一种基于VLF/LF闪电数据的多算法融合雷暴系统识别技术

A MULTI-ALGORITHM FUSION THUNDERSTORM SYSTEM RECOGNITION TECHNOLOGY BASED ON VLF/LF LIGHTNING DATA

关键词:  VLF/LF闪电定位  参数自适应  多算法融合  雷暴识别  密度聚类

Key words:  VLF/LF lightning locating  parameter adaptation  multi-algorithm fusion  thunderstorm identification  density clustering

摘要:

为深入挖掘VLF/LF三维闪电监测网在雷电临近预警中的应用效益，提出一种基于VLF/LF闪电数据的多算法融合雷暴系统识别技术，首先对OPTICS聚类法进行参数自适应改进，即基于VLF/LF闪电数据的空间分布特征，利用双总体t检验法实时计算初始输入参数Min_ρ，使其自适应于待处理的闪电数据；而后与K-Means聚类法融合应用，利用K-Means法通过迭代计算聚类中心的优势，改进仅凭OPTICS法无法准确定位雷暴体的中心点的缺陷。经不同尺度雷暴天气过程检验，该技术有效改善OPTICS法等常规的密度聚类法无法适用于三维闪电数据聚类处理、无法识别区分密度差异不大的相邻雷暴体的缺陷，能实现任意尺度形状雷暴体的识别及中心点的精准计算，可为雷暴追踪及趋势外推工作提供优于传统聚类技术的雷暴系统识别结果，也能为雷暴灾害天气的预报及雷电预警模型的训练提供有效的知识库支撑。

Abstract:

In order to deeply explore the benefits of a VLF/LF 3D lightning monitoring network in lightning proximity warning, this paper proposes a multiple algorithm fusion thunderstorm system identification technique based on VLF/LF lightning data. First, the OPTICS clustering algorithm was improved to make it parameter-adaptive, i.e., the initial input parameters Min_ρ was calculated in real time by using the double total t-test method and the spatial distribution characteristics based on VLF/LF lightning data, making it adaptive to the lightning data to be processed. Then it was applied in combination with the K-Means clustering algorithm, whose advantages was able to calculate the cluster center iteratively to improve the defect that the center of thunderstorms cannot be accurately located by the OPTICS algorithm. Tested by thunderstorm weather processes on different scales, this technique effectively solves the defects of the conventional density clustering algorithms, e.g., the OPTICS algorithm can neither be applied to VLF/LF lightning data clustering processing nor distinguish adjacent thunderstorms with little difference in density. It identifies arbitrary scale shape thunderstorms and accurately calculates the centroid. Therefore this technique provides better thunderstorm system identification results for thunderstorm tracking and trend extrapolation compared with traditional clustering techniques, and also serves as an effective knowledge base support for severe thunderstorms forecasting and lightning warning model training.