METANET Model Validation using a Genetic Algorithm*

This paper aims to describe a black box approach to traffic model validation. A genetic algorithm is utilised with a least-squares method for parameter identification. The procedure developed requires no knowledge of the model, just the model output and some reference data. Testing is done by applying the method to the second order macroscopic simulator METANET. This produces a valid model fitted to real data for a stretch of UK motorway. The optimal parameter set found is valid for data from different days and remains accurate when turning rates are estimated either as a 15-minute average or a constant rate.