車輛偵測器數據補償與正規化研究

Abstract

智慧型運輸系統 (Intelligent Transportation Systems, ITS) 常依賴車輛偵測器 (Vehicle Detector) 自動化收集交通資料，供交通管理決策參考。所以偵測器本身是否設計精良，或是否被擺置於適當位置上，可能影響車輛偵測器資料的完整性，或可參考性。研究鑑於漏失資料有礙ITS子系統之先進交通管理系統(Advanced Traffic Management Systems, ATMS) 運轉順暢，特別是即時動態交通控制系統，所受影響最著。又參考價值低的車流資料，不僅可能導致所得之交通管理決策缺乏效率，更無法彰顯投資設置車輛偵測器之價值，故本研究利用中央極限定理 (Central Limit Theorem) 、傅立葉轉換 (Fourier Transforms) 、卡曼濾波器 (Kalman Filter) 及基因演算法 (Genetic Algorithm) 構建資料補償機制；其中插補估計值之計算，同時參考歷史資料及漏失資料發生稍早的資料走勢；並且，歷史資料符合大樣本集中趨勢。漏失資料插補實驗共計測試四種插補方法，其結果顯示：所實驗之四種方法，其插補績效大多控制在平均絕對標準誤 (Mean Absolute Percentage Error, MAPE) 小於 20% 之範圍內。如此結果，依統計學家Lewis (1982) 所提出的衡量標準，本研究之插補方法績效良好；而且，含有卡曼濾波器的方法，因其考量了漏失資料發生稍早以前的資料走勢，可提高資料漏失比率小於 30% 情況下的插補績效，達到精確的水準。而，再可靠的車輛偵測器，若未能佈設在合適的位置上，其所得之交通參數則可能帶有交通號誌控制的干擾，造成資料難以為系統求解交通號誌時制計畫或做成相關管理決策所使用。故本研究利用最佳控制理論 (Optimal Control Theory)，發展車輛偵測器資料正規器 (Regulator)，令車輛偵測器能自動感知其所量測到的資料是否存在交通號誌控制雜訊，並且自動化解調受干擾的資料。實驗分析結果顯示，比較解調前後之車流參數，速率資料之差異率最高達91.45%。如此差異將導致應用解調前後之速率資料所得之黃燈及全紅時段長度相差54% 之多。經檢定，由於經解調後之速率資料幾與未受號誌干擾之速率資料並無顯著差異；足見經解調之車流參數將令系統更能有效得出適合交通實況的交通號誌之黃燈、全紅、清道時間、損失時間及交通號誌時制週期。同時可利用解調後之速率與占有率資料反推未受號誌干擾之車流流率 (Flow)。合上述，本研究嘗試克服二種偵測器於應用面上可能遭遇的問題。為交通資料時間數列提供了穩健的插補機制，支援動態的交通管理系統持續不斷的運作，也為落於受交通號誌控制干擾範圍內之車輛偵測器提供一套數據補救措施，以利動態交通控制系統得出適應交通狀況的交通號誌時制計或相關控制策略。期望所得結果能為先進交通管理系統和ITS做出貢獻。

Collecting traffic data for advanced use in Intelligent Transportation Systems (ITS) heavily relies on the performance of vehicle detectors. The completeness and the accuracy of data measurement depend on the quality of vehicle detector and whether it is installed at appropriate position. Missing data and inaccuracy traffic parameters prevent Real-time Traffic Control Systems (RTCS) from continuous working and decrease control performance,. These disadvantages decrease the value of vehicle detector investment. This dissertation develops a novel compensation mechanism for missing data by integrating Central Limit Theorem, Fourier Transforms, Kalman Filter, and Genetic Algorithm. The interpolation considers the central tendency of historical data of large number and immediate past data in the current day. All experiment results show that the proposed compensation methods perform well under the Lewis (1989) criteria. In addition, the interpolation methods with Kalman Filter perform an accurate result when the missing rate is under 30% by fusing historical data and the immediate past data in the current day. Even though the employed vehicle detectors can work reliably and reply traffic data continuously, installation at improper positions shall make the measurement disturbed by the nearby traffic signal. This dissertation utilizes optimal control theory to dynamically track whether the position of vehicle detector is proper and to automatically regulate the noised data to be similar to those collected in a proper site. The results show that the difference between regulated and non-regulated speed data is up to 91.45%. The difference leads yellow and all-red intervals obtained by non-regulated data differ from those obtained by regulated ones to 54%. The test proves that the regulated speed and occupancy do not significantly differ from those not disturbed by traffic signals. The findings reveal that the regulation is needed for the system to obtain real traffic conditions without the noise from signal control and then to decide the adaptive yellow and all-red intervals, clearance, lost time, and timing plan. In summary, this dissertation proposes efficient methods to compensate missing data and regulate the data disturbed by traffic signal due to improper positions of vehicle detectors. The proposed methods will support RTCS to continuously obtain the data similar to those without the noise from traffic signal. The findings shall contribute to ITS.