A hybrid dynamic programming - Tabu Search approach for the long-term hydropower scheduling problem

Abstract

The long-term energy scheduling of a large hydroelectric power system is studied in this paper. The problem aims at defining a policy that provides the best trade-off between energy conservation into the reservoir for future revenues and current energy sales with a risk of system failure in the future. The policy should take into account the uncertainty of energy inflows for the next decades. Energy inflows are obtained from water inflows using an energy aggregation process and therefore behave like hydrological time series. Long-term persistence, present in the energy inflows, especially with multiyear sequences of low and high inflows, poses a serious threat to the system’s reliability. A Shifting Level hydrological model is used to capture precisely the annual and interannual dynamic of the energy inflows. However, this model is challenging to include in the framework required by state-of-the-art optimization methods that mostly rely on the dynamic programming principle and Markovian processes. We propose a method combining stochastic dynamic programming and Tabu Search to solve the long-term energy scheduling problem without the need to find an appropriate Markovian approximation of the Shifting Level model. The policies resulting from this hybrid method are compared with stochastic dynamic programming policies coupled with a Hidden Markov Model. The results show that the hybrid method retains more energy in the reservoirs, thus reducing the volume of possible energy deficits. Overall, the objective value obtained by the hybrid method policies is higher than the value returned by the stochastic dynamic programming with the Hidden Markov Model, suggesting a better trade-off between a low risk of energy deficits and revenue maximization through high energy sales.