Meta-heuristic Ensemble Feature Selection (MEFS) and Stacking Deep Ensemble Classifier (SDEC) Model for Weather Prediction and Renewable Energy Demand Forecasting

Wind speed and solar radiation varies is a predictable part of generating electricity from renewable resources. However, renewable energy data is unpredictable and disorganized is a difficult process. Deep learning is a technique for identifying the high-level invariant structures and intrinsic nonlinear features in the dataset. Enhancing the effectiveness of forecasting models requires feature selection, which is the process of identifying and choosing the most relevant elements from the dataset. In this paper, Meta-heuristic Ensemble Feature Selection (MEFS) model, and Stacking Deep Ensemble Classifier (SDEC) model are presented for feature selection and classification of wind speed and weather forecasting. The dataset, which was collected from Kaggle, contains four years’ worth of Spanish weather, price, generating, and electrical consumption data. The dataset was sourced from the Transmission Service Operators (TSO) public portal, the European Network of Transmission System Operators for Electricity (ENTSOE). Min-max normalization was applied to pre-process the dataset. Entropy Binary Dragonfly Algorithm (EBDA), Adaptive Weight Dung Beetle Optimization (AWDBO), and Inertia Weight Wild Horse Optimizer (IWWHO) were combined into MEFS. SDEC is an ensemble method that combines the procedure of several models, such as the Peephole Long Short-Term Memory Network (PLSTM), Conditional Generative Adversarial Network (CGAN), and Lagrange Contractive Auto Encoder (LCAE)). SDEC model improves forecasting performance by training a meta-learner (Weighted Averaging Ensemble (WAE)) on the outputs of the underlying models. PLSTM gates can enter the cell since they have direct links or peepholes to the cell state. CGAN creates a corresponding number of samples to predict electricity from renewable sources including forecasted and actual samples. LCAE is an unsupervised Artificial Neural Network (ANN) with a regularization term using a Lagrange multiplier for forecasting prediction. Finally, WAE is used to merge the results of individual models. Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Pearson Correlation Coefficient (r), Nash Sutcliffe Efficiency (NSE), precision, recall, f-measure, and accuracy metrics is used to evaluate the efficiency of forecasting approaches.