<p>The global transition to renewable energy is vital for meeting sustainability objectives, but renewable resources, including solar, wind, and hydroelectric power, are intermittent and variable, thereby significantly challenging the accuracy of power forecasting and grid stabilization. Hence, this paper puts forward a hybrid model, which combines CatBoost regression, the advanced evolutionary optimization algorithms of Genetic Algorithm (GA), Differential Evolution (DE), and Bayesian Optimization (BO), and a DT-based real-time correction approach to improve the forecast accuracy as well as the system adaptability. The approach combines machine learning for initial prediction, evolutionary algorithms to dynamically fine-tune model hyper-parameters and a Digital Twin architecture which iteratively updates predictions by ingesting live sensor data feed from renewable energy assets. Empirically, it is observed that Bayesian Optimization offers the best predictive performance (i.e., lowest Root Mean Squared Error of 9.34&#xa0;MW and highest R² = 0.9774). Further, the mean absolute error is reduced by 58.5 percent by use of the Digital Twin module to provide significant improvements in forecast accuracy and rapid response. These breakthroughs allow an effective management of energy resources, thus reduce the need for fossil-fuel backup systems and the level of CO₂ emission to about 215,000 tons in simulation scenarios. The combined method provides an effective, flexible, and scalable solution toward renewable energy forecast, which contributes to better grid stability and sustainable energy policymaking. The capability of the framework to adapt to temporal environmental changes and operational deviations highlights its promising role in promoting resilient and intelligent renewable energy infrastructures on a global scale.</p> Graphical abstract <p></p>