考虑已有的灰色预测模型主要能对指数型发展系统或幂函数型发展系统进行模拟预测,本文构建了一种不仅能够模拟指数型和幂函数型的发展系统,并且能够体现出二者之间的相互作用关系的离散灰色幂模型;并针对初始条件对离散灰色幂模型模拟精度的影响,首先给出了离散灰色幂模型的建模步骤,然后以平均相对误差最小化为目标、参数之间的关系为约束条件,构建了离散灰色幂模型初始条件的优化模型,实现对离散灰色幂模型初始条件的优化。结果表明,优化的离散灰色幂模型使得平均相对误差在理论上达到了最小化,其模拟精度和预测精度都高于传统模型。最后,通过中国网络购物人数数据预测和仿真数据分析,说明了本文优化方法的有效性和适用性。
A large number of practical systems have the characteristics of incomplete information, it is prevalent in the real world of the small sample, and the uncertainty of poor information system provides a very rich resource for the study of gray system theory. Grey prediction model has provided a useful tool for the small sample data predict. But, the existing research suggests that grey forecasting model can better simulate exponential function changes system and power function system, but the existing prediction model cannot better reflecting the exponential and power function combined effects to the system, so it is difficult to apply these model to predict the small sample data which effect by complex multifactorial impact. Based on this consideration, a new grey forecasting model-discrete grey power model is proposed.The new model add one time power function in the existing discrete grey model, and it also allows the power parameters by endue any value. So, the time-responsive features of this model can reflect the exponential and power function changes system, and includes the interaction characteristics of power function and exponential changes in the system. Based on the new model, taking into account the effect of the initial condition in the discrete grey power model, two optimization models are constructed with the objective of minimum average relative error, the constraints of relationships between parameters in order to optimize the initial condition. The example of online shopper from 2006 to 2012 in China is used to compare the simulation and prediction results of GM(1,1) model, discrete GM(1,1) model and the new method, the results show that the optimized model has better simulation and prediction accuracy than other two grey models. Additionally, in order to test adaptive of the new model, under the conditions of a given power exponent, six kinds of situations data Low-growth data、Medium-growth data、Rapid-growth data、Volatility data、Disturbance-growth data and Attenuation data are genercated by using randomly generated method, and the new model is used to simulate and predict the rand data, the results also show that the new model has better stability, which is further illustrated the validity and applicability of the new model. Therefore, the new prediction model constructed in this paper not only enriches the theory of grey forecast model system, but also provides a more rich set of tools for the prediction of small sample system under the combined effects of multiple factors.
[1] Deng J L.Introduction of grey system[J]. The Journal of Grey System, 1989,1(1):1-24.
[2] Liu Sifeng, Lin Yi. Grey systems theory and applications[M]. Berlin:Springer-Verlag, 2010.
[3] 刘思峰,邓聚龙.GM(l,l)模型的适用范围[J].系统工程理论与实践,2000, 20(5):121-124.
[4] 党耀国,王正新,刘思峰.灰色模型的病态问题研究[J].系统工程理论与实践,2008, 28(1):156-160.
[5] 熊萍萍,党耀国,姚天祥,等.灰色Verhulst模型背景值优化的建模方法研究[J].中国管理科学,2012,20(6):154-159.
[6] 王道平,张学龙,赵相忠.具有灰色随机动态特征的供应链牛鞭效应的鲁棒性分析[J].中国管理科学,2013,21(1):57-62.
[7] Xie Naiming, Liu Sifeng. Discrete grey forecasting model and its optimization[J]. Applied Mathematical Modeling, 2009, 33(2):1173-1186.
[8] 王正新. 含可变参数的缓冲算子与GM(1,1)幂模型研究.南京:南京航空航天大学,2010.
[9] 王正新,党耀国,赵洁珏.优化的GM(1,1)幂模型及其应用[J].系统工程理论与实践, 2012,32(9):1973-1978.
[10] 王正新,党耀国,练郑伟.无偏GM(1,1)幂模型及其应用[J].中国管理科学,2011,19(4):144-151.
[11] 王正新,党耀国,刘思峰.GM(1,l)幂模型的病态性[J]. 系统工程理论与实践, 2013,33(7):1859-1866.
[12] 王正新.基于傅立叶级数的小样本振荡序列灰色预测方法[J]. 控制与决策,2014,29(2):270-274.
[13] Wang Jianzhou, Zhu Suling, Zhao Weigang, et.al. Optimal parameters estimation and input subset for grey model based on chaotic particle swarm optimization algorithm[J].Expert Systems with Applications,2011,38(7):8151-8158.
[14] Zhao Ze, Wang Jianzhou,Zhao Jing,et al. Using a grey model optimized by differential evolution algorithm to forecast the per capita annual net income of rural households in China[J].Omega, 2012,40(5):525-532.
[15] Evans M. An alternative approach to estimating the parameters of ageneralized Grey Verhulst model:An application to steel intensity of use in the UK[J]. Expert Systems with Applications,2014,41(4):1236-1244.
[16] Dang Yaoguo, Liu Sifeng. The GM models that x(n) be taken as initial value[J]. Kybernetes, 2004, 33(2):247-255.
[17] Wang Yuhong, Dang Yaoguo, Li Yueqing,et.al. An approach to increase prediction precision of GM(1,1) model based on optimization of the initial condition[J]. Expert Systems with Applications, 2010, 37(8):5640-5644.
[18] Li D C, Lin L S. A new approach to assess product lifetime performance for small data sets[J].European Journal of Operational Research,2013,230(2):290-298.
[19] Zhang Yongjun, Zhang Xitian, Li Qinhao, et.al. Gray theory based energy saving potential evaluation and planning for distribution networks[J].Electrical Power and Energy Systems, 2014,57:298-303.
[20] Piotr F, Bellegem V, Sachs R. Forecasting non-stationary time series by wavelet process modeling[J].The Institute of Statistical Mathematics, 2002,55(4):737-764.
[21] 王革丽,杨培才,毛宇清. 基于支持向量机方法对非平稳时间序列的预测[J].物理学报,2008,57(2):714-719.
[22] Yang Peicai, Wang Geli, Bian Jianchun, et al. The prediction of non-stationary climate series based on empirical mode decomposition[J]. Advances in Atmospheric Sciences, 2010,27(4):845-853.
[23] Li D C, Chang C J, Chen C C, et.al. A grey-based fitting coefficient to build a hybrid forecasting model for small data sets[J]. Applied Mathematical Modeling,2012,36(10):5101-5108.
[24] 刘思峰,曾波,刘解放,等.GM(1,1)模型的几种基本形式及其适用范围研究[J].系统工程与电子技术,2014, 36(3):501-508.
