Application Of LSTM Model Based On Improved Particle Swarm Optimization In Stock Price Prediction

As a kind of negotiable securities,stock has always been one of the most favored investment directions of investors,and the price prediction of stock has naturally become a long-term concern of investors.However,due to the high noise,dynamic,non-parametric,nonlinear and other characteristics of data,it is quite difficult to predict stock prices.In recent years,the epidemic and war around the world have aggravated the instability of the stock market and put forward higher requirements for price prediction.With the development of artificial intelligence technology,various machine learning models have been widely applied in more and more fields,and their ability to fit nonlinear problems has become stronger and stronger,which has been widely used in stock price prediction.In this thesis,we choose the CSI 500 index.First,we use Back-Propagation neural network to predict the prices,then we use Long Short-Term Memory network model to make predictions.Subsequently,the traditional particle swarm optimization algorithm is introduced on the basis of the LSTM model to optimize the process of parameter selection,and construct Particle Swarm Optimization-Long Short-Term Memory network model for stock price prediction.In order to obtain better prediction results,the own principles of the particle swarm optimization algorithm that are unfavorable to the process of optimization are improved by determining the optimal dimensionality of the particle swarm,introducing the nonlinear variation of inertia weights and learning factors,applying the adaptive variation operation in the genetic algorithm,and proposing a new position update formula to obtain the Improved Particle Swarm Optimization-Long Short-Term Memory network model.To illustrate the superiority of the proposed improvement,the previously proposed DMPSO-LSTM model is used to predict the same stock data.A comparison of the five models in terms of three types of error indicators and decision coefficients shows that the IPSO-LSTM model proposed in this thesis has significant advantages when applied to the field of stock price prediction.