Research On Air Quality Prediction Based On Deep Learning Hybrid Model

Data-driven statistical methods which can build air quality prediction models quickly with moderate accuracy have been studied a lot in recent years.However,air quality prediction involves many factors,such as the concentration of pollutants and meteorology.In particular,the meteorological conditions can cause large fluctuations in the concentration of pollutants and increases the difficulty of prediction.In order to deal with this problem,this thesis proposes a deep learning hybrid model WaveNetLSTM that combines deep convolutional neural network(WaveNet)and long-shortterm memory network(LSTM)to simultaneously use pollutant concentration and meteorological factors to predict air quality.The main research contents of this thesis are as follows:First of all,the thesis adds meteorological data into the univariate autoregressive air quality prediction models based on air pollutants,and uses Granger causality test method to study the different contributions of various meteorological factors in predicting air quality indicators.Moreover,this thesis establishes a multivariate air quality prediction model based on Back Propagation Neural Network(BPNN),Recurrent Neural Network(RNN)and LSTM.Compared with the univariate air quality models,the multivariate prediction models incorporating meteorological data have better accuracy.Secondly,since the simple models are difficult to generalize the underlying relationship between various factors,this thesis further proposes a hybrid model WaveNet-LSTM that combines WaveNet and LSTM.The WaveNet-LSTM model can effectively extract the underlying relationship between air pollutants and meteorological data as well as the global features and local features.Finally,the thesis employs the differential evolution algorithm,which aims to make the network structure optimization of the hybrid model automatically,rather than determine the network structure manually,thereby improving the robustness of the WaveNet-LSTM model.The results showed that the hybrid model based on differential evolution algorithm further improved the performance of air quality prediction.