Research And Application Of Stock Data Analysis Technology Based On Deep Learning

With the rapid development of China's economy,the stock market is also constantly strengthening its construction,and the types of investment are constantly enriched.The core of stock research has been shifted from profit to effective risk aversion.In order to better guide investment with modern portfolio theory,construct portfolio and disperse non-systematic risk,this paper explored the following work in the research of stock data:Firstly,this paper constructed the time series matrix of stock price data.Stock price data has the typical characteristics of discrete time series:high dimension,large amount of data,inconsistent length and sparse data.Based on Tu-Share's financial data interface,this paper obtained the closing price data of stocks from 2017 to 2018 and normalized the stock price based on time series.The data after processing has the problem of data sparsity,which was solved by filling the missing value.Finally,the time series matrix could be used as input of the model.Secondly,this paper constructed a clustering model based on Auto-Encoder for stock trading data.Auto-Encoder feature dimension reduction was used to extract higher dimension features,which weakened the problem of excessive noise influence when extracting feature rules in clustering process.Experiments showed that this model could obtain high cohesion and low coupling clustering results,and proved the feasibility of the model.At the same time,the new model could achieve better performance under large data volume,and the convergence speed was faster than the original model.Then this paper established a short-term prediction model of stock cluster rally based on deep belief network.In previous studies,it was found that the deep belief network,as one of the commonly used models of deep learning,has better performance in forecasting.Therefore,based on the deep belief network,a short-term forecasting model of stock cluster for stock time series data was constructed,and the experimental results were compared with the BP neural network model.The experimental results showed that the network has better forecasting accuracy,which proved the validity of the stock clustering model indirectly.Finally,this paper applied the theoretical results of the study to the real scene.The result is only a small part of the real scene.However,it will be a big step in the field of stock engineering to effectively aggregate the results of stock.