Urban Area PM2.5 Prediction with Machine Methods: An On-Board Monitoring System

Recently, particulate matter 2.5 (PM _2.5 ) has drawn more and more attention due to the pursuit of life quality. In Taiwan, PM _2.5 concentration data mostly come from the limited static stations, and their locations are far from streets where people walk or drive by. This might cause the underestimation of PM _2.5 concentration. In this paper, an on-board monitoring system is established to record the PM _2.5 concentration in the surrounding areas where people live. The recorded PM _2.5 concentration data are more accurate than the data from the static stations. In order to get the future PM _2.5 concentration trend, the prediction model is established in this study. A long short-term memory (LSTM) and a gated recurrent unit (GRU) are used as the core of the prediction model due to their ability to analyze time series data such as the PM _2.5 concentration data. And the research results show that the root mean square error (RMSE) of the prediction model using LSTM and GRU is 3.57 and 3.67 in the testing set. The prediction results can provide important air pollutant information to the public and can be used to make better air pollution control policies.