Simulated Annealing in a Co-Evolutionary, Agent-Based Transport Modeling Framework - The Example of Ride-pooling Driver Supply Optimization

Abstract

This paper introduces an integrated simulated annealing optimization method within the coevolutionary agent-based transport modeling framework MATSim, using a small illustrative ridepooling service as an example to optimize driver shift supply for a given and static demand. Simulated annealing is a metaheuristic optimization algorithm that has already been employed in a wide range of problems and domains. MATSim makes use of a co-evolutionary design in which individual agents try to optimize their daily schedule by finding optimal transport options. The iterative nature of both simulated annealing and MATSim’s co-evolutionary design makes the implementation straightforward and compatible. The outcomes validate the feasibility of the approach in optimizing specific components of the transport model and indicate its potential for future use in comparable applications. The presented case of driver supply optimization may help to design scenarios for new services and to better assess the efficiency and costs of such a service.