A review of the application status of multi-agents in urban transportation systems
1 Introduction
This article refers to the address: http://
Agent is an entity with high self-control ability running in dynamic environment, with autonomy, distribution, coordination and certain learning and reasoning ability. The multi-agent system expresses system functions and behavioral characteristics through communication, cooperation, coordination and control between agents. The urban transportation system is a natural, distributed, complex, dynamic, and large-scale system. The use of multi-agent technology to model the urban transportation system can provide a good solution for traffic decision makers and users. From the end of the 1980s, some scholars introduced multi-agent technology into the modeling of urban transportation systems, using their collaboration, storage, intelligence and autonomy to provide users with online decision support, real-time traffic control, or utilization. It accurately simulates the objective world and conducts traffic system operation simulations, discovers problems, laws, or verifies new theories and algorithms in the transportation system. This paper discusses the application status of Agent technology in urban transportation system from three aspects : advanced traffic management system (ATMS), advanced traveler information system ( ATI S) and advanced public transport system (APTS). Finally, it points out the problems and development trends of multi-agent technology in the application of urban transportation system.
2 Multi-agent application in ATMS
In ATMS, Multi-Agent technology is mainly used to provide real-time decision support and appropriate management control. According to the two models of Agent's cautious and reactive models, there are two ideas for constructing an agent-based urban traffic management system: hierarchical hierarchical and fully distributed.
2.1 Hierarchical hierarchical structure
Each level of the hierarchical hierarchical structure is composed of agents with similar functions and structures. The agents at the same level can coordinate with each other. The superior agent can be used as the coordination unit corresponding to the lower agent, and the lower agent transmits the local system environment to the upper agent. The feedback information of the system control provides the decision basis for the superior agent. The earliest hierarchical hierarchical systems were KITS and TRYS.
KITS was developed in 1992-1994 to decompose traffic domain knowledge into a set of elements that match the road network topology, providing a specialized reasoning mechanism for traffic decision making and management. As shown in Figure 1, the underlying Agent performs traffic monitoring and management tasks through collaboration. The Actor is the traffic evaluation and management unit that directly corresponds to the problem area. The Supervisor is responsible for global road condition analysis, interpretation, and synthesis of global action plans. The success of KITS shows that knowledge-based models can be combined with multi-agent technology to improve the monitoring and management capabilities of urban transport systems.
Figure 1 KITS architecture diagram
TRYS is a real-time adaptive traffic management decision system established between 1991 and 1994. As shown in Figure 2, the structure of TRYS is similar to that of KITS. The real-time collected road condition data is accessed by Agent, and the data is analyzed and processed by the internal knowledge base and inference engine. The coordinator is responsible for coordinating the work of each agent to form a global solution. . Unlike KITS, the problem area in TRYS is overseen by an independent, powerful agent.
Figure 2 TRYS architecture diagram.
Based on the theory of hierarchical control structure and the structural characteristics of urban transportation system, Qi Gaoshou and Choy proposed a four-layer hierarchical hierarchical structure: decision-making layer (urban traffic control decision-making system) and strategic control layer (several regional coordination). Control system), tactical control layer (several intersection control systems), execution layer (detector, signal controller, signal light, etc.).
The coordinated control strategy of the structure was further decentralized to the intersection level on the basis of TRYS, and the intersection agent was established. Each intersection became an intelligent knowledge system, which can timely deploy and adjust the control strategy according to the traffic conditions at the intersection. It adapts to the dynamic and real-time characteristics of the traffic system, and has a good adaptability and adjustment ability for sudden changes in traffic flow.
2.2 Fully distributed structure
In a fully distributed structure system, Agent relies on its own knowledge and intelligence to coordinate with adjacent area agents to complete the intersection control. The initial application was the Spanish TRYSA2 system, as shown in Figure 3. The TRYSA2 Agent has a set of control plans, each of which is given a utility value that reduces traffic pressure. The system can synthesize the optimal solution for the system by evaluating the planned utility values ​​of the relevant Agents. Oliveira, Cheng Xiangjun, Yang Zhaosheng and other scholars have also proposed a fully distributed control structure with the intersection agent as the basic control unit . The Agents in the system all have certain storage, matching and intelligent computing functions, which can rely on a good coordination algorithm. Realize coordination and cooperation between multi-agents to achieve overall optimization and control.
