基于动态回归模型的组合模型研究

发布时间：2018-04-01 02:34

本文选题：时间序列　切入点：ARIMA模型　出处：《大连海事大学》2016年硕士论文

【摘要】：时间序列分析方法是概率和数理统计学科应用领域中的一个分支,根据数理统计学的基本原理,可对获得的新数据进行实时调整,从而确定模型的参数,提高模型的预测精度。但是,由于受到各个因素的影响,它们只能对数据的整体趋势进行分析和预测,并不能进行全面地分析。本文是基于ARIMA模型和动态回归模型的组合模型研究,结合1994年到2013年的浙江省入境旅游人数数据和《基于ARIMA的组合模型问题研究》数据进行深入的研究,体现组合模型更高的预测效率。首先,深入研究了时间序列模型的相关理论和预测的相关知识,透彻地分析了ARIMA模型和动态回归模型的建模步骤,给出了在模型建立过程中常见问题解决方法。例如异常点处理办法,对于数据是否能够建立模型,以及建立模型的合理性给出依据。其次,对给出的数据建立ARIMA模型和动态回归模型,验证建立模型的合理性,并对数据进行了预测,根据相对百分比误差、平均绝对误差和平均绝对百分比误差评价标准进行评价。将动态回归模型与指数平滑模型进行对比,体现出动态回归模型具有更高的预测效率。最后,本文选出4种组合模型的方法,进行模型的选取工作。利用3组组合数根据SSE、MSE、MAE、MAPE和MSPE5个指标进行评价,并与单一模型进行了对比,证明了组合模型在模型选取方面的突破性成绩。同时,对4种模型进行横向的对比,发现不同组的组合模型的选取不是统一的,需要进一步研究,选出最佳的组合模型。
[Abstract]:The method of time series analysis is a branch of the application field of probability and mathematical statistics. According to the basic principle of mathematical statistics, the new data can be adjusted in real time, and the parameters of the model can be determined. Improve the prediction accuracy of the model. However, because of the influence of various factors, they can only analyze and forecast the overall trend of the data. This paper is based on the combination model of ARIMA model and dynamic regression model. Based on the data of inbound tourist population of Zhejiang Province from 1994 to 2013 and the data of "combination Model based on ARIMA", this paper makes a deep study to show that the combination model has higher forecasting efficiency. The related theories of time series model and the related knowledge of prediction are deeply studied, the modeling steps of ARIMA model and dynamic regression model are thoroughly analyzed, and the solutions to common problems in the process of modeling are given, such as the method of dealing with outliers, The basis of whether the data can be established and the rationality of the model are given. Secondly, the ARIMA model and the dynamic regression model are established to verify the rationality of the model, and the data are predicted. According to the relative percentage error, average absolute error and average absolute percentage error, the evaluation criteria are evaluated. The dynamic regression model is compared with the exponential smoothing model, which shows that the dynamic regression model has higher prediction efficiency. In this paper, four methods of combinatorial models are selected to select the models. The three groups of combination numbers are used to evaluate SSEMS-MSE MAEMAE MAPE and MSPE5, and the results are compared with those of the single model. It is proved that the combination model is a breakthrough in the selection of the model. At the same time, the horizontal comparison of the four models shows that the selection of the combination model with different groups is not uniform, and needs further study to select the best combination model.
【学位授予单位】：大连海事大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：O212.1

【相似文献】