摘要：AbstractCalibration of car-following models based on empirical trajectory data is a fundamental problem in traffic simulations. As a common practice, optimization problems are formulated to calibrate the model parameters under the objective of minimizing the “difference” between observed vehicle movements and their simulated correspondences. Oftentimes, high degrees of nonlinearity, nonconvexity, and non-smoothness are embedded inside the cost functions. In consequence, derivative-based optimization methods are found to be impractical due to the laboriousness of gradients computations. To overcome this obstacle, derivative-free approaches are widely adopted for seeking the optimal model parameters. In this paper, we explore a new derivative-free cost-minimization methodology called Parameterized Derivative-free Optimization (PDFO) for tackling the car-following model calibration problem. The accuracy and efficiency of the PDFO are examined with the NGSIM data sets. Besides, comparative studies are conducted to investigate the performance differences between the PDFO and other frequently utilized derivative-free optimization techniques including the Genetic Algorithm (GA), Particle Swarm Optimization (PSO), and Simulated Annealing (SA).
关键词：KeywordsCar-following modelderivative-free optimizationintelligent driver model (IDM)
