| Summary: | 碩士 === 國立臺北科技大學 === 電資國際專班 === 107 === Air pollution is an essential issue in recent years, especially in most of the populated area with high density. People have anxiety about increasing air pollution. Many factors like emission from the factories, emission from the vehicles, building constructions, wildfires, wood-burning devices, coal-fired power plants, and others cause increasing air pollution. The commonly seen dangerous air pollutants we know are Nitrogen Dioxide (NO2), Ozone (O3), Carbon Dioxide (CO2), Particulate Matter 10 (PM 10) and Particulate Matter 2.5 (PM 2.5). This research has more concern about PM 2.5 because it is the particulate matter that has an aerodynamic diameter less than or equal to 2.5 μm. Size of this pollutant is tiny so that it can be inhaled by human easily and lodge deep in the lungs even may cross into the bloodstream. It can trigger some health problems like asthma, respiratory inflammation and jeopardizes lung function, lung cancers, and others. This research is developed to forecast next hour PM 2.5 based on air pollution concentrations and meteorological conditions. This research also uses station location data to cluster the area and determines the neighbors for each station. This research forecasts next hour PM 2.5 not only using air pollutant concentrations and meteorological conditions in an area but also considering air pollutant concentrations and meteorological conditions in the neighboring areas. The first step in this research is to cluster and determine neighboring areas using station location data. The clustering process uses X-means clustering method to cluster the stations into several clusters. After that, we determine the neighboring areas of each station based on the clustering result. Next, this study forecasted the PM 2.5 in an area not only basing on air pollutant concentrations and meteorological conditions in that area but also taking air pollutant concentrations and meteorological conditions in the neighboring areas. This research uses Long Short-Term Memory (LSTM) as forecasting engine. The result shows that the proposed approach can effectively forecast next hour PM 2.5.
|
|---|
