Time series forecasting using ensemble learning methods for emergency prevention in hydroelectric power plants with dam

In hydroelectric plants, the responsibility for the operation of the reservoirs typically lies with the national system operator, who controls the level of the reservoirs based on a stochastic problem for the economy of the potential energy available in the reservoir. However, in an emergency, the responsibility for the operation and control of the reservoir becomes the plant's management. To have a faster decision-making process, it is important to have a forecast of water affluence in relation to the turbine capacity and use of the spillway. With the objective of evaluating the forecast increase in the level of the reservoir of a hydroelectric plant, this paper compares the use of the bagging, boosting, random subspace, bagging plus random subspace, and stacked generalization ensemble learning models to analyze this problem. The case study is based on data from a 690 MW hydroelectric plant, which has a 94 km reservoir and a 185 m high dam. The random subspace and stacking models had the best results for lower error, with a low time required for convergence in relation to the other models. The ensemble models resulted in greater accuracy for the assessed problem than long short-term memory.