组合预测模型研究及应用

发布时间：2018-08-20 14:27

【摘要】：随着社会经济的快速发展和科学技术的不断进步,预测方法正发挥着它不可替代的作用,在预测过程中,它所运用的方案、建立的模型、使用的算法等关键方面也在日益成熟。简单地讲,预测就是由已知的信息推测出未来可能发生或者不可能发生的事情,其本质上是一种对事物的发展和变化趋势的必要认知与理性分析的过程。到目前为止,预测方法的种类已经达到几百种,并且多数方法在实践中都得到了很好的应用。在预测过程中,由于每一种预测模型都是从不同角度去提取有用的信息,最后所采集的数据信息是完全不一样的。在关于预测的研究上,许多学者都为此付出了很多的努力,其中Bates.J.M.和Granger.C.W.J.两位学者的贡献最为突出。1969年,他们在大量研究分析每一个单项预测模型特性的基础上,提出了以某种合适的准则将这些模型合理有效的结合在一起的组合预测模型概念。换句话说,当一个模型的预测误差很大时,我们不能将其舍弃,而是提取这个模型的系统独立信息,并进行分析。通过多方面研究可以发现,现有的传统组合预测模型存在着一些不足,已经无法满足社会需求。由于各单项预测模型是不相同的,它采用的方法是选取适当的加权平均系数,使得这些模型按照设定好的配置方式进行组合,很显然,有时候这样的做法是不符合现实要求的。在整个预测模型中,如何求解加权平均系数是重中之重,论文旨在选取正确的方法进行求解,使得整个预测模型的预测精度得到显著提高。本文首先对课题背景及意义进行调查研究,了解了问题的产生并找到解决方案。其次,详细了解了相关预测技术理论,系统学习并掌握了确定组合预测模型权重的方式方法。随后,阐述了基于误差指标的线性和非线性组合预测模型,选取多个合适的实际案例分别进行仿真实验,进行了对比分析。并引入人工蜂群算法来确定最优组合预测模型的权重,其目的是解决确定权重时工作量大和无法保证所求权重恒大于零的问题。最后,通过引入IOWHA算子,很好地解决了传统最优不变权组合预测模型所存在的问题。以此为基础,阐述了二阶预测有效度和几何距离这两种概念,并尝试与IOWHA算子结合,组成了两种新型模型,并分别对其进行研究与说明,最后,结合实际案例,进行详细分析。在预测的实际运用中,我们所面临的预测对象很有可能是一个极为复杂的系统,所建立的模型具有很强的不确定性,会大大增加预测的风险。通过多个实例分析表明,本文所建立的组合预测模型是一个性能优良的模型,它不仅克服了传统组合预测模型的缺陷,而且提高了预测精度,能够很好的应用于实际。
[Abstract]:With the rapid development of social economy and the continuous progress of science and technology, the forecasting method is playing an irreplaceable role. In the process of prediction, the key aspects such as the scheme, the model and the algorithm used are becoming more and more mature. In short, prediction is a process of cognitive and rational analysis of the development and changing trend of things, which is to speculate from known information that may or may not happen in the future. Up to now, there are several hundred kinds of prediction methods, and most of them have been well applied in practice. In the process of prediction, because each prediction model extracts useful information from different angles, the data information collected is completely different. Many scholars, including Bates.J.M., have made a lot of efforts in the research of forecasting. And Granger.C.W.J. In 1969, on the basis of a large number of studies and analysis of the characteristics of each individual prediction model, they put forward the concept of combining these models reasonably and effectively with some appropriate criteria. In other words, when the prediction error of a model is very large, we can't abandon it, but we can extract the independent information of the model and analyze it. It can be found that the existing traditional combinatorial forecasting model has some shortcomings and can not meet the needs of the society. Because the individual prediction models are different, its method is to select the appropriate weighted average coefficient, so that these models are combined according to the set configuration mode. Obviously, sometimes this method does not meet the practical requirements. In the whole prediction model, how to solve the weighted average coefficient is the most important. This paper aims to select the correct method to solve the problem, so that the prediction accuracy of the whole prediction model can be improved significantly. In this paper, the background and significance of the subject are investigated, and solutions are found. Secondly, the related theory of prediction technology is understood in detail, and the method of determining the weight of combined prediction model is systematically studied and mastered. Then, the linear and nonlinear combined prediction models based on error index are described, and several suitable practical cases are selected to carry out the simulation experiments, and the comparative analysis is carried out. Artificial bee colony algorithm is introduced to determine the weight of the optimal combination prediction model. The purpose of the algorithm is to solve the problem that the weight can not be guaranteed to be equal to zero. Finally, by introducing IOWHA operator, the problem of traditional optimal invariant weight combination prediction model is solved well. On this basis, two concepts of second-order predictive effectiveness and geometric distance are expounded, and two new models are constructed by combining with IOWHA operator. Finally, combined with practical cases, two new models are analyzed in detail. In the practical application of prediction, the prediction object we are facing is probably a very complicated system, and the model established has strong uncertainty, which will greatly increase the risk of prediction. Through the analysis of many examples, it is shown that the combined prediction model established in this paper is a model with good performance. It not only overcomes the defects of the traditional combined prediction model, but also improves the prediction accuracy and can be applied to practice.
【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP18

【参考文献】