Abstract:

Aiming at the problem of stock price forecasting, this paper proposes a stock price forecasting model which combines deep learning and decomposition algorithm. In this model, the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise （CEEMDAN） algorithm is introduced to extract the features of stock price time series on time scale, and the weight of input feature parameters is captured by attention mechanism, and stock price prediction is carried out by combining the GRU neural network. The results show that compared with RNN, LSTM and other models, the proposed model can effectively reduce the prediction error and improve the fitting ability of the model.

Key words: stock forecasting, attention mechanism, Gated Recurrent Unit（GRU） neural network, signal decomposition algorithm

摘要：

针对股票价格预测问题，实现对非平稳、非线性股票价格序列的预测，提出一种结合深度学习和分解算法的股票价格预测模型。该模型引入自适应噪声的完整集成经验模态分解（CEEMDAN）算法提取股票价格时间序列在时间尺度上的特征，利用注意力机制捕获输入特征参数的权重并结合门控循环单元（GRU）网络进行股票价格预测。实验对苹果、贵州茅台等国内外四家公司的股票价格和上证指数进行预测，结果表明与RNN、LSTM等模型相比，所提模型能有效减少预测误差，提高模型拟合能力。

关键词: 股票预测, 注意力机制, 门控循环单元（GRU）神经网络, 信号分解算法