Development of lane changing decision model for heterogeneous traffic operation using adaptive neuro-fuzzy interface system (ANFIS)

Abstract

Lanc-changing behavior models are important components of microscopic traffic simulation tools. Particularly, heterogeneity of traffic can significantly affect lanechanging behavior. The objective of this thesis is to analyze lane changing factors and related decision making sequences to develop a framework for modeling the lanechanging behavior in local traffic scenario. The intention is to provide an improved lane-changing model with a generalized and flexible structure that will be capable of providing the lane-changing behaviors of drivers in differing situations. The proposed model is generalized to overcome to a certain extent the limitations of the existing lane-changing models and fill the present gap in this arena. It is also intended to provide means to train and validate the model parameters using local traffic and behavioral data and also for cross-checking and calibration facilities. The proposed modcl is built using the Adaptive Neuro-Fuzzy Interface System (ANFIS). Various membership functions of lanc changing parameters were documented. Assignment of mcmbership values for each class was done such that the obtained output gave a realistic measure of the likely outcome of different factor combinations to reflect truly as much as practicable driver levcl decision towards lateral movement while moving in a powered-nonpowered mix traffic stream. To train the ANFIS model the field data acquisition method was also highlighted which used varbalisation technique to quantify drivcr's intention towards lateral movement. Using series of input/output data set, the model constructs a fuzzy inference system (FIS) whose membership function parameters are tuned (adjusted) in combination of a least squares type of method and a backpropagation algorithm. This allows fuzzy systems to leam from the data they are modeling. In this regard, the validation, checking and calibration methodologies are also discussed. This model could be used as an embedded tool for traffic simulation software, particularly to mimic lane changing bchavior utilizing local data. This model could also be used for the estimation of risks particularly for road accidents relating to lane change and also for assessing the relationships among lane change maneuvers, traffic delay and congestion phenomena.