PM2.5 Prediction Based On LSTM Neural Network

In recent years,with the rapid development of China's economy and the acceleration of industrialization and urbanization,the air pollution problem focusing on PM2.5 has become increasingly prominent,seriously affecting people's production and living.Therefore,achieving accurate prediction of PM2.5 has important practical significance and social value.This paper designs an improved LSTM(long and short-term memory)circulatory neural network PM2.5 prediction model because the formation mechanism and process of PM2.5 are more complex,with many components,nonlinearity in time and space,and multiple data dimensions..LSTM(long-short memory type)circulatory neural network structure is complex,it can nonlinearly fit PM2.5 related data,effectively consider the timing of the input data,and achieve time series data encoding and decoding.The LSTM(long-and-short-term memory)circulatory neural network model,through deep learning of large data samples and self-characteristic selection,can better reveal the essential relationship between PM2.5 and influencing factors,and improve the prediction accuracy of PM2.5.This paper designs a PM2.5 prediction model with missing value data.The forecast uses air pollutants and meteorological data as the impact factors.In the absence of input features,short-term PM2.5 predictions,missing values can be used in timesteps.Instead of the moving average of 7,the long-term PM2.5 prediction,the missing value can be replaced by a moving average with a time step of 20,and the experimental results are highly accurate.Based on the Tensorflow platform,the paper improves the PM2.5 prediction model of LSTM(Long-term and short-term memory)and selects experimental data from two cities in Beijing and Guangzhou and converts it into standardized data.By setting and changing experimental LSTM(long and short-term memory)cyclic neural network model parameters,the optimization effect of deep learning can be achieved.This paper designs three groups of comparative experiments:SVR,random forest algorithm,and LSTM(long and short memory)cyclic neural network model.The experimental results show that compared with the SVR algorithm and the random forest algorithm,the LSTM(long and short-term memory)neural network model adds the seq2seq model,and the prediction rate is the highest,and the precision can reach within 8 hours.In this paper,the LSTM(long-short-term memory)neural network is used to realize the prediction of PM2.5,and the PM2.5 prediction method is innovated.