Generation hybrid forecasting for frequency-modulation hydropower station based on improved EEMD and ANN adaptive switching

To master the generation evolution principle of a frequency-modulation hydropower station (FMHS), exhibiting a non-linear, nonstationary time series, a hybrid model consisting of signal decomposition and adaptive switching between artificial neural networks (ANNs) is established. The ensemble empirical mode decomposition (EEMD) is improved to prevent end effects to obtain ‘purer’ sub-series, which reduces the complexity in prediction to a substantial extent. Three popular ANN algorithms show different performances in various periods, so an adaptive switching strategy among extreme learning machine (ELM), backpropagation neural network (BP) and general regression neural network (GRNN) is introduced to increase accuracy. We explain how to operate the hybrid model to forecast the generation of Pubugou Hydropower Station (PBG), a typical FMHS in Sichuan, China. More than 35,000 samples with strong fluctuation are set as a target. The results show strong applicability in terms of forecasting accuracy, curve fitting, and computational efficiency. Five scenarios are set to discuss the necessity and advantage of such a decomposition strategy and ANN adaptive switching. As a result, the proposed hybrid model is shown to be superior both on typical days and over the whole year. When the frequency of a power system fluctuates on a large scale, the proposed model is a tool for determining dispatch schedules for generator in cascade hydropower stations and grid managers.