Surrogate‐based optimization of a periodic rescheduling algorithm

Abstract

Periodic rescheduling is an iterative method for real‐time decision‐making on industrial process operations. The design of such methods involves high‐level when‐to‐schedule and how‐to‐schedule decisions, the optimal choices of which depend on the operating environment. The evaluation of the choices typically requires computationally costly simulation of the process, which—if not sufficiently efficient—may result in a failure to deploy the system in practice. We propose the continuous control parameter choices, such as the re‐optimization frequency and horizon length, to be determined using surrogate‐based optimization. We demonstrate the method on real‐time rebalancing of a bike sharing system. Our results on three test cases indicate that the method is useful in reducing the computational cost of optimizing an online algorithm in comparison to the full factorial sampling.

A generalized disjunctive programming model for the static bike sharing rebalancing problem with demand intervals

“…The model is based on assumptions that the pickup and return rates are constant within 15-min time intervals and that arrival times of customers are independent, leading to an assumption of Poisson distributed time intervals between bike pickups/returns. Graphical illustrations of the transient probabilities are shown in the paper by Ikonen et al (2022).…”