外文翻译---多智能体系统优化城市交通

《外文翻译---多智能体系统优化城市交通》由会员分享，可在线阅读，更多相关《外文翻译---多智能体系统优化城市交通（14页珍藏版）》请在毕设资料网上搜索。

1、PDF外文：http:/ Multiagent System for Optimizing Urban Traffic John France and Ali A. Ghorbani Faculty of Computer Science University of New Brunswick Fredericton, NB, E3B 5A3, Canada  Abstract  For the purposes of managing an urban traffic system, a hierarchical multiagent system that consis

2、ts of several locally operating agents each representing an intersection of a traffic system is proposed. Local Traffic Agents (LTAs) are concerned with the optimal performance of their assigned intersection； however, the resulting traffic light patterns may result in the failure of the system when

3、examined at a global level. Therefore, supervision is required and achieved with the use of a Coordinator Traffic Agent (CTA).A CTA provides a means by which the optimal local light pattern can be compared against the global concerns. The pattern can then be slightly modified to accommodate the glob

4、al environment, while maintaining the local concerns of the intersection.  Functionality of the proposed system is examined using two traffic scenarios: traffic accident and morning rush hour. For both scenarios, the proposed multiagent system efficiently managed the gradual congestion of the t

5、raffic. 1 Introduction The 20th century witnessed the worldwide adoption of the automobile as a primary mode of transportation. Coupled with an expanding population, present-day traffic networks are unable to efficiently handle the daily movements of traffic through urban areas. Improvements to road

6、 networks are often confined by the boundaries of existing structures. Therefore, the primary focus should be to improve traffic flow without changing the layout or structure of the existing roadways. Any solution to traffic problem must handle three basic criteria, including: dynamically changing t

7、raffic patterns, occurrence of unpredictable events, and a non-finite based traffic environment 2. Multiagent systems provide possible solutions to this problem, while meeting all necessary criteria. Agents are expected to work within a real-time, non-terminating environment. As well, agents can han

8、dle dynamically occurring events and may posses several processes to recognize and handle a variety of traffic patterns 3, 5. Although several approaches to developing a multiagent traffic system have been studied, each stresses the importance of finding a balance between the desires of the local op

9、timum against a maintained average at the global level 4. Unfortunately, systems developed to only examine and optimize local events do not guarantee a global balance6. However, local agents are fully capable of determining their own local optimum. Therefore, a more powerful approach involves the cr

10、eation of a hierarchical structure in which a higher-level agent monitors the local agents, and is able to modify the local optimum to better suit the global concerns 7.  The remainder of this paper is organized as follows. Section 2 examines the problems of urban traffic. The design of a hiera

11、rchical multiagent model is given in Section 3. The experimental results are presented in Section 4. Finally, the conclusions of the present study are summarized in Section 5.  2 Urban Traffic Congestion Improvements to urban traffic congestion must focus on reducing internal bottlenecks to the

12、 network, rather than replacing the network itself. Of primary concern is the optimization of the traffic lights, which regulate the movement of traffic through the various intersections within the environment. At present, traffic lights may possess sensors to provide basic information relating to t

13、heir immediate environment. This includes road and clock sensors, measuring the presence and density of traffic and providing the time of day to the traffic light. A solution to the urban traffic problem using agents is to simply replace all decision-making objects within the system by a correspondi

14、ng agent. Even the most basic system will consist of several agents, leading to the creation of a multiagent environment. In this case, the traffic environment is broken down into its fundamental components, with one agent for each of the traffic lights within the system. To maintain organization an

15、d cooperation between the Local Traffic Agents (LTA), a Coordinator Traffic Agent (CTA) exists to monitor global concerns and maintain order. 3 Hierarchical Multiagent Model for Urban Traffic To achieve a balance between the local and global aspects of an urban traffic system, a multiagent system ba

16、sed on a hierarchical architecture is proposed. LTAs and CTAs make up the fundamental levels of the hierarchy, in which the LTAs meet the needs of the specific intersection, and the CTAs determine if the chosen patterns of a LTA are suited to meet any global concerns. A solitary Global Traffic Agent

17、 (GTA) may exist for networks of sufficient size, and an Information Traffic Agent (ITA) provides a central location for the storage of all shared information within the system. For each agent, the variables necessary to organize and maintain the hierarchy are listed. The development of this system,

18、 in which several LTAs work under the guidance of a single CTA, represents the backbone to a hierarchical structure of agents within the system. The CTA provides the bonds between itself and the LTAs of the system, requiring that the CTA store a list of the neighboring intersections for each of the

