Stochastic Scheduling with Incomplete Information
This chapter treats a class of scheduling models subject to machine breakdowns with incomplete information. Under this class of models, the repeated processing times between breakdowns are dependent via a latent random variable. This leads to partially available information on the processing times during the process, and the information is gradually accumulated from previous processing experience and adaptively incorporated into the decision making for processing remaining jobs. Section 8.1 formulates the model and discusses the probabilistic characteristics of the repetition frequency and occupying times, and the impact of incomplete information. The optimal restricted dynamic policies for this model are derived in Section 8.2 based on posterior Gittins indices. Finally, Section 8.3 discusses an interesting case in which the posterior Gittins indices can be represented by the one-step rewards rates of the jobs.
KeywordsPosterior Distribution Optimal Policy Probabilistic Characteristic Reward Rate Decision Epoch
- Bast, H. (1998). Dynamic scheduling with incomplete information. Proceedings of the tenth annual ACM symposium on parallel algorithms and architectures, Puerto Vallarta (pp. 182–191).Google Scholar
- Chen, C. Y. I., George, E. I., & Tardif, V. (2001). A Bayesian model of cycle time prediction. IIE Transactions, 33(10), 921–930.Google Scholar
- Gittins, J. C. (1989). Multi-armed bandit allocation indices (Wiley-Interscience series in systems and optimization). Chichester: Wiley. ISBN:0-471-92059-2.Google Scholar
- Glazebrook, K. D., & Boys, R. J. (1995). A class of Bayesian models for optimal exploration. Journal of the Royal Statistical Society, Series B (Methodological), 57, 705–720.Google Scholar
- Ross, S. M. (1996). Stochastic processes (2nd ed.). New York: Wiley.Google Scholar