This article examines the problem of optimizing a tram route network. A network configuration that maximizes overall passenger comfort is sought. The tram infrastructure is represented as a directed weighted graph. The comfort criterion is formalized based on a sociological survey using conditional logistic regression and takes into account waiting time, car occupancy, and the number of transfers. To find the optimal rolling stock distribution, seven methods are analyzed: a greedy algorithm, column generation, route pair enumeration, a genetic algorithm, competitive and cooperative multi-agent models, and simulated annealing with controlled noise. The best result was achieved using the simulated annealing algorithm; the cooperative approach yielded the most balanced network without critical dips in individual sections. To reduce the computational load, surrogate modeling based on an ensemble of machine learning methods was implemented. Optimized configurations significantly reduce headways on key highways and eliminate load imbalances. The developed methodology and quantitative comfort model can serve as a tool for strategic planning of tram systems in cities with complex transportation infrastructure.

modeling, reinforcement learning, genetic algorithm, simulated annealing algorithm, transportation planning, public transportation

Artificial intelligencе and big data