摘要
考虑到股票价格具有高噪声和非线性的特点,构建了一个基于门控循环单元-注意力机制(Gated Re-current Unit-Attention Mechanism,GRU-AM)的股票价格预测模型.首先将经过处理后的股票数据输入到门控循环单元(Gated Recurrent Unit,GRU)中,使其能够充分地学习数据并挖掘更深层次的数据特征.接着将从GRU中学习到的特征向量传入注意力机制(Attention Mechanism,AM).其次通过这种方式,可以学习到不同时间特征状态的权重,从而更有效地捕捉时间特征的重要性.最后通过与七种基准模型在多个数据集上的对比,验证了模型的有效性.
Abstract
This research constructs a stock price prediction model based on Gated Recurrent Unit-Attention Mechanism(GRU-AM)considering the characteristics of high noise and non-linearity of stock price.The researchers first input the processed stock data into the Gated Recurrent Unit(GRU)to fully learn the data and mine deeper data features,and then introduced the feature vectors learned from GRU into the Attention Mechanism(AM)to learn the weights of different time feature states for capturing the importance of time features more effectively.This research fi-nally verified the effectiveness of the model by comparing with seven baseline models on multiple datasets.
NETL
NSTL
万方数据