【作者】 高宏岩；

【导师】 刘法胜；

【作者基本信息】 山东科技大学 ， 控制理论与控制工程， 2011， 博士

【摘要】 随着交通需求的快速增长,高速公路交通拥挤日益严重。为保障高速公路的安全、高效运行,本研究利用手机定位技术构建了高速公路交通参数采集平台,在已有的研究成果基础上,较为深入地研究了高速公路交通拥挤识别、预警和调控方法,为高速公路管理工作提供相关的理论依据和决策支持。在利用手机定位技术构建的高速公路交通参数采集平台上,分析了随时空变化的移动信号流和高速公路交通流之间的关系特性,提出了一种新的手机定位地图匹配方法,以图论聚类方法提取了观测路段的独立车辆数,给出了独立车辆车型识别方法以及平均行程速度、密度和交通量参数的估计算法,结合仿真实验分析了影响交通参数采集精度的因素。根据高速公路交通参数周期性变化特性,建立了交通参数季节自回归求和移动平均预测模型、等维新息灰色预测模型和广义回归神经网络预测模型。综合三种预测模型的优点,建立了最小方差组合预测模型,实验表明最小方差组合预测模型精度优于三种单一预测模型。针对高速公路交通状态划分中的模糊性和随机性,提出了一种高速公路交通状态云识别模型。在确定云识别模型输入参数基础上,利用云合成理论和云相似度的定义,计算了待识别状态的拥挤度,给出了拥挤确认和拥挤类型判别条件。仿真实验结果表明：云识别模型不仅能够准确地对高速公路交通拥挤状态进行判别,而且能够反映出交通拥挤的程度和拥挤的变化过程。此外,将云识别模型和交通参数多步预测模型结合,实现了对常发性拥挤发生时间、路段、可能性以及拥挤程度的预测。总结分析了影响高速公路交通事件持续时间的因素,分别基于粗糙集理论和最小二乘支持向量机建立了事件持续时间和事件影响因素之间的关系模型,实现了事件持续时间的滚动预测；在事件持续时间滚动预测和交通参数多步预测基础上,考虑了车辆到离率随机变化情况,分别采用随机排队理论和冲击波理论建立了排队长度和拥挤持续时间滚动预测模型,给出了两种预测模型的具体实现步骤和流程图,并通过实例比较了两种模型的预测效果；在事件持续时间和排队长度滚动预测基础上,研究了适用于解决偶发性拥挤的调控分流方法,以最大排队长度作为警戒指标,分别采用随机排队理论和冲击波理论推导了调控分流量和调控作用施加时刻之间的数学关系。分析了高速公路入口匝道定时调节、感应调节和匝道协调调节的特点和原理,提出了适用于解决局部常发性拥挤的单点入口匝道模糊控制方法,设计了控制规则易于调整的修正因子模糊控制器。针对入口匝道定时协调调节的不足,建立了由宏观交通流模型、匝道排队长度模型和拥挤路段排队长度模型组成的匝道动态协调控制模型,采用粒子群优化算法对协调控制模型进行了求解。通过仿真实验对上述控制方法的有效性进行了验证。更多还原

【Abstract】 With the rapid growth of traffic demand, freeway traffic congestion becomes increasingly serious. In order to ensure safety and efficiency of freeway, this research constructs the platform for freeway traffic parameters collection with mobile phone location. Based on the prior founding and knowledge, research on freeway traffic congestion identification, early-warning and control are carried out and developed, which provides theoretical base and decision support for freeway management.Based on the platform for freeway traffic parameters collection, relationship between mobile signal flow which varies with space-time and freeway traffic flow is analyzed. A new map matching method for mobile phone location is proposed. The methods to collect vehicle counts with graph clustering, to identify vehicle type and to estimate mean travel speed, traffic density and volume are presented. The factors which affect traffic parameters collection accuracy are further discussed through simulation.According to characteristics of cyclical changes for traffic parameters, seasonal autoregressive integrated moving average prediction model, grey prediction model and general regression neural network prediction model are established. Integrating advantages of three prediction models, minimum variance combination model is set up. Case study shows that minimum variance combination model predicts more accurately than the single model.In view of the fuzziness and randomness of classification for freeway traffic state, a cloud identification model for freeway traffic state is proposed. Based on determination of input parameters, congestion degree for the identified state is calculated according to synthesized cloud theory and cloud similarity definition. Besides, the conditions for congestion verification and congestion type discrimination are also given. The simulation result shows that the proposed approach not only identifies the traffic state accurately, but also reflects degree of congestion and process of congestion change. Furthermore, occurrence time, section, possibility and congestion degree for recurring congestion is predicted by combining cloud identification model with traffic parameters multi-step prediction model.By analyzing factors which affect freeway incident duration, the relationship model between incident duration and its influential factors are set up respectively based on rough theory and least square support vector machine to predict incident duration dynamically. Based on incident duration dynamic prediction and traffic parameters multi-step prediction, considering random variation of arrival and departure vehicles, stochastic queue and Shockwave theory are respectively used to formulate dynamic prediction models for queue length and congestion duration. The step and flow chart of the two prediction models are presented in detail, and their prediction results are compared through case study. On the basis of incident duration and queue length dynamic prediction, the diversion method suitable to solve nonrecurring congestion is developed. It takes maximal queue length as warning index and respectively uses stochastic queue and shockwave theory to deduce the mathematical relationship between diversion volumes and diversion exerted time.The characteristics and principle of local fixed-time metering, local responsive metering and coordinated fixed-time metering is analyzed. An on-ramp fuzzy control approach suitable to solve local recurring congestion is proposed. A fuzzy controller with correction factor which can adjust control rules is designed. In view of weakness of coordinated fixed-time metering, a dynamic coordinated control model composed of macroscopic traffic flow model, queue length model of on-ramp and queue length model of congestion section is established and solved by particle swarm optimization. The effectiveness of the proposed control approaches are further verified with intensive simulation.更多还原