基于分层动态贝叶斯网络的股市趋势扰动推理算法  

Inference Algorithm for Stock Market Trend Disturbance Based on Hierarchical Dynamic Bayesian Network

作　　者：姚宏亮[1] 贾虹宇 杨静[1] 俞奎[1] YAO Hongliang;JIA Hongyu;YANG Jing;YU Kui(School of Computer Science and Information Engineering,He-fei University of Technology,Hefei 230601)

机构地区：[1]合肥工业大学计算机与信息学院,合肥230601

出　　处：《模式识别与人工智能》2022年第4期363-373,共11页Pattern Recognition and Artificial Intelligence

基　　金：国家重点研发计划项目(No.2020AAA0106100);国家自然科学基金面上项目(No.61876206,62176082)资助。

摘　　要：当前研究者主要通过对历史交易数据的学习生成预测模型,由于影响市场的因素动态可变,因此训练好的模型在实际应用中预测效果远不及预期.针对现有预测模型适应力较弱的问题,文中提出基于分层动态贝叶斯网络的股市趋势扰动推理算法,对股市趋势进行实时预测.首先,针对稳定性较高的均线数据,通过马尔可夫毯对均线特征进行融合,提取为均线能量,得到均线的量化特征.由于多根均线之间存在结构关系,这种结构关系具有较强的抗噪能力和稳定性,因此利用分层动态贝叶斯网络对单根均线内部结构和多均线之间结构关系进行建模.然后,对顶层网络中多结点状态进行扰动,通过动态灵敏性分析实时计算结点状态变化对于股市趋势影响力.最后,基于灵敏分析的结果,利用联合树对股市趋势进行动态推理.在实际数据上的实验证明文中算法的有效性.The current research mainly focuses on the forecasting models generated by the learning of historical transaction data.Due to the dynamic variation of the factors affecting the market,the forecasting effect of the trained model in practical applications is much worse than the expected.To solve the problem of weak adaptability of the existing forecasting models,a disturbance inference algorithm based on hierarchical dynamic Bayesian network(DA-NEC)is proposed to predict stock market trends in real time.Firstly,for the moving average data with high stability,the energy of the moving average is extracted through the Markov blanket fusion of the moving average features,and the quantitative characteristics of the moving average are generated.Since the structural relationship among multiple moving averages possesses strong anti-noise ability and stability,the hierarchical dynamic Bayesian network is employed to model the internal structure of a single moving average and the structural relationship among multiple moving averages.Then,the state of multiple nodes in the top-level network is disturbed,and the state changes of the nodes are calculated in real time through dynamic sensitivity analysis.In the end,based on the results of sensitive analysis,the junction tree is applied for dynamic inference on the stock market trend.Experimental results on actual data show the effectiveness of the proposed algorithm.

关 键 词：动态贝叶斯网络 灵敏性分析 联合树 马尔可夫毯 股市趋势预测 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]