摘要
金融时间序列预测遵循不同的模式,由于用户行为的改变或环境本身的改变,这些模式可能随着时间的推移而改变.股票走势预测作为金融时间序列预测中最具挑战性的任务之一,目前的研究主要集中在公开市场数据上,而没有充分考虑行情局部趋势特征模式和交易行为相关性分析.本文提出了一个融合历史交易数据和关联市场信息的面向局部特征模式的深度神经网络趋势预测模型.首先,通过改进的Zigzag技术指标识别算法识别金融时间序列的重要点,并对局部趋势特征进行建模;然后,利用知识图谱和图嵌入技术来融合市场信息和行情交易特征信息,并与感知重要点等K线指标信息进行多特征融合.最后,将上述这些信息输入到基于注意力的双向长短期记忆网络进行股价走势预测.实验结果表明,所提出的模型具有较好的有效性、可用性与稳健性.
Financial time series forecasting follows different patterns.These patterns may change over time due to changes in user behavior or changes in the environment itself.Stock trend forecasting is one of the most challenging tasks in financial time series forecasting.The current research is mainly focused on open market data,without fully considering the local trend characteristic patterns and correlation analysis of trading behavior.This paper proposes a deep neural network trend prediction model based on local feature pattern,which integrates historical transaction data and associated market information.Firstly,our method uses the improved Zigzag technical indicator recognition algorithm to identify important points in the financial time series,and model local trend characteristics.Then,we use knowledge graph and graph embedding technology to fuse market information and market transaction feature information,and integrate it with K-line indicator information such as important points of perception for multi-feature fusion.Finally,the associated dates are fed to the attention-based bidirectional long-short term memory network for predicting stock price trend.The experimental results show the effectiveness,usability and robustness of the proposed model.
作者
蒋慧敏
陈锋
JIANG Hui-min;CHEN Feng(School of Information Science and Technology,University of Science and Technology of China,Hefei 230027,China)
出处
《小型微型计算机系统》
CSCD
北大核心
2022年第9期1793-1800,共8页
Journal of Chinese Computer Systems
基金
国家重点研发计划项目(2017YFC0840206)资助.
