Research On Air Quality Prediction Based On WT-WOA-LSTM Model

With the rapid industrialization and urbanization in China in recent years,the consumption of fossil fuels has increased rapidly,and a large number of atmospheric pollutants have spread into the air,causing serious environmental and air pollution.Serious air pollution not only harms the health of the people will also cause huge social and economic losses.As an important indicator to measure the air condition,the air quality index provides an accurate reference for people’s life and travel,and provides accurate suggestions for protecting people’s bodies from the hazards of air pollution.Therefore,the prediction research of the air quality index has important practical significance for the treatment of air pollution.In this context,this thesis selects the air quality index data and PM2.5,PM10,NO2,SO2,CO,O3 six air pollutant concentration data of Wuhan City from January 2018 to June 2021 as the research object,put forward the idea of"decomposition-optimization-reconstruction"on the basis of long and short-term memory neural network to build WT-WOA-LSTM’s air quality prediction model.Firstly,the data are decomposed into several sub-sequences by wavelet decomposition.Then,LSTM neural network is selected as the air quality index prediction model,and the whale optimization algorithm is used to optimize the six parameters of learning rate,discard rate,number of iterations,batch size,and the number of neurons in the two hidden layers in the LSTM network.Finally,the prediction results of each sub-sequence are integrated through wavelet reconstruction to obtain the final predicted value of the air quality index.In order to verify that the WT-WOA-LSTM combined model proposed in this thesis has better prediction performance and stability,this thesis uses the same set of data,through multiple linear regression,SVR,ARIMA,BP neural network five classical prediction models and single LSTM neural network,WT-LSTM combined model for comparative analysis,and MAPE,RMSE,R~2 are selected as model evaluation indicators.The research results show that the prediction error of the WT-WOA-LSTM combined model is only 9.69,which is more than50%lower than the classic prediction models,and 12%lower than the WT-LSTM combined model,and its prediction accuracy reaches 89.49%,which has increased by nearly 50%compared with the classic forecasting model,and it has increased by 4.5%compared with the WT-LSTM combined model.The experimental results confirmed the rationality of the WT-WOA-LSTM combined model and its superiority in prediction accuracy.