Development and application of decision support systems for improved planning and operation of large dams along the White Nile.

Abstract

In this study the regulation of Lakes Victoria, Kyoga and Albert in East Africa are investigated with the objective of maximising hydropower production subject to system constraints for existing and future planned dams along the Upper White Nile in Uganda. A Decision Support System (DSS) has been assembled and applied to search for efficient lake-reservoir operating rules for this basin. Elements of the DSS include power plant functions, a simulation model of the Upper Nile Equatorial Lake Basin, the Stochastic Analysis Modelling and Simulation (SAMS) computer software package for analysing hydrologic time series and the Colorado State University Dynamic Programming (CSUDP) model for solution of the optimisation problem. A concurrent record of observed lake levels and outflows for the three lakes during the reference period 1899 – 2008 has been constructed from various long term monitoring stations and utilised to derive net basin supply or net inflow time series at a monthly and annual time scale. Statistical tests confirmed the non-stationarity of the annual lake net basin supply time series. A justification to model the stochastic process of the monthly inflows as a Markov process was also reached. A Univariate Shifting Mean model was fitted to the annual historical data in tandem with a model for temporal disaggregation of annual to monthly net basin supplies for the purposes of generating synthetic flow series. The model performed well in terms of preserving the statistical characteristics of the historical reference set for each lake. The synthetic time series are considered to be a useful reference data set for future research in generating reservoir operating rules. Two Dynamic Programming (DP) models that may be used to generate reservoir operating rules were investigated. The desired scope of optimization was however curtailed by the well-known dimensionality problem of DP. Application of the deterministic method of Incremental Dynamic Programming (IDP) to the optimisation problem could only be carried out on a monthly time step and for single years separately. Annual time step optimization could only be carried out for the historic net inflows. The 1000 stochastically generated time series of net basin supplies could not be utilized within the implicit framework of deriving operating rules due to impractical computational requirements. The IDP however, yielded a realistic set of optimal operating policies at an annual time scale for the historical reference period (1898 – 2008). The beginning of year lake levels and annual release magnitudes obtained were compared against similar data for natural unregulated lake conditions. It is concluded that, in general, lake regulation would yield desirable benefits in terms of hydropower generation but would lead to marked deviation from natural lake levels and more variable outflows. The Stochastic Dynamic Programing (SDP) model was only applied to Lake Victoria in single reservoir optimization scheme due to limitations imposed by the large dimensionality of the problem and difficulty of simultaneously incorporating multiple lake reservoir transition probability matrices in the model. Application of the model for Lake Victoria showed that, it was feasible to define final storage levels for discretized initial storage and previous period inflow class combinations. The results from the study indicate that realistic heuristic operation rules can be inferred from the results of applying the IDP models and SDP algorithm.