基于时间序列分析的汽车销量预测研究

发布时间：2018-01-01 04:43

本文关键词：基于时间序列分析的汽车销量预测研究　出处：《合肥工业大学》2017年硕士论文　论文类型：学位论文

【摘要】：中国已经成为全球最大的汽车生产和汽车消费市场。汽车工业的发展带动了诸多周边产业的发展,对于拉动内需和促进出口增长都具有很大的积极意义。因此准确地预测汽车销量有助于政策制定者从宏观上整体掌控市场发育与成长态势,有助于汽车制造厂商从微观上研究市场行情以制定营销策略。现有汽车销量预测研究大多面向整体汽车市场而非单一汽车品牌,而且缺乏对网络大数据的深入挖掘,忽略了大数据对于汽车销量的预测能力。针对上述问题,提出一种基于网络大数据和传统统计学时间序列分析的考虑品牌情感的汽车销量预测BOAR模型,对于单一品牌,在考虑该品牌不同时间窗历史销量基础上,结合用户在线评论中挖掘的用户情感值进行汽车销量预测。对多个汽车品牌的实验表明,综合历史同期销量、前期销量和品牌情感的BOAR模型平均预测误差为5.93%,比自回归模型降低8.59个百分点,BOAR模型可以准确预测单一汽车品牌的销量,具有更好的稳定性。与此同时,我们意识到BOAR模型及传统的时间序列分析模型都存在以下问题:(1)汽车销量影响因素众多,不同品牌的汽车其销量的影响因素本身可能也是不一致的,对于不同的汽车品牌采用一致的影响因素作为预测的解释变量是不合理的(2)需要事先假定影响因素和销量之间呈线性关系,对于高度非线性的关系难以准确用数学模型拟合。因此进一步提出一种通用的销量预测MISF模型,基于MARS变量选择过程和BP神经网络相结合的方式预测单一汽车品牌的月度销量。实验结果表明MISF模型的预测误差平均为4.04%,比BOAR模型进一步降低了1.49个百分点。这也验证了变量选择和神经网络在汽车销量预测研究中具有重要作用。本文提出的BOAR模型和MISF模型准确预测了单一汽车品牌粒度的月度销量,研究结果不仅对于汽车领域的销量预测研究具有一定的借鉴意义,同时可以为汽车制造厂商生产规划和控制提供更有效的决策支持。
[Abstract]:China has become the world's largest auto production and automobile consumption market. The development of automobile industry has driven the development of many peripheral industries. It has great positive significance for stimulating domestic demand and promoting export growth. Therefore, accurate prediction of car sales will help policy makers to control the overall market development and growth situation from the macro perspective. It is helpful for automobile manufacturers to study the market from the micro level in order to formulate marketing strategy. Most of the existing research on automobile sales forecast is aimed at the whole automobile market rather than a single automobile brand. And the lack of the network big data in-depth mining, ignoring big data's ability to predict car sales. In response to the above problems. This paper presents a BOAR model of automobile sales forecasting based on network big data and traditional statistical time series analysis. For a single brand, considering the historical sales of the brand in different time windows. Combined with the user emotional values mined in online reviews to predict the car sales. The experiments on multiple automobile brands show that the integrated historical sales volume in the same period. The average prediction error of pre-sales and brand emotion BOAR model is 5.93%, which is 8.59% lower than that of autoregressive model, which can accurately predict the sales volume of a single car brand. At the same time, we realized that both the BOAR model and the traditional time series analysis model have the following problems: 1) there are many factors affecting car sales. The factors affecting the sales of different brands may also be inconsistent. For different automobile brands, it is unreasonable to use consistent influencing factors as predictive variables.) it is necessary to presuppose linear relationship between influencing factors and sales volume. For highly nonlinear relationships, it is difficult to fit the mathematical model accurately. Therefore, a general sales forecasting MISF model is proposed. Based on the combination of MARS variable selection process and BP neural network, the monthly sales volume of a single automobile brand is predicted. The experimental results show that the average prediction error of the MISF model is 4.04%. It is proved that variable selection and neural network play an important role in the prediction of automobile sales volume. The BOAR model and MISF model proposed in this paper are more important than the BOAR model. Type A accurately predicts the monthly sales volume of single automobile brand granularity. The results can not only be used for reference in the field of automobile sales prediction, but also provide more effective decision support for the production planning and control of automobile manufacturers.
【学位授予单位】：合肥工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F426.471

【相似文献】