Published Paper Details:

Wind power has become a pillar of the worldwide transition to renewable energy. Yet, the very nature of wind as being variable and intermittent creates serious challenges for assured grid integration, energy storage, and electricity market operations. Reliable forecasting of the wind speed, especially on short and medium time horizons, is thus imperative. Classic models like ARIMA and persistence methods, though interpretable, are unable to handle the non-linearities and the huge volumes of contemporary high-frequency wind data. Deep learning architectures like LSTM offer better accuracy but require large computational resources and huge amounts of training data, which restrict their applicability in real-time environments.This paper suggests a pattern sequence similarity-based forecasting model applied in a big data setting with Hadoop and Spark. Building on Dynamic Time Warping (DTW) and Euclidean distance, the approach locates preceding periods with maximal similarities under current circumstances and projects them forward to estimate upcoming wind speed. Empirical verification was done on an empirical database consisting of 34,080 observations taken every 15 minutes spanning 2022. Cleaning, normalization, and lag feature extraction were the preprocessing activities.The model was compared with ARIMA and LSTM. Results show that the similarity-based method is better than these models, improving RMSE by about 15-20% and MAE by 10-12%, and especially performing well in short-term predictions (1-6 hours ahead). In addition to better accuracy, the approach proves scalable and efficient, essential for real-time use.The research theoretically contributes by generalizing sequence similarity methods to renewable energy prediction, and practically contributes by providing a strong, scalable, and precise method to enable wind power scheduling, smart grid operation, and energy trading platforms.