應用資料探勘技術於市區幹道旅行時間預測

台灣大道為台中市重要幹道之一，由於中短途旅次較多，在上下班時段抑或是週休假日時段常有大量車流湧入，不僅導致市區道路呈現壅塞狀況，且在進、出城之方向也易造成等候的車隊，進而影響到其他道路之道路行駛品質。因此，在未來車聯網時代來臨，車與車及車與路側設備能互相通訊之下，開發出能即時提供用路人旅行時間資訊之預測模式，使民眾能以即時旅行時間資訊來選擇正確的路徑，達到降低旅行時間、疏導壅塞交通流量於替代路徑及改善路網品質之效果將日益重要。本研究蒐集台中市之車輛Tag資料，針對台灣大道忠明南路至台灣大道文心路口出城方向之四輪小汽車建立平假日旅行時間資料庫，並以本研究建立之旅行時間過濾模式將異常資料過濾，並將資料分為訓練階段資料以及測試階段資料。本研究將訓練階段資料以決策樹之分類與迴歸樹(Classification and Regression Tree, CART)之迴歸樹建立旅行時間預測模式，並採用之貝氏最佳化結合交叉驗證之K-fold方法，進行30次10折交叉驗證後，找尋出最佳參數(最小葉節點大小)以決定出合理的選擇模型的複雜度(迴歸樹的迴歸樹的深度)。本研究在平日之全日預測結果皆較假日之全日預測結果為佳，而不論平日或假日，尖峰時段之預測表現皆較離峰時段來的較差。整體之預測表現，於凌晨時段在MAPE評估標準中，皆可達高精準預測或相當接近高精準預測。不論平日或假日之各個時段，其預測表現在MAPE評估標準中皆屬於優良預測之範圍內。

關鍵字

旅行時間預測；資料探勘；分類與迴歸樹；交叉驗證

並列摘要

Taiwan Boulevard is one of the most important roads in Taichung City. Due to the large number of short and medium range trips, there is often a large of traffic flow during commute hours or weekends, which leads to congestion in urban roads. Therefore, with the coming era of car-to-car connection , it will be more and more important to develop real-time travel time prediction model. With the model, people who driving on the road can choose the right path to reduce their traffic time and divert traffic congestion in the alternative path, thereby improving the quality of road network. In this study, we collected the Tag information of Taichung City, and established the travel time database for the four-wheeled car from the intersection of Zhongming South road and Taiwan Boulevard to the intersection of Wenxin road and Taiwan Boulevard. In this study, we build the travel time fix model to filter the abnormal data., and the data is divided into training phase information and test phase information. In this study, the training phase data were set up with the decision tree of Classification and Regression Tree to establish the travel time prediction model, and use the bayesian optimization combined with the cross validation K-fold method. With 30 times 10 fold cross validation, we find out the best parameters to determine the complexity of a reasonable selection model. The overall predictive result in the early morning session in the MAPE evaluation criteria can reach high precision prediction or very close to high precision prediction. Regardless of weekdays or holidays, the prediction performance is within the range of good predictions in the MAPE evaluation criteria.

並列關鍵字

Travel Time Prediction ； Data Mining ； Classification and Regression Tree ； Cross Validation

參考文獻

[51] 張哲寧，建立風險決策模式於路段機車空間管制，臺灣大學土木工程學研究所學位論文，民國105年

[6] Yang, D.B, ”Application of The ARIMA Models to Urban Roadway Travel Time Prediction”, Systems, Man and Cybernetics, SMC IEEE International Conference, pp.2529-2534, 2006.

[9] Chen, M., and I.J. Chien, ” Dynamic Freeway Travel Time Prediction Using Probe Vehicle Data: Link-based vs. Path-based”, Transportation Research Board 80th Annual Meeting, 2001.

[12] Coifman, B, ”Vehicle Re-Identification and Travel Time Measurement in Real-Time on Freeways Using Existing Loop Detector Infrastructure”, Transportation Research Reacord, pp.181-191, 1998.

[14] Kwon, J., B. Coifman, and P. Bickel,“Day-to-Day Travel-Time Trends and Travel-Time Prediction from Loop-Detector Data” ,Transportation Research Record: Journal of the Transportation Research Board, Vol.1717,pp.120-129,2000.

國際替代計量

應用資料探勘技術於市區幹道旅行時間預測

全文下載

主題瀏覽