多维时间序列突变点检测方法研究

Study on change point detection method of multidimensional time series

水文系列的变异性检验是非一致性水文频率分析过程中的重要环节,其中跳跃性变异是非一致性的重要表现形式,准确识别突变点对认识水文过程发生的变化及开展实际水文水资源工作具有重要意义。对于多维时间序列而言,可能会出现各单维变点不统一的情况,为此,将启发式分割算法(BG算法)与线性加权综合统计量结合,应用于多维水文时间序列的变异点检测,通过综合各维度的统计结果,确定最终的突变点。统计试验结果表明:该法有利于排除多维系统中虚假突变点或弱突变点的干扰,得到更为合理的检测结果。对汉江黄龙滩入库洪水极值系列变异性检测的应用结果表明,洪峰和最大7 d洪量的单变量检测突变点分别为1990年和1985年,通过综合统计量判定最终的突变点为1985年,为解决多维水文时间序列突变点的统一检测问题提供参考。

The variability test of hydrological series is an essential part of non-stationary hydrological frequency analysis,in which jumping variation is an important component of inconsistency.Accurate identification of change point is of great significance for understanding the changes of hydrological process and carrying out the practical work of hydrology and water resources.For multidimensional time series,it is possible that the single dimensional change points are not uniform.Therefore,in this paper,the heuristic segmentation algorithm(BG algorithm)is combined with the linear weighted comprehensive statistics to detect the change point in multidimensional hydrological time series,which determines the final change point by synthesizing the statistical results of each dimension.The statistical test results show that this method can eliminate the interference of false or weak change points in multi-dimensional system,and get more reasonable detection results.The application results of variability detection of Huanglongtan inflow flood extreme series in Hanjiang River show that,the change points of flood peak and maximum 7 d flood volume are 1990 and 1985 respectively,and the final change point is 1985 according to comprehensive statistics,which provides a reference for the unified detection of change points in multidimensional hydrological time series.