Power System Inertia Forecasting Using a Hybrid Deep Learning Model

The increasing integration of renewable energy sources, particularly wind and solar, has introduced substantial challenges to the stability of power systems due to their inherently low inertia, leading to fluctuations in system frequency. Frequency containment reserves are crucial for counteracting these fluctuations, but their effective deployment relies heavily on accurate inertia forecasts. Several forecasting models like time series, machine learning, and deep learning are used for inertia forecasts. Deep learning models show superior performance among these models. However, a single deep-learning model may not offer the desired accuracy. Also, the deep learning model’s accuracy highly depends on its hyperparameters. This necessitates hybrid deep learning models where one or two deep learning models will be combined. Also, effective hyperparameter tuning necessitates a suitable optimization algorithm. Therefore, this paper proposes a hybrid deep-learning model (CNN-GRU-BO) using two deep-learning algorithms: Convolutional Neural Network (CNN) and Gated Recurrent Units (GRU) models. The hyperparameters of the proposed model are optimized by Bayesian Optimization (BO). The CNN model is proficient in extracting spatial features, while the GRU model excels at capturing temporal dependencies from the data. The BO ensures the optimal configuration of the hyperparameters. The performance of the CNN-GRU-BO model is compared with different reference models. The results show that the CNN-GRU-BO model outperforms the reference models.