Parallel Bayesian Optimization of Agent-Based Transportation Simulation

MATSim (Multi-Agent Transport Simulation Toolkit) is an open source large-scale agent-based transportation planning project applied to various areas like road transport, public freight regional evacuation, etc. BEAM (Behavior, Energy, Autonomy, and Mobility) framework extends enable powerful scalable analysis of urban systems. The agents from the simulation exhibit ‘mode choice’ behavior based on multinomial logit model. In our study, we consider eight mode choices viz. bike, car, walk, ride hail, driving transit, walking hail pooling. ‘alternative specific constants’ for each choice are critical hyperparameters in a configuration file related particular scenario under experimentation. We use ‘Urbansim-10k’ (with 10,000 population size) all experiments. Since these affect complex ways, manual calibration methods time consuming. present parallel Bayesian optimization method with early stopping rule achieve fast convergence given multi-in-multi-out problem its optimal configurations. Our model HpBandSter package. This approach combines hierarchy several 1D Kernel Density Estimators (KDE) cheap evaluator (Hyperband, single multidimensional KDE). has also incorporated extrapolation rule. With model, could 25% L1 norm fully autonomous manner. To best knowledge, work first kind multi-agent simulations. can be useful surrogate modeling scenarios very large populations.