Analysis And Application Of Deep Learning Long-term And Short-term Memory Algorithms And Monte Carlo Method

In recent years,the process of digitization has developed rapidly.It is better to discover its essential characteristics and changing rules by digitizing all kinds of information.Time series is a kind of representative data.For most of the actual data,it is often non-linear time series data because of many influencing factors.Therefore,it is very important to analyze and study the prediction problem of non-linear time series.In this paper,the Monte Carlo method and the long-term and short-term memory algorithm are discussed and analyzed.The experimental data are cleaned up.Based on the long-term and short-term memory algorithm,a model for predicting the non-linear time series data is established,and the measurement criteria for predicting the effect are set up according to the principle of the Monte Carlo method.Firstly,the principles and advantages and disadvantages of Monte Carlo method and short-term memory algorithm are explored,and the causes of over-fitting problems in supervised machine learning and the corresponding solutions are summarized and analyzed by combining examples.Secondly,it preprocesses the collected population data and stock data(deletion and filling,normalization of abnormal data),and shows the processing method of abnormal data in the process of data cleaning with an example of Ping An Bank stock data.It preliminarily determines the Keras framework structure of the prediction algorithm,and sets the calculation method of prediction accuracy according to the principle of Monte Carlo method.Finally,a set of randomly generated sinusoidal data points is used to test the actual prediction effect of the algorithm framework.The population data of Harbin after cleaning up are forecasted by using different structure algorithms.By comparing the forecasting effect of different structure algorithm frameworks,the frame structure of the final long-term and short-term memory algorithm is determined.The forecasting analysis and accuracy calculation of the five characteristic data of Dongfeng Automobile Stock and Ping An Bank Stock are carried out.The empirical analysis of this paper chooses the daily data of Pingan Bank stock and Dongfeng Automobile stock from 2002 to 2018.The empirical results show that the combination of short-term and long-term memory algorithm and Monte Carlo prediction framework has a good effect on the prediction of non-linear time series data,and can learn the overall trend and detailed characteristics of the data well.The accuracy of setting and calculating by the principle of Monte Carlo method is more close to the actual prediction effect of the data,which is conducive to adjust the internal parameters within the framework of the algorithm.