19、LTAs. However, the computational capabilities of a single CTA are limited, and a road network of sufficient size may require the use of multiple CTAs to handle all of the LTAs within the system. In this circumstance, the network will be subdivided into regions controlled by a single CTA, with a top-

20、level Global Traffic Agent (GTA) linking the CTAs together. The GTA is an optional agent, existing only if the network is sufficiently large that it is required. A LTA interacts at a global level by sending a message containing the calculated optimal local light pattern to its supervising CTA. The C

21、TA will find the appropriate neighboring intersections, and then determine what the global optimum for the handled LTA will be. To calculate the global optimum, the CTA will require all information relating to each of the neighboring intersection. The CTA will request the information from the ITA by

22、 providing a list of the intersections the CTA is concerned with. Once this information is retrieved, a CTA calculates the global optimum and determines if a variance exists between the local and global traffic light patterns. If a significant difference is found, a balance between the local and glo

23、bal optimums must be negotiated, and then returned to the LTA. 4 Implementation The proposed urban traffic multiagent system has been implemented using the JACK Development Environment, utilizing JACK Intelligent AgentsTM. JACK uses the Belief Desire Intention (BDI) model. Under this framework,“the

24、agent pursues its given goals (desires), adopting appropriate plans (intentions) according to its current set of data (beliefs) about the state of the world.” 1. Agents created under the JACK environment are event-driven, and can respond to internal or external events occurring within the system The

25、 first phase of implementing the multiagent system involves the creation of LTAs. Each of these agents are tailored to meet the requirements of its corresponding intersection.For the purposes of this project, the traffic network consists of six intersections. Each intersection consists of two roads

26、crossing over one another. Each approaching road posses two lanes, a left-turning lane, and a straight/rightturning lane. The decision-making capabilities of the LTAs is developed in the second phase. The first round of decisions by a LTA are concerned with finding the local optimum, with no conside

27、ration for neighboring intersections. A basic expert system divides the sensor inputs into a corresponding light pattern. The resulting light pattern consists of an eight-element array, which can be broken down into two elements for each of the North, East, South and West directions. Odd elements of

28、 the array (zero is the first index) specify the duration of the advanced green state for each of the appropriate directions, while even elements indicate the time of the straight/right-turning lanes. This light pattern is always in the same format, and once calculated, stored by the LTA. The values

29、 contained within the array consist of strings, indicating the duration of the traffic light. The values of the strings are as follows: Red: Red light, lanes remain in a stopped state. Short: Green light, most frequently occurring, 30-seconds in duration for straight directions, 15 seconds for leftt

30、urning lanes. Medium: Green light, often for above average traffic densities,45-seconds in duration for straight directions, 25 seconds for left-turning lanes Short: Green light, indicating a high traffic density, 60-seconds in duration for straight directions, 35 seconds for left-turning lanes. Onc

31、e the optimal local traffic light pattern is calculated,the LTA sends a message event to the CTA. The traffic light pattern is passed to the CTA, allowing the CTA to adjust the LTAs light pattern to better meet any global concerns. Stored within the CTA is a vector of neighbors for each LTA within t

32、he system. When a CTA receives a message event from a LTA, the CTA gathers all information relating to the neighbors of the currently handled LTA from the ITA. The CTA will use this information within its own expert system, comparing the local optimum light pattern against the current densities of t

33、he neighboring intersections. If a significant difference is found between the local optimum and the essence of the global optimum, the traffic light pattern to be implemented is altered to reduce the difference between the two optimums. The new traffic light pattern is returned to the LTA for imple

34、mentation within the traffic light. 4.1 Experiments This sections presents some of the experiments carried out for two fixed state scenarios. In each experiment, a list of variables is provided to initialize the current state of the environment. Once the state of the environment is established, each

35、 LTA goes through the process of changing the state of their traffic light to accommodate the other direction. The resulting traffic light pattern for each intersection is recorded, and the number of vehicles passing through the intersection, N, in the available time indicated by the traffic light p

36、attern is calculated as N = T/( + )where  and  represent the ideal amount of time required for a vehicle to pass through a traffic intersection and the latency increase to the ideal length of time due to unexpected events, respectively. An advanced form of this calculation would allow the

37、latency value of _ to increase by a constant factor for each additional segment of the waiting vehicles. This can be demonstrated by using  to represent each of the latency groups, imposing a maximum number of vehicles that exist within each latency group. Let the number of vehicles found in latency group k is calculated as,