| Summary: | Accurate short-term prediction of traffic condition on freeways and major arterials has become increasingly important because of its vital role in traffic management functions and various trip decisions. Given the dynamic nature of freeway traffic, this study proposed two stochastic model approaches, Hidden Markov Model (HMM) and One-Step Stochastic Model, for short-term traffic prediction during peak periods. The data used in the study are real-time traffic monitoring data gathered over 6 years on a 40-mile segment of Interstate-4 in Orlando, Florida. Both approaches are based on the traffic state transition probabilities. The HMM approach defines traffic states in a two dimensional space using both first and second order statistics of traffic parameters. For a sequence of traffic speed observations, the HMMs estimated the most likely corresponding traffic states sequence. Model performance was evaluated using the relative length of the distance between the predicted state and true state to the possible largest distance away from the true state in the two dimensional space. The one-step stochastic model uses traffic speed as the traffic condition indicator. The cumulative negative/positive transition probabilities and expected values were derived from the transition probabilities. The conditional expected value of the most likely transition trend is taken as the predicted speed, which is associated with a probability indicating the chance of such transition happening. The model performance was evaluated using Root Mean Square Errors (RMSE). Relatively small prediction errors were obtained for both approaches (less than 10% for HMMs and around 5 mph for One-Step Stochastic Model), and the model performance was not remarkably affected by location, travel direction, and peak period time. It is concluded that the stochastic properties are the characteristics of freeway traffic by nature and the stochastic approaches are appropriate for short-term traffic condition prediction during peak periods.
|
|---|
