Adaptive State Space Partitioning for Dynamic Decision Processes

Authors

Ninja SoeffkerFollow
Marlin W. Ulmer
Dirk C. Mattfeld

Document Type

Research Paper

Abstract

With the rise of newbusiness processes that require real-time decision making, anticipatory decision making becomes necessary to use the available resources wisely. Dynamic real-time problems occur in many business fields, for example in vehicle routing applications with stochastic customer service requests expecting a fast response. For anticipatory decision making, offline simulation-based optimization methods like value function approximation are promising solution approaches. However, these methods require a suitable approximation architecture to store the value information for the problem states. In this paper, an approach is proposed that finds and adapts this architecture iteratively during the approximation process. A computational proof of concept is presented for a dynamic vehicle routing problem. In comparison to conventional architectures, the proposed method is able to improve the solution quality and reduces the required architecture size significantly.

Recommended Citation

Soeffker, Ninja; Ulmer, Marlin W.; and Mattfeld, Dirk C. (2019) "Adaptive State Space Partitioning for Dynamic Decision Processes," Business & Information Systems Engineering: Vol. 61: Iss. 3, 261-275.
Available at: https://aisel.aisnet.org/bise/vol61/iss3/3