An Advanced Hybrid Meta-Heuristic Model for Solar Power Generation Forecasting via Ensemble Deep Learning

The increasing adoption of solar power as a renewable and eco-friendly energy source necessitates precise forecasting of solar power generation. Accurate predictions are crucial for effective grid management and the seamless integration of renewable energy into the power grid. This study proposes a novel hybrid meta-heuristic optimization framework, empowered by an ensemble deep learning model, to enhance the accuracy of solar power generation forecasting. The proposed methodology comprises several methodical phases: data pre-processing, feature extraction, feature selection, and deep learning-based forecasting. Initially, the collected raw data undergo a pre-processing stage involving data cleaning and standardization via the z-score method. Subsequent feature extraction transforms the pre-processed data into a reduced set of representative features, leveraging Linear Discriminant Analysis (LDA), measures of central tendency (Weighted arithmetic mean, Winsorized mean, standard deviation), statistical dispersion (Interquartile range (IQR), Median absolute deviation (MAD)), and Information Theoretic measures (Mutual Information and Information Gain). The optimal features are then selected through a newly proposed hybrid optimization approach, the Gorilla Customized Teaching Learning-Based Optimization (GC-TLBO) Algorithm, an innovative combination of the Artificial Gorilla Troops Optimizer (GTO) and the Teaching-Learning-Based Optimization (TLBO). Solar power forecasting is accomplished using a novel ensembled deep learning model, which integrates optimized Recurrent Neural Network (O-RNN) with a Deep Belief Network (DBN) and a Deep Convolutional Neural Network (DCNN). The final outcome is derived from the O-RNN, which inputs the results from the DBN and DCNN, respectively. The DBN and DCNN are trained using the optimal features derived from the GC-TLBO, while the weights of the RNN are fine-tuned using the same algorithm. The proposed model was implemented in Python (Google Colab), and its performance was evaluated using several metrics: Normalized Mean Square Error (NMSE), Mean Squared Relative Error (MSRE), Mean Squared Error (MSE), Mean Absolute Percentage Error (MAPE), and Root Mean Squared Error (RMSE). The results demonstrate that the proposed model outperforms existing models, offering superior forecasting performance.