基于压缩感知的国际油价预测方法研究

发布时间：2018-05-01 08:07

本文选题：原油价格预测 + 压缩感知　；参考：《北京化工大学》2015年硕士论文

【摘要】：国际原油价格自从1970年开始频繁波动。它不仅受到基本的供求影响,同时也受到许多其他因素的综合影响,包括天气、存货水平、经济增长、政治因素、心理期待甚至是一些非常规突发事件。多因素的影响导致原油价格序列呈现出非线性、非平稳性、季节性、不规则性能一系列复杂特征。在这种情况下,本文引入了一种新的基于压缩感知的人工智能预测方法来预测原油价格。具体来说,该方法引入了压缩感知的两种技术,分别为压缩感知去噪方法和稀疏分解方法并将其作为原油价格序列的预处理方法,然后基于这两种数据处理方法分别构造了两种模型。一种是基于压缩感知去噪的人工智能预测模型,另一种是基于稀疏分解的分解集成预测模型。基于压缩感知去噪的人工智能预测模型是基于去噪预测的思想,首先采用压缩感知去噪方法来对原始油价序列进行去噪的预处理,以减少噪声数据对人工智能预测方法建模效果的影响,然后使用智能预测算法对去噪后的数据进行建模和预测。基于稀疏分解的分解集成预测模型是基于分解集成预测的思想,首先根据原油价格序列所表现出来的多种特性,构造了一个过完备字典,并基于该字典将原油价格序列分解为不同的特征分量,然后使用前馈神经网络对分量进行建模和预测,最后将各个预测结果集成为最终的预测结果。本文对基于压缩感知的两种模型分别使用了WTI的日度原油价格和月度原油价格来进行实证分析,并得出以下结论：一方面,基于压缩感知的两种预测模型与基准模型比较时均获得最高的预测精度,表明基于压缩感知的预测方法的有效性；另一方面,在不同数据集预测中,该框架下的两种预测模型也都获得了最高的预测精度,验证了该方法的稳定性。此外,实证结果也表明该方法在预测具有复杂非线性特征时间序列方面特别有效。
[Abstract]:International crude oil prices have fluctuated frequently since 1970. It is affected not only by basic supply and demand, but also by many other factors, including weather, inventory level, economic growth, political factors, psychological expectation and even some unconventional emergencies. Because of the influence of many factors, crude oil price series presents a series of complex characteristics, such as nonlinear, non-stationary, seasonal and irregular. In this case, a new artificial intelligence prediction method based on compression perception is introduced to predict crude oil price. Specifically, this method introduces two kinds of compression sensing techniques, namely, compressed perception denoising method and sparse decomposition method, which are used as pretreatment methods of crude oil price sequence. Then two models are constructed based on these two data processing methods. One is an artificial intelligence prediction model based on compressed perceptual denoising, the other is a decomposed integrated prediction model based on sparse decomposition. The artificial intelligence prediction model based on compressed perceptual de-noising is based on the idea of denoising and forecasting. Firstly, the compressed perceptual de-noising method is used to pre-process the original oil price sequence. In order to reduce the influence of noise data on the modeling effect of artificial intelligence prediction method, the model and prediction of de-noised data are modeled and predicted by using intelligent prediction algorithm. The decomposition integrated prediction model based on sparse decomposition is based on the idea of decomposing integrated prediction. Firstly, an overcomplete dictionary is constructed according to the characteristics of crude oil price series. Based on the dictionary, the crude oil price series is decomposed into different characteristic components, and then the feedforward neural network is used to model and predict the components. Finally, the prediction results are integrated into the final prediction results. In this paper, we use WTI's daily crude oil price and monthly crude oil price to analyze the two models based on compression perception, and draw the following conclusions: on the one hand, The two prediction models based on compressed perception obtain the highest prediction accuracy when compared with the reference model, which indicates the validity of the prediction method based on compressed perception. On the other hand, in different data sets, The two prediction models under this framework also obtain the highest prediction accuracy and verify the stability of the method. In addition, the empirical results show that this method is especially effective in predicting time series with complex nonlinear characteristics.
【学位授予单位】：北京化工大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：F416.22;F764.1

【共引文献】