異常交通狀況下之高速公路基因模糊邏輯速限控制模式

Abstract

高速公路上經常因事件導致道路容量急劇的減少，例如連續的施工區域也會對上游交通產生負面的影響，甚至引起事故發生。為了避免已產生交通壅塞之區域與上游高速接近之車輛間產生衝擊波，可以合理的事先以某一間距逐漸降低上游車輛的速限。然而，目前大多數的速度控制策略在研究時的改善目的僅改善交通的安全性而犧牲效率且在多數先前的研究中速度限制的下降比例皆是主觀設定，缺乏最佳化的機制。就此而言，無法保證速限控制的績效。因此，必須去發展一個動態最佳化速限控制模式，能夠將流量最大化而碰撞危險度最小化。 基此，本研究提出一個基因模糊邏輯控制(genetic-fuzzy logic controller ，GFLC)-基礎模式，能在異常交通狀況下找出最佳化的可變速限控制模式。兩個主要的目標分別是流量和安全性最大化。第一個目標是事件持續時間前後60分鐘內所通過之總車輛數，第二個目標是碰撞危險度，其計算是參考Abdey-Aty et al. (Sources: Accident Analysis & Prevention Vol.38, 2006)所提出之包含五種變數的數學函數：(1)影響時間前5-10分鐘事故地點的平均佔有率，(2)影響時間前10-15分鐘事故地點下游1英哩的平均佔有率，(3)影響時間前5-10分鐘事故地點下游1英哩的流量標準差，(4 )影響時間前5-10分鐘事故地點下游0.5英哩的平均流量，(5)影響時間前10-15分鐘事故地點上游0.5英哩的平均流量。 GFLC模式是一個反覆演化的模式，由Chiou and Lan (source: Fuzzy Sets and System Vol.152, 2005)提出，在本研究之運用是由從即時的上游交通狀況及每分鐘判斷一次之事件嚴重性來決定最佳化的速限降低程度。三個狀態變數包括平均速度、流率及事件嚴重程度(計算事件佔用總車道數之比例)，控制變數為上游每一公里設置之可變速限號誌其所增加的速限降低程度，皆為語意等級五之變數。為了評估學習後的邏輯規則及隸屬函數之績效，採用格位傳送模式(cell transmission model，CTM)，一個中觀的車流模式由Daganzo (Source: Transportation Research Vol.28B, 1994)所提出，運用於模擬交通流體力學的行為。事件的發生和速限改變對交通行為產生的影響可能都會降低道路的容量，流量-密度關係圖和方程式都會因此而修改並影響車輛從一個格位傳送到其它格位的行為。為了確保修正後的CTM模式不管是在正常或異常交通狀況下都能模擬實際交通行為，本篇將三種案例應用於台灣國道一號高速公路並加以驗證，結果顯示修正後的CTM模式能夠精確的預測下游的交通流量，平均絕對百分比誤差(MAPE)分別在自由流、交通壅塞和車道封閉等情形下皆低於13%。 為了完成邏輯規則之學習和隸屬函數之校估，總共有六種不同嚴重程度和交通狀況之事件被隨機產生，並將這六個實際事故資料分類成三種嚴重程度(佔用一、二和三車道)和兩種交通狀況(尖峰和離峰)去驗證本研究所提出模式之適用性和績效。經由與無速限控制策略比較之結果顯示，本研究提出之GFLC-基礎可變速限控制模式可以降低碰撞危險度1.02%~14.49%而增加總流量-2.24% ~ -0.91% 。值得注意的是，在三車道的高速公路上，提出的模式在尖峰時段事件佔用一車道時大部份能夠有效降低碰撞危險度，而尖峰時段事件佔用兩車道也是如此。然而，事件佔用三車道時提出之模式只能輕微減少碰撞危險度，研判是由於如此嚴重之事故，最終所有車輛都會停止下來，所以改善績效有限。同時，在大部分事件中，提出之模式都會輕微減少總流量。 關鍵字：速限控制、格位傳送模式、基因模糊邏輯控制

