基於時間序列、迴歸和正規化的快速預測PM2.5方法

Abstract

隨著城市化和全球化的不斷推進，空氣污染已經成為一個全球性的問題。儘管研究人員一直在試圖找到解決空氣污染的辦法，但空氣汙染中還是有許多問題無法解決，因此透過資料科學進行預測達到預防空氣污染變成了自我保護的重要關鍵。 本研究透過了線性迴歸、正規化、時間序列以及布建在臺灣地區大量的PM2.5感測器預測未來五小時內的PM2.5數值，不同於其他多數預測方法參考了許多參數，我們只參考了PM2.5歷史資料這單一資料。透過上述的觀念我們設計了一項適定性迭代法 (Adaptive Iterative Forecast) 進行預測，能夠根據歷史資料的變化，快速預測出未來數小時PM2.5的數值。 整體研究致力的方向為更快速地建立出準確的預測模型，經由各種比較分析在最後的實驗結果中我們證實已達到了上述成果。我們也將整體研究成果建置成一套預測系統廣泛應用於全臺各地，讓使用者能透過預測出的結果進行個人空氣品質的防護作用。

Air pollution has become a global problem because of the continuous urbanization and globalization. Although, researchers have been trying to come up with solutions to tackle air pollution but still there are many loopholes that need be addressed to have an effective air pollution monitoring system. One way to tackle this problem by using - data science. In this research, we use linear regression and regularization, to forecast the PM2.5 values for the next five hours using PM2.5 data obtained from large scale PM2.5 sensors deployment in Taiwan. Our method is a data centric method and we use historical PM2.5 data to do the forecast. In our work , we designed an Adaptive Iterative Forecast (AIF) method for forecasting, AIF can rapidly forecast the PM2.5 based on the changes in historical data. The goal of the research is to develop efficient and accurate forecast models. Through various comparative analyses, we have proved that our model can achieve significant results. Based on the results, we have also built a forecasting system which is widely used throughout Taiwan. Such a system allows the users to stay aware of the air quality and plan their day to day life.