基于改进的DBSCAN方法和多项式拟合的雷电短时预测  

Lightning forecast based on the improved DBSCAN and polynomial fitting

作　　者：冯万兴[1] 朱晔[1] 郭钧天[1] 张晓庆[2] 刘娟[2] 

机构地区：[1]国网电力科学研究院,湖北武汉430074 [2]武汉大学计算机学院,湖北武汉430072

出　　处：《计算机工程与科学》2014年第10期2028-2033,共6页Computer Engineering & Science

基　　金：国家自然科学基金资助项目(61272274)

摘　　要：结合雷电数据自身特征改进DBSCAN方法,提出了一种基于DBSCAN和多项式拟合的雷电预测方法,提高预测的准确性。首先对某一时间段内的雷电数据按密度进行聚类并将每类所有雷电数据的平均坐标作为该类的中心点;然后在下一个时间段使用上一时间段的中心点作为初始选择点进行DBSCAN聚类,重复上述过程直到所有历史数据处理完毕,得到一系列不同时间段不同类别的雷电中心点;最后使用多项式拟合预测接下来的雷电可能发生的中心位置。对雷电监测网提供的雷电数据进行测试,结果表明,在数据充分的情况下,基于DBSCAN方法和多项式拟合的雷电预测准确率较令人满意,实际雷电中心点与预测中心点坐标误差约为0.1(±0.1)。Using the improved DBSCAN algorithm considering the original characteristics of the lightning data, a lightning forecast method based on the improved DBSCAN and polynomial fitting is pro- posed to improve the prediction accuracy. Firstly,the lightning data during a span of time is clustered according to the density, and the average coordinates of all of the lightning data are designated as the central points of the cluster. Secondly, the central points in the last span of time are chosen as the initial se lection points in the current span of time to perform DBSCAN clustering. The two-step procedure is iter- ated until all history data are processed. Finally, the polynomial fitting method is used to process the central location of each category. After dealing with the data provided by the Lightning Detection Net- work, the estimated and the observed performance data are presented and compared. The results are en couraging since the deviation between the actual data and the forecasted data is around 0.1 in most situations.

关 键 词：雷电预测 DBSCAN方法 多项式拟合 

分 类 号：TP274[自动化与计算机技术—检测技术与自动化装置]