Generators are critical components in many industrial and power systems, where unexpected failures can lead to major operational and economic losses. Traditional predictive maintenance techniques often lack accuracy and timeliness. This paper presents a new method for generator fault prediction using Gate Recurrent Units (GRU), a type of recurrent neural network that is potentially effective in capturing temporal dependence in sequential data. Our method leverages GRU to analyze historical operational data, identifying patterns that indicate potential errors. We performed extensive experiments on real generator datasets, comparing the performance of GRU with other regression models, such as standard RNN. The results show that GRU achieves superior prediction accuracy and robustness, with lower evaluating benchmarks. This method not only improves the reliability of generator operations but also contributes to cost-effective maintenance strategies. Our findings highlight the potential of GRU to improve predictive maintenance practices across a variety of industries.