都市交通資訊預測藉由Elman類神經網路與交通網路模型

Urban Traffic Information Prediction using Elman Neural Network and Traffic Network Models

徐瑋(HSU, WEI)

指導教授 ： 熊博安

國立中正大學/工學院/資訊工程學系/碩士(2016年)

摘要

隨著人口的增加，汽機車數量也逐漸增加，這導致交通議題受到重視。因此，智慧運輸系統(Intelligent Transportation Systems, ITS)被提出來改善交通問題與提高運輸效率。在ITS中有個子系統為先進用路人資訊系統(Advanced Traveler Information Systems, ATIS)，ATIS藉由資訊技術與通訊技術，可隨時隨地提供用路人所需的交通資訊，作為路線選擇的參考。
在論文中我們提出一個都市交通資訊預測系統(Urban Traffic Information Prediction System, UTIPS)，此系統包含一個方法與一個模型，分別為交通資料與交通網路資料預處理方法和交通資訊預測模型，其中我們用Elman類神經網路來作為預測模型。交通歷史資料則使用開放資料，我們希望藉由上述的預處理方法與預測模型以及歷史資料來預測未來交通資訊，並在訓練模型及預測時，將上游道路的交通資訊也考慮進來，藉此提高預測的精準度。
在車流預測的實驗結果顯示，我們的預測模型在有參考上游道路時平均絕對百分比誤差(MAPE)可低於10%，不參考上游道路時的MAPE也可以低於12%。

關鍵字

智慧運輸系統 ； Elman類神經網路 ； 開放資料 ； 交通資訊預測

並列摘要

With the increase in population, the number of vehicles on the road has increased rapidly, the traffic issues are becoming the focus of attention. Therefore, Intelligent Transportation Systems (ITS) were proposed to improve the traffic problems and increase the transport efficiency. There is a subsystem of ITS called Advanced Traveler Information Systems (ATIS). ATIS depend on advanced information technology and communication technology to provide the traffic information to drivers in real time. The traffic information can be the references of route choices.

In this Thesis, we proposed an Urban Traffic Information Prediction Systems (UTIPS). This system contains traffic data and network data pre-processing method and traffic information prediction model, which we use Elman Neural Network to be the prediction model. We adopt the open data to be our historical traffic data, and the historical traffic data is used to predict future traffic information with above method and model. In addition, in the training and prediction process of prediction model, we can consider the traffic information of upstream sections to improve the accuracy of prediction results.

Experiments show that the Mean Absolute Percentage Error (MAPE) of our traffic volumes prediction considering the traffic information of upstream section is less than 10%, and the MAPE of traffic volumes prediction without considering that of upstream section is also less than 12%.

並列關鍵字

Intelligent Transportation Systems ； Elman Neural Network ； Open data ； Traffic information prediction

參考文獻

[2] J. You and T. J. Kim. Development and evaluation of a hybrid travel time forecasting

model. Transportation Research Part C: Emerging Technologies, 8(1-6):

231–256, September 2000.

Conference on Geoinformatics, pages 1–6, August 2009.

[4] C. Qun. Dynamic route guidance method based on particle swarm optimization