基于多源数据的甲醇产品价格预测与可视分析

发布时间：2018-09-01 13:45

【摘要】：随着经济的不断发展,我国在短短时间内迅速发展成工业大国,甲醇在中国工业发展中扮演着重要角色。甲醇产品价格随时间变化,具有不稳定性和波动性,受季节因素、月度产量、国家宏观经济等因素的影响,受到国内外的高度关注。传统的预测方法主要基于专家经验和统计学等方面,很难对波动的价格进行准确有效地预测。大宗商品关乎国家战略利益,经济的兴衰,越来越多的专家投身于甲醇产品的预测中。针对多维因子预测,本文结合了统计学与专家经验对这些因素进行筛选,提出一种融合网络情感值和专家经验值等多源数据的预测模型。首先,对历史数据进行预处理和相关性分析,选择适当预测模型对其进行预测,同时比较预测误差,建立最优GARCH和ARMA相结合的长期预测模型。其次,通过挖掘甲醇产品的相关网络数据信息,对其进行预处理、统计分析,构造针对甲醇产品的情感词典,获取网络数据情感值。通过长时间的预测,设计各个行业的专家调查问卷,量化调查问卷并获得专家经验值。最后,结合网络数据情感值和专家经验值模拟新的预测模型,并且评估原模型和新模型的短期和长期预测误差。进一步地,基于甲醇多源数据的预测,设计了交互性强、具有层次结构的可视化分析系统。由于甲醇价格受到多维因素的影响,本系统采用时间序列图来反映历史甲醇价格走势和情感走向,并且有效地将多维数据的预测结果进行展示；通过动态饼状图来呈现预测误差；通过搜索框对历史数据情感文本进行有效的搜索；同时专家也可以通过用户交互界面将自己的情感值、观点融入到本系统中,通过量化数据动态更改预测结果,显示出融合专家经验的预测值。通过动态时间序列图和交互技术等可视化分析技术,解决数据的预测、显示。通过实验跟踪分析以及用户使用调查,发现该方法提高预测准确性,本系统具有较强的实用性。
[Abstract]:With the development of economy, China has developed rapidly into a large industrial country in a short time. Methanol plays an important role in the industrial development of China. The price of methanol products varies with time, which is unstable and fluctuating, which is influenced by seasonal factors, monthly output, national macroeconomic and so on, and is highly concerned at home and abroad. The traditional forecasting methods are mainly based on expert experience and statistics, so it is difficult to predict the fluctuating price accurately and effectively. Commodities are a matter of national strategic interest, economic rise and fall, more and more experts are engaged in methanol product forecast. Aiming at multidimensional factor prediction, this paper combines statistics and expert experience to screen these factors, and puts forward a prediction model which combines network emotion value with expert experience value and other multi-source data. Firstly, the preprocessing and correlation analysis of historical data are carried out, and the appropriate prediction model is selected to predict it. At the same time, the prediction error is compared, and a long-term prediction model combining optimal GARCH and ARMA is established. Secondly, through mining the related network data information of methanol product, preprocessing and statistical analysis, the emotion dictionary for methanol product is constructed, and the emotional value of network data is obtained. Through long-term prediction, the questionnaire of experts in various industries is designed, the questionnaire is quantified and the experience of experts is obtained. Finally, the new prediction model is simulated with the emotional value of network data and expert experience value, and the short-term and long-term prediction errors of the original model and the new model are evaluated. Furthermore, based on the prediction of methanol multi-source data, a visual analysis system with high interactivity and hierarchical structure is designed. Because the price of methanol is influenced by multi-dimension factors, this system adopts time series diagram to reflect the trend of historical methanol price and emotional trend, and effectively displays the forecast results of multidimensional data. The prediction error is presented by dynamic pie chart; the emotional text of historical data is searched effectively by searching box; at the same time, experts can integrate their emotional values into the system through the user interface. By changing the prediction result dynamically by quantifying data, the prediction value of fusion expert experience is shown. Dynamic time series diagram and interactive technology are used to solve the problem of data prediction and display. It is found that the method can improve the accuracy of prediction and the system has strong practicability.
【学位授予单位】：华东师范大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：F767

【相似文献】