The abrupt capacity reduction caused by incident sites such as tandem work zones would negatively affect the efficiency of upstream traffics and even induce accidents. To avoid the shockwaves generated from the conflicts between approaching high-speed traffics and gridlock traffics, it is reasonable to gradually lower down the speed-limits of upstream traffics at a certain distance ahead. However, most existing speed-control strategies implemented in practice aim only at improving traffic safety at the expense of efficiency and the strategies to determine the reduction of speed-limit proposed by most previous studies are subjectively set without an optimization mechanism. As such, the performance of speed-limit control can not be guaranteed. Thus, it is essential to develop a dynamic optimal speed-limit control model which can maximize the throughput while minimizing the crash likelihood. Based on this, this study proposes a genetic-fuzzy logic controller (GFLC)-based model for optimal variable speed-limit control under abnormal traffic conditions. Two objectives respectively representing throughput and safety maximization are considered. The first objective is the total number of vehicles passing through the incident site within 60 minutes before-and-after the time of interest. The second objective is the crash likelihood, calibrated by Abdey-Aty et al. (Sources: Accident Analysis & Prevention Vol.38, 2006), a function of five variables: (1) the log of average occupancy at the station of interest 5–10 min before the time of interest, (2) the log of average occupancy 1 mile downstream of the station of interest 10–15 min before the time of interest, (3) the standard deviation of volume 1 mile downstream of the station of interest 5–10 min before the time of interest, (4) the average volume 0.5 mile downstream of the station of interest 5–10 min before the time of interest, and (5) the average volume 0.5 mile upstream of the station of interest 10–15 min before the time of interest. The GFLC model with an iterative evolution algorithm, proposed by Chiou and Lan (source: Fuzzy Sets and System Vol.152, 2005), is then employed to optimally determine the reduction of speed-limit depending on the real-time upstream traffic condition and estimated severity of the incident at every 1 minutes. Three state variables include average speed, flow rate, and severity degree of the incident (represented by the number of lanes blocked), each with five linguistic degrees. The control variable is the incremental reduction of speed-limit at every variable speed-limit sign upstream (at a distance of 1 kilometer). In order to evaluate the performance of learned logic rules and tuned membership function, cell transmission model (CTM), a mesoscopic traffic flow model proposed by Daganzo (Source: Transportation Research Vol.28B, 1994), is employed to approximate the traffic hydrodynamic behaviors. To account for the capacity reduction due to the incident occurrence and the effect of speed-limit change to the traffic behaviors, the fundamental diagrams and the equations governing traffics moving from one cell to another are revised accordingly. To ensure the revised CTM model can replicate the freeway traffic behaviors under normal or abnormal traffic conditions, three field cases on Taiwan Freeway No. 1 are examined. The results show that the revised CTM model can accurately predict downstream traffic flow rate with mean absolute percentage error (MAPE) less than 13% under free-flow, gridlock, and lane-blocked conditions, respectively. To perform learning the logic rules and tuning the membership functions, a total of six incidents with various severity degrees and traffic conditions are randomly generated. A total of six real accidents of three severity degrees (one, two and three lane blocked) and two traffic conditions (peak and off-peak) are then collected to validate the applicability and performance of the proposed model. The results show that, compared with the strategy without variable speed-limit control, the proposed GFLC-based variable speed-limit control model can curtail crash likelihood by 1.02%~14.49% and increase total throughput by-2.24% ~ -0.91%. It is worth noting that for a three lane freeway, the proposed model can most effectively curtail crash likelihood for one lane blocked incidents under peak traffic condition, followed by two lanes blocked incidents under peak traffic condition. However, for three lanes blocked incidents, the proposed model only slightly curtails crash likelihood, since eventually all vehicles will be stopped under such a severe accident. Meanwhile, in most of incidents, the proposed model will slightly reduce total throughput.

Keywords: Variable speed-limit control, genetic fuzzy logic controller, cell transmission model