A mathematical modeling framework for scheduling and managing multiple concurrent tasks

Graduation date: 2003Occurrence of human error in highly complex systems, such as a cockpit, can be\ud disastrous and/or overwhelmingly costly. Mismanagement of multiple concurrent\ud tasks has been observed by researchers to be a type of repetitive human error in\ud previous studies of accidents and incidents. This error may occur in the form of\ud wrong selection of a strategy to attend to tasks, and/or wrong assessment of a task's\ud priority at each moment.\ud The desire to prevent such errors forms two essential questions: 1) Is there any\ud (near) optimal method of managing multiple concurrent tasks? 2) How optimally\ud do human operators manage these tasks? To answer the first question, operations\ud research as it is applied to single machine scheduling was used. The operator was\ud assumed to be a single resource that attended to different tasks, one at a time. To\ud answer the second question, a software environment was developed to measure the\ud human's multitasking performance, which was then compared with the answer to\ud question one.\ud In this research, the operator's quality of performance was maximized as opposed\ud to the number of tasks accomplished, which was considered by previous\ud researchers. A metaphor of 'Juggler and spinning plates' along with a graphic bar\ud illustration was used to resemble an operator (a juggler) who manages several tasks\ud (plates on vertical poles) concurrently.\ud Several mixed (binary) integer-linear programming models were developed\ud discretely over time. One model was selected and solved by the means of tabu\ud search heuristic method. In tabu search, the significance of different initial solution\ud finding mechanisms and different applications of long-term memory was\ud investigated. A conjecturing method, within the tabu search, was introduced for\ud solving problems with very large planning horizons. In all cases, tabu search gave\ud good quality solutions in a much shorter time than branch-and-bound.\ud Under five different scenarios, ten subjects were studied while managing multiple\ud concurrent tasks in the software environment. None of the subjects could gain a\ud score better than tabu search in any of the scenarios. Subjects' patterns of\ud attendance to tasks were analyzed and compared against the pattern suggested by\ud tabu search, and similarities/differences were identified