| The development of the stock market is closely related to the economic construction of our country.Enterprises can raise funds through stocks,and investors can realize wealth appreciation by investing in stocks.But uncertainty about the return on investment in stocks comes with the risk of huge losses.Therefore,reasonable analysis and prediction of the future trend of the stock market can help investors avoid risks and invest scientifically.At present,the method of realizing stock trend prediction based on time series analysis is a major research hotspot.Among many time series prediction methods,deep learning models have won the favor of researchers due to their strong self-learning ability and complex feature extraction ability.This paper improves the existing time series model by combining the characteristics of the stock market and the method of deep learning,and completes the research and implementation of the stock trend prediction model.This thesis firstly investigates the characteristics of stock data and the collection method of stock historical transaction data,and then obtains various types of information in the A-share market through the financial data interface provided by the Tushare Pro platform,and finally constructs a multi-feature stock data set.Then,according to the phenomenon of "both up and down" in our country’s stock market,this thesis proposes a stock clustering method based on the similarity of stock trends.Firstly,this thesis analyzes several common unsupervised clustering algorithms and possible problems in stock clustering,and then improves the AP clustering algorithm,and proposes a stock trend similarity clustering for large data sets.The algorithm finally realizes the trend similarity analysis and stock clustering of thousands of stocks in the A-share market.At the model level,based on common time series prediction methods,this thesis analyzes the shortcomings of existing stock trend prediction models,and a stock trend forecasting model based on multi-scale time series feature fusion is proposed from the perspective of feature extraction.The model is trained by learning the combined effect of global coarsegrained features and local fine-grained features and improves the accuracy of stock trend prediction results.Finally,based on the clustering results,this thesis constructs a multifeature data set with clusters as a unit,and on this basis,trains the model by configuring the experimental parameters hierarchically.Then,the validity of the clustering algorithm is proved by the exploration experiment,and the validity of the model proposed in this thesis is proved by the comparison experiment with other models.Finally,by formulating a stock selection trading strategy,combined with a quantitative trading platform for backtesting analysis,it is proved that the model proposed in this paper has a certain reference value in assisting investors’ trading decisions. |
