Data Assimilation Technique for Social Agent-Based Simulation by Using Reinforcement Learning

This paper presents a data assimilation technique for social agent-based simulation to fit real world data automatically by a reinforcement learning method. We used the hidden Markov model in order to estimate the states of the system during the reinforcement learning. The proposed method can improve simulation models of the social agent-based simulation incrementally when new real data are available without total optimization. In order to show the feasibility, we applied the proposed method to a housing market problem with real Korean housing market data.