Figure 3 TRYSA2 architecture diagram.
2.3 Performance Comparison of Two Architectures
The hierarchical hierarchy fully reflects the organic combination of centralized and decentralized control. Considering the overall interests, coordination can be carried out purposefully, but the implementation of regional agents and master agents is slightly more complicated. Fully distributed has the characteristics of rapid response and strong flexibility, which can give full play to the autonomy and coordination of the agent. However, due to the limited capabilities of the agent and the scattered knowledge of the system, the ability to solve global problems is slightly insufficient. Coordination mechanisms can have a large impact on system performance. In terms of scalability, fully distributed, the new agent can be expanded into the current agent group by registering the new agent with other agents and modifying the corresponding solution and knowledge base, while the hierarchical stepping needs to integrate the regional control center. And the main control center, re-grant the priority relationship of each agent. In the complexity of collaboration, hierarchical hierarchical selection selects a local optimal scheme from each agent control scheme, and it is completely distributed in all agents to find the best solution through search strategy, so the latter has a large workload. .
2.4 Coordination Control and Optimization of Multi-Agent
Multi-Agent realizes the distributed parallel operation of the system by coordination, and improves the execution efficiency of the task. In the multi-agent based ATMS, there are three coordination modes: 1 to establish a special coordination agent; 2 to distribute the coordination behavior to each agent, and the agent to complete it autonomously; 3 the combination of concentration and distribution, the agent itself can complete Some coordination actions can be accepted by the high-level agent. Currently commonly used coordination methods include blackboard models, game models, coordinators, and exchange of opinions.
The blackboard model has a large amount of information transmission and has certain requirements for the stability of information transmission, and is suitable for simple distributed multi-channel control. The game theory model is applicable to the coordination between the upper and lower agents of the hierarchical hierarchical structure and the peer agents of the fully distributed structure. However, due to the complex equilibrium point convergence control required in the repeated game process, the calculation based on the traffic information game Larger. The coordinator can synthesize proposals generated by peers and subordinate agents into global proposals based on certain goals. The coordinator reduces the communication complexity of the system and the implementation complexity of other agents, but increases the design complexity and computational complexity of the coordinator agent itself. The exchange of opinions method has great requirements on the stability of system communication. When a communication failure occurs in a single Agent node, the system will not work normally.
It can be seen from the analysis of the above several methods that the coordination process needs to transmit a large amount of data, so it is easy to cause congestion of the transmission network. At present, many scholars use intensive learning methods to optimize local traffic information. The reinforcement learning method is based on the enhanced signal provided by the environment as feedback for performance evaluation, and completes the mapping from state to behavior, especially suitable for dealing with the changing road network environment. Baher and Ou Haitao have studied real-time adaptive traffic signal control based on reinforcement learning, reducing the large number of communication requirements between intersection nodes and enhancing the reliability of decision making.
2.5 Related Application Studies
Ronald studied the possibility of dynamic traffic management equipment collaborating by modeling separate and independent transportation facilities into agents that can cooperate with each other. Filippo implements a multi-agent-based traffic management system, CARTESIUS, which demonstrates good collaborative reasoning and conflict resolution capabilities in the analysis of sporadic obstructions and online development of integrated control schemes. It can coordinate traffic management personnel across multiple regions. Road network congestion in fast lanes and ground streets provides real-time decision support.
Bo Chen et al. integrated mobile agent technology into traffic management system, enhanced the ability to deal with uncertain events and dynamic changes of the environment, and proposed a real-time traffic detection and management system based on flexible agent.
3 Multi-Agent Application in ATI S
ATIS can influence travel behavior and enhance road network performance. Currently using Agent technology to study ATIS is mainly to build a variety of intelligent travel information systems for different travel needs, to provide high-quality travel information and navigation services for travellers; to study the behavior of travellers under ATIS conditions and ATIS for urban transport Impact.
