Research On Air Quality Prediction Based On Deep Learning

Air quality has become a worldwide concern due to the adverse effects of environmental degradation on human health,economy and livelihood.Numerous studies have shown that air pollution can be very harmful to human health and therefore measures are needed to prevent and control the phenomenon of air pollution and reduce the damage it causes to humans and the planet.With the rapid development of artificial intelligence,deep learning technology can play a great role in the context of the big data era.In view of this,this paper uses deep learning to achieve the prediction of air quality,with the following main work.(1)Processing and analysis of information from multiple sources.Missing important data and unreasonable dimensionality of the data can weaken the representational power of the data set,making the prediction results introduce certain errors.In this paper,different filling schemes are developed for the data according to the distribution of missing data,and the Pearson correlation coefficient method is used to analyse the air quality data and meteorological data to explore the degree of contribution of different characteristic factors to air quality indicators.Finally,the air quality data and meteorological data are integrated on the basis of time series and spatial correlation to complete the dataset.(2)Exploring deep learning-based air quality prediction models.As air quality data is non-linear time series data,traditional prediction models have difficulty in capturing the deep patterns among data through large amount of data,resulting in unsatisfactory prediction results.Therefore,this paper applies deep learning to the air quality prediction task and compares it with machine learning algorithms to explore the advantages of deep learning for air quality prediction problems in the era of big data.Through analysis of the model structure and comparison of experimental metrics,it is found that the gated recurrent neural network GRU can still guarantee prediction accuracy with fewer parameters and faster training speed.Therefore,the GRU model is chosen as the prediction model for air quality,and the attention mechanism is incorporated to propose a GRU-attention model with better extraction capability for important features,which provides the basis for the subsequent proposal of an improved air quality prediction model incorporating spatial features.(3)The GCN-GRU-ATT model for spatio-temporal information fusion is proposed to address the problem of spatio-temporal dependence in air quality prediction.The graph convolution module is used to solve the spatial feature extraction problem of irregular graph structure,and the historical air quality data and meteorological data of different stations in Shenzhen are constructed as spatio-temporal information,and the graph convolution neural network is used to mine the spatial correlation between different monitoring stations for spatial feature extraction;then the gated recurrent neural network is used to capture the observations at different times to achieve spatio-temporal information fusion,and combined with the attention mechanism to further improve the accuracy of air quality prediction.