Research On Financial Time Series Prediction Based On Self-Organizing Map

Time series analysis has a wide range of applications in many fields such as astronomy,biology,engineering,and finance.Among them,a large amount of time series data generated from financial markets such as stock,futures and foreign exchange is a typical type of financial time series with significant characteristics such as strong nonlinearity,large volatility,and high noise.The predictive analysis of financial time series and the study of financial market volatility characteristics,avoiding market risks and assisting investment decisions are of great significance,and have long received extensive attention from researchers at home and abroad.In recent years,along with the rapid development of artificial intelligence,the use of intelligent models for financial time series forecasting has become an emerging research hotspot.In this paper,from the perspective of financial time series with differential volatility characteristics,intelligent learning models are constructed based on self-organizing map to realize the application of prediction of financial time series,and the main research contents include:(1)For financial time series with high noise and large volatility characteristics,Fuzzy Logic is introduced into the deep learning model,and a new dual(multi)-layer deep fuzzy self-organizing map(DFSOM)is proposed,which can combine multiple time scales to learn the volatility characteristics of time series data,to realize the clustering analysis of different volatility types.Based on the above learning model and clustering results,different predictors are further constructed to achieve differentiated prediction of different volatility types of financial time series.(2)To address the fact that financial time series fluctuations in Euclidean space are usually strongly nonlinear,the graph attention mechanism and constant curvature Riemannian manifold were introduced into the self-encoder model to capture the overall fluctuation information of the multidimensional time series data of financial segments and project the features onto the manifold space.Then,a new class of Mask Self-Organizing Map(Mask-SOM)is constructed,and the clustering process is used to analyze the clustering of financial time series features on the manifold space in a progressive manner from coarse to detailed clustering.Then,based on the clustering results,the prediction analysis of the volatility direction of financial time series is realized.(3)Based on the above proposed intelligent learning model to build the corresponding intelligent trading system,applied to the foreign exchange,futures,and other financial markets.The feasibility and effectiveness of the model are verified by real financial time series.