CONFINED SPACE FUZZY PROPORTION MODEL FOR SHORT-TERM TRAFFIC FLOW PREDICTION

Abstract

This paper develops a confined space fuzzy proportion (CSFP) model to predict the short-term traffic flow dynamics. The rationale for prediction reasoning is based on the "fuzzy proportion" inference from similar historical trajectories that are within a "confined space" defined by a spatial threshold in the reconstructed state space. To perform the inference, we assign to each trajectory with unequal weight that is proportional to the relative distance to the latest (present) data vector. The United States I-35 Freeway one-minute flow data are used to construct and validate the proposed model. We employ Theil inequality coefficient (U) and its decomposed proportions to evaluate the prediction performance and to analyze the sources of prediction errors. The empirical results have demonstrated extremely high predictive capability of this CSFP model. The small bias and variance proportions of U statistic further indicate that the prediction model can successfully capture the trends and conspicuous fluctuations of one-minute traffic flow variations.