Article Preview
TopIntroduction
The shop-around behavior model is also known as the MultiPurpose-MultiStop (MPMS) model and since the 1980s it has been developed and studied in such fields as geography and urban planning, not only for its practical application, e.g., downtown revitalization and town center management, but also for its theoretical interest in the field of spatial analysis (Kelly, 1981). By the 1990s, the application of the logit model that combines data-fitting and approximate utility-maximization, helped establish the ’Markov-chain type’ models that make up transition probability OD-matrices. A typical microsimulation is Linked Logit and Poisson Model (LLPM), with a Poisson assumption on visitor arrival times. However, in the era of agent modeling, limitations pointed out concerning the Markov property, which ignores the personal history of downtown visitors, led to new approaches being explored.
The Logit model can be interpreted as used in an approximate estimation of the random utility, thus LLPM is considered to be a rational model. In other hands, the agent model would be considered as a bounded rational model, so there are at least two types of bounded rational models. One type is a rule-based approach that employs heuristics, which can be interpreted as an expression of ‘procedural rationality,’ as referred to by H. A. Simon. Implementation technologies such as the production system in knowledge engineering and advanced researches had already been made into this approach.
The other type is the assumption-relaxation approach, which relaxes the assumption of perfect rationality with perfect information by adopting the concepts of satisficing or the constraint satisfaction principle. This approach is also based on the ‘satisficing principle’ of mathematical models proposed by Simon and followers (e.g., Rubinstein, 1998).
In agent modeling research the daily activity-travel model has taken the lead in such fields as transportation planning (Table 1). Albatross (Arentze et al., 2001) is formulated as a rule-based system that guarantees data-fitting by employing a data-mining tool to generate heuristic rules (binary tree). Aurora (Arentze, Pelizaro, & Timmermans, 2005) is formulated as a utility-based theoretical model that generates a schedule by combining each activity (errand) that has non-linear S-shape utility and employing genetic algorithms; in addition, in response to an unexpected event such as congestion, the model carries out re-scheduling.
Table 1. Existing agent modeling researches
| Albatross | Aurora | Logit Model Approach (ex. LLPM: Linked Logit Poisson Model, Transition Matrix Models) | Kurose's Approach | ASSA ver.3 |
Deals with | Daily Activity-Travel | Daily Activity-Travel | Daily Activity-Travel / MultiPurpose-MultiStop in Shopping District | MultiPurpose-MultiStop in Shopping District | MultiPurpose-MultiStop in Shopping District |
Principle of Modeling | Idea of The System | Heuristic-Rule Based | Attempts to Keep a Utility- Maximized Schedule under Constraints/Events | Utility-Maximization Based | Heuristic-Rule Based | Utility/Constraints- Satisfaction Based |
Model Type | Bounded Rational | Bounded Rational but Adaptation/Intelligent Functions | Rational | Bounded Rational | Bounded Rational but Adaptation/Intelligent Functions |
Adaptation / Intelligent Functions | Schedule Planning Function | YES but Conditional Rule Expression | YES | NO | YES but Conditional Rule Expression | YES |
Re-Scheduling | NO | Incremental- Type (Triggered by Congestion) | NO | NO | Recalculation-Type (Triggered by Errand- Failure, etc.) |
Preference Updating | NO | NO | NO | NO | Reinforcement Learning (to District State Change) |
Knowledge Extension of Mental Map | NO | Yes (Long-term Adaptation) | NO | NO | Not Yet, but Possible (Long-Term Adaptation) |
Practicality | Data-Fitting Methods | by Machine-Learning (C4.5), Automatic Decision-Tree Forming | by GA, Non-Linear Utility Shape Estimation | by Classical Statistical Analysis, Utility and Probability Estimation | by Conditional classification | Some by Statistical Analysis, Some by Experiments, Other by Applying Hypothesis |
Case Study | Real Cases inc. a Benchmarking Case (of Hendrik-Ido-Ambacht and Zwijndrecht) | Numerical Illustration | Many Real Cases for Practical Uses | Real Case (of Veldhoven) | Real Case (of Ohsu, Kanayama, Nagoya) |