查询结果:   韩山杰,谈世哲.基于TensorFlow进行股票预测的深度学习模型的设计与实现[J].计算机应用与软件,2018,35(6):267 - 271,291.
中文标题
基于TensorFlow进行股票预测的深度学习模型的设计与实现
发表栏目
算法
摘要点击数
837
英文标题
DESIGN AND IMPLEMENTATION OF DEEP LEARNING MODEL FOR STOCK FORECASTING BASED ON TENSORFLOW
作 者
韩山杰 谈世哲 Han Shanjie Tan Shizhe
作者单位
中国海洋大学信息科学与工程学院 山东 青岛 266100     
英文单位
College of Information Science and Engineering, Ocean University of China, Qingdao 266100, Shandong, China     
关键词
TensorFlow 人工智能 数据分析 MLP 股价预测
Keywords
TensorFlow Artificial intelligence Data analysis MLP Stock price prediction
基金项目
作者资料
韩山杰,硕士生,主研领域:智能信息系统。谈世哲,副教授。 。
文章摘要
基于谷歌人工智能学习系统TensorFlow,构建多层感知器MLP(Multi-layer Perceptron)神经网络模型,用于预测每日收盘股价。将苹果公司的每日开盘股价作为数据集输入到神经网络,收盘价格作为神经网络学习的样本,并在训练过程中不断调整权值和阈值及网络结构,最终得到具有较高预测精度的神经网络模型。并就股价预测问题将TensorFlow与传统BP(Back Propagation)神经网络进行性能对比:(1) TensorFlow所构建的神经网络的均方误差RMSE(Root Mean Square Error)=0.624 5,而BP神经网络的RMSE=0.894 2,显示出TensorFlow具有更好的预测准确度;(2) 同样的学习样本数量,TensorFlow的预测耗时=1.221 s而BP神经网络的预测耗时=2.483 s,TensorFlow在分析效率及收敛速度上更有优势;(3) TensorFlow具有更友好的编程接口支持。证明了TensorFlow具有加快神经网络建模以及编程速度,提高数据分析效率的作用。通过对TensorFlow的开发流程的介绍,为进一步使用TensorFlow构建复杂的神经网络并进行数据分析提供了依据。
Abstract
Based on TensorFlow, a Google artificial intelligence learning system, a multi-layer perceptron (MLP) neural network model was constructed to predict the daily closing stock price. Apple’s daily OPEN stock price was entered into the neural network as a data set, and the closing price was used as a sample of neural network learning. In the course of training, weights, thresholds, and network structure were constantly adjusted, and a neural network model with high prediction accuracy was finally obtained. TensorFlow and traditional BP(Back Propagation) neural network performance comparisons for stock price prediction: (1) the root mean square error(RMSE) of the neural network constructed by TensorFlow = 0.624 5, and the RMSE of the BP neural network = 0.894 2, which showed that TensorFlow had better prediction accuracy; [JP](2) With the same number of learning samples, TensorFlows predicted time-consuming is 1.221 s and BP neural networks predicted time-consuming is 2.483 s, TensorFlow had more advantages in analyzing efficiency and convergence speed; (3) TensorFlow had a more friendly programming interface support. It was proved that TensorFlow had the function of accelerating neural network modeling and programming speed and improving the efficiency of data analysis. The introduction of TensorFlows development process provided the basis for further using TensorFlow to build complex neural networks and perform data analysis.
下载PDF全文