A parameter identification algorithm for the METANET model with a limited number of loop detectors

This paper analyzes the parameter identification of the macroscopic traffic flow model METANET. In previous papers, this calibration has been done by minimizing numerically the difference between the data and the prediction of the model. Results indicate that this optimization usually falls in quite suboptimal local minima, especially when there are sensors only available in some segments. The authors propose an identification procedure where the calibration is done in 3 main steps: Firstly, the parameters of the fundamental diagram for each segment with data available are found. Subsequently, the parameters of the speed equation are computed using an optimization procedure considering the values of the fundamental diagram as known. Lastly, a global optimization is run in order to improve the final identification. Moreover, a new mathematical definition of the fundamental diagram is proposed. The identification algorithm was tested over a section of the I-210 West in Southern California, using several days of loop detector data collected during the morning rush-hour period. The results shows a good estimation of both densities and speeds with a mean error of 10.95% and 11.35%, respectively.