利用灰色理論於偵測器遺失資料插補之研究

Abstract

本研究目的為了告知用路人準確的旅行時間預估資訊，最直接的資料蒐集來源為車輛偵測器(Vehicle Detector, VD)，其所回傳的資訊為流量(volume)、速度(speed)與佔有率(occupancy)等。但在實務上常會發生收集的資訊不完全的情況，若忽略偵測器遺漏的情況則會造成旅行時間預估模式發生問題。為了確保其模式預估的正確性，需針對資料遺失值作處理，本研究利用灰色理論不需要符合統計分配的優點，發展出有效的資料遺失值插補方法。在實證分析上以國道三號車輛偵測器為對象，利用灰預測法GM(1,1)和最小循環式殘差修正法（Minimum Recursive Residual GM(1,1), MRRGM(1,1)）在不同資料遺失比例和不同遺漏情境（任意時間段、尖峰時間段、離峰時間段）下，比較兩種演算法的插補結果。 研究結果證實當遺失比例高和多重插補次數多時，以MGGRM(1,1)其插補績效優於GM(1,1)法。在建立插補模式時亦較其他插補理論簡單，且插補績效相當良好。

Purpose of this study to inform road users accurate estimates of travel time information, the most direct source of data collection for vehicle detectors (Vehicle Detector, VD), it returns the information to flow, speed and occupancy and so on. But often occurs in practice the information collected is not entirely the case, if the detector ignores the case of missing travel time prediction model will result in problems. In order to ensure the accuracy of their model estimates, the value of data loss need to be targeted for treatment, this study do not meet the statistical distribution of gray theory, the advantages of developing an effective value of the loss of data interpolation methods. In the empirical analysis on the National Highway No. 3 vehicle detectors for the object, using gray prediction method GM (1,1) and minimum cyclic residual correction method (Minimum Recursive Residual GM(1,1), MRRGM (1,1)) in different proportions and different missing data loss situations (arbitrarytime, peak time, off-peak time), the result of comparing the two interpolation algorithms. The results confirmed that a high proportion of loss when the number of long and multiple interpolation to MGGRM (1,1) the interpolation performance is better than GM (1,1) method.Interpolation mode than in the establishment of other interpolation theory is simple and very good interpolation performance.