An agent-based model for continuous activity planning of multi-week scenarios

Abstract

Computer modeling and computer simulation provide valuable support in the decision making process for many disciplines. In trac analysis, trac management, and trac forecast, microscopic simulation models have become increasingly popular and are finding their way into practice. In contrast to aggregated models, they model the decision making process on an individual level and produce person specific outputs. This gain on informational value comes at the cost of computational demand and memory requirements. Most existing microscopic models introduce restrictive constraints to avoid these problems. For instance, they predefine the simulation horizon to a single day, force agents to commit to a predefined day plan, or limit the maximal number of simulated people.</br>The aim of this work is to demonstrate ways around the limitations of existing microscopic models and facilitate a continuous simulation of large scale multi-week scenarios. C-TAP (Continuous Target-Based Activity Planning - the model presented in this work) allows for an open simulation horizon by implementing a nonrecurrent decision scheme, i.e., agents are able to make decisions about upcoming activities on-thefly considering dierent planning horizons simultaneously. C-TAP uses customizable parameters which have an intuitive interpretation and which facilitate a direct influence on behavioral eects. These parameters allow for a clear separation between agents’ behavioral elements and elements modeling agents’ interaction environment. This intuitive interpretation and clear separation makes the model comprehensible for users as well as extensible for developers, which in turn should facilitate its acceptance by the modeling community as well as its enhancement in order to satisfy future modeling needs. Additionally, the presented method also allows for an automatic model configuration based on existing long-duration diaries. The combination of all the features presented in this dissertation is a step forward to a travel demand generation framework that allows for a continuous simulation of large scale multi-week scenarios.