Abstract: In the existing algorithms of piecewise linear representation for time series，local extremum points always are regarded as segment points，but the changement of time series states can’t be depicted completely only by them.So according to the idea of piecewise linear of time series，new segment points are brought in after local extremum points are fixed，which can acquire high fitting precision.The difference of slopes is used to reflect state changements and by comparing with slopes of adjacent sequence segments，new segment points can be confirmed.Lastly this method leads to much less fitting error and better practicability.

Key words: piecewise linear representation, extremum points, slope, fitting error, data mining

摘要： 现有的时间序列分段线性表示一般把局部极值点作为分段点，但是局部极值点并不能完全刻画时间序列的状态变化，根据时间序列线性分段的基本思想，提出在确定局部极值点的基础上引入斜率差值大的分段点，以便得到较高的拟合精度。新的分段点通过比较相邻序列段的斜率实现，斜率差值越大，该点的状态改变越明显。实验证明该方法拟合误差小，有很好的实用性。

关键词: 分段线性表示, 极值点, 斜率, 拟合误差, 时间序列挖掘