Two directions of research are to be investigated for controlled stochastic reward processes which model diverse phenomena in engineering. The first direction augments the usual expected value criteria with variability considerations. Efficient methods are sought to compute mean-variance tradeoffs, to compute the effects of a "little" risk sensitivity, to characterize intertemporal variability, and to extend such notions to sequential games. General results would be applied to structured models which arise in contexts such as inventory, production, maintenance, reservoirs, and storm water basins. The second direction of research exploits preliminary results for models with certain affine properties. The properties significantly reduce the complexity of dynamic programming problems and, in many cases, permit the dynamic model to be replaced with a one-period model. Efficient computational methods would be sought for higher-order models which frequently arise in practice. General results would be combined with specially structured models such as in queuing control, capacity expansion, inventory, multi-stage manufacturing, multi-reservoir river basins, aquaculture, and agricultural crop production combined with pest control. This research concerns sequential decision processes. In particular, two topics are being investigated: (1) risk sensitivity and (2) models which have certain affine properties allowing the development of efficient computational procedures. It is expected that the results of this work will be applied to a wide range of engineering problems such as multi-stage manufacturing and capacity expansion.