3.1 Agent-based typical travel information system framework
In order to achieve effective coordination between road network managers and travellers, it is necessary to effectively allocate road networks based on time and space based on the use preferences (travel type, route selection, departure/arrival time, etc.) that do not seriously affect individual travellers. . Based on this, Adler and Blue studied the Intelligent Travel Information System (IT IS), which provides travel plans and navigation assistance information for travellers, and proposes a vehicle-mounted intelligent navigation agent that can travel, define and calibrate routes and travel. Plan preferences. On this basis, they also proposed a conceptual framework based on Multi-Agent Traffic Management and Path Navigation Collaboration (CTMRGS), which enables effective coordination and communication between road network managers, information providers and travellers. The system uses principle negotiation to guide the interaction between the travel agent Agent and the information provider Agent, finds a time-space optimal travel plan, and finally points out that more intelligence will be used to capture and present the true intentions and behavior of the traveler.
3.2 Research on Traveler Behavior Based on Multi-Agent Simulation under the Influence of ATIS
The effectiveness of ATIS depends on the ability of the system to provide information and the traveler's response to travel information. Therefore, it is especially important to understand the behavior of travellers and their decision-making process under travel information, which will help to design an efficient ATIS. At present, many scholars at home and abroad use Agent simulation method to study travellers in the ATIS environment. behavior.
Dia first proposed using multi-agent simulation to study driver behavior under the influence of real-time traffic information. Through the investigation of driver behavior (characteristics, psychology, knowledge, preferences, etc.), BDI (belief-wage-intention) structure modeling is used, and traffic simulation components are used to evaluate the impact of real-time traffic information on driver behavior. Based on the BDI architecture, Rossetti proposed a multi-agent extension model based on DRACULA (a dynamic path assignment model combining user learning and micro-simulation) to model travellers, allowing travellers to make rational choices about travel and departure times.
Driver behavior can affect the benefits of the ATIS system and the overall performance of the system. Rossetti models the traveler agent based on the predicate logic expression, which makes more psychological factors for the traveler. The simulation results show that the overall performance of the system will be affected by the travel information demand and the traffic network topology. When the travel information is provided to the individual alone, the overall impact can be greatly improved.
Joachim models the traveler as an agent, and analyzes the traveler's path selection behavior in the ATIS environment based on two parallel path road networks. The research indicates that the characteristics of travel information greatly affect the potential benefits of ATIS. On the basis of Joachim, Zhao Wei observes the macroscopic features of the system "emergence" by establishing an agent-based simulation model for the microscopic behavior in the system. The simulation results show that ATIS has certain influence on the travel planning before the commuter travels. As the traffic volume increases, the uncertainty of the traffic system also increases, and the ATIS system revenue will increase.
Zargayouna proposed an Agent-based traveler information service center architecture. By instantiating a large number of traffic entities, an environment-based service, information resource and pedestrian active interaction support model were established, allowing entities to establish their own interesting interactions.
Wahle proposed a real-time traffic flow online simulation and prediction framework based on multi-agent. The combination of historical data and current dynamic data can provide short-term prediction of path selection behavior and traffic direction. Wang Jian uses the decision tree method in data mining to obtain the traveler information demand, and uses the Agent technology to establish a mobile service-based information service network framework. Chou built a multi-agent based parking navigation negotiation network, which modeled the car, parking lot and navigation system into Agents. Through the cooperation of each agent, the car park with the best price and route was selected for the driver.
4 Multi-agent application in APTS
4.1 Multi-Agent based bus operation status detection
The detection of the running status of the bus is of great significance for ensuring the punctuality and operation of the bus system. Using the AVM system to obtain bus operation data for disturbance (delay and super-line) detection lacks an overview of the global road conditions and has poor stability, and it is difficult to provide a state of the road condition based on space-time two-dimensional. Therefore, Flavien proposed the use of multi-agent technology to diagnose bus disturbances and to detect the continuity of transmission data. The bus and the station are modeled as Agent, and the station agent is equipped with a bus operation schedule, which is responsible for calculating the dispatch after the bus arrives at the station; the bus agent is responsible for reporting the actual status of the road network to the STOP Agent, and the theoretical time for the STOPAgent to arrive at the vehicle. The current actual time is compared to detect bus disturbances. On this basis, they dynamically model the entire life cycle of the disturbance, and integrate the disturbance model into the multi-agent decision support system to study the influence of disturbance on the road network activity. The model consists of three information areas: the successor area (delayed bus follow-up site), the key area (the station where the bus is delayed), and the previous area (the bus station's predecessor station). As shown in Figure 4, the lowest level STOP Agent receives the information from the BUS Agent, and the STOPAREA Agent in the middle layer is responsible for collecting information from the STOP Agent to synthesize traffic evaluation, passenger flow information, road condition progression coefficients, etc. The top-level INCIDENT Agent forms a comprehensive Real-time scheduling decisions.
