Methodology of Parameter Calibration for Macroscopic Traffic Flow Models

To improve the coherent whole precision of traffic flow stream models, the methodology of parameter calibration was studied. The joint-prediction model parameter estimate method was developed considering the correlation among density-speed, density-volume and speed-volume relationship, and the optimal functions were constructed in the light of principle of minimum sum of squares of estimation deviation and the roles of different dependent and independent variables in the traffic flow stream models. To demonstrate the feasibility and efficiency of joint-prediction model calibration method on the estimate precision, the Castillo-Benítez's model and Van Aerde's model were selected and the field data on the Beijing 2nd ring were used to calibrate the model parameters. The weighted determination coefficient was suggested and combined with mean absolute percentage error (MAPE) and root mean squared error (RMSE) were used to evaluate the model parameter estimate efficiency. Results show that for the Castillo-Benítez's model, the MAPEs of speed and volume calculated from single prediction model are 19.8% and 18.7%, respectively. The values calculated from joint-prediction model are 10.0% and 10.0%, respectively, and the determination coefficients change from 0.913 to 0.910. For the Van Aerde's model, the MAPEs of density and volume calculated from single prediction model are 16.4% and 16.3%, respectively. While the values calculated from joint-prediction model are 114.2% and 14.2%, respectively, and the determination coefficients change from 0.732 to 0.749.