Research On Construction Of Air Quality Forecast Model Based On Neural Network

Air quality issues have a huge impact on people's physical and mental health.In recent years,with the increasing living standards of urban and rural residents,people have higher awareness and requirements for the living environment,and more attention to air quality.The establishment of air pollution prediction models can promptly issue early warning information,and provide strong technical support for the steady improvement of environmental quality.Early warning information for heavy pollution weather and abnormal air quality is issued in advance,and the pollution control period and area are accurately presented,and a heavy pollution process analysis report is formed in time;targeted remediation and pollution traceability can provide directions for governance.After the pollution process is over,the predicted value of the model and the actual monitoring result are compared to scientifically evaluate the effectiveness of the management and control of the heavy pollution process to ensure continuous improvement of air quality.It has certain research value.In this paper,the concentration of PM2.5 pollutants in Baotou City is used as the prediction object,and the atmospheric environmental pollution is used as the research background.The historical monitoring data of China's online air quality monitoring and analysis platform is used to build a prediction model of air quality based on neural network in Baotou City.This article builds a prediction model based on the experience of the predecessors based on the knowledge learned at the graduate level.Since the concentration of air pollutants has a strong correlation with time,the use of LSTM(Long Short Term Memory)long-term and short-term memory neural network can well deal with this kind of memory-related problems.The data in the experiment selected data from December 2,2013 to September 30,2019,a total of 2127 data,respectively from different aspects and multiple sets of comparative experiments,using relevant model evaluation indicators to evaluate the model's advantages and disadvantages.The experimental results show that while controlling certain variables to remain unchanged,setting different types of parameter values to the prediction model has a relatively obvious impact on the final prediction performance.Next,we conducted experiments on the optimization methods in the neural network.The experimental results show that the more advanced optimization methods may not be suitable for this experiment,which may be related to the data size and historical data itself in this experiment.After the parameters and optimization methods of the neural network model are determined,they are brought into the model for final prediction.Through the analysis of the experimental prediction results,the prediction accuracy of the prediction model built using the LSTM neural network will decrease with the increase of the prediction time step.The LSTM neural network model is only suitable for predicting the concentration of pollutants in the short term.But the prediction effect for a long period of time is not very good.Finally,we also built an ARIMA(Auto Regressive Integrated Moving Average Model)integrated moving average autoregressive model based on time series prediction.By comparing with the model built using LSTM neural network,the experimental results show that the prediction model built based on LSTM The prediction error and accuracy are significantly better than the ARIMA prediction model.