Data-driven Online Modeling and Parameter Estimation of Highway Traffic Flow

Traffic flow models are an important tool to describe highway traffic conditions. With the advent of the big data era, it is necessary to develop a data-driven model to identify the spatial-temporal characteristics of highway traffic flow from partially observed and noisy traffic data. For such a purpose, we propose a data-driven modeling method of highway traffic flow using Dynamic Mode Decomposition (DMD) algorithms. To accurately describe highway traffic flow, a recursive method for updating model parameters is proposed. A small weight term for the historical data and a penalty term for the cost function are considered, which can be more accurate and suitable. Finally, extensive numerical simulations are conducted to validate the effectiveness of the recursive data-driven modeling method. The results show that the proposed recursive algorithm has high accuracy in reconstructing traffic states compared with classical DMD. It may also have the potential to predict the highway traffic flow in real time.