基于模糊信息粒化和SVM优化模型的上证指数实证分析

发布时间：2018-02-21 22:48

本文关键词： 模糊信息粒化隶属函数支持向量(回归)机上证指数　出处：《兰州大学》2017年硕士论文　论文类型：学位论文

【摘要】：大数据时代的到来,给数据挖掘带来了新的挑战,如何在海量的数据中挖掘有用信息成为一大课题。模糊信息粒化模型通过将信息粒化成模糊粒子,能够大幅度降低信息的复杂度,简化问题求解过程,减少运算量。隶属函数作为模糊粒化约束规则直接影响信息粒化效果。本文将基于不同隶属函数的模糊信息粒化模型和支持向量(回归)机模型结合,组建混合模型,对上证指数进行回归分析,寻找使混合模型预测效果最佳的隶属函数。上证指数实证分析的结果表明,基于不同隶属函数的混合模型对模糊粒子中值R的预测结果相同,其中基于非对称抛物线型隶属函数的混合模型的范围预测结果精确度和可靠度最高。然而,常见的以时间t为自变量的混合模型,预测误差与数据波动情况有关,模型预测效果受数据时间性的影响,范围预测结果的精确度和可靠度达不到理想要求。同时,相邻两个时间窗口范围预测的周覆盖比CCR和周可靠比CRR变动较大,模型预测的稳定性较差。为了剔除模型受数据时间性的影响,本文摒弃常见的以时间t作为自变量的假设,引入上证指数每日最高点、每日最低点、每日收盘点、每日成交量(万手)和每日成交额(亿)作为自变量,重新组建混合模型。从实证分析的结果来看,该种做法能够有效排除数据时间性对预测结果的影响,大幅提高范围预测的精确度和可靠度。同时,该种混合模型得到的相邻两个时间窗口范围预测的周覆盖比CCR和周可靠比CRR变动较小,模型预测的稳定性较高。
[Abstract]:The arrival of big data has brought a new challenge to data mining. How to mine useful information in mass data becomes a major topic. Can greatly reduce the complexity of information, simplify the process of solving the problem, This paper combines fuzzy information granulation model based on different membership functions with support vector (regression) machine model to form a hybrid model. The regression analysis of Shanghai Stock Exchange index is carried out to find the membership function that makes the best prediction effect of the hybrid model. The empirical analysis of Shanghai Stock Exchange Index shows that the mixed model based on different membership functions has the same prediction results for the median R of fuzzy particles. The range prediction accuracy and reliability of the hybrid model based on asymmetric parabola membership function are the highest. However, the common mixed model with time t as independent variable, the prediction error is related to the fluctuation of data. The prediction effect of the model is affected by the time of data, and the accuracy and reliability of the range prediction results are not up to the ideal requirement. At the same time, the circumferential coverage ratio (CCR) and the cycle reliability of the adjacent two time windows are more variable than that of CRR. The stability of model prediction is poor. In order to eliminate the influence of data timeliness, this paper discards the assumption that time t is the independent variable, and introduces the daily highest point, daily low point and daily closing point of Shanghai stock index. Daily turnover (10 million hands) and daily turnover (100 million) are taken as independent variables to reconstruct the mixed model. From the results of empirical analysis, this method can effectively eliminate the impact of data timeliness on the forecast results. The accuracy and reliability of range prediction are greatly improved. At the same time, the predicted coverage ratio (CCR) and reliability ratio (CCR) of the two adjacent time windows obtained by this hybrid model are smaller than those of CRR, and the stability of the model prediction is higher.
【学位授予单位】：兰州大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F832.51;F224

【参考文献】