聚类算法在四个超级单体强对流过程雷电预警中的应用

Application of Clustering Algorithm in Lightning Early Warning for Four Supercell Strong Convective Processes

选取2020年5月3日邢台出现四个超级单体的强对流过程,以6分钟的间隔对三维闪电数据进行划分,使用DBSCAN聚类算法删除离散点,使用K-means聚类算法对三维闪电数据进行聚类分析。选取轮廓系数最大的K值,并与雷达回波拼图数据进行对比,识别四个超级单体并计算聚类中心和聚类最大半径,使用趋势外推法对四个超级单体的运动轨迹进行预测。分析表明:DBSCAN聚类算法可以有效删除离散点,操作性强;四个超级单体的K-means算法聚类中心和30 dBZ以上的强回波区域一致性较好,可以获取聚类中心运动轨迹和聚类最大半径;使用临近三个时次数据进行趋势外推,MSE最小,该方法对雷电预警信号发布有参考价值。

It selects four supercell strong convective processes that occurred in Xingtai on May 3,2020,the three-dimensional lightning data is divided at 6 minutes intervals.The DBSCAN clustering algorithm is used to remove discrete points,and the K-means clustering algorithm is used to perform clustering analysis on the three-dimensional lightning data.It selects the K value with the highest contour coefficient and compare it with the radar echo mosaic data to identify four supercells and calculate the clustering center and clustering maximum radius.Trend projection is used to predict the motion trajectories of the four supercells.Analysis shows that the DBSCAN clustering algorithm can effectively delete discrete points and has strong operability.The K-means algorithm for four supercells has good consistency in clustering centers and strong echo areas above 30 dBZ,and can obtain the movement trajectory of clustering centers and the maximum radius of clustering.Using data from nearly three time periods for trend projection,with the minimum MSE,this method has reference value for the release of lightning early warning signals.