The Application of Grey GM(1, 1) Forecasting Model with Neural Network Residual Modification to Oil

Consumption

Qiu-Ping WANG1, Jian-Bo YAN2

1)

School of Sciences, Xi’an University of Technology, Xi’an 710054 (E-mail: wqp566@yahoo.com.cn)

2)

Zhangjiakuo Bridge East District Wu-yi Road Office, Hebei 075000

Abstract—The annual total oil consumption time series is a non-steady time series whose developmental changes have trends of increase and stronger random fluctuation. Using the Grey GM(1, 1) Model and BP neural network optimized by L-M algorithm, this paper builds grey BP Neural Network combination model. This model has combined the advantages of grey forecasting and BP neural network forecasting, it has overcomed the influence by little raw data and high data fluctuation to precision of forecasting, and it has also enhanced the self-adaptability of forecasting. At last, the validity and applicability of the model is demonstrated by a simulation of annual oil consumption .

Keywords—GM(1,1) model; Levenberg-Marquardt algorithm; BP neural network; Grey residual sequence; Oil consumption; forecasting

灰色神经网络模型在石油消费预测中的应用

王秋萍1，闫建波2

1)2)

西安理工大学理学院 西安710054

张家口桥东区五一路办事处 河北 075000

摘 要 石油消费总量的时间序列具有增长趋势性和较强的随机波动性。本文利用灰色GM(1, 1)模型与L-M算法优化的BP神经网络，建立了BP神经网络残差修正的灰色组合模型。此组合模型兼有灰色预测和BP神经网络预测的优点，既克服了原始数据少，数据随机波动大对预测精度的影响，也增强了预测的自适应性。最后通过对石油消费总量的仿真验证了模型的有效性及可应用性。

关键词 GM(1, 1)模型；L-M算法；BP神经网络；灰色残差序列；石油消费；预测

1．引言

邓聚龙教授于1982年提出的灰色系统理论，已被成功用于工程、经济、物理控制等领域[1]。为了提高GM(1, 1)模型的预测精度，邓聚龙教授提出了残差辩识的方法[2]。残差辨识就是将预测值与原始值之差再建立GM(1, 1)模型，称之为残差GM(1, 1)模型，并利用残差的预测值修正原来的预测值，以提高预测精度。然而残差序列不总是适合直接建立GM(1, 1)模型，比如残差序列中存在负项时。邓聚龙采用高阶残差[2]的方法来实现残差项的非负，以便建立GM(1, 1)模型。不少学者已经证实，利用灰色模型建立的残差修正模型是不能保证适用于解决具有波动变化规律的时间序列[3]。石油作为一种战略资源，其消费需求既

…… 此处隐藏3881字 ……

[6] 刘思峰，党耀国，方志耕.灰色系统理论及其应用[M].北京：

科学出版社，2004.

[7] 飞思科技产品研发中心.神经网络理论与MATLAB7实现

[M].北京：电子工业出版社，2006.

[8] Junfeng Li, Aiping Yang , Wenzhan Dai. Modeling Mechanism

of Grey Neural Network and Its Application[C]. Proceedings of 2007 IEEE International Conference on Grey Systems and Intelligent Services, Nanjin, 2007: 404-408.