Renewable energy resources are increasingly important for sustainable power generation, but accurately forecasting their production remains challenging due to non-stationary data and fluctuating environmental conditions. This study presents a novel hybrid model combining Variational Mode Decomposition with advanced forecasting techniques to improve the precision and reliability of photovoltaic power predictions. By evaluating data from two grid-connected photovoltaic power plants in Algeria, the model demonstrates significant improvements over traditional methods, including Long Short-Term Memory, 1D-Convolutional Neural Network, and Gated Recurrent Unit. Key results show a substantial reduction in Root Mean Square Error by up to 89.39%, highlighting the effectiveness of the proposed approach.