Stock Trend Forecasting Model Based On Deep Learning

In recent years,computer technology has made great progress in many fields,from scanning code to aerospace,all of which use computers.Stock market research is an interdisciplinary subject formed by the combination of computer and finance.The background of stock participants are different,which makes the investment difficult to operate.Therefore,how to accurately study the stock market is an important task for scientists.This topic introduces the quantification method of stock factors,constructs a set of quantification system,constructs the influencing factors that reflect the company's operation and stock price's change,discusses how to ensure the high quality of quantification factors,and establishes a data visualization system for the stock market prediction model,laying a data foundation for the construction of the stock market forecasting model.Based on the stock data,the main research contents of this paper include the following two aspects:A stock trend prediction model based on EMD-Transformer.The paper describes the time series relationship of the stock market data,analyzes the advantages and disadvantages of the existing deep models,introduces the model Transformer in the NLP field,and improves its network structure to suit time series tasks.It solves the problem of limited performance of LSTM in processing long sequences,overcomes the defect that Transformer cannot effectively capture the sequence.There are many investors in the stock market and a wide range of influencing factors.These factors introduce a lot of randomness.The existence of randomness will make the model fit away from the real data and reduce the effect of the model.In order to reduce the bad influence of noise,it needs a reasonable response and selecting a specific wavelet base,but the wavelet base cannot be changed during the analysis and is unable to adapt to the different distribution of data itself.This topic uses empirical mode decomposition(EMD)to denoise and change the lack of adaptability of the wavelet transform basis function,constituting the EMD-Transformer model.The topic verifies the model's effect through comparative experiments.A stock trend prediction model based on EE-Transformer.The paper analyzes the problem of limited samples' number in stock forecasting,studies the feasibility of providing additional information for model training,analyzes the development of Embedding technology,proposes a knowledge transfer method based on stock gain Embedding,and establishes network structure to represent the gain value,together with the improved Transformer proposed above,forms an EE-Transformer network,and the model's effect is verified through experiments.