Dynamic Reliability Of Stochastic Transportation Network Under Incident Condition

Road network’s performance presents stochastic under the influence of external factors. These factors include recurrent factors and non-recurrent factors. Recurrent factors include capacity degradation and traffic demand variation due to day-to-day congestion, the character is continuous for a long time; Non-recurrent factors include capacity degradation due to accidents and significant meetings, the character is that it only influences network during duration time.Non-recurrent congestion caused by traffic incident and other emergencies has great impacts on the network service level. Incident often brings random capacity degradation for partial links, breaks the original equilibrium of network, leads to sharp fluctuations in network performance and declines network efficiency. From the perspective of travelers, non-recurrent congestion caused by incident and other emergencies has weak predictability and is difficult to respond before trip, which may lead to travel time increase greatly. Therefore, analyzing interaction between network travel time and route choice probability under incident condition has important significance in real-time traffic management and control. Definitions of dynamic travel time reliability and capacity reliability are presented to describe responsive ability to incident based on travel behavior. One hand, reliability data can be issued to travelers for basis of route choice before departure, and route choice behavior considering reliability can decrease travel time variation due to incident; the other hand, for network managers, reliability can be used to analyze vulnerable links, assess network performance and provide guidance for traffic manage and incident prevention.In order to compute network reliability under incident condition, analysis of road network condition and route choice behavior is necessary component. Therefore, Travel behavior model under incident and definitions of dynamic reliability are constructed in this dissertation in order to provide theoretical support for network performance analysis during incident. The main contributions of this dissertation include:(1) The existing models on dynamic network loading are reviewed including speed-density loading model, cell transmission model and link transmission model. Speed-density loading model is based on the function of speed and density; the other models are based on the function of flow and density. All the three models include two parts——link model and node model. Speed-density loading model has easier computation and larger errors due to the assumption that vehicles distribute homogeneous on the whole link. Cell transmission model has computational complexity due to calculation for every cell and link transmission has storage complexity.(2) Improvement for dynamic network loading is presented and a quasi-dynamic model is established based on Link Transmission Model and logit principle. For speed-density loading model, diverge-merge model and speed-density function are adopted to calculate link accommodated vehicles and travel time of non-incident links firstly. Then through analyzing traffic flow evolution of incident link, Kinematic Wave Theory is used to compute queue length and travel time of incident links.Methods of computing travel time and flow density based on cell transmission model and link transmission model are presented, interaction and evolution rule between network travel time and route choice probability is presented. The results indicate that:route travel time and route choice probability present concussed during incident period and queue dispersing period; the queue spot transfers during incident duration period and clearing period.(3) In order to analyze en-route behavior, mixed-logit model is adopted to compute route conversion probability at nodes. Combined with link transmission model, influence factor value is calculated and route conversion probability value is computed by improving node model of LTM. Through analyzing interaction and evolution rule between network travel time and route choice probability under en-route condition, the results indicate that:route travel time and route choice probability present concussed, the differences exist that:The amplitudes of route travel time and route choice probability reveal smaller due to compliance of original routes before departure.(4) In order to compute network reliability under incident, a definition of dynamic travel time reliability is presented. Distributing incident duration time, through establishing a quasi-dynamic model based on Link Transmission Model and logit principle, the number of arrived vehicles and travel time is obtained; Setting incident duration time, incident impact and incident spots as stochastic variables, Monte-Carlo method is adopted to calculate travel time reliability. The results indicate that:larger travel demand corresponds to lower reliability; larger time threshold corresponds to higher reliability; larger mean value of incident duration time corresponds to lower reliability; change trend of reliability corresponding to duration time variance presents ascent and descent with different time threshold. Lower reliability corresponds to more serious incident and reliability changes with different incident spots.(5) In order to compute network reliability under incident, a definition of dynamic capacity reliability is presented. Through establishing a quasi-dynamic model based on Link Transmission Model and logit principle, the vehicle number of entering into network is obtained; setting incident duration time, incident impact and incident spots as stochastic variables, probability analysis method is adopted to calculate capacity reliability. The results indicate that:Larger capacity demand corresponds to lower reliability; larger mean value of incident duration time corresponds to lower reliability; change trend of reliability corresponding to duration time variance presents ascent and descent with different time threshold.