• Journal of the Korea Society of Computer and Information
• /
• v.26 no.12
• /
• pp.29-43
• /
• 2021

In this paper, a new approach to solve the constrained dynamic programming is proposed by using the constraint programming. While the conventional dynamic programming scheme has the state space augmented with states on constraints, this approach, without state augmentation, represents states of constraints as domains in a contraining programming solver. It has a hybrid computational mechanism in its computation by combining solving the Bellman equation in the dynamic programming framework and exploiting the propagation and inference methods of the constraint programming. In order to portray the differences of the two approaches, this paper solves a simple version of the long-term car rental financing problem. In the conventional scheme, data structures for state on constraints are designed, and a simple inference borrowed from the constraint programming is used to the reduction of violation of constraints because no inference risks failure of a solution. In the hybrid approach, the architecture of interface of the dynamic programming solution method and the constraint programming solution method is shown. It finally discusses the advantages of the proposed method with the conventional method.