Bridging the Micro and Macro: Calibration of Agent-Based Model Using Mean-Field Dynamics

Area of Science:

• Computational modeling
• Complex systems analysis
• Agent-based modeling (ABM)

Background:

• Agent-based models (ABM) require calibration to accurately represent distributed systems.
• Traditional calibration methods are iterative, time-consuming, and prone to slow convergence.
• A gap exists in efficient methods for linking agent micro-level behaviors to system-level observations.

Purpose of the Study:

• To propose a novel and efficient approach for calibrating agent-based models (ABM).
• To establish a direct link between agent microbehavioral parameters and systemic macro-observations.
• To reduce computational complexity and improve accuracy in ABM calibration.

Main Methods:

• Formulating agent behavior as a high-order Markovian process.
• Searching for optimal transfer probability via a macrostate transfer equation.
• Employing mean-field approximation and principal component analysis (PCA) for parameter computation and state space compression.

Main Results:

• The proposed method demonstrates higher accuracy compared to machine-learning surrogates and evolutionary optimization.
• Achieved significantly lower computational complexities in validation scenarios.
• Successfully validated in population evolution and urban travel demand analysis.

Conclusions:

• The novel ABM calibration approach effectively links micro-parameters to macro-observations.
• Offers a computationally efficient and accurate alternative to traditional calibration techniques.
• Applicable to diverse distributed systems requiring ABM calibration, such as population dynamics and urban planning.