Predictive, stochastic and dynamic extensions to aversion dynamics scheduling

An extensive research agenda on real-world scheduling practices has reported major gaps between decades of mainstream production scheduling research and scheduling in practice (McKay, 1987, 1992, 2001; McKay, et al. 1988, 1989, 1992, 1995, 1999). Real schedulers' decisions often deal with perceived risks and associated impacts. They proactively anticipate and reactively mitigate risky events by altering the formal schedule to minimize impacts. These risk mitigation concepts are referred to as aversion dynamics (AD). AD describes the aversion that jobs exhibit to impacts associated with risky events in dynamic and/or unstable production environments. But traditional scheduling research has largely ignored these aspects by employing simplifying assumptions and, thus, diluting the problem formulation to such an extent that it lacks real-world fidelity and is of limited application in real-world settings.; The broad goal of this exploratory research is to expand the scheduling problem to incorporate relevant real-world phenomena and to show that, by doing so, significant performance advantages are obtained under favorable circumstances and insignificant performance degradation is incurred under unfavorable circumstances. This research extends McKay and Morton's preliminary aversion dynamics heuristic, Averse-1, to capture more of the dynamics observed in the field (McKay, et al. 1997, 2000). Specifically, it adds predictive and stochastic (i.e., proactive) components, redefines impacted processing time and incorporates these extensions within a dynamic job arrival environment. The resultant heuristic is deemed Averse-2. Subsequently, Averse-2 is further extended to consider aversion to jobs that will be in process when the risky event is predicted to occur. The resultant heuristic is deemed Averse-3.; After theoretical concepts underlying the extensions are presented, simulation experiments are conducted to test their performance. Research hypotheses include the following: Averse-2 will significantly outperform competing heuristics when the event occurs, Averse-2 will not significantly underperform competing heuristics when the event does not occur, Averse-3 will not significantly underperform Averse-2 when the event occurs, and Averse-3 will significantly outperform other competing heuristics when the event occurs. Subsequently, these hypotheses are validated to support the concept of aversion dynamics as a mechanism to increase the reality of the scheduling problem while maintaining favorable performance trends.