A hybrid algorithm for fast learning individual daily activity plans for multiagent transportation simulation

This paper propose a hybrid learning algorithm based on the competing risk duration model and the cross entropy method for generating complete all-day activity plan in multiagent transportation simulation. We formulate agent's activity scheduling problem as a sequential Markov decision process. By initially generating individual's activity type and duration sequence from empirical data based on the competing risk duration model, the obtained plans can be efficiently improved by reinforcement learning technique towards near-optimal activity plan. We apply the cross entropy method to efficiently learn near-optimal activity plan. The numerical result shows that the proposed method generates consistent daily activity plans for multiagent transportation simulation.