CN-121999590-A - Thunderstorm early-warning model training and thunderstorm early-warning method and device

CN121999590ACN 121999590 ACN121999590 ACN 121999590ACN-121999590-A

Abstract

The invention relates to the technical field of meteorological disaster early warning, and discloses a thunderstorm early warning model training method and a thunderstorm early warning method and device, wherein the thunderstorm early warning model training method comprises the steps of acquiring a historical thunderstorm data set of a region to be predicted; the method comprises the steps of respectively analyzing thunderstorm record data of each thunderstorm to obtain thunderstorm characteristics of each thunderstorm, wherein the thunderstorm characteristics comprise spatial thunderstorm characteristics at a plurality of moments and time sequence thunderstorm characteristics at a plurality of characteristic positions, the spatial thunderstorm characteristics are constructed according to distances between each monitoring site and the coordinates of the center of mass of the thunderstorm and monitoring data sets of each monitoring site, the time sequence thunderstorm characteristics are constructed according to the monitoring data sets of the characteristic positions at different moments, a training set is established according to the thunderstorm characteristics of each thunderstorm and thunderstorm early-warning parameters, and a preset model is trained by using the training set to obtain a thunderstorm prediction model. According to the invention, through double analysis of thunderstorm data in time sequence and space, accuracy of thunderstorm prediction is increased, and further, the thunderstorm early warning efficiency is improved.

Inventors

Hu Sige
MA DONG
ZHANG ZHE
CHEN YUWEN
CHEN YINSHENG
Kong Zekun
NIU SHIKAI
XU SIDA
LI YUNFEI

Assignees

华电电力科学研究院有限公司

Dates

Publication Date: 20260508
Application Date: 20251217

Claims (10)

