Accurate wind energy forecasting is essential for grid stability, energy-demand balance, and the efficient use of renewables. Shallow learning methods are favored for their scalability and generalization ability, yet their performance strongly depends on proper hyperparameter tuning. This study introduces an enhanced XGBoost model with a Gradient-Dynamic Tree Pruning (GDTP) mechanism to control tree complexity adaptively, optimized through a novel DE-Fly hybrid algorithm that integrates Differential Evolution, Firefly Algorithm, and Mayfly Algorithm. Experimental validation using real-world wind power data demonstrates that the proposed DE-Fly–optimized GDTP-XGBoost model achieves superior forecasting accuracy and significantly faster computation than conventional approaches. Beyond predictive performance, the framework provides practical benefits by supporting condition-based maintenance, enabling earlier anomaly detection, minimizing downtime, and enhancing the overall reliability of wind farm operations.