基於機器學習之臺北捷運粉塵檢測

Abstract

本研究探討了機器學習模型在預測台北捷運系統中的懸浮微粒（PM）濃度，特別是PM2.5和PM1.0的應用。這項研究的主要動機是由於人們對地鐵系統的依賴增加，並需要評估可能對通勤者造成重大影響的粉塵暴露危害。通過使用一個配備了PM和環境因素（如溫度、濕度和風速）傳感器的自製裝置，在捷運系統的不同位置（包括入口、售票口和月台）收集數據。 研究中使用並評估了四種機器學習模型——K-近鄰演算法（KNN）、隨機森林（RF）、深度神經網絡（DNN）和循環神經網絡（RNN），以均方根誤差（RMSE）作為衡量標準。在這些模型中，RNN模型表現最佳，具有最低的損失，四種模型預測PM2.5的RMSE為 RNN:0.51~1.53、DNN:1.14~3.38、RF: 1.54~4.12和 KNN:1.76~3.43，預測PM1.0的RMSE為 RNN:0.43~0.79、DNN:1.23~2.36、RF: 1.45~2.25和 KNN:1.53~2.83，證明了其在處理環境監測典型的時間序列數據時的有效性。 研究結果表明，RNN模型在地鐵環境中進行即時預測和監測PM濃度方面最為有效。展望未來，將該模型與自動控制系統相結合，可以實現地鐵站內空氣質量的主動管理。通過主動調節環境來維持安全的空氣質量標準，從而增強公共健康和安全。

This study investigates the application of machine learning models to predict particulate matter (PM) concentrations, specifically PM2.5 and PM1.0, in Taipei's Metro Rapid Transit (MRT) system. The primary motivation behind this research is the increased reliance on subway systems and the need to assess potential dust exposure hazards that significantly affect commuters. By utilizing a custom-made device equipped with sensors for PM and environmental factors such as temperature, humidity, and wind speed, data were collected at various station locations including entrances, ticket gates, and platforms. Four machine learning models—K-Nearest Neighbors (KNN), Random Forest (RF), Deep Neural Networks (DNN), and Recurrent Neural Networks (RNN)—were employed and evaluated using Root Mean Square Error (RMSE) as the measurement criterion. Among these, the RNN model demonstrated superior performance with the lowest loss, proving its efficacy in handling the sequential nature of the time-series data typical of environmental monitoring. The study concludes that the RNN model is the most effective for real-time prediction and monitoring of PM levels in subway environments. Looking ahead, integrating this model with automated control systems could enable proactive management of air quality within subway stations. Such advancements hold the potential to enhance public health and safety by not only providing timely alerts to commuters when unsafe PM levels are detected but also by actively regulating the environment to maintain safe air quality standards. This research paves the way for future innovations in smart transportation systems, combining machine learning with environmental engineering to create safer, healthier public transit environments.