Noise-intensification data augmented machine learning for day-ahead wind power forecast

Abstract

The day-ahead wind power forecast is essential for the designation of dispatch schedules for the grid and rational arrangement for production planning by power generation companies. This paper specifically investigates the effect of adding noise to the original wind data for forecasting models. Linear regression, artificial neural networks, and adaptive boosting predictive models based on data-intensification white noise and uniform noise are evaluated in detail and their superiority over the original data-based models is compared. The results demonstrate that solely injecting noise into the dataset can statistically boost the performance of all forecasting models with learning algorithms. The findings of this study suggest a fresh perspective for developing wind power prediction models and carry certain wind energy engineering merits.