Time Series Prediction Based On Similarity Search And Its Application

Time series is a series of data points sampled at a specific time interval,which widely exists in various fields of real life.Stock trading data is a typical time series.The analysis of stock trading data has important practical significance,which helps people understand the knowledge behind the data and helps investors make decisions.The stock market has the characteristics of high risk and high return.Stock price forecasting will reduce investment risk and increase income.It has always been a hot topic in financial data analysis.This thesis studies the characteristics and prediction methods of stock market and proposes a time series prediction method based on similar pattern search.It is different from the traditional stock analysis method,which can predict complex stock data.It can predict the trend and confidence of multidimensional time series.A useful forecasting software is designed.It will find out all historical stock trading data similar to current stock price pattern and use that to predict future prices.Finally,the effectiveness of this software is verified by real stock trading data.Main research contents:(1)The characteristics of original stock data are analyzed.Time series data,classified data and event data are collected and stored in document database.The data acquisition strategy and logical storage structure are designed.Analyze data preprocessing requirements and develop a set of command line tools for preprocessing.(2)Similarity measures and normalization methods of time series data are studied.This paper defines similarity score and similarity measure method of multidimensional time series.Pattern search and prediction algorithm are described in detail.This algorithm searches for similar patterns in one-dimensional or multidimensional time series and calculates future values and confidence.Design data filtering function to support data set selection by multiple conditions.The data is segmented and processed in parallel to improve the efficiency of the search algorithm.(3)Study the influencing factors and forecasting methods of stock.Based on pattern search and prediction algorithm,stock analysis software is designed and developed.This software supports price matching,volume-price matching and multi-stock price matching in searching and forecasting stock prices.The effectiveness of this software is verified by real stock trading data.