Comparing the Johansen VECM Model to an AI Hybrid Model for the Short Term Photovoltaic Power Forecasting

Photovoltaic (PV) power systems have witnessed a massive growth of over the years. However, owing to the intermittent nature of solar energy, PV systems pose threats to the reliability, operational planning, stability, and the economic profitability of the electric grid system. Therefore, an accurate solar power forecasting is of utmost importance to stabilize and secure grid operation. In this paper, a method composed of wavelet packet decomposition (WPD), Long Short-Term Memory (LSTM) Network and multi-layer perceptron (MLP) is developed for 15-minutes ahead PV power forecasting. Two years of historical PV power data, recorded from a 20 kW PV system installed in the Reduit region, were used for training and testing. Comparisons between the proposed model and the Johansen VECM model were made in order to determine its effectiveness. Results suggest that the proposed forecasting framework outperformed the Johansen model under sunny, cloudy and rainy weather conditions. The proposed WPD-LSTM-MLP model achieved a MAE, RMSE and R2 of 90.0217, 113.7639 and 0.9992, respectively.