9701403 Ryan The research component of this CAREER award addresses several important topics in the area of sequential decision making. Decision makers are often faced with making a sequence of decisions about a system over a possibly infinite horizon, with inherent uncertainty in the environmental conditions influencing the system state. Standard multiperiod decision-making techniques are often inadequate in integrating forecast information for future time periods with the decision made in the current time period. This research is focused on integrating a general and flexible forecasting approach with a rolling horizon procedure that iq modeled as a Markov decision process. The research objectives are to: (1) determine appropriate lengths and formulations of the finite deterministic subproblems to account for forecast uncertainty, (2) assess the effect of each level of approximation on the long term cost of the decisions, and (3) gauge the effect of forecast revisions. The proposed educational activities are intended to improve student understanding and retention at the undergraduate level through the use of active and cooperative learning strategies, and enhance teaching effectiveness by participation in a project on the peer review of teaching. The use of sequential decision-making for long- (or infinite-) term planning projects under uncertainty arise in a number of different contexts. For example, electric utilities must deliver power to customers over an infinite horizon, with uncertainty in the demand for electricity over time. Manufacturing concerns must allocate funds for capacity expansion over time based on uncertain demand estimates. This research should have an impact on our ability to better make decisions over long planning horizons. Moreover, the educational component should have a positive impact on the engineering infrastructure, encouraging interdisciplinary and team-thinking among undergraduate students.
