Parallel Computing, Intelligent Optimization And Hybrid Model Research For Traffic Simulation System

Intelligent transportation system (ITS) is a main method to solve the transportation problem. As an important part of ITS, traffic simulation system is a very important tool and experitmental instrument to traffic management, control, optimization and so on. But the computation speed problem of traffic flow model in traffic simuluation system prohibits the taffic simulation system using in the large traffic scene, especially in micro-traffic simulation system. With the further progress of ITS, traffic control system needs to be applied in complex traffic scene. The demand for real-time of traffic control system is higher and higher. As the kernel of traffic control system, traffic optimization algorithm also has computation speed problem. Parallel computation is available to solve computation speed problem, which is fast developing recently. It also brings new reaserch problem to traffic simulation system and optimization algorithm in traffic control system.To account for this reaserch problem, the author explores the key theory and technology of parallel computing for traffic simulation and optimization algorithm in traffic control system from three aspects. The first is setting up a parallel computing framework for microscopic traffic flow model. The second is presenting a parallel hybrid traffic flow (PHTF) model which is enlightened on the research of parallel computing framework for microscopic traffic flow model. The third is putting forward the parallel ant colony optimization algorithm based on the layered construction graph decomposition (PACO-LCGD algorithm).Furthermore, traffic network model is the base of traffic simulation system. It still needs more research in traffic network model which can describe real micro-traffic network elaborate, flexible and unify macro-traffic network and micro-traffic network in a same framework. From this view, a lane-based hybrid traffic network model is presented. This model has been successfully achieved in the simulation and analysis system for urban mixed traffic (SASUMT) which is developed by the intelligent transportation system research center of Zhejiang University.In this thesis, the author also expounds the recent research findings of SASUMT and discusses some software and hardware implements related to parallel computing for traffic simulation system and optimization algorithm in traffic control system.The major research works done in this thesis are listed as follows:1. A new parallel computing framework is presented according to the parallel computing framework for cellar automatic (CA) model, which is in order to reduce the complexity of parallel computing for microscopic traffic flow model. The framework includes three parts: grid domain decomposition, boundary buffer model and improved synchronization strategy. The grid domain decomposition is used to achieve distributing the workload in parallel computing and the boundary buffer model is presented to supply the interactive data. The improved synchronization strategy uses the barrier to achieve synchronization based on the property that the messages can be parallel transferred in the cluster, which also refers to the synchronization strategy of CA model. A large scale transportation scenario is used in the experiment to prove this framework. And the LogGP model is used to theoretical analysis. The theoretical analysis and experiment indicate that the parallel computing for continuous trafficflow model is easy to accomplish by using this framework and the computation speed of continuous traffic model is improved. It means that this framework meets the demand for real-time, high-efficiency and dynamic simulation of a large-scale traffic network in the intelligent transportation systems.2. A parallel hybrid traffic hybrid flow model is presented in response to idea of reducing the sizes of message, which is based on the parallel computing for microscopic traffic flow model and hybrid traffic flow model. It also uses the parallel computing framework for the microscopic traffic flow model. In each slave computer, it still uses the microscopic traffic flow model for computation. The difference is that the transferred messages are changed into macro-statistics of boundary buffer, which are the data of each vehicle in the parallel computing for microscopic traffic flow model. When another computer receives this macro-statistics, it will generate new cars based on these data and computation using microscopic traffic flow model. Because the PHTF model can reduce the message sizes, it will have higher computation speed and expansibility.3. A PACO-LCGD algorithm is presented here. The computer capacity of ant colony optimization algorithm is exponentially increased if the stage number and decision variable dimension of complex multi-stage decision problem are increasing. It caused that the ant colony optimization algorithm couldn't be computed by a single computer. PACO-LCGD algorithm can figure out this problem. The parallel ant colony optimization algorithm decomposes the construction graph into some parts and each part is placed on different computer. The whole computation task is accomplished by mutual cooperation in the computers which join in the computation. Experiment and theoretical analysis by the LogP model verifies that it can solve this problem and improve the computation efficiency.4. A lane-based hybrid traffic network model is presented here. It uses lanes as the primitive topological objects of data model, of which geometry depends on itself. The basic topology of traffic network is composed of a set of lanes and nodes. The complex topologies between lanes are represented by the feature-based modeling method. Intersection in the traffic network is described as the polygon feature that has many virtual lanes in it. The micro traffic network and macro traffic network are integrated in a same framework. In the last SASUMT version, it adopts this model.5. The recent research findings of SASUMT are expounded in this thesis. Parallel computing for traffic simulation system and optimization algortihm is discussed in it. It discusses and accomplishes the software and hardware implement of the parallel computing for traffic flow model and PACO-LCGD algorithm. It also presents a principium framework of the hardware implement for traffic simulation and optimization system to achieve parallel computing.6. The work of this dissertation is summarized and the prospective of further research orientation is discussed.