Comparison of artificial intelligence approaches for estimating wind energy production: A real-world case study

Abstract

• Global Wind Energy Growth : Wind energy capacity reached 920.736 GW by 2023, with Morocco contributing 1.52 GW. • Challenges in Integration : Managing wind speed variations is crucial for grid stability, maintenance, and profitability. • Forecasting Methods : The study explores machine learning algorithms like Support Vector Machines (SVM) and Recurrent Neural Networks for wind energy forecasting. • Data and Evaluation : The study used two years of 10-minute interval data from a Moroccan wind farm to assess forecasting accuracy. • RNN Performance : RNN demonstrated superior performance in daily wind energy predictions. The global installed capacity of wind energy has experienced substantial expansion during the last ten years, reaching 920.736 GW by 2023, with 1.52 GW located in Morocco. Given the growing integration of wind power into the grid, effectively handling the inherent uncertainties and variations in wind speeds has become a crucial task. The precise prediction of wind power is essential not only for the smooth integration into the power grid but also for the optimization of unit commitment, maintenance scheduling, and the improvement of power traders' profitability. The present work investigates several forecasting methodologies for wind energy by employing sophisticated machine learning algorithms, including Support Vector Machines and Recurrent Neural Networks. The analysis utilizes temporal data obtained at 10-minute intervals over a span of two years from a wind farm located in Morocco. For performance robustness assessment, three distinct error metrics were employed to evaluate accuracy in weekly, monthly, and yearly forecasting scenarios, using the persistence model as a reference point. The results illustrate the efficacy of data-driven approaches in daily wind energy prediction, specifically emphasizing the higher performance of the RNN model. These results emphasise the need of accurate prediction for the efficient operation and maintenance of contemporary wind turbines and provide direction on choosing the most appropriate prediction techniques considering variables such as time frames, input properties, and processing needs.