基于数据流学习的国际石油价格预测研究

发布时间：2018-01-26 06:21

本文关键词： 石油价格预测模型数据流学习机器学习　出处：《中国地质大学(北京)》2016年博士论文　论文类型：学位论文

【摘要】：石油是地球上最重要的能源之一。目前,全球有接近三分之一的能源消耗源于石油。国际石油价格对全球的环境和经济有重大影响。无论是工业界、政府,还是个人,对石油价格都很关注。预测国际石油价格已成为学术界和工业界的热点研究话题。迄今为止,预测石油价格的方法已有很多。但是,由于国际石油价格波动频繁,准确预测国际石油价格仍极具挑战。本文提出了一种基于数据流学习的新方法来预测国际石油价格。伴随着新的国际石油价格数据,基于数据流学习的国际石油价格预测模型可时时更新。因此,油价预测模型会不断演变,能更好地捕捉到国际石油价格的变换模式。基于数据流学习的新方法既弥补了传统计量经济学方法的不足,即无法捕捉到国际石油价格的非线性特征;也克服了传统机器学习方法的缺陷,即无法有效地预测非平稳的国际石油价格。因此,这种新方法能够更加准确地预测未来的国际石油价格。另外,这种新方法能够有效地处理连续不断的国际石油价格数据,即便这些数据产生的速度极快。对于每个新的数据点,仅需很短的常量时间和定量内存就可更新基于数据流学习的国际石油价格预测模型,而无需使用所有的新旧训练数据点重新训练预测模型。本文将数据流学习方法应用到预测国际石油价格中,这在国内乃至国际是首创的。这一创新得到了实验的有效支持。在评估基于数据流学习的国际石油价格预测模型性能的过程中,本文将其与不变化模型、人工神经网络模型和一种前沿的混合预测模型做了比较。本文采用了多时间段定量分析方法。预测时间段包括一个月、三个月、六个月、九个月和十二个月。本文采用了两种性能指标:一种是均方预测误差,另一种是方向性精度。实验结果表明,对于1992年1月至2012年9月美国西德克萨斯中质原油现货价格和美国进口原油的炼油成本价格这两种国际石油价格,基于数据流学习的国际石油价格预测模型在绝大多数时间段(一个月、六个月、九个月和十二个月)预测精度最高(均方预测误差最低,方向性精度最高)。本文为能源预测领域提出了一种新的强有力的计算和分析方法。
[Abstract]:Oil is one of the most important sources of energy on the planet. At present, nearly 1/3 of the world's energy consumption comes from oil. International oil prices have a significant impact on the global environment and economy. Both industry and government. The prediction of international oil prices has become a hot topic in academia and industry. Up to now, there are many methods to predict oil prices. Due to frequent fluctuations in international oil prices. It is still a challenge to predict the international oil price accurately. This paper proposes a new method based on data flow learning to predict the international oil price, which is accompanied by new international oil price data. The international oil price forecasting model based on data flow learning can be updated from time to time. Therefore, the oil price forecasting model will evolve continuously. The new method based on data stream learning not only makes up for the shortcomings of the traditional econometrics method, but also can not capture the nonlinear characteristics of the international oil price. It also overcomes the shortcomings of the traditional machine learning method, that is, it can not effectively predict the non-stationary international oil prices. Therefore, this new method can more accurately predict the future international oil prices. This new approach can effectively process a continuous flow of international oil price data, even if they are generated at an extremely rapid rate. For each new data point. The model of international oil price prediction based on data flow learning can be updated in a very short amount of constant time and quantitative memory. Without using all the old and new training data points to re-train the prediction model, this paper applies the data stream learning method to predict the international oil price. This is the first at home and even internationally. This innovation has been effectively supported by experiments. In the process of evaluating the performance of international oil price prediction model based on data flow learning, this paper will not change the model. The artificial neural network model is compared with a hybrid prediction model. In this paper, a multi-time quantitative analysis method is used. The prediction period includes one month, three months and six months. For nine months and twelve months, two performance indexes are used: one is mean square prediction error, the other is directivity accuracy. The experimental results show that. From January 1992 to September 2012, the spot price of West Texas Intermediate crude oil and the refining cost price of American imported crude oil are two kinds of international oil prices. The international oil price prediction model based on data flow learning has the highest prediction accuracy in most time periods (one month, six months, nine months and twelve months) (mean square prediction error is the lowest). The directivity accuracy is the highest. This paper presents a new powerful calculation and analysis method for energy prediction.
【学位授予单位】：中国地质大学(北京)
【学位级别】：博士
【学位授予年份】：2016
【分类号】：F416.22;F764.1

【相似文献】