep learning module.

[0061] The deep learning module takes terrain features, prediction point location information, historical temperature, humidity and air pressure as input data, and the historical wind data collected at each prediction point as output data to build and train a deep learning model. After the training is completed, the trained deep learning module will be put into forecasting use.

[0062] The real-time monitoring module monitors and collects temperature, humidity and air pressure data of several test areas in real time, and inputs the data into the trained deep learning model. The deep learning model analyzes the data and outputs the convective gale prediction results at different locations. The test areas that are about to be hit by convective gale weather are marked with a signal representing "forecast required" and transmitted to the forecast module.

[0063] When the forecast module receives the data analysis completed by the deep learning model and determines that there is convective strong wind weather, after identifying the "forecast required" mark, it will forecast the convective strong wind intensity and arrival time t that the area will suffer in real time. x .

[0064] After completing the forecast of convective strong winds, the real-time data processing module uses the collected real-time data as data for updating and training the deep learning model, continuously updates and optimizes the deep learning model, extracts the wind speed distribution map of the target area, and screens out strong wind areas that need enhanced protection.

[0065] Obviously, the above embodiments are merely examples for the purpose of clear explanation, and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived therefrom are still within the scope of protection of the invention.

Claims

1. A short-term forecasting method for convective gale based on deep learning, characterized in that: The following steps are involved: S1. Collection and preprocessing of historical meteorological data: First, the target area is divided into several test areas, and several prediction points are set in the test areas. Then, the terrain feature data and the location information of the prediction points in each test area are collected respectively, and each test area is grouped according to the terrain feature data. Then, the wind speed, temperature, humidity and air pressure data corresponding to each test area when encountering convective strong winds are collected respectively, and several temperature, humidity and air pressure data are integrated and summarized for data preprocessing; S2. Construction and training of deep learning model: The terrain features and prediction point location information as well as historical temperature, humidity and air pressure are used as input data, and the historical wind speed data collected at each prediction point is used as output data to construct and train the deep learning model; S3. Real-time monitoring and data collection: Real-time monitoring of temperature, humidity and air pressure data of several test areas, and input of the data into the trained deep learning model, which analyzes the data and outputs the convective gale prediction results at different locations; S4. Forecast result output: When the deep learning model completes the data analysis and determines that there is convective strong wind, it outputs the convective strong wind area and convective strong wind level; S5. Post-processing and extraction and screening of real-time data: After completing the forecast of convective strong winds, the collected real-time data will be used as data for updating and training the deep learning model. The deep learning model will be continuously updated and optimized, and the wind speed distribution map of the target area will be extracted to screen out the strong wind areas that need enhanced protection.

2. The method for short-term forecasting of convective gale based on deep learning according to claim 1, characterized in that: In S1, several areas to be tested are classified according to terrain feature data, and the classification groups include wind speed reduction terrain, wind speed enhancement terrain and wind speed influence neglected terrain.

3. The method for short-term forecasting of convective gale based on deep learning according to claim 2, characterized in that: The classification and grouping are carried out according to the following description: after the collection and processing of terrain feature data, it basically includes canyons and mountain passes, plains, lakes and oceans, hilly basins, urban industrial areas, and river valleys and rivers. Canyons and mountain passes and river valleys and rivers are wind speed enhanced terrain, plains and lakes and oceans are wind speed neglected terrain, and urban industrial areas and hilly basins are wind speed slowed terrain.

4. The method for short-term forecasting of convective gale based on deep learning according to claim 3 is characterized in that: In S1, the main levels of air pressure data collection for each region include: 850hPa, 700hPa and 500hPa.

5. The method for short-term forecasting of convective gale based on deep learning according to claim 4, characterized in that: In S1, data preprocessing specifically includes removing duplicate data and invalid data, filling missing values ​​with interpolation or mean substitution, and individually identifying and correcting outliers.

6. The method for short-term forecasting of convective gale based on deep learning according to claim 5, characterized in that: In S2, the deep learning model is constructed and trained, and the time when convective gales arrive at any predicted point in a certain tested area and the gale intensity in the historical data are collected as test and correction data.

7. The method for short-term forecasting of convective gale based on deep learning according to claim 6, characterized in that: In S3, the short-term forecast of convective gale needs to report the time when the gale arrives in each area. The arrival time of convective gale in each area is calculated by the following formula: Where x is the number of each area, t x To predict the time required for convective winds to reach each area, S x is the distance between each area and the prediction point, n is the type code of wind speed mitigation terrain, wind speed enhancement terrain and wind speed influence neglected terrain, V n is the displacement speed of convective gale of corresponding type in each area; The convective wind displacement velocity V predicted by the deep learning model is n , calculated by the following formula: V n =y*V Where V is the moving speed of convective strong winds predicted by the deep learning model, y is the increase or decrease ratio of the displacement speed of convective strong winds corresponding to different terrain types. In the terrain area with reduced wind speed, y is 0.96-0.98, in the terrain area with enhanced wind speed, y is 1.02-1.04, and in the terrain area with ignored wind speed, y is 1.

00.

8. The method for short-term forecasting of convective gale based on deep learning according to claim 7, characterized in that: The convective wind displacement velocity V in each region is calculated based on the deep learning model. n , V in wind-slowing terrain areas n If the wind speed is greater than 17.5m / s, a convective gale forecast is required. The wind speed has negligible impact on terrain areas V n If the wind speed is greater than 17.0m / s, a convective gale forecast is required. n If the wind speed is greater than 16.5m / s, a convective gale forecast is required.

9. The method for short-term forecasting of convective gale based on deep learning according to claim 8, characterized in that: In S5, the data is screened specifically by screening out observation data with wind speeds greater than or equal to 15 m / s, and using the threshold segmentation method to extract all connected areas with wind speeds greater than the threshold from the screened wind speed data, and marking these areas as potential high wind areas.

10. A convective gale short-term forecasting system based on deep learning, operating based on the convective gale short-term forecasting method according to any one of claims 1 to 9, characterized in that: include: A data collection module, used to collect characteristic data of the target area; The data preprocessing module is used to identify several areas to be tested as wind speed enhanced terrain, wind speed neglected terrain and wind speed slowed terrain according to the terrain feature data collected by the data collection module, and then search and collect wind speed, temperature, humidity, air pressure and distance from the prediction point of several categories of areas when encountering popular strong winds, integrate the data into the data blocks corresponding to each area, and finally remove or identify and correct the duplicate data, invalid data and erroneous data in the collected data; A deep learning module is used to construct and train a deep learning model using terrain features, prediction point location information, and historical temperature, humidity, and air pressure as input data and historical wind data collected at each prediction point as output data; The real-time monitoring module is used to monitor and collect temperature, humidity and air pressure data of several test areas in real time, and input the data into the trained deep learning model. The deep learning model analyzes the data and outputs the convective gale prediction results at different locations. The test areas that are about to be subject to convective gale weather are marked with a signal representing "need for forecast" and transmitted to the forecast module; The forecast module is used to receive the data analysis completed by the deep learning model and determine the presence of convective gale weather. When the "forecast required" mark is identified, the convective gale intensity and gale arrival time t that the area will suffer are forecasted in real time. x ; The real-time data processing module is used to use the collected real-time data as data for updating and training the deep learning model after completing the forecast of convective strong winds, continuously update and optimize the deep learning model, extract the wind speed distribution map of the target area, and screen out the strong wind areas that need enhanced protection.

Images