Discussion and Conclusion
This chapter introduces a theoretical framework to modelling multidimensional travel behaviour based on artificially intelligent agents, search theory and bounded rationality. For decades, despite the number of heuristic explanations for different results, the fact that 'almost no mathematical theory exists which explains the results of the simulations' (Herbert, 1999) remains as one of the largest drawbacks of agent- based computational process approach. This is partly the side effect of its special feature that 'no analytical functions are required'. Among the rapidly growing literature devoted to the departure from rational behaviour assumptions, this theoretical framework makes an effort to embed a sound theoretical foundation for computational process approach and agent-based microsimulations. The theoretical contributions are three-fold:
• A pertinent new theory of choices with experimental observations and estimations to demonstrate agents with systematic deviations from the rationality paradigm. Modelling components including knowledge, limited memory, learning and subjective beliefs are proposed and empirically estimated to construct adaptive agents with limited capabilities to remember, learn, evolve and gain higher payoffs. All agent-based models are based on empirical observations collected via various data collection efforts.
• Modelling procedural and multidimensional agent-based decision-making. Individuals choose departure time, mode and/or route for their travel. Individuals also choose how and when to make those choices. A behaviourally sound modelling framework should focus on modelling the procedural decision-making processes. This study seeks answers to questions that largely remain unanswered including but not limited to: (1) When do individuals start seeking behaviour changes? (2) How do they initially change behaviour? (3) How do they switch behaviour adjustment dimensions? (4) When do they stop making changes?
• The transformation from the static user equilibrium to a dynamic behavioural equilibrium. Traditional solution concepts are based on an implicit assumption that agents have complete information and are aware of the prevailing user equilibrium. However, a more realistic behavioural assumption is that individuals have to make inferences. These inferences can be their subjectively believed search gain (or perceived distributions of travel time and travel cost), the multidimensional alternatives they subjectively identify, and the heuristics they employ to evaluate alternatives. It is the process of making inferences that occupies each individual in making a decision. With search start/stop criteria explicitly specified, this process should eventually lead to a steady state that is structurally different to user equilibrium.
The estimation of the proposed agent-based models usually needs additional behaviour process data. Whether or not the increased data needs can be justified by improved model realism and model performance in applications can be a subject for further examination. This chapter empirically estimates the models using data collected from a stated adaptation survey, a similar but different survey structure compared to stated preference experiments. This survey method effectively captures adaptations in response to changing attributes or context and can record behaviour process if implemented in an iterative manner (see e.g. Khademi, Arentze, & Timmermans, 2012). The observed behaviour process actually is a search path possessed by each respondent. This historical information can be applied to further calibrate the knowledge model or the search cost models. Another future research direction may explore how advanced data collection technologies such as GPS- surveys, smartphone applications and social network data can improve the affordability and quality of behaviour process data and further support the proposed modelling framework.
The numerical example presented in the paper highlights the capabilities of the proposed theory and models in estimating rich behavioural dynamics, such as multidimensional behavioural responses, day-to-day evolution of travel patterns, and individual-level learning, search and decision-making processes. The computational efficiency of the proposed models needs further exploration through real-world implementations using agent-based simulation techniques. It is believed that the flexible framework, computational efficiency and more realistic assumptions can make the proposed modeling tool extremely suitable for integrated large-scale multimodal planning/operations studies which typically have to cope with millions of agents. This work is primarily exploratory in its conceptualization of a descriptive theory, estimation of quantitative models and demonstration in an agent-based microsimulation. In an era of big-data access, multicore processors and cloud computing, the ambition of transportation demand modellers has never been greater. The hope is that the preliminary findings in this chapter could raise interest in the behavioural foundation of multidimensional travel behaviour as well as in microsimulating people's complex travel patterns in the time – space continuum. Extensive examination of the proposed tool on a larger and more representative survey sample and for real-world studies is necessary before we can conclude that the tool is fully practice-ready.