基于DMD-LSTM模型的股票价格时间序列预测研究

针对股票市场关系复杂导致的有效特征提取困难、价格预测精度低等问题，提出一种基于动态模态分解—长短期记忆神经网络（DMD-LSTM）的股票价格时间序列预测方法。首先通过DMD算法对受市场板块联动效应影响的关联行业板块样本股数据进行分解计算，提取包含整体市场和特定股票走势变化信息的模态特征；然后针对不同市场背景，采用LSTM网络对基本面数据和模态特征进行价格建模预测。在鞍钢股份（SH000898）上的实验结果表明，该方法相较于传统预测方法，在特定的市场背景下能实现更高的价格预测精度，更为准确地描述股票价格的变化规律。

Research of stock price prediction based on DMD-LSTM model

Aiming at the problems of low prediction accuracy and feature extraction difficulty in complicated stock market, this paper proposed a stock price prediction method based on dynamic mode decomposition and long short-term memory neural network(DMD-LSTM). Firstly, it used the DMD algorithm to decompose the industry specific stock in the background of plate linkage phenomenon, and extracted the mode feature which included stock trend information. Then, it built the LSTM network to establish the relations between stock price and the feature of mode and basic index in different market conditions. The experimental results on Angang Steel(SH000898) show that, the proposed method has the higher forecasting precision compared with the traditional ways in specific condition, which can characterize the trend of stock price changes better.