摘要
针对观测数据时间序列,综合组合预测与投影寻踪学习网络的优点,提出一种新的预测模型。即采用静态预测法提取多组趋势项部分,自回归模型提取周期项部分,将它们都作为投影寻踪学习网络的输入部分,然后利用PPLN具有逼近复杂非线性函数的能力,通过网络学习与训练解决传统方法定权困难的问题。沉降预测的实验结果表明,与传统的曲线拟合法、变权重组合预测法相比较,该预测模型精度更高、具有实用性。
Aiming at observed time series data,a new forecast model which has the advantages of combination prediction and Pursuit Projection Learning Network(PPLN) is put forward.The model uses several static forecasting models to obtain the tendency part and uses the regression model for periodic part,then takes them as input signals of PPLN,calculates the weights between the models by PPLN which has the ability to approximate the complex nonlinear function.The model resolves a difficult problem of combination prediction.The experimental results show that new forecast model has a higher accuracy than traditional curve-fitting or combination forecasting method with variable weight,can be applied to other data prediction problems in the subject of surveying and mapping.
出处
《大地测量与地球动力学》
CSCD
北大核心
2010年第3期105-109,共5页
Journal of Geodesy and Geodynamics
基金
国家教育部重点项目(108085)
关键词
投影寻踪学习网络
组合预测
时间序列
变权重系数
沉降
pursuit projection learning network
combination forcasting
time series
variable weight coefficient
subsidence