基于“序列相对贴近度” 的组合预测权值分配.pdfVIP

2009 12 D ecem ber 2009 35 12 Journal of Be ijing U n iversity o fA eronautics and A stronautics V o .l 35 N o12 1 吕永乐 郎荣玲 谈展中 ( , 100191 ) : 在时间序列的组合预测权值分配问题上, 为克服传统的均方误差倒数加权 熵权和最优化方法之不足, 从预测值序列与评价样本序列间的贴近性出发, 提出新方法综合衡 量单一参与模型的适用性, 并据此分配权值 详细给出了序列相对贴近度( SRND, Sequence Rel ative N earness Degree)及与之相关的 序列趋势关联度 和 尺度区间熵 的概念, 并提出基于 SRND 的权值分配方法 将 SRND 权值分配方法应用于航空发动机排气温度裕度参数时间序列 的联合自回归滑动平均模型函数系数自回归模型和径向基函数网络预测, 有效地提高了预测 准确度, 获得优于均方误差倒数加权和熵权方法的组合性能, 且运算量远小于最优化方法 : 时间序列分析; 组合预测; 建模; 性能 : O 23; TP 18 : A : 2009) W eight allocation of combination pred iction based on sequence re lative nearness degree LYong le Lang Rongling Tan Zhanzhong ( Schoo l ofE lectron ics and In form ation Eng ineering, Beijing Un iversity ofAeron autics and As tronau tics, B eijing 100191, Ch ina) Abstract: A im ing at the w e ight allocation problem s o f comb ination pred iction for a tim e series, a new m ethod w as proposed to evaluate the applicability o f the em ployed m ode ls and a llocate w eights, based on the nearness betw een the test sequence and the corresponding prediction value sequence, wh ich overcam e the shortages of ex isting m ethods such asm ean square erro r rec iproca lw e ight ( 1 /M SE ), entropy w eight and opti m iza tion w e igh.t The de fin itions of sequence relative nearness degree ( SRND), re lated sequence trend assoc i ation and sca le interval entropy w ere g iven and w ell d iscussed, as w ell as the w eight a lloca tion expre