Prediction Of Air Quality Based On Deep Learning

Air quality has occupied more and more attention in daily discussions in China.The concentration of pollutants has exploded.People go out and wear masks.People feel many inconveniences caused by pollution,but we must know that air pollution is a global Problems,but when environmental governance and economic growth are interrelated,the solution of pollution problems often becomes a neglected problem.Respiratory diseases,pneumonia,bronchitis,stroke,chronic diseases,etc.are all hazards caused by air pollution to people’s health.The number of deaths caused by pollution worldwide has also increased sharply each year,and nearly 7 million people have died due to diseases caused by air problems.This data is also supported by data reports from global authorities.Therefore,whether it is to improve the air quality of the individual’s home or to look at the global atmospheric governance,this must be a long-term and arduous task.As a very important branch of artificial intelligence technology,deep learning is very necessary to apply its powerful feature extraction and fitting ability to nonlinear time series data in air quality prediction.This article hopes to predict the future air quality through the numerical values of the main pollutants in the air in the past few days and the corresponding air quality index.We take into account this kind of time series,each input does not exist separately,Before and after are interrelated,past data has an impact on the future,we choose to use a cyclic neural network to make predictions based on historical data,because the cyclic neural network can connect the data before and after,It is equivalent to having memory.It is very suitable to process time series data.In addition to other indicators of the day,the air quality has previous data.The data dimension is increased.The features are extracted and combined,and the CNN in deep learning can be used.In this paper,a mixed model of convolutional neural network and recurrent neural network is used to establish an air quality prediction model.For very long sequences that cannot be processed by recurrent neural networks,we use convolutional neural networks to preprocess long time series,then the down sampling process can convert the original long sequence into a shorter but featureextracted sequence,The extracted data is transferred to the recurrent neural network,which can memorize the information carried in the long-term history and can effectively combine with the current input information,so as to make a good prediction for the future air pollution.At the same time,for the air quality prediction problem we studied,we used machine learning algorithms such as support vector regression algorithm,XGboost algorithm and two variants of recurrent neural network LSTM,GRU to perform regression prediction on the air pollution data we collected.These models are compared to reflect the superiority of the CNN-LSTM hybrid neural network model we created.