Highway traffic state estimation using improved mixture Kalman filters for effective ramp metering control

In this paper, we use a cell transmission model based switching state-space model to estimate vehicle densities and congestion modes at unmeasured locations on a highway section. The mixture Kalman filter algorithm, which is based on sequential Monte Carlo method, is employed to approximately solve the difficult problem of inference on a switching state-space model with an unobserved discrete state. We propose a scheme to prevent the risk of weight underflow and to introduce forgetting. The estimation results show that comparable accuracies can be achieved using either a small or a large number of sampling sequences, thus make it possible to carry out efficient online filtering. Underflow prevention and forgetting improves estimation accuracy in our examples. On average, a mean percentage error of approximately 10% is achieved for the vehicle density estimation. The estimation performance is consistent with data sets from various days.