报 告 人：Apostolos Kotsialos博士
报告题目：Traffic Flow Model Validation Using METANET, ADOL-C and RPROP:The cases of the Sheffield and Manchester networks
报告时间：星期三（6月29日） 上午9:30-10:30
报告地点：安中大楼A322
报告简介：
Macroscopic traffic flow model calibration is an optimisation problem typically solved by a derivative-free population based stochastic search methods. In this presentation we report the use of a gradient based algorithm using automatic differentiation. The ADOL-C library is coupled with the METANET source code and this system is embedded within an optimisation algorithm based on RPROP. The result is a very efficient system which is able to be calibrate METANET’s second order model by determining the density and speed equation parameters as well as the fundamental diagrams used. Results from two different test sites are reported. A 22 km site is considered near Sheffield, UK and the motorway network around the city of Manchester, UK. Manchester orbital network consists of a total road length of 187 km considering traffic flow on both and opposite directions of each modelled motorway. For these networks a single optimisation problem is formed for calibrating METANET, i.e. for identifying all its parameters, and subsequently solved. Three different data sets are used in each case and the corresponding optimal parameter sets are obtained. The results show that the combined METANET-RPROP-ADOL-C package is able to calibrate a large scale motorway network with very good accuracy. The optimal parameter sets where the optimisation algorithm converged for a particular data set are verified by running METANET simulations using the other two datasets not used during the corresponding calibration. Results show the expected degradation of the parameter set's quality, but the essential network wide dynamics of congestion are retained. A byproduct of using a gradient based optimisation algorithm is the calculation of the system's Jacobian matrix, containing information about the sensitivity of the speed with respect to the model parameters. This information obtained from the Jacobian provides extra insight into the system dynamics and interesting shockwave patterns emerge.
报告人简历：
Dr. Apostolos Kotsialos is a lecturer at the School of Engineering and Computing Sciences in Durham University, United Kingdom. He has extensive experience in traffic flow modelling and control systems. He has developed integrated traffic control strategies based in feedback and model predictive control methods as well as tools for macroscopic traffic flow model validation using numerical optimisation. His research interests include traffic flow forecasting, numerical optimisation and control supporting intelligent transportation systems.
欢迎广大师生踊跃参加!
联系人：王亦兵
交通工程研究所