適應性基因模糊邏輯號誌控制系統

Abstract

基因模糊邏輯控制（GFLC）可克服傳統模糊邏輯控制須主觀設定邏輯規則及隸屬函數的缺點，因此可提昇模糊邏輯控制（FLC）的應用性。基此，本研究建構一個獨立路口的適應性基因模糊邏輯號誌控制系統，以流量、停等車輛數為狀態變數，綠燈延長時間為控制變數，路口總延滯為控制績效指標，並以流體近似法估算之。當每一週期之最小綠燈時間結束或每次推論之綠燈延長時間結束時，則進行綠燈延長時間之再次推論；若所推論的綠燈延長時間為0或已達最大綠燈限制時，則進行時制轉換。 本研究依流量高低（分為低、中、高）及流量型態（分為均勻到達及尖離峰型態）設計六種不同情境，進行GFLC號誌控制系統之績效測試，並與韋伯(Webster)定時號誌比較。為進一步了解GFLC的控制績效，本研究以窮舉法分別求解一階段最佳化定時號誌時制（在模擬時間內已知交通狀況下僅有一種時制），以及多階段最佳化定時時制（在模擬時間內因流量型態具有尖離峰特性而有不同之時制）。情境分析顯示，GFLC模式較韋伯定時號誌減少總延滯0~13.1%。當流量型態為均勻到達時，GFLC模式的總延滯較一階段最佳化時制高0~5.66%；當流量為尖離峰型態時，GFLC模式的總延滯則較一階段最佳化時制減少0.81~4.68%，較多階段最佳化時制高1.78~13.50％。GFLC控制績效較韋伯定時號誌及流量具尖離峰型態之一階段最佳化時制為佳，但比流量均勻到達之一階段最佳化時制及尖離峰型態之多階段最佳化時制為差，顯示本GFLC模式之控制績效仍有進一步提升之空間。 此外，本研究以台北市中正路文林路號誌化交叉路口為例，以驗證本模式之應用性，結果顯示GFLC模式較現場調查之時制控制減少19％總延滯，亦比韋伯定時號誌減少16％總延滯。

Genetic fuzzy logic controller (GFLC) can overcome the drawbacks of conventional fuzzy logic controller (FLC) which has to subjectively set the logic rules and membership functions. Thus, GFLC can greatly enhance the applicability of FLC. This thesis attempts to construct an adaptive genetic fuzzy logic control model for an isolated intersection signal timing control. This GFLC model uses traffic flow and queue length as state variables and extended green time (EGT) as control variable. The intersection total delay, estimated by fluid approximation method, is used to evaluate the control performance. At the moment of ending of a minimum green time or of the previous EGT inference, the GFLC model is activated for another inference of EGT. If the value of EGT is zero or a maximum green time is reached, then the signal switches to the competing direction. Based on three flow volumes (low, medium and high) and two traffic patterns (uniform and varying arrivals), a total of six scenarios are designed and compared with the Webster pre-timed signal control model to verify the robustness of this GFLC model. In order to further validate the control performance of the GFLC model, a fully enumerative method is employed to solve, respectively, for the optimal single timing plan (only one set of signal timing for a given traffic condition during the simulation period) and for the optimal multiple timing plans (several sets of signal timings depending on varying traffic patterns). The scenario analysis shows that GFLC model can reduce total delay by 0~13.1% in comparison with the Webster’s model. Under uniform arrivals, the total delay of GFLC model is slightly higher than the optimal single timing plan by 0~5.83%. Under varying traffic patterns, the total delay for GFLC model is 0.81~4.68% less than the optimal single timing plan but 1.78~13.5％ higher than the optimal multiple timing plans. It indicates that the proposed GFLC model has better control performance than the Webster model and the optimal single timing plan under varying traffic patterns. However, the GFLC model is inferior to the optimal single timing plan under uniform arrivals as well as the optimal multiple timing plans under varying traffic patterns. It suggests that our proposed GFLC model can still be improved, which deserves to be explored. To validate the applicability of our GFLC model, a field study at the signalized intersection of Zhong-Zheng Road and Wen-Lin Road in Taipei City is conducted. The results show that the total delay of GFLC model is respectively 19% and 16% less than that of the current timing plan and Webster’s model.