從空氣品質監測站的歷年監測資料來看,台灣空氣污染指標以懸浮微粒為主。懸浮微粒是直徑小於10微米的可吸入顆粒物,會對身體健康產生威脅。 本研究蒐集影響空氣品質最有相關性的四種因素(氣溫、風速、風向、相對溼度)。把資料庫中有異常值去掉並擷取有價值的因素資料後,所用資料及合計有109訓練樣本及78測試樣本。再用繼承式基因演算法(Inheritable Bi-objective Genetic Algorithm, IBCGA)及支持向量機(Support Vector Machine, SVM)來預測空氣品質,IBCGA是以智慧型基因演算法(Intelligent Genetic Algorithm, IGA)加入繼承式機制並最佳化適應性函數。本文從中選出影響懸浮微粒最重要的特徵,以建立懸浮微粒的模型,然後可用來預測空氣品質。經實驗結果分析,TEMP(氣溫)與RH(相對溼度)為影響空氣品質最重要的兩個因素,並且使用此方法對於空氣品質預測的準確率可達74.5%。 由本文提出的方法可降低空氣品質監測站所耗費的時間和人力等資源,目的使空品站的效率提高,形成一個完整的防護網路。
According to the statistics from sensing data for numerous years, the concentration of aerosols is the major index of Taiwan's air pollution. The aerosols refer to suspended particles smaller than 10 microns in diameter which are respirable particulate matter, a threat to health. In this study, we collected four most relevant factors (temperature, wind speed, wind direction, and relative humidity) affecting the air quality. The used dataset consists of 109 and 78 samples for training and test respectively, obtained by removing outliers and extracting informative factors in database. The study applied an inheritable genetic algorithm (IBCGA) and support vector machine (SVM) to establish a mathematic model of suspended particles for predicting air quality and analyzing the most important factors.The prediction accuracy of independent test was 74.5%. Using factor analysis, the most important two factors affecting air quality were temperature and relative humidity.The proposed method can help the air quality monitoring stations reduce time cost, manpower, and other resources to advance efficiency of the air quality monitoring stations in Taiwan, and establish a complete air protection network.