Integrating Deep Learning Algorithms for Fault Prediction and Thermal Efficiency Optimization in Gas-Steam Combined Cycle Power Plants
DOI:
https://doi.org/10.31185/wjes.Vol14.Iss3.977Keywords:
Combined cycle power plant, Deep learning, Fault prediction, Thermal efficiency optimization, Long Short-Term Memory, Convolutional Neural Network, Transformer, Heat recovery steam generator, Predictive maintenanceAbstract
Gas-steam combined cycle plants (CCPPs) are the leading technology for producing large-scale thermal electric power, they have thermal efficiencies greater than 60% and employ a multi-system architecture. However, CCPPs' superior thermodynamic properties, they are vulnerable to complex, evolving fault modes that are difficult to detect at an early stage using conventional monitoring strategies based on thresholds. In conjunction with this, real-time optimization of thermal efficiencies requires constantly changing multiple variables outside of the capability of manual or traditional controls. This paper proposes and implements a comprehensive deep learning framework that addresses both issues at the same time. The framework consists of three complementary components: (1) a one-dimensional Convolutional Neural Network (1D-CNN) that extracts local spatial-temporal fault signatures from high-dimensional sensor array data; (2) a Bidirectional Long Short-Term Memory (BiLSTM) network that models long-range temporal degradation dynamics; and (3) a multi-head Transformer attention mechanism that adaptively weights time intervals that are diagnostically significant. The second module of the framework employs an LSTM-Transformer encoder-decoder for real-time thermal efficiency prediction and constrained set-point optimization with a 30-minute prediction horizon. Performance was validated using 36 months' worth of operational data from a 750 MW industrial CCPP with a total of 187 sensor channels classified into 14 categories of fault. The proposed CNN-BiLSTM-Attention model has achieved a macro-averaged fault detection accuracy of 94.3% and a macro-averaged F1-score of 0.921 with a 31.4% reduction in false alarms, compared monitoring done using threshold-based methods. The thermal efficiency prediction module demonstrated a root mean square error (RMSE) of 0.41 pp and a mean absolute percentage error (MAPE) of 0.68%.
References
[1] A. Mehrpanahi, M. Akbari Vakilabadi, S. Nikbakht Naserabad, and M. H. Ahmadi, “Multiobjective optimization of heat recovery steam generator in a combined cycle power plant using genetic algorithm,” Energy Science & Engineering, vol. 11, no. 11, pp. 4224–4237, 2023. doi: 10.1002/ese3.1575. DOI: https://doi.org/10.1002/ese3.1575
[2] A. Nekoonam and M. Montazeri-Gh, “Noise-robust gas path fault detection and isolation for a power generation gas turbine based on deep residual compensation extreme learning machine,” Energy Science & Engineering, vol. 11, no. 11, pp. 3864–3885, 2023. doi: 10.1002/ese3.1576. DOI: https://doi.org/10.1002/ese3.1576
[3] P. Y. Aisyah, T. Sochartanto, R. D. Noriyati, and F. Chilmi, “Analysis of heat recovery steam generator performance through simulation based on artificial neural network,” AIP Conference Proceedings, vol. 2580, no. 1, p. 040012, 2023. doi: 10.1063/5.0122817. DOI: https://doi.org/10.1063/5.0122817
[4] A. T. W. K. Fahmi, K. R. Kashyzadeh, and S. Ghorbani, “A comprehensive review on mechanical failures causing vibration in gas turbines of combined cycle power plants,” Engineering Failure Analysis, vol. 134, p. 106094, 2022. doi: 10.1016/j.engfailanal.2022.106094. DOI: https://doi.org/10.1016/j.engfailanal.2022.106094
[5] J. Su, H. Song, F. Song, et al., “Fault diagnosis of steam power system based on convolutional neural network,” Chinese Journal of Ship Research, vol. 17, no. 6, pp. 96–102, 2022. doi: 10.19693/j.issn.1673-3185.02616.
[6] S. Tang, S. Yuan, and Y. Zhu, “Deep learning-based intelligent fault diagnosis methods toward rotating machinery,” IEEE Access, vol. 8, pp. 9335–9346, 2020. doi: 10.1109/ACCESS.2020.2964871. DOI: https://doi.org/10.1109/ACCESS.2019.2963092
[7] H. Xu, G. Chen, and M. Li, “Electrical power output prediction of combined cycle power plants using a recurrent neural network optimized by waterwheel plant algorithm,” Frontiers in Energy Research, vol. 11, p. 1234624, 2024. doi: 10.3389/fenrg.2023.1234624. DOI: https://doi.org/10.3389/fenrg.2023.1234624
[8] M. Yoo, J. Kim, G.-S. Oh, and H.-Y. Lee, “Hybrid explainable anomaly detection framework of gas turbines for feature selection and fault localization,” Structural Health Monitoring, vol. 0, no. 0, pp. 1–17, 2025. doi: 10.1177/14759217251333312. DOI: https://doi.org/10.1177/14759217251333312
[9] J. Wang, H. Zhang, X. Li, and Y. Zhao, “Optimizing combined-cycle power plant operations using an LSTM-attention hybrid model for load forecasting,” Journal of Mechanical Sciences and Technology, vol. 39, no. 1, pp. 1–14, 2025. doi: 10.1007/s12206-025-0961-3. DOI: https://doi.org/10.1007/s12206-025-0961-3
[10] X. Wen, S. Li, and G. Wei, “Time series prediction based on LSTM-attention-LSTM model,” IEEE Access, vol. 11, pp. 48324–48331, 2023. doi: 10.1109/ACCESS.2023.3277554. DOI: https://doi.org/10.1109/ACCESS.2023.3276628
[11] Y. Zhu and L. Bi, “An optimization approach for improving steam production of heat recovery steam generator,” Scientific Reports, vol. 15, p. 3841, 2025. doi: 10.1038/s41598-025-87715-z. DOI: https://doi.org/10.1038/s41598-025-87715-z
[12] L. Chen, M. Benbouzid, and Y. Cheng, “Advancing predictive maintenance for gas turbines: An intelligent monitoring approach with ANFIS, LSTM, and reliability analysis,” ISA Transactions, vol. 129, pp. 429–441, 2023. doi