【作者】 李勇；

【导师】 冯宏伟；

【作者基本信息】 西北大学 ， 软件工程（专业学位）， 2019， 硕士

【摘要】 股票市场受经济市场、政策等因素的影响,其内部变化规律极其复杂。随着中国股票市场的快速发展和投资者规模的扩大,股票市场产生了大量的交易数据,使获取有价值的信息变得更加困难。由于深度神经网络善于处理数据量大,非线性映射关系复杂的预测问题,本文基于深度神经网络设计了一个股票智能预测系统。主要工作内容如下:(1)股票选股模型研究:从股票财务指标和股票变化趋势层面,研究了多因子量化选股的问题,提出一种基于股票趋势识别算法构建选股模型的方法。该方法根据改进的股票趋势识别算法,对由财务指标和技术指标构建的选股指标集进行趋势标识,并使用主成分分析法对选股指标进行相关性处理,将处理后的股票数据输入BP神经网络中,训练出股票选股模型。相比于现有的选股方法,提出的选股模型在选股精度上平均提高了5.09%。(2)股票价格预测模型研究:针对股票交易数据量大,非线性关系复杂,难以准确预测股票价格的问题,基于LSTM(Long Short Term Memory)深度神经网络对股票价格的预测进行研究。从股票选股模型的输出结果中选择可投资的股票。通过研究影响股票价格变动的因素,选取股票交易基本数据和股票技术指标数据,融合成股票训练数据。采用LSTM深度神经网络构建股票价格预测模型,对影响模型效果时间序列长度和网络结构等参数进行调优。相比于选取股票交易基本数据训练的股票价格预测模型,决定系数提高了2.4%,均方根误差降低了0.12。(3)股票智能预测系统的设计与实现:基于股票选股模型和股票价格预测模型,设计并实现了股票智能预测系统。运用Django框架、Scikit-learn机器学习库和Keras深度学习库完成股票智能预测系统各个功能。该系统可以实时地进行股票选股和股票价格预测,给投资者提供实时、有效的投资决策建议,从而降低投资者投资的风险,并获得稳定投资收益。更多还原

【Abstract】 The stock market is affected by various factors such as economic markets,political,which lead to the complex of its internal changes.With the rapid development of China’s stock market and the expansion of investor scale,a large amount of transaction data of the stock market has been generated,from which it’s difficult to obtain valuable information.The deep neural network has certain advantages in dealing with large amount of data and complex nonlinear mapping.Therefore,based on deep network technology,an intelligent stock forecasting system has been designed.The main work is summarized as follows:(1)Stock selection model: The problem of multi-impact factor quantitative stock selection is studied based on the stock financial indicators and stock change trend,and a stock trend identification algorithm is proposed to construct stock selection model.According to the improved stock trend identification algorithm,the method identifies the trend of the stock selection indicators set by the financial indicators and technical indicators,and it utilizes the principal component analysis method to deal with the stock selection indicators,then it inputs the processed stock data into the BP neural network to train the stock selection model.Compared with the existing stock selection methods,the proposed stock selection model has improved the accuracy of the stock selection by an average of 5.09%.(2)Stock price forecasting model: Because of the large amount of stock trading data and the complex nonlinear relationship,it forecasts stock price based on LSTM(Long Short Term Memory)deep neural network.We select investable stocks from the output of stock selection model.By studying the factors affecting stock price changes,we select stock trading basic data and stock technical indicator data to form the stock training data.The LSTM network is then used to construct the stock price forecasting model,and the parameters such as the effect time series length and network structure are optimized.Compared with the stock price forecasting model constructed using the basic stock trading data,the coefficient of determination of the stock price is improved by 2.4%,and the root mean square error is reduced by 0.12.(3)Designing and implementing of intelligent stock forecasting system: Based on stock selection model and stock price forecasting model,the intelligent stock forecasting system is designed and implemented.Django framework,Scikit-learn machine learning library and Keras deep learning library are used to build the intelligent stock forecasting system and complete the various functions of this system.The system can perform stock picking and stock price forecasting in real time,providing investors with meaningful investment decision advice,therefore,the risk of investors’ investment is reduced and the high investment returns are obtained.更多还原