Research On Stock Prediction Based On LSTM Neural Network

As one of the most popular financial management methods,stocks have attracted more and more investors.The risk of stock investment is relatively high,so how to reduce the risk and increase the profit has become the most concern of investors.Researchers have also begun to use various methods to predict stocks,which has provided some guidance for investors to trade stocks.The rapid development of machine learning has also promoted the development and application of neural networks.Numerous studies have confirmed that the use of neural networks can make better predictions for stocks.This article focuses on the application of BP neural network,recurrent neural network and LSTM neural network in stock prediction.The main research content includes the following three aspects:First,the research and application of BP neural network,recurrent neural network,LSTM neural network and their optimization methods on stock prediction are reviewed.Second,the principles and characteristics of BP neural network,recurrent neural network and LSTM neural network are analyzed.For the linear inseparable problems existing in the perceptron,the BP neural network is used to solve the nonlinear problems;the BP neural network cannot reflect the time-series relationship between the data,and the cyclic neural network can be used to solve the problem,which can solve the shortterm timing Perform memory;for the problem of gradient disappearance in recurrent neural networks,LSTM neural networks are used to solve,which can mine long-term dependencies in time series.The results of simulation experiments prove that the prediction performance of BP neural network,recurrent neural network,and LSTM neural network for stocks is enhanced in turn.Third,an encoder-decoder model using an attention mechanism is added with attention mechanisms in terms of features and time.Both the encoder and the decoder use LSTM neural networks.This method solves two problems in time series prediction: The first problem is that multiple input features have different effects on the target sequence.Using the feature attention mechanism to deal with this problem can obtain the weight of different input features,Obtain a more robust feature association;the second problem is that the data before and after the sequence has a strong temporal correlation.Using the temporal attention mechanism to deal with this problem can obtain the weights at different time points and obtain a more robust timing dependency.Simulation experiment results show that the introduction of attention mechanism can obtain lower prediction errors,which proves the effectiveness of the model in dealing with stock forecasting problems.