基于LSTM算法与股票历史数据的股票价格预测

Stock Price Prediction Based on LSTM Algorithm with Historical Stock Data

摘要: 股票价格的准确预测一直是金融领域关注的焦点之一。传统的股票预测方法通常依赖于k线图、十字线等方法，然而仅依靠传统方法分析前一天的价格来预测股票走向存在较大的局限性。为了进一步提升股票价格预测的精度和可靠性，本文提出了一种基于长短时记忆神经网络(Long Short-Term Memory Networks, LSTM)的股票价格预测模型。通过LSTM算法处理时序数据的优势捕捉价格数据中的复杂模式和非线性关系，挖掘股票历史价格数据中隐藏的波动信息，为股票价格预测领域的研究和实践提供了新的思路和方法。此外，为进一步获取最优的模型参数量，本文以平安银行2019年1月至12月的开盘价和收盘价数据为样本进行实验。实验结果表明，基于LSTM算法的股票价格预测准确率可达78.2%；历史股票价格序列数为5时，模型的预测精度最高。

Abstract: Accurate prediction of stock prices has always been a focal point in the financial sector. Traditional stock prediction methods often rely on techniques such as k-line charts and crosshairs. However, relying solely on these traditional methods to analyze the previous day’s price for predicting the direction of a stock is subject to significant limitations. To enhance the accuracy and reliability of stock price prediction, this study puts forth a model based on Long Short-Term Memory Networks (LSTM). By leveraging the capabilities of the LSTM algorithm for processing time-series data and capturing complex patterns and nonlinear relationships in price data, this study uncovers hidden fluctuation information in historical stock price data, offering novel insights and methodologies for research and practice in the field of stock price prediction. Additionally, to determine the optimal number of model parameters, this study utilizes the opening and closing price data of Ping An Bank for the period between January and December 2019 as experimental samples. The experimental results demonstrate that the LSTM algorithm achieves a stock price prediction accuracy of 78.2%. Moreover, the model exhibits the highest precision when utilizing 5 historical stock price sequences.