Generating complete all-day activity plans with genetic algorithms

Activity-based demand generation contructs complete all-day activity plans for each member of a population, and derives transportation demand from the fact that consecutive activities at different locations need to be connected by travel. Besides many other advantages, activity-based demand generation also fits well into the paradigm of multi-agent simulation, where each traveler is kept as an individual throughout the whole modeling process. In this paper, we present a new approach to the problem, which uses genetic algorithms (GA). Our GA keeps, for each member of the population, several instances of possible all-day activity plans in memory. Those plans are modified by mutation and crossover, while 'bad' instances are eventually discarded. Any GA needs a fitness function to evaluate the performance of each instance. For all-day activity plans, it makes sense to use a utility function to obtain such fitness. In consequence, a significant part of the paper is spent discussing such a utility function. In addition, the paper shows the performance of the algorithm to a few selected problems, including very busy and rather non-busy days