Power generation forecasting based on hybrid deep learning models for a multi-energy power generation system

The accurate power generation forecasting (PGF) of the multi-energy complementary power generation system (MECPG) is great significance which lays the foundation for coordinated energy matching and optimization. This paper proposed a hybrid artificial intelligence (AI) model for the forecasting of the output in the multi-energy complementary power generation system which is designed in the prior research and the dataset primarily comprises meteorological in Lhasa, Tibet, China and operational parameters which were collected per hour, totaling 4461 samples. The hybrid model comprises the cross-correlation Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CC-ICEEMDAN) method for denoising, Convolutional Neural Network (CNN) and Gated Recurrent Unit (GRU) for forecasting. This structure captures multi-dimensional features to boost the accuracy and develop learning speed and robustness. Compared with other five typical deep learning models, the proposed model is the most suitable for PGF of MECPG with Mean Squared Error (MSE) of 0.0005, Mean Absolute Percentage Error (MAPE) of 2.8, Mean Absolute Error (MAE) of 0.014, and Root Mean Squared Error (RMSE) of 0.02. This study provides a theoretical basis for the realization of artificial intelligence application which involves the real-time peak forecasting and optimal allocation for MECPG.