OPTIMAL OPERATION OF A RESERVOIR SYSTEM USING FORECASTS
Abstract
The present study concerns the optimal operation of an existing multipurpose multireservoir system using forecasts. An optimization-simulation model was constructed for a system of four flood control reservoirs in the Green River Basin, Kentucky, having recreation and low-flow augmentation as secondary objectives. The resulting model, called GRBOSM, is designed for use in real-time as well as in long-run operations studies. GRBOSM consists of descriptive and prescriptive components. Its descriptive component consists of a segmented model comprising nine multi-input linear (MIL) models of the river system downstream of the four reservoirs. The prescriptive component is largely in the form of an operating policy algorithm that involves repeated solution of a mathematical program. The optimization technique employed to solve the mathematical program is Linear Programming. Each of the nine MIL model components of the segmented model represents a routing model for a reach between control stations. These models accept reach inflow, gaged tributary inflow(s), and rainfall over the ungaged side inflow area as inputs. Their outputs are the reach outlows. Three approaches to MIL-model construction were investigated. The results showed that a constrained linear systems estimation method gave better results than the ordinary least-squares method. The addition of an error model further improved the forecasting performance of the models. The development of the operating policy component involved an investigation of goals and priorities for reservoir operation. To that end the state variables of the system associated with the reservoirs and with the control stations were divided into time-varying target or ideal values or ranges. Deviations from the ideal state vector were then divided into various zones and different penalties were associated with different zones. By aggregating these penalties over an operating horizon, most of the goals and priorities for reservoir operation were commensurated into an overall measure of effectiveness of the system. The use of GRBOSM model in long-run reservoir operations studies has been demonstrated; system's responses under various operating and hydrologic conditions were obtained. In particular, trade off curves between various objectives of the system were generated both for the winter and summer seasons by using historical hydrologic as well as synthetic model input data. The use of GRBOSM in real-time operations has been discussed also. It involves the use of forecast data during an operating horizon and the sensitivity analyses of operating decisions with respect to changes in those forecasts. The results obtained from the model are represented in an integrated set of graphs and tables that appropriately shows the recommended optimal release decisions and the resulting state of the system at the reservoirs and control stations throughout the operating horizon. The GRBOSM model, besides being computationally efficient, is flexible to allow the examination of the operation of a complex reservoir-river system for a variety of operational policies. Specific to the approach in designing the GRBOSM algorithm is that penalty weights and target levels and zones are among the input variables for each time step. This provides for the flexibility needed in real-time operations.
Degree
Ph.D.
Subject Area
Civil engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.