Research On Financial Time Series Forecast Based On LSTM Neural Network

With the continuous development of the financial market,the amount of time series data in the financial market is getting larger.The speed of data generation and accumulation are going fast.The traditional measurement model cannot meet the processing requirements of large data for nonlinear and complex data.Machine learning method could exploit data characteristics,learn characteristics of the historical data and apply to a subsequent determination.Among them,neural networks based on machine learning methods have the ability to process huge data sets.It can perform good nonlinear fitting and have unparalleled advantages in financial time series processing.In this paper we uses model prediction of dynamic financial time series full of noise and non-linear changes based on LSTM neural network to make up for the lack of subjective evaluation of the basic analysis method and provide reference for quantitative trading work and provide artificial intelligence methods in financial time The application of sequence problems provides practical guidance.This paper based on LSTM neural network for regression prediction of stock index daily data.Three types of stock price sample index,comprehensive index and classification index are selected.The specific prediction methods include full-sequence index regression prediction and single-step stock index regression prediction,Multi-step stock index regression prediction.The experimental process includes data preprocessing,network error evaluation indicators,network model structure construction and loss function image drawing.Analyze the performance of different prediction methods on different sample data sets,and summarize the relevant laws.This paper based on the LSTM neural network selection single-step prediction network structure model applied to the test of individual stock historical data.Through the volume-price relationship model,technical index model and combined PCA model,the high-frequency closing price of each stock is predicted.Observe the model fitting situation and compare the advantages and disadvantages of the fitting effects between different models.Investigate the effect of unsynchronized length on the model fitting effect,and summarize the relevant laws according to the fitting situation of each model.