GENOSIM is a generic traffic microsimulation parameter optimization tool that uses genetic algorithms and was implemented in the Port Area network in downtown Toronto, Canada. GENOSIM was developed as a pilot software as part of the pursuit of a fast, systematic, and robust cali- bration process. It employs the state of the art in combinatorial paramet- ric optimization to automate the tedious task of hand calibrating traffic microsimulation models. The employed global search technique, genetic algorithms, can be integrated with any dynamic traffic microscopic sim- ulation tool. In this research, Paramics, the microscopic traffic simulation platform currently adopted at the University of Toronto Intelligent Transportation Systems Centre, was used. Paramics consists of high-per- formance, cross-linked traffic models that have multiple user adjustable parameters. Genetic algorithms in GENOSIM manipulate the values of those control parameters and search for an optimal set of values that min- imize the discrepancy between simulation output and real field data. Results obtained by replicating observed vehicle counts are promising. Numerous problems in engineering require either maximizing a sys- tem’s performance or minimizing a misfit function. “Misfit” is used here to denote the error between the model output and observations from the real system (1). An overall system model consists of a set of cross-linked submodels that depend on a number of model param- eters. Optimization of model performance involves the selection of the “best” set of values for the parameters. Best values, loosely inter- preted as optimal, can be obtained using genetic algorithms (GAs), which achieve a combinatorial optimization of parameters of the target system (microsimulation) by minimizing a misfit function. Problems such as these pose a great challenge because of the large parametric space to be searched. The search space is usually multi- dimensional: the values of the parameters can be conceived of as coordinates, and the “fitness” represents goodness of fit as a hilly surface. The process of seeking an optimum point, either a global or the best attainable local optimum, should involve some systematic search method to avoid ad hoc selection of the model parameters and to ensure robustness of the results. A very common optimization challenge is how to thoroughly traverse the whole search space to reach a global peak in the case of unevenly distributed, nonuniform, multiple-peak space. To solve such problems, researchers probably resort to either traditional ana- lytical gradients or numerical search methods. The traditional meth- ods may fail to achieve good results, particularly because of potential entrapment in local minima. In recent years, GAs have gained pop- ularity as a generic, systems-independent optimization tool (2, 3) and have been shown to do quite well. INTELLIGENT TRANSPORTATION SYSTEMS, MICROSIMULATION, AND PARAMICS Intelligent transportation systems (ITS) and modern urban road network design require more insight and knowledge of the consti- tutive properties of traffic systems (4). Such knowledge is needed to predict—through numerical simulations—the behavior of the vehi- cles under physical conditions. Reliable predictions allow the engi- neering requirements to be met at a lower cost. Therefore, traffic models have been developed to describe the driving behavior in as much detail as possible. Paramics, a comprehensive visualization and microscopic traffic simulation model, is one of the recently introduced suites of simu- lation tools that display distinguished features, and it shows poten- tial as an advanced, ITS-ready modeling suite. Paramics is a suite of high-performance, cross-linked traffic models that interact to repli- cate the urban and highway road network and to simulate the move- ment and behavior of individual vehicles in the virtual replica of a physical roadway network. Paramics models a traffic network micro- scopically at the individual driver and vehicle levels using such tech- niques as car-following, lane-changing, and gap-acceptance models. These models have multiple user adjustable parameters by which the modeler may adjust the simulated driving behavior and evaluate the performance of the network under different scenarios. Any change in one control parameter may cause networkwide cross effects through the linkage chains and can result in different model output. Paramics takes the virtual physical network and decisions parameter values as inputs, and it outputs the corresponding networkwide vehicular move- ments. To accomplish a realistic traffic simulation, parameters that govern vehicle movement and routing behavior in the simulation models need to be calibrated using a systematic approach. MOTIVATION OF GENOSIM DEVELOPMENT Our interest in using GAs to tackle combinatorial parametric opti- mization problems is motivated by the need for a systematic approach to calibrate the increasingly popular traffic microsimulation models for ITS applications and thus offer an automated and robust opti- mization tool for microscopic simulation parameters. Manual exper- imentation with model parameters appears to be impractical because Genetic Algorithm-Based Optimization Approach and Generic Tool for Calibrating Traffic Microscopic Simulation Parameters Tao Ma and Baher Abdulhai Intelligent Transportation Systems Centre and Testbed, Department of Civil Engineering, University of Toronto, 35 St. George Street, Toronto, Ontario, Canada M5S 1A4. 6 ■ Transportation Research Record 1800 Paper No. 02- 2131