【作者】 王会琦； 马洪； 周永道；

【Author】 WANG Hui-Qi MA Hong ZHOU Yong-dao (Mathematic College, Sichuan University, Chengdu 610064, China)

【机构】 四川大学数学学院；

【摘要】 本文首先把一般的AR模型转化成通信信号输入输出的递推矩阵形式,然后利用通信系统的观点,通过收集各个时延的预测分量信息,设计了一种对异常点有一定免疫性的时间序列RAKE预测模型。这种基于信息融合思想得到的预测模型在并行与分布式计算中具有良好的性能。最后通过理论分析和计算机模拟结果表明,这种RAKE预测模型效果在不出现异常点时与最佳线性预测效果接近,但当时间序列中混有异常点时,RAKE预测效果更优。本文更大的特点在于沟通了一般经济时间序列预测与电子通信系统信号检测的桥梁。更多还原

【Abstract】 This paper transforms the ordinary AR model into matrix form of communication signals I/O. And then we design a time series RAKE forecasting model which is immune to outliers by collecting forecasting information in each time delay channel. This forecasting model based on information fusion will have good performance in parallel and distributed computation. Finally, the theories analysis and the computer simulations also demonstrate that the RAKE forecasting model and common forecasting model have the same performance in the time series without outliers; but the RAKE forecasting model performs better in the time series with outliers.更多还原