摘要
为构建更精准的股票价格预测模型,提出具有局部信息挖掘功能的DNN加权算法对eplion-TSVR模型进行改进,并对改进模型的求解进行推导,针对DNN算法对于参数的选取太过随意,提出使用网格搜索法确定DNN的最优参数以确定最优DR域。搜集中国上证A股中的15支股票的日价格和高频5分钟价格数据并计算其技术指标,对20天以及20分钟后的收盘价进行实证预测。预测结果显示,改进模型在高频股票数据上具有很好的预测能力和泛化性能。
Using the DNN(dependency nearest neighbor)algorithm leads to the random selection of parameters.The grid search method was used to determine the optimal parameters of the DNN and the optimal DR(dependency region)domain was determined.The improved DNN-eTSVR(eplion twin support vector regression)model was used to deal with the daily price data and high frequency 5 minutes data including the original price data and technical indicators data of 15 stocks in China SSE A shares,and to predict the closing price after 20 days or 20 minutes.The results of the evaluation indicators show that the proposed prediction model has good predictive power and generalization performance on stock data,especially for high-frequency minute data,which shows that the improved model can better extract the local information of the sample and obtain excellent prediction results.
作者
张冰
王传美
贺素香
ZHANG Bing;WANG Chuan-mei;HE Su-xiang(School of Science,Wuhan University of Technology,Wuhan 430070,China)
出处
《计算机工程与设计》
北大核心
2019年第11期3241-3246,共6页
Computer Engineering and Design
基金
国家自然科学基金项目(71473186)
中央高校基本科研业务费基金项目(2018IB016)
关键词
D近邻加权算法
孪生支持向量回归机
股价预测
高频数据
网格搜索
weighted dependent nearest neighbor
twin support vector regression
prediction of stock price
high frequency
grid search
