一种基于残差熵的动态模型检验方法

Dynamic model checking method based on residual entropy

针对时间序列的模型检验与定阶问题,构造了一种新的统计量“残差熵”。该物理量融合了随机序列的相关性和信息论中信源熵的思想,综合反映了残差中所残留的信息量。在正态假设的基础上推导了模型“残差熵”的估算公式,并给出了具体的应用方法。实例计算表明:“残差熵”可用于模型的检验和定阶,极小化残差熵是一种有效的模型检验准则,并与经典的模型检验准则做了性能比较。

For model checking and order determination, another statistical quantity named “residual entropy” is presented. The quantity fuses correlation of stochastic process and idea of entropy in information theory, it indicates the amount of information left in residual series. Then, the estimation expressions of residual entropy on normal distribution hypothesis are given, and its application approaches are illustrated. The result of experiment shows that residual entropy is a useful diagnostic quantity in time series modeling. Compared with the classical methods in performance, minimum residual entropy can be applied as an effective checking criterion.