摘要
随着机器学习与深度学习的发展,传统的时间序列模型已经不能满足人们对于股票预测准确性的要求。因此,本文引入深度学习中基于PyTorch框架的LSTM循环神经网络模型对创业300指数的收盘价进行预测,通过设置迭代次数、遗忘门偏置值以及LSTM单元数,对比模型的预测误差。研究结果表明,迭代次数为200、LSTM单元数为2、遗忘门偏置值为0.4的LSTM模型对创业300指数收盘价走势的拟合误差最小,平均绝对百分比误差达到0.0109,为进一步使用PyTorch框架构建循环神经网络准确预测股价提供了依据。
With the development of machine learning and deep learning, the traditional time series model has been unable to meet people’s requirements for the accuracy of stock forecast. Therefore, this paper introduced the LSTM recurrent neural network model based on PyTorch framework in deep learning to predict the closing price of Entrepreneurship 300 Index, and compared the forecast error of the model by setting the number of iterations, the value of the forgetting gate bias and LSTM unit numbers. The research shows that the LSTM model with 200 iterations, 2 LSTM units and 0.4 forgetting gate bias value has the smallest fitting error for the closing price trend of Entrepreneurship 300 Index, and average absolute percentage error reaches 0.0109, providing a basis for further using PyTorch framework to construct recurrent neural network to accurately predict stock price.
出处
《运筹与模糊学》
2021年第2期137-146,共10页
Operations Research and Fuzziology