1. A thunderstorm warning model training method, the method comprising: Acquiring a historical thunderstorm data set of an area to be predicted, wherein the historical thunderstorm data set comprises thunderstorm record data of a plurality of thunderstorms, and the thunderstorm record data comprises monitoring data sets of a plurality of monitoring stations at different moments, and thunderstorm centroid coordinates and thunderstorm early warning parameters of the thunderstorms; Respectively analyzing the thunderstorm record data of each thunderstorm to obtain thunderstorm characteristics of each thunderstorm, wherein the thunderstorm characteristics comprise spatial thunderstorm characteristics at a plurality of moments and time sequence thunderstorm characteristics at a plurality of characteristic positions, the spatial thunderstorm characteristics are constructed according to the distance between each monitoring site and the coordinates of the thunderstorm centroid and the monitoring data set of each monitoring site, and the time sequence thunderstorm characteristics are constructed according to the monitoring data sets of the characteristic positions at different moments; and building a training set according to the thunderstorm characteristics and the thunderstorm early-warning parameters of each thunderstorm, and training a preset model by utilizing the training set to obtain a thunderstorm prediction model, wherein the thunderstorm prediction model is used for predicting the thunderstorm early-warning parameters according to the thunderstorm characteristics.
2. The method of claim 1, wherein the monitoring dataset of the monitoring site includes monitoring data for a plurality of monitoring items, the step of constructing a spatial thunderstorm feature comprising: calculating the distance between each monitoring station and the coordinate of the thunderstorm mass center; fitting the monitoring data of each monitoring item according to the sequence of increasing the distance to obtain a spatial characteristic change curve corresponding to each monitoring item, wherein the spatial characteristic change curve is used for representing the corresponding relation between the distance and the monitoring data; And integrating the spatial characteristic change curves of the monitoring items to obtain the spatial thunderstorm characteristics of the monitoring data set at the corresponding acquisition time.
3. The method of claim 2, wherein the step of constructing a time series thunderstorm feature comprises: Segmenting the spatial characteristic change curve at each moment according to a preset distance interval and determining the characteristic positions corresponding to the segments respectively; Calculating characteristic values in the spatial characteristic change curves of the characteristic positions at different moments according to end point coordinates of head and tail end points of each segment in the spatial characteristic change curves, wherein the abscissa of the end point coordinates represents the distance, and the ordinate represents the monitoring data; fitting characteristic values in the spatial characteristic change curves of the characteristic positions at different moments according to time sequences for each monitoring item to obtain a time sequence characteristic change curve of the monitoring item at each characteristic position; integrating time sequence characteristic change curves of the same monitoring item at each characteristic position to obtain time sequence thunderstorm characteristics of the monitoring items; and integrating the time sequence thunderstorm characteristics of the monitoring items to obtain the time sequence thunderstorm characteristics.
4. A thunderstorm warning method, the method comprising: Acquiring a current monitoring data set and a current thunderstorm centroid coordinate of each monitoring station in an area to be predicted; Generating a thunderstorm feature of the current thunderstorm according to the current monitoring data set and the current thunderstorm centroid coordinates, wherein the thunderstorm feature comprises a space thunderstorm feature at a plurality of moments and a time sequence thunderstorm feature at a plurality of feature positions, the space thunderstorm feature is constructed according to the distance between each monitoring site and the thunderstorm centroid coordinates and the monitoring data set of each monitoring site, and the time sequence thunderstorm feature is constructed according to the monitoring data sets of the feature positions at different moments; Inputting the thunderstorm characteristics of the current thunderstorm into a pre-trained thunderstorm prediction model to obtain thunderstorm early-warning parameters of the current thunderstorm, wherein the thunderstorm prediction model is obtained by training according to the thunderstorm early-warning model training method of any one of claims 1-3; And carrying out thunderstorm early warning on the area to be predicted according to the thunderstorm early warning parameters.
5. The method of claim 4, wherein performing a thunderstorm warning on the area to be predicted according to the thunderstorm warning parameter comprises: determining a thunderstorm grade and corresponding early-warning intensity based on the thunderstorm intensity in the thunderstorm early-warning parameters, and determining an early-warning range based on the thunderstorm radius; Integrating the early warning intensity and the early warning range, and generating and issuing early warning information.
6. A thunderstorm warning model training device, the device comprising: The system comprises a historical data acquisition module, a data processing module and a data processing module, wherein the historical data acquisition module is used for acquiring a historical thunderstorm data set of an area to be predicted, the historical thunderstorm data set comprises a plurality of thunderstorm record data of thunderstorms, and the thunderstorm record data comprises monitoring data sets of a plurality of monitoring stations at different moments, and thunderstorm centroid coordinates and thunderstorm early-warning parameters of the thunderstorms; The thunderstorm feature acquisition module is used for respectively analyzing the thunderstorm record data of each thunderstorm to obtain the thunderstorm feature of each thunderstorm, wherein the thunderstorm feature comprises a space thunderstorm feature at a plurality of moments and a time sequence thunderstorm feature at a plurality of feature positions, the space thunderstorm feature is constructed according to the distance between each monitoring site and the coordinate of the thunderstorm centroid and the monitoring data set of each monitoring site, and the time sequence thunderstorm feature is constructed according to the monitoring data sets of the feature positions at different moments; The thunderstorm early-warning model training module is used for building a training set according to the thunderstorm characteristics and the thunderstorm early-warning parameters of each thunderstorm, training a preset model by utilizing the training set to obtain a thunderstorm prediction model, and the thunderstorm prediction model is used for predicting the thunderstorm early-warning parameters according to the thunderstorm characteristics.
7. A thunderstorm warning device, the device comprising: The current monitoring data acquisition module is used for acquiring a current monitoring data set and a current thunderstorm centroid coordinate of each monitoring station in the area to be predicted; The current thunderstorm feature acquisition module is used for generating thunderstorm features of the current thunderstorm according to the current monitoring data set and the current thunderstorm centroid coordinates, wherein the thunderstorm features comprise spatial thunderstorm features at a plurality of moments and time sequence thunderstorm features at a plurality of feature positions, the spatial thunderstorm features are constructed according to the distance between each monitoring site and the thunderstorm centroid coordinates and the monitoring data set of each monitoring site, and the time sequence thunderstorm features are constructed according to the monitoring data sets of the feature positions at different moments; The thunderstorm early-warning parameter determining module is used for inputting the thunderstorm characteristics of the current thunderstorm into a pre-trained thunderstorm prediction model to obtain the thunderstorm early-warning parameters of the current thunderstorm, and the thunderstorm prediction model is obtained by training according to the thunderstorm early-warning model training method of any one of claims 1-3; And the thunderstorm early warning module is used for carrying out thunderstorm early warning on the area to be predicted according to the thunderstorm early warning parameters.
8. An electronic device, comprising: A memory and a processor, the memory and the processor being communicatively connected to each other, the memory having stored therein computer instructions, the processor executing the computer instructions to perform the thunderstorm warning model training method of any one of claims 1 to 3, or to perform the thunderstorm warning method of any one of claims 4 or 5.
9. A computer-readable storage medium having stored thereon computer instructions for causing a computer to perform the thunderstorm warning model training method of any one of claims 1 to 3 or to perform the thunderstorm warning method of any one of claims 4 or 5.
10. A computer program product comprising computer instructions for causing a computer to perform the thunderstorm warning model training method of any one of claims 1 to 3, or to perform the thunderstorm warning method of any one of claims 4 or 5.

