Predictions Of PM2.5 Concentration And Stock Closing Price Based On Time Series Prediction Model

With the rise of Internet of Things,medical digitization and smart city,time series data analysis and prediction are becoming more and more significant.In the coming years,continuous monitoring and data collection will become more common,the quantity of time series data will grow rapidly,and time series analysis requiring statistical and machine learning techniques will continue to increase.In this paper,experiments are carried out on Beijing air quality data set and stock data set respectively,ARIMA model,GBDT model,LSTM model,LSTM+Attention model,DA-RNN model and Transformer model to achieve single-step prediction of PM2.5concentration and stock closing price,forecast results show that LSTM+Attention model has the smallest single-step prediction error on PM2.5 concentration and stock closing price,MAE,RMSE and MSE of LSTM+Attention model for singlestep prediction of PM2.5 concentration are 0.099,0.122,0.015,MAE,RMSE,MSE of LSTM+Attention model for single-step prediction of stock closing price are 0.018,0.022 and 0.0005 respectively;LSTM model,LSTM+Attention model,Transformer model achieve multi-step prediction of PM2.5 concentration and stock closing price,and the same prediction results show that LSTM+Attention model has the smallest multi-step prediction error on PM2.5 concentration and stock closing price.Compared with statistical prediction model,such as ARIMA model,the Recurrent Neural Network prediction model based on machine learning can capture the timedependent relationships in the time series,however,can only deal with short-time order prediction problems,Long Short-Term Memory Neural Networks compared to the Recurrent Neural Network is more suitable for processing longer time sequences.The researchers in order to explore the input features of impact degree to the prediction results,the attention mechanism is proposed and combined with the Recurrent Neural Network to improve the prediction effect.In this paper,the LSTM+Attention model combined with the Attention mechanism and the Recurrent Neural Network was used to calculate the concentration of PM2.5 and stock closing price,and the prediction error of the concentration of PM2.5 and stock closing price is minimal.