運用時序性資料分析於PM2.5

Abstract

PM2.5影響人類健康，台灣需要減少。因此，Airbox傳感器網絡正在人們圍繞台灣建設，可以測量PM2.5。然而，Airbox傳感器可能會失敗，檢測哪些傳感器無法防止數據中的噪聲很重要。在這項工作中，我們引入了基於動態時間包裝的創新預測模型和新的研究課題。我們收集了Airbox Edimax網頁和台灣環保局的數據。使用它，我們設計了一個新的預測模型，該模型考慮了環境保護署（EPA）傳感器的記錄，並通過使用動態時間包裹（DTW）建立了它們與Airbox傳感器之間的相似性。基於我們的預測模型，我們提出了三種方法來檢測哪些傳感器有故障，並可以從Airbox網絡中移除。此外，重要的是找到我們傳感器周圍的PM2.5顆粒最常見的擴散軌跡。為此，我們設計了兩個模型，通過考慮該地區的Airbox傳感器的歷史數據，可以在確定的區域中找到最常見的擴展軌跡。兩種模式的良好表現都是基於我們實現的廣泛實驗。

PM2.5 affects human health and Taiwan needs to reduce it. Thus, Airbox sensor network is being built around Taiwan by people and it can measure the PM2.5. However, Airbox sensors can fail and it is important to detect what sensor is failing to prevent the noise in the data. In this work,We introduce an innovative predicting model based on dynamic time wrapping and a new research topic. We collected the data from Airbox Edimax web page and Taiwan's Environmental Protection Administration. Using that, we devise a new predicting model which considers the records given by Environmental Protection Administration (EPA) sensors and establish a similarity between them and Airbox sensors by using Dynamic Time Wrapping (DTW). Based on our prediction model, we propose three methods to detect which sensors are faulty and can be removed from the Airbox network. Furthermore, it is important to find the most common spread trajectories for the PM2.5 particles around our sensors. For that purpose, we devise two models that can find the most common spread trajectories in a determined area by considering the historical data of Airbox sensor in that region. The good performance of both models is based on the extensive experiments we realized.