STOCHASTIC OPTIMIZATION MODELS FOR LONG TERM PLANNING AND REAL TIME OPERATION OF RESERVOIR SYSTEMS

BITHIN DATTA, Purdue University

Abstract

For efficient planning and real time operation of a reservoir system, it is desirable to formulate optimization models to serve as tools for decision making. In order to be a realistic representation of the decision making process, these models should also be able to incorporate the stochastic nature of streamflows as system inputs. The time steps considered in such models are of prime importance because the related aggregation or disaggregation of the inputs has significant effects on the uncertainties involved in the decision making process and directly influences the specification of optimal criteria in such models. This study is limited to two optimization models, one for longterm planning and seasonal operation and the other for real time daily operation of a reservoir system. Both of these models use chance constraints, assume Linear Decision Rules (LDR), and incorporate reliability criteria for different performance requirements. Both of these models are also capable of using multiple Linear Decision Rules, either based on different intervals of streamflow events or for different ranges of short term streamflow forecasts. To determine the effectiveness of a multiple LDR model as an aid to longterm planning and operation, its performance is evaluated for various 'dimensions' of the model and different input conditions. This model accounts for the Markovian dependence structure of streamflows by incorporating their distribution functions, conditioned on the streamflows in other seasons. The solutions obtained are tested in simulation of actual operation. The results are also compared with those for the original single LDR model. For real time daily operations, a model using similar concepts is proposed and evaluated. This model overcomes the myopic nature of short term operation, and uses distributions of actual streamflows conditioned on the forecasted values, thus accounting for uncertain forecasts. This model is also solved for different input conditions and the solutions are tested in a simulation of actual operation. The results of the evaluations and motivations for the results are discussed. Some of the basic characteristics of the decision making process for planning and management of reservoir systems are also explored.

Degree

Ph.D.

Subject Area

Civil engineering

COinS