Description

Thunderstorm early-warning model training and thunderstorm early-warning method and device Technical Field The invention relates to the technical field of meteorological disaster early warning, in particular to a thunderstorm early warning model training and thunderstorm early warning method and device. Background In recent years, urban areas have extreme weather disasters and secondary and derivative disasters each year, and weather disasters such as storm and thunder and lightning all form serious threats to the development of the whole area economy and society, the life and property safety of people and the ecological environment. In addition, with the acceleration of the urban process, urban population is continuously gathered, the bearing load of urban infrastructure is continuously increased, the sensitivity and vulnerability of the city to weather and the influence of the weather and the derivative disasters are obvious, and urban inland inundation becomes a great hidden trouble for urban development. Many times of heavy storm water causes water accumulation in most low-lying road sections and road culverts of cities, and the traveling, production and life of residents are seriously affected. The existing thunderstorm early warning system is used for carrying out early warning by analyzing the time sequence change characteristics of the thunderstorm related data, and has the advantages of convenience in acquiring the time sequence data, lower analysis difficulty and stronger practicability in the whole state evaluation of the thunderstorm. The method has obvious limitations that on one hand, when the time sequence characteristic disassembly analysis is carried out on the related data of different areas in the thunderstorm, the operation difficulty is high, the local evolution difference in the thunderstorm is difficult to accurately capture, on the other hand, the method only depends on single monitoring data time sequence characteristic to carry out early warning, the complete rule of thunderstorm formation and development cannot be comprehensively described, and finally the accuracy of the early warning of the thunderstorm is low. Disclosure of Invention The invention provides a thunderstorm early-warning model training method and a thunderstorm early-warning method and device, which are used for solving the problem that the accuracy of the thunderstorm early-warning is low due to the fact that the related technology only analyzes the time sequence characteristics of monitoring data. In a first aspect, the invention provides a thunderstorm warning model training method, which comprises the following steps: The method comprises the steps of acquiring a historical thunderstorm data set of an area to be predicted, wherein the historical thunderstorm data set comprises a plurality of thunderstorm record data of the thunderstorms, the thunderstorm record data comprises a plurality of monitoring data sets of monitoring stations at different moments, and thunderstorm centroid coordinates and thunderstorm early-warning parameters of the thunderstorms, analyzing the thunderstorm record data of each thunderstorm respectively to obtain thunderstorm characteristics of each thunderstorm, wherein the thunderstorm characteristics comprise space thunderstorm characteristics at a plurality of moments and time sequence thunderstorm characteristics at a plurality of characteristic positions, the space thunderstorm characteristics are constructed according to the distance between each monitoring station and the thunderstorm centroid coordinates and the monitoring data sets of each monitoring station, the time sequence thunderstorm characteristics are constructed according to the monitoring data sets at different moments of the characteristic positions, a training set is established according to the thunderstorm characteristics and the thunderstorm early-warning parameters of each thunderstorm, a preset model is trained by using the training set to obtain a thunderstorm prediction model, and the thunderstorm prediction model is used for predicting the thunderstorm early-warning parameters according to the thunderstorm characteristics. According to the thunderstorm early warning model training method provided by the invention, through acquiring the historical data set containing the multi-thunderstorm full period monitoring data, the thunderstorm centroid coordinates and the early warning parameters, a plurality of monitoring items are covered, and the analysis standard is defined by the distance between the thunderstorm centroid coordinates and each monitoring station point, so that the basic data with complete dimension and accurate standard are provided for subsequent feature extraction and model training, and analysis deviation caused by single data or key information loss is avoided. Further, space thunderstorm features reflecting the thunderstorm space distribution rule are constructed based on the distance, time