應用隸屬函數及倒傳遞網路於都市通勤旅程預估模式之研究

Application of Membership Function and Back-Propagation Network on Urban Commute-Journey Forecasting Model

黃俊勳(Chun-Hsun Huang)

指導教授 ： 廖祐君

中原大學/工學院/土木工程研究所/碩士(2001年)

https://doi.org/10.6840/cycu200100444

摘要

個人一日旅程型式之決策受到能力限制、聯結限制及權限限制之影響而決定旅程活動之排程。個人為使其旅程效益最佳化，在實際旅程中傾向以旅次鏈結方式安排旅程。本研究以旅次鏈為分析單位，並以不同分析方法建構都市通勤旅程預估模式。
以迴歸方法分析及建構模式時，往往必須滿足諸多基本假設；但都市旅運行為受到外在環境、個人家戶社經特性及家戶成員互動影響，其複雜性通常很高，實際旅運資料很難滿足迴歸分析之種種基本假設。近年來部分研究應用倒傳遞網路探討旅運行為[5][15][16]，藉由隱藏層與各神經元間之相互鍵結以反映實際旅運行為，其推估結果大致皆較迴歸方法理想。本研究將倒傳遞網路結合模糊理論隸屬函數，將部分影響較顯著之語意變數透過隸屬函數之處理，進一步考慮旅運行為中所存在之模糊特性，以提昇模式推估能力。本研究分別以邏輯斯迴歸、倒傳遞網路及倒傳遞網路結合隸屬函數等方法建構都市通勤旅程預估模式，並對模式推估結果進行比較評估。
利用81年台北都會區住戶交通旅次調查之資料所建構之模式顯示，透過隸屬函數與倒傳遞網路之結合，以輸入及輸出變數模糊化模式之建構結果較其他模式好。透過隸屬函數之結合，對反映旅運行為中所存在之複雜性及模糊性有相當助益。此外，性別特性亦使旅運決策存在顯著之差異性。研究中分別建立男性與女性之行為模式，結果顯示，女性之模式推估結果優於男性及整體資料所建立之模式。

關鍵字

旅運行為 ； 邏輯斯迴歸 ； 倒傳遞網路 ； 隸屬函數 ； 旅次鏈

並列摘要

The arrangement of daily journeys and activities depends on capability constraints, coupling constraints and authority constraints faced by individuals. In order to optimize the activities out of journeys, travelers tend to chain trips along the way to or from work. In this study, we use trip chains as analysis units and establish urban commute-journey forecasting models based on various methods.
By using the regression method, numbers of assumptions need to be satisfied. However, the activity-travel behaviors are often affected by external environments, personal/household social-economics characteristics and the household-member interrelationships. The complexity makes the assumptions hard to be fulfilled. Recently, some studies apply back-propagation networks (BPNs) to simulate travel behaviors. By links of hidden layers and neurons, BPN models are capable of reflecting travel behaviors to a certain degree and generating better results than regression methods. In this study, back-propagation networks are combined with fuzzy membership functions to reflect the fuzziness in travel behaviors in order to improve the forecasting ability of models. Logistic regressions, back-propagation networks (BPN) and back-propagation networks combined membership functions (FBPN) are utilized separately to establish forecasting models on urban commute journeys.
By using the travel data collected in Taipei metropolitan area in 1992, the forecasting results generated from the combination of back-propagation networks and membership functions are better than other models. Furthermore, the input and output fuzzification models (FBPN-all) seem to perform well among FBPNs. By using the membership functions, it is helpful to reflect the complexity and fuzziness existing in travel behaviors. In addition, certain personal characteristics, especially gender, make significant differences during the decision-making process of travel. In the latter section of this study, separate models based on genders are also established for comparison. Following the model evaluation, the results indicate the models by female travelers generate much better forecasting than the counterpart.

並列關鍵字

travel behavior ； trip chaining ； logistic regression ； membership function ； back-propagation network

參考文獻

6.邱裕鈞、藍武王，民國86年，「模糊類神經網路學習演算法之推導與應用－以鋪面診斷為例」，運輸計劃季刊，第26卷，第2期，頁233-252。

27.Akiyama, T. and C. F. Shao, (1993) “Fuzzy mathematical programming for traffic safety planning on an urban expressway,” Transportation Planning and Technology, Vol.17, pp.179-190.

28.Avineri, E., J. Prashker and A. Ceder, (2000) “Transportation projects selection process using fuzzy sets theory,” Fuzzy Sets and Systems, Vol.116, pp.35-47.

29.Bhat, C. R. and F. S. Koppelman, (1993) “A conceptual framework of individual activity program generation,” Transportation Research A, Vol.27A, No.6, pp.433-446.

30.Buckley, J. J. and Y. Hayashi, (1994) “Can fuzzy neural nets approximate continuous fuzzy functions? ” Fuzzy Sets and Systems, Vol.61, pp.43-51.