車輛偵測器資料漏失之在線插補技術研究

Online Imputation for Missing Data in Vehicle Detectors

本研究利用傅立葉轉換及卡曼濾波器發展車輛偵測器漏失資料在線插補技術。插補作業僅以資料本身型態進行有效率的插補，資料包括較其他研究更為充分的歷史資料，以及漏失資料發生稍早之資料走勢，期符合統計學上對不偏估計的要求，並獲得良好的插補績效。實驗結果顯示，本研究所提出之方法應用於不同資料漏失程度的插補作業皆有良好的績效表現，其主要貢獻為提供穩定的在線資料插補技術，讓工程師們可依應用上之需求及限制，參考本研究之分析結果選擇適當的方法，為連線式交通控制系統提供一套防範因資料漏失造成系統失靈（failure）的即時漏失資料插補技術，支持此類系統持續運作。

This paper combines the Fourier transform and Kalman filter to develop a novel real-time technique to impute missing data in an online traffic control system. The proposed imputation technique uses only yet abundant historical data, including short-term and long-term historical data and the trend prior to the missing values, to meet the requirement of unbiased estimation in statistics. The results indicated that the proposed technique can effectively impute missing data and thus repair specific series of traffic parameters under different levels of missing rates. The contribution of this paper is to provide a reliable online imputation technique, which allows traffic engineers to sustain the operation of online traffic control architectures without being interrupted by missing data.