適應性類神經模糊推論系統於跟車模式之應用

Abstract

傳統跟車理論探討駕駛人刺激與反應間的確定性關係，並假設所有的駕駛 人具相同的駕駛行為，即在相同的刺激下，會有相同的加減速反應，此種 確定性關係及相同之駕駛行為可能與實際車流行為不符。為改善此缺失， 本研究嘗試應用適應性類神經模糊推論系統(ANFIS)，依跟車間距將駕駛 人區分為高、中、低三種冒險度，在分別建立其加減速之模糊推論規則， 並結合類神經網路之學習能力來修正駕駛人對於跟車間距及前後車相對速 度之隸屬函數。本研究觀測行駛於中山高速公路４２位駕駛人之跟車行為 ，利用ANFIS模擬不同狀況下之駕駛人跟車間距變化，結果發現各類冒險 度駕駛人均能得到一趨於穩定的跟車間距，且達到穩定跟車間距的時間與 駕駛人的冒險度成反比關係。 Conventional car-following theory discusses the deterministic relationship between drivers'stimulus and response.It assumes all drivers are homogeneous which suggests that different drivers have indentical acceleration under the same stimulus. Such determiniistic relationship and homogeneous driving assumption do not well describe the real world traffic conditions.In order to make the conventional theory more realistic,this study first treats the drivers as non-homogeneous by classifying them into three groups based on their distance headways.Fuzzy inference rules of acceleration for each group are then established.The drivers'membership functions of distance headways and relative speeds are modified via the learning ability of neural network. Forty-two drivers car- following behavior are observed on the mainline of Sun-Yat Sen national freeway.The variation of distance headways under different conditions are simulated by adaptive neuro-fuzzy inference system(ANFIS).It is found that a nearly stable traffic condition will be obtained for each driver group.The more aggressive drivers are,the less time they reach a stable car- following distance headway.