Prediction of short-term freeway traffic volume using recursive least squares and lattice filtering

Abstract

Estimating and predicting the dynamic variation of traffic variables such as volume and speed is becoming increasingly important for intelligent transportation systems applications. In this paper, we present a linear model for the short-term (30 second) prediction of freeway traffic volumes using a recursive least squares algorithm with a lattice filter. An innovative feature of the model is that all of the parameters, such as the optimal weights and the model order, are time-varying and are automatically updated in real-time so that unexpected traffic variations can be addressed by varying the parameters. Recursive filtering algorithms are applied in order to save computation time and storage space. The results of the performance analysis show that the proposed model works well under different conditions, including multiple locations and recurrent and non-recurrent congestion.