本文为芬兰于韦斯屈莱大学（作者：Jouni Helske）的毕业论文，共87页。

今天收集的大量数据都是以时间序列的形式出现的。为了根据时间序列预测做出真实的推断，除了点预测之外，还应确定预测间隔或其他不确定性度量。不确定性的多种来源往往被忽视，因为正确核算这些不确定性是非常复杂的。本文对这些问题进行了回顾，并提出了一些新的解决方案。提出了一种高效、灵活的时间序列预测框架，可以将多种类型的传统时间序列与其他模型相结合。

A large amount of data collected today isin the form of a time series. In order to make realistic inferences based ontime series forecasts, in addition to point predictions, prediction intervalsor other measures of uncertainty should be presented. Multiple sources of uncertaintyare often ignored due to the complexities involved in accounting themcorrectly. In this dissertation, some of these problems are reviewed and somenew solutions are presented. A state space approach is also advocated for anefficient and flexible framework for time series forecasting, which can be usedfor combining multiple types of traditional time series and other models.

1. 引言

2. 连续计数数据的时间序列模型

3. 模型的未知参数

4. 预测中的其他问题

更多精彩文章请关注公众号：