Prediction of PM10 Concentrations Using Long Short-Term Memory

• Main Authors: Guan-Ting Chen, 陳冠廷
• Other Authors: Chuen-Sheng Cheng
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/b3z3n6

碩士 === 元智大學 === 工業工程與管理學系 === 107 === Air pollution has become one of the most serious problems in the world. The early and accurate prediction of the air pollutant concentrations would be of great value for effective air quality management and public warning. In this study, we proposed using Long Short-Term Memory (LSTM) networks to develop hourly and daily average PM10 prediction models. LSTM can learn long-term dependency information, which is suitable for addressing the characteristic of long residence time of air pollutants. The predictors selected for the prediction model include meteorological variables and some traffic-related air pollutant concentrations. Data used in this study were collected from TAQMN (Taiwan Air Quality Monitoring Network) site of Nanzih District located in Kaohsiung City, Taiwan. The Nanzih District was selected as it is the most-polluted region by particulate matter concentrations in Taiwan. The developed models were based on data collected during 1/1/2013 and 12/31/2017. Various performance metrics were calculated to compare the performance of the proposed model with other competing methods. The results from an extensive comparative study demonstrate that LSTM is a promising method for predicting hourly and daily PM10 concentrations.