Stock Trading Decision Based On Deep Reinforcement Learning

With the development of artificial intelligence technology,artificial intelligence has been widely used in more and more fields.In the field of economics and finance,the role of artificial intelligence has become increasingly prominent.In recent years,there have been a large number of studies based on deep learning methods for stock price fluctuations.Existing research on stock price volatility generally stays on the prediction of stock prices,and such research can achieve high accuracy.For a certain day’s stock price,the forecast accuracy is lower than or higher than a certain price,but this corresponds to two completely different results of profit or loss for stock trading.At the same time,using neural networks to predict stock prices has a high degree of delay,the neural network only needs to repeat the previous day’s data to obtain extremely high accuracy,which is meaningless for the prediction of stock prices.Moreover,the existing methods are instructive for the prediction of stock prices,and still require people to make stock trading decisions based on the predicted prices,and cannot realize automatic trading decisions of stocks.Inspired by the Alpha Go game theory,this paper adopts the method of deep reinforcement learning to realize the automatic stock trading decision without human participation.Compared with the forecasting method,the method of this paper is more meaningful for stock investors.The research content of this paper is as follows:1.Establishing a stock trading decision-making model: analyzing the shortcomings of the existing stock trading forecasting model,mathematically modeling the changing law of stock data,and demonstrating the feasibility of using reinforcement learning to solve the stock trading decision problem from the mathematical model of stock data change.A stock trading decision model based on deep reinforcement learning is proposed,and the specific structure of the stock trading decision model is given.2.The environmental model of stock trading decision: find the stock data on the Internet,and find high-quality data sources and obtain stock data according to the size of the data,the quality of the data,and the difficulty of obtaining the data.After the acquisition,the data of the continuous missing data is manually screened out,and the remaining data is guaranteed to be sufficient for training.After the data set is cleaned and pre-processed,manual feature extraction is performed.According to experience,the key data of the stocks,such as the daily average line,are added to the historical data of the stock to construct the training data set.The specific content of the environmental model of stock trading decision is given.3.Agent model of stock trading decision: Discuss the mathematical theory basis and principle of reinforcement learning algorithm,introduce the mainstream deep reinforcement learning algorithm,choose the appropriate according to the advantages and disadvantages of different reinforcement learning algorithms and their adaptability to stock trading problems.Reinforce learning algorithms.According to the characteristics of time series characteristics of stock data,Long Short Term Memory(LSTM)is selected to extract data features.In view of the characteristics of stock trading decisions,the reinforcement learning algorithm is improved to adapt to the characteristics of stock trading.The specific content in the agent model of stock trading is given.4.Stock trading decision based on Deep Q Network(DQN): Construct a model of stock trading decision based on improved DQN network and LSTM network,and target different input data volume,different network layers and different neurons.The number and different regularization methods compare the models,and the network with the current parameters is obtained,and good decision-making effects are obtained.In the research of this paper,a new concept of stock trading decision is proposed for stock trading,and a model of stock trading decision making with reinforcement learning is proposed.The stock trading decision model is implemented based on DQN and LSTM network.Using the stock historical data for data preprocessing and manual feature extraction as the input of the model,the better stock trading decision-making effect is obtained,and it has important guiding significance for further research on this issue in the future.