Research on Power Plant Equipment Failure Early Warning Technology Based on Data-Driven and Deep Learning

In order to ensure the safety and reliability of power plant equipment operation, we propose a power plant equipment failure early warning model based on data-driven and deep learning. Our method deeply explores the data characteristics of historical operation information and captures key factors affecting equipment failure condition through feature selection and feature fusion. Furthermore, a deep learning model is constructed, which combines Bi-directional Long Short-Term Memory (BiLSTM) model with attention mechanism, to predict the future change tendency of the measuring points. Finally, our model introduces a failure threshold function to issue an alarm message before the threshold is likely to be exceeded, thereby providing early warning before the failure actually occurs. Experiment results show that our proposed method has high prediction accuracy and significantly outperforms the baseline models, which has great promotion and application value.