许多金融、气象等数据都具有长相依性，且含有变点，分析这些数据时，必须先估计出其中的变点，否则会得出错误结论。从模型优化的角度精确分析数据中的多变点问题是目前统计学中的研究热点之一。本项目将利用lasso和带惩罚的模型选择两种模型优化方法研究长记忆时间序列各类变点的估计和在线监测问题。主要研究目标有：提出估计平稳长记忆时间序列均值、方差、时间趋势项及分整自回归模型结构多变点的lasso方法和带惩罚模型选择方法；提出推断长记忆参数变点个数、位置及变化方向的带惩罚模型选择方法；基于带惩罚的模型选择方法实现在线长记忆时间序列各类变点的监测和精确估计。最后利用提出的方法分析汇率等一些金融数据及青海省和桂林市的某些气象和水文数据中的变点问题。这些研究成果不仅具有重要的理论价值和应用价值，还突破了现有各类变点精确分析方法只能分析固定数据中突变点问题的局限，并解决了渐近方法分析多变点问题时遇到的一些困难。

For many financial and meteorological data exhibit long range dependencies and contain change points, we must estimates these change points before do other analysis. If not, we may got wrong conclusions. Exactly estimate the multiple change points in different types of data is now one of popular topic in statistics. This project will study multiple change points exact estimation and online monitoring problems in long memory time series via two different types of model optimization based methods, which are lasso and model selection method with penalty. The project will propose a lasso method and a model selection method with penalty to estimate the mean, variance, time trend change points in stationary long memory time series and the structure changes in fractional integration autoregressive models, and inference the number, location and direction of the long memory parameter's change points via the model selection method with penalty. The model selection method with penalty which will also be used to illustrate the change point monitoring and exact estimation problems in online long memory time series. Finally, we will apply our proposed methods to analyze the change points in financial data such as exchange rate data, the meteorological and hydrological data in Qinghai province and Guilin city. These results are not only very important both in theory and application, but also break the limitation of the available model optimization based methods which are restricted to analyze sharp change points in offline data, and overcome some drawbacks when using the asymptotic based method to analyze multiple change points.