Abstract: Full-head 3D modeling and dynamic division of vibration zones are key technologies for ensuring the long-term safe and stable operation of hydroelectric generating units. Traditional methods rely on manual experiments and empirical judgments, making it difficult to achieve dynamic updates and intelligent predictions. Utilizing the massive historical data accumulated during the operation of hydroelectric generating units, this paper selects monitoring data such as water head, active power, and runout to propose a deep learning prediction model based on Gated Recurrent Units (GRU). Through data preprocessing, feature extraction, and sequence modeling, the model achieves visual analysis and trend prediction of three-dimensional vibration distribution and vibration zones across the full water head range. This method can provide data-driven decision support for the intelligent maintenance, vibration warning, and operational optimization of hydroelectric generating units.