Intelligent Investment Models And Methods Based On Deep Learning

In recent years,with the rapid development of artificial intelligence technology and the continuous improvement of China's financial market,more and more people are beginning to study how to use computer science technology to deal with financial problems.As an important part of the financial market,stock investment has always been valued.At the same time,as one of the common financial management methods commonly used by ordinary people,its importance is self-evident.The main content of this research is how to apply deep learning to stock investment.This topic first introduces the construction of stock data sets,implements a stock data acquisition system based on Scrapy framework,and discusses how to verify the correctness of data and pre-processing data.Combining the multi-factor model in the stock selection model,the stock historical data and the financial data of the listed company are analyzed to find out the effective information contained in the massive data.Based on the stock dataset,the primary achievements of this thesis include the two fields hereinafter:Stock trend prediction model based on Attention-LSTM.This paper analyzes the timing relationship of stock data and discusses the long dependence of time series data.Compared with traditional recurrent neural network and long-term and short-term memory network(LSTM),it analyzes in detail how long-and short-term memory networks solve the problem of long dependence in time series data.At the same time,the shortcomings of encoder-decoder in LSTM networks are analyzed,and the physiological principles of Attention mechanism are discussed.The feasibility of adding Attention mechanism to LSTM networks is demonstrated.In the project,a double Attention mechanism is designed and implemented to solve the problem of degraded model performance when the sequence data is too long.Finally,the Attention-LSTM model is compared with other models to analyze the capabilities of the model in various aspects.Stock trading model based on deep reinforcement learning.The model mainly learns the stock investment strategy through the training deep Q network(DQN),achieves the goal of the strategy network to automatically operate the stock.A series of DQN models are introduced in detail,and the experience replay principle and Q network training process are analyzed.At the same time,the problem of overestimation of Q value in DQN network is analyzed.It is considered that using double-layer network structure to update parameters can reduce the risk of overestimation of Q value.At the data processing level,the hybrid network is used to analyze time series data.The model combines CNN network and LSTM network to more effectively mine the nonlinear relationship hidden in the data.Finally,through a series of technical indicators and comparisons with other models,the feasibility of deep reinforcement learning for stock strategy research is verified.