Abstract Snehal Sahu (22129016) M.Tech, SEE department OFFSHORE WIND SPEED PREDICTION USING DEEP LEARNING MODELS Thesis Supervisor: Pradip Swarnakar April 2025 Accurate wind speed prediction stands at the forefront of renewable energy management, essential for maintaining the structural integrity of wind turbines and ensuring the seamless integration of wind power into the electrical grid. This thesis presents an extensive analysis of various sophisticated machine learning models ranging from LSTM, BiLSTM, and Stacked LSTM to ConvLSTM, CNN LSTM, GRU, and CNN each tested for their ability to forecast wind speeds across a diverse set of hub heights, from the lower altitude of 80 meters up to the lofty height of 200 meters. Initial stationarity verification through the Augmented Dickey-Fuller (ADF) test forms the basis for a series of experiments designed to pinpoint the most accurate model in terms of predictive power. The investigation reveals nuanced differences in model performance, with the Conv LSTM model demonstrating notable accuracy at intermediate hub heights and the BiLSTM showing its strengths at greater elevations, suggesting a complex interplay between model architecture and atmospheric dynamics. Meanwhile, the Stacked LSTM, CNN LSTM, and GRU models each contribute unique insights into the temporal and spatial aspects of wind speed data. The importance of these findings extends beyond theoretical analysis; they are critical for the proactive monitoring of turbine health, which directly impacts their operational longevity and efficiency. Additionally, accurate forecasts are instrumental in managing the grid's load balance, where the stability of energy supply hinges on the predictability of wind power input. Looking to the future, this research opens the door to integrating real-time data feeds and refining machine learning algorithms to not only enhance the precision of wind speed forecasts but also to fortify the role of wind energy as a reliable and integral component of global energy systems. v