Keynote: Air pollution source identification by using Neural Network with Bayesian Optimization

Abstract

In recent years, many researchers and experts have engaged themselves in air pollution research, mainly focusing on how air pollution impacts human health, how to accurately predict air pollution concentration and sources, and how to reduce and avoid air pollution? Actually, these problems need to be urgently solved for human being’s living environments. If we can know the sources of air pollution, no matter whether it is produced by local manufacturing factories, vehicles or other pollution sources outside our environment, it will be easier for us to push these sources to reduce pollution and/or avoid causing pollution. Many researchers use machine learning techniques to predict the concentration of air pollution and enhance the accuracies of their models so that people can acquire accurate pollution information in advance to avoid exposing themselves in the polluted environment. To the best of our knowledge, currently, there is no research which predicts air pollution in a wide area. In fact, it is not easy to locate contamination sources according to the distribution of pollution concentration in such an environment. Therefore, in this research, we will do this. We first create an air-pollution sensing network in a relatively smaller grid area to collect environmental data, such as air pollution concentration, wind speed and wind direction. Data are tuned when necessary to prevent the model built by using diffusion models of pollutants from being seriously affected by outliers and other unstable factors, like wind direction. The purpose is to more accurately find out the sources, and identify possible distribution of pollution. After that, the built model is applied to predict the sources of air pollution in a wide area. The tensorflow is utilized to establish three neural network analytical models, with which to find the sources of air pollution. Source identification accuracies of these neural networks are compared with other air diffusion models, aiming to develop one which is suitable for predicting the sources of air pollution.