Research Progress And Application Of LSTM Recurrent Neural Network

With the advent of the era of artificial intelligence,people are paying more and more attention to the potential value of data,and the ability to mine and use data is getting stronger and stronger.In the era of big data,neural network technology has received great attention.RNN has strong performance ability in processing time series data,but the classic neural network has insurmountable shortcomings.RNN is prone to gradient vanishing and gradient explosion during training.Although the problem of gradient vanishing can be solved by optimizing the activation function,but it is not a very good solution.In 1997,the long short-term memory neural network(LSTM)was proposed,the memory units and gate mechanisms to the layer nodes can not only memorize long-term information,but also effectively avoid the problem of gradient vanishing in RNN,and the problem of gradient explosion can be avoided by adding regular terms.Later,more scholars continue to explore and optimize the structure of LSTM,variants of LSTM with various structures have been formed,which have a wide range of applications in different fields.This article mainly discusses the origin of LSTM in depth,by understanding why LSTM is proposed and the working principle of its network structure to understand the model,and enumerate the research progress of LSTM network in recent years.Futures market data has the characteristics of nonlinear and high noise.It is difficult for general models to accurately extract data features to control the trend of data.However,LSTM networks have long-term memory capabilities and have excellent learning capabilities for time series data.Therefore,in this paper,LSTM neural network is used to predict the closing price of soybean meal futures,and good prediction results have been achieved.Compared with the traditional machine learning algorithm-support vector machine(SVM),it is found that the prediction effect of LSTM is better than SVM.