组合模型在我国社会消费品零售总额预测中的应用研究

发布时间：2018-05-12 16:50

本文选题：社会消费品零售总额 + 指数平滑方法　；参考：《兰州交通大学》2017年硕士论文

【摘要】：社会消费品零售总额是衡量我国人民消费水平的重要指标,也是影响国民经济的重要因素。因此,研究我国社会消费品零售总额的发展趋势对于我国经济的发展具有重要的意义。社会消费品零售总额是一组时间序列时序,根据经典的时间序列预测理论,本文的具体工作包括:首先,构建了X-12-ARIMA模型(加法和乘法),同时对这两个模型进行比较分析,结果表明X-12-ARIMA乘法模型的拟合效果高于X-12-ARIMA加法模型,乘法模型的MAPE较小,拟合程度较高。最后对模型AR、MA、ARIMA进行了比较,结果表明X-12-ARIMA乘法模型的拟合程度较高,对预测具有一定的优势。其次,构建了状态空间模型下的指数平滑方法(ETS方法),对状态空间模型下的指数平滑理论进行了系统的研究,并给出常用指数平滑方法的点预测推导形式。通过实证分析,最优拟合模型为ETS(M,MD,M),结果表明ETS模型的拟合程度和预测精度都比较高,模型的MAPE较小,模型对原始时序的季节性、趋势性和周期性因素拟合较好。ETS模型能够充分的剔除原始时序中所包含的各项信息。再次,根据单项预测模型的拟合效果,本文构建了组合预测模型。并在此基础上引入了两种优化权重系数算法,分别为非线性规划方法和混沌粒子群优化算法。根据实证分析,结果表明基于混沌粒子群算法权重优化的组合模型拟合程度更高,且拟合效果均高于单项预测方法。应用混沌粒子群算法来优化权重系数,较大程度上提高了模型的拟合精度和预测精度。组合模型充分的应用了各个单项模型的优点,同时将单项预测模型的优势结合到了一起。最后,根据上述单项模型和组合模型的研究结果,进行分析比较结果表明基于混沌粒子群算法来优化权重系数的组合模型的拟合程度较高,对我国社会消费品零售总额的拟合预测程度较好,MAPE较小。并应用本文所建立的两类权重优化方法的组合模型对我国社会消费品零售总额时序进行了数据的拟合和预测对比分析,同时对未来的我国社会消费品零售总额时序进行了预测。综上所述,组合模型的拟合精度均高于单项预测模型的拟合精度,而在组合结构中,应用混沌粒子群算法对组合权重系数进行优化能够进一步提高组合模型的拟合精度,因此本文所建立的组合预测模型是有效的,具有一定的实用价值和指导意义。
[Abstract]:The total retail sales of consumer goods is an important index to measure the consumption level of Chinese people, and also an important factor affecting the national economy. Therefore, it is of great significance to study the development trend of retail sales of consumer goods in China. The total retail sales of consumer goods is a group of time series. According to the classical time series prediction theory, the specific work of this paper includes: firstly, the X-12-ARIMA model (addition and multiplication) is constructed, and the two models are compared and analyzed. The results show that the fitting effect of the X-12-ARIMA multiplication model is higher than that of the X-12-ARIMA addition model. The MAPE of the multiplication model is smaller and the fitting degree is higher. Finally, the comparison of the model ARGMAA and ARIMA shows that the fitting degree of the X-12-ARIMA multiplication model is high, and it has some advantages for prediction. Secondly, the exponential smoothing method under the state space model is constructed and the ETS method is constructed. The exponential smoothing theory under the state space model is systematically studied, and the point prediction derivation form of the commonly used exponential smoothing method is given. Through the empirical analysis, the optimal fitting model is ETS MMDM. The results show that the ETS model has higher fitting degree and prediction accuracy, the MAPE of the model is smaller, and the model is seasonal to the original time series. The trend and periodicity factors fit well. The ETS model can fully eliminate the information contained in the original time series. Thirdly, according to the fitting effect of the single prediction model, the combined prediction model is constructed in this paper. On this basis, two kinds of optimization weight coefficient algorithms are introduced, which are nonlinear programming method and chaotic particle swarm optimization algorithm. According to the empirical analysis, the results show that the combined model based on the weight optimization of chaotic particle swarm optimization has higher fitting degree, and the fitting effect is higher than the single prediction method. Chaotic particle swarm optimization algorithm is used to optimize the weight coefficient, which greatly improves the fitting accuracy and prediction accuracy of the model. The combined model fully utilizes the advantages of each single model and combines the advantages of the single prediction model. Finally, according to the research results of the above single model and combination model, the results of analysis and comparison show that the combination model based on chaotic particle swarm optimization has higher fitting degree. The fitting prediction of the total retail sales of consumer goods in China is better than that of MAPE. Using the combined model of the two kinds of weight optimization methods established in this paper, the time series of the total retail sales of consumer goods in China is fitted and compared, and the future time series of the total retail sales of consumer goods in China is forecasted. In conclusion, the fitting accuracy of the combined model is higher than that of the single prediction model. In the combinatorial structure, the optimization of the combined weight coefficient by using chaotic particle swarm optimization algorithm can further improve the fitting accuracy of the combined model. Therefore, the combined prediction model established in this paper is effective and has certain practical value and guiding significance.
【学位授予单位】：兰州交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F224;F724

【参考文献】