State Dependence in Lane-Changing Models

Abstract

Lane changes have a substantial impact on traffic flow characteristics. The lane-changing model is therefore an important element in microscopic traffic simulations. Lane changing is commonly modeled in two steps: lane choice, which captures the desire to change lanes, and the decision about whether a desired lane change can be completed, which is captured by gap acceptance models. Most current models assume that these decisions are repeated at every time step of the simulation independently of previous decisions. However, it may be more realistic to assume that drivers persist in their lane choices, and so their desired lane at any time point depends on earlier choices. To capture persistency in lane-changing behavior, a model that integrates a hidden Markov model (HMM) structure is presented. The evolution of lane choices, which are the underlying hidden states, is modeled using a Markovian process. The observed lane-changing actions depend on these hidden lane choices. An important difficulty that arises with this model structure is the problem of unobserved initial conditions on the hidden states. A method to address this problem is proposed. Estimation results of the resulting model are presented and compared with a model that does not incorporate state dependence.