基于LSTM模型的股价预测研究——以保利发展为例

Stock Price Forecasting Research Based on LSTM Model—Taking Poly Development as an Example

股票市场受到国际形势、市场行情和国家政策等多因子影响,简单的K线图与机器学习预测很难抗击股价的波动随机性,导致股价预测精度不能达到投资参考水平。本文采用LSTM神经网络模型仿真保利发展(600048) 2018年2月13日至2022年3月30的时间序列收盘价,通过调节隐含单元数进行多次仿真,得到误差较小、可行性较强的结果。分析实验结果得到,向后预测的收盘价数据,得到的误差比极低。因此,本文实验得到LSTM模型能对长期随机波动的时间序列数据做出预测,且模型的精度高、可推广性强。

The stock market is affected by multiple factors such as international situation, market conditions and national policies. It is difficult for simple K-graph and machine learning prediction to resist the randomness of stock price fluctuations so that the accuracy of stock price prediction cannot reach the investment reference level. This paper uses LSTM neural network model to simulate the time series closing price of Poly Development (600048) from 13 February 2018 to 30 March 2022. By adjusting the hidden unit number for many times, the results of small error and strong feasibility are obtained, and by analyzing the closing price data, the error ratio is very low. Therefore, the LSTM model can predict the time-series data with long-term random fluctuations, with high accuracy and strong generalizability.