With the growing energy demand and climate change, wind power is considered a promising renewable energy solution. However, the inherent fluctuations and intermittency of wind generation affect grid integration. Therefore, wind power forecasting and related data decomposition methods need to be reviewed to categorize them by factors, including model architecture. Precise wind power prediction requires a systematic review of the complex interplay between high-fidelity sensor data, appropriate decomposition methods, and advanced model architectures. By exploring the evolution of wind power prediction and categorizing it into statistical, physical, and combined techniques, we compared deep learning models (e.g., convolutional neural networks, long short-term memory, and transformers) to enhance the accuracy of sensor-data-derived power generation prediction. The results of this study serve as a guide for engineers and researchers in developing the next-generation sensors and supervisory control and data acquisition systems to collect the specific, high-resolution data streams required by modern combinatorial forecasting models.