面向组合投资预测的大数据生成算法  

Big data generation algorithm for portfolio investment forecasting

作　　者：赵会群[1] 曲艺 ZHAO Hui-qun;QU Yi(School of Information,North China University of Technology,Beijing 100144,China)

机构地区：[1]北方工业大学信息学院,北京100144

出　　处：《计算机工程与设计》2021年第2期388-395,共8页Computer Engineering and Design

基　　金：国家自然科学基金项目(61672041)。

摘　　要：为满足组合投资预测对数据的需求,提出一种基于增量式贝叶斯网络模型的大数据生成方法。使用时间序列生成算法对未来各项数据进行部分生成;结合新生成数据对历史数据训练的贝叶斯网络模型进行更新,使更新后的贝叶斯网络能够体现该时间段内新旧金融数据中各项变量之间的关系及蕴含的规律;在贝叶斯网络中通过路径搜索算法生成投资组合路径的集合,根据其中各路径的概率分布,生成具有真实数据特征的可用于组合投资预测的大数据集。实验结果表明,该方法具有可行性,保证了一定的准确度。To meet the data demand of portfolio investment forecasting,a big data generation method based on incremental Baye-sian network model was proposed.The time series generation algorithm was used to generate part of the future data.The Baye-sian network model which was trained by historical data was updated with the newly generated data,so that the updated Bayesian network reflected the relationship and the laws of the new and old financial data in the time period.The path search algorithm was used to generate the portfolio path set from the Bayesian network,according to the corresponding probability distributions,the big data set with real data features was generated which could be used to the portfolio investment forecasting.Experimental results show that the proposed method is feasible and guarantees a certain accuracy.

关 键 词：组合投资 贝叶斯网络模型 时间序列模型 路径搜索 大数据生成 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]