Research On Method And Application Of Stock Sequence Data Visualization Based On DenseNet

Stock is a kind of financial product produced with the development of financial market.The issuance and circulation of stocks has accelerated the concentration and accumulation of capital,and has gradually become a "barometer" of a country’s economic development.There are many factors influencing the stock market,and its forecasts are also very complicated.However,many scholars have found that in the real financial market,constructing a trading strategy can obtain excess returns.With the development of computer technology,the concept of quantitative investment has emerged.Through the use of mathematical models,it can objectively and effectively help maximize the benefits of stock investment and minimize the risks.Therefore,it has attracted more and more attention.In recent years,the field of artificial intelligence has developed vigorously,and deep learning has appeared in various fields.More and more scholars and investors are trying to use artificial intelligence technology to use deep learning network models to build high-return and low-risk investment strategies.How to use deep learning networks to make better stock forecasts also has great research value and practical significance.In the past research in this field,the main focus was on processing time series data directly,and training the series data by means of machine learning or neural networks.In the financial analysis and forecasting,researchers have developed some stock time series forecasting models based on machine learning and deep learning networks,and have achieved good accuracy.However,due to the large number of parameters,the network structure is more complex,which consumes time during training.A large number of computing and time resources,shallow features are not fully utilized,and some of the available information may be lost,so that there are still certain problems in scalability and practicality.In major competitions in recent years,Dense Net,with its excellent feature extraction capabilities and feature utilization capabilities,has greatly reduced the number of parameters,lowered the calculation cost,and achieved excellent practical results in the field of image processing.This paper uses the superiority of densely connected networks in image classification,and selects four stocks of Shanghai Pudong Development Bank,China Merchants Bank,China Everbright Bank,and Bank of China from January1,2010 to December 31,2019(excluding holidays and weekends).,And suspension,etc.)of the closing price,highest price,and lowest price and convert them into RGB images(the data images in this article),image the stock sequence data,and close the last day and the next day of the time window The rise and fall of prices are set as labels,which turns into an image classification problem.Moreover,in deep learning network training,when the data contains more usable information,it is more conducive to train a more robust network.Based on this idea,this paper proposes to separate the data images in order to increase the feature information of the data images.Three kinds of diversification processing:reverse processing diversification,random processing diversification and Gaussian diversification are carried out.Using Dense Net network,simulation experiments show that the reverse processing diversification is optimized on the basis of the original sequence data image,and the comparison is obtained.In this paper,benchmark strategies and stock trends have better high-yield and low-risk strategy algorithms.At the same time,the yield curve shows a stable growth trend,which has obtained a relatively good forecasting effect.On the other hand,in the stock market,due to the different nature of each stock in each industry,it is generally impossible to find a network or strategy algorithm to obtain excellent forecasting results for multiple stocks.In order to obtain better generalization ability,this paper proposes A method of constructing stock data is constructed.Two sets of different stock data are constructed and verified by experiments.A predictive network model that satisfies the three stocks is obtained.Under certain conditions,a better predictive network model is obtained.The forecasting effect,and at the same time,to reduce the risk,the strategy is optimized.The quantitative trading realized by this can obtain higher returns under lower risks.The yield curve shows a steady growth trend,which has certain research significance and practical significance..