A hybrid approach based machine learning models in electricity markets

In recent years, integrating renewable and non-renewable energy sources has transformed electric grids, presenting new challenges in predicting energy data due to varying levels of variability. Accurate prediction of both types of energy data is crucial for smart grid technology development and effective renewable energy integration into existing grids. We have introduced an innovative hybrid approach for forecasting both renewable and nonrenewable data. This method employs a sophisticated ensemble empirical mode decomposition (EEMD) algorithm, carefully selecting intrinsic mode functions (IMFs) to dissect the original data into distinct IMFs and residuals. The IMFs are predicted utilizing support vector regression (SVR), while the residual series is forecasted using bidirectional long short-term memory with an attention mechanism (BiLSTM-AM). In pursuit of enhanced predictive accuracy, our approach employs an ensemble summation methodology to merge forecasted sub-series effectively. We conducted experiments using two distinct wind speed datasets, generating 24-h forecasts. In comparison to the second-best model, EEMD combined with BiLSTM-AM, our approach demonstrated significant improvement, reducing mean absolute error, root mean square error, and peak percentage of threshold statistics by 7.87 %, 3.91 %, and 23.51 %, respectively. The proposed model accurately captured peak and valley occurrences' timing and amplitude, surpassing existing models.