Expectation-Maximization Based Parameter Identification for HMM of Urban Traffic Flow

摘要

This paper concerns on modeling of traffic flow as a hybrid approach that combines continuous and discrete dynamics in a system. The model is chosen as simple as possible such as a Hidden Markov Model (HMM). Traffic flow can be classified into two-states and switching between two states controlled by first-order Markov chain with a certain probability. The model is characterized by several Gaussian parameters and estimated by using Expectation-Maximization (EM) technique. Actual traffic flow data on City of Jakarta and Bandung is used to model through EM estimation parameter and to validate the results by using particle filter. The results confirm that the proposed model gives satisfactory results which capture the variation of traffic flow. This work is easily extended to Jump Markov Model as a more general model especially relating to the development of traffic control design based upon queue length.