Abstract The systems in power plants exhibit intricate thermal and mechanical behaviours, making early fault detection and accurate forecasting crucial for maintaining safety, efficiency, and operational reliability. Existing methods face issues such as inconsistency among multi-source data, poor representation of thermal dynamics, and lack of ability to model long-term dependencies, which result in the delay or inaccuracy in fault detection. In response to these challenges, an AI-oriented multi-source data fusion framework is proposed in this study that combines advanced deep learning techniques for better thermal control, fault detection, and forecasting of future system behaviour in the power plant area. In this study, the dataset was obtained from the Kaggle repository of Power Plant Data: Steam Turbine and Boiler Metrics which contains normal as well as faulty operational conditions. Once the data was collected, extensive preprocessing was performed consisting of missing value handling, Minmax normalization, and noise reduction to keep the data high-quality. The extraction of features was based on analysis, domain knowledge, and time-series representations, which reinforced the thermal characterizations. A combined deep learning model of CNNs for spatial feature extraction and LSTMs for temporal pattern learning was developed for the purpose of fault detection and remaining useful life (RUL) prediction. The new method not only traditional ways but also gave very high measurement values such as 0.9988 for accuracy, 0.9968 for precision, and 0.9955 for recall and 0.9961 for F1-score. Besides, the remaining useful life prediction had so small errors the MAE, RMSE, and MAPE all indicated convergence. Thus, the results support the hybrid model’s power and reliability in prediction. This research presents a highly effective and adaptable approach for the continuous monitoring, proactive maintenance, and enhanced decision-making of modern power plant operations.
Li et al. (Mon,) studied this question.