Reservoir optimization requires the ability to produce inflow scenarios that are consistent with the available climatic information. We approach stochastic inflow modelling with a Markov-switching model where inflow anomalies are described by a mixture of autoregressive models with exogenous input, each corresponding to a hidden climate state. Climatic information is used as exogenous input and to condition state transitions. We apply the model to the inflow of the Daule Peripa reservoir in western Ecuador, where El Niño events cause anomalously heavy rainfall. El Niño–Southern Oscillation (ENSO) indices constitute the climatic input of the inflow model. The Daule Peripa reservoir serves a hydropower plant and a downstream water supply facility. Based on ENSO forecasts, which are available with 9 month lead time, monthly inflow scenarios are generated to perform stochastic optimization of reservoir releases with monthly time-steps. To account for inflow uncertainty, we generate multiple synthetic inflow time series and apply a multi-objective genetic algorithm to evaluate the objective functions. The results highlight the advantages of using a climate-driven stochastic model to produce inflow scenarios and forecasts for reservoir optimization, and show significant potential improvements with respect to the current reservoir management.
Keywords: El Niño, genetic algorithms, inflow forecast, reservoir optimization