Figure 4 is based on a hierarchical multi-agent bus disturbance detection framework.
4.2 Multi-Agent based bus system operation simulation
The operation simulation of the bus system can be used to adjust bus scheduling, evaluate the structure of the bus network, and formulate strategies. David used the multi-agent simulation method to describe the operation of the bus system, modeling buses and travelers as Agents, all models combined with bus operations, traveler behavior and road traffic load. In this paper, a number of logit models are used to meet the traffic demand, and the utility of walking, car and bus modes is evaluated. Based on the utility model, the travel routes and traffic behaviors of the travellers are formulated. The simulation focused on bus passenger load and passenger waiting time.
The simulation results show that by modeling the bus and traveler as Agent, it is convenient to simulate various conditions (saturation, shortage) that may occur during the operation of the bus, and to effectively formulate the occurrence of special events (accidents, obstructions). Scheduling strategy.
5 Conclusions and prospects
In the future, ITS will be covered by various intelligent and autonomous agents in the entire transportation system. Through the Internet, wireless network or self-organizing network connection, the information will be collected continuously to make intelligent decisions, and finally the transportation system will be thoroughly intelligent. In order to make the Agent play a greater role, it is necessary to fully consider the characteristics of the urban transportation system and its embedded entities (travel mode characteristics, traffic rules, road network structure, travel mind) in practical applications, alleviate communication needs and reduce the amount of calculation. And coordinate complexity, optimize system organization, and enhance system stability and security. The future research direction of multi-agent in urban transportation system should focus on the following aspects:
(1) Information fusion of multiple Agent systems, such as sharing information between traffic management systems, travel information systems, navigation systems, and parking systems, and improving the operational efficiency of road networks and the quality of travel information services by coordinating the work of multiple systems;
(2) In view of the problems existing in the urban transportation system, study the multi-agent system structure, coordination algorithm and organization optimization technology for specific application fields, and form a standardized technical system, including communication environment, modeling method, evaluation method, etc.
(3) Introducing more new Agent technologies into the design of urban transportation systems, such as mobile agent, Agent specification, Agent architecture, Agent communication and language, Agent organization and alliance, Agent learning and planning, Agent negotiation and coordination, etc. New technology;
(4) The application of Agent technology theory in urban transportation has formed a certain scale. How to more effectively utilize the characteristics of Agent to make it more closely combined with urban transportation will become a new research hotspot;
(5) The wide application of Agent will introduce more artificial intelligence, system engineering, control theory, optimization algorithm and distributed computing technology into the actual traffic problem solving, and provide more new ideas for the specific application of Agent.
1. Indoor LED Curtain Poster Display with price Advantage: We have the best competitive price in the market, we also have the same or even better quality.
2. Indoor LED Curtain Poster Display with quality advantage: From material to finished product, from design to production, we control all processes.
3. Indoor LED Curtain Poster Display service advantage: We also provide best service after sales and we prioritize every customer's needs.
4. We ensure that our LED Curtain Poster Display are CE certified, ROHS compliant, FCC identified.
5. Indoor LED Curtain Poster DisplayWith specially designed fast lock system, easy to install and dismantle.
6. The new structure design enablesIndoor LED Curtain Poster Display diversified installation to meet the hanging-up and stacking requirements.
7. Full Color Indoor LED Curtain Poster Display with High contrast rate up to 5000:1 by using black LED.
8. Indoor LED Curtain Poster Display with high refresh rate with NOVA control system, no scan line when photographed.
Indoor LED Curtain Light Box,LED Curtain Slim Light Box,LED Curtain Poster.
Shenzhen Macion Optoelectronics Technology Co.,Ltd. , https://www.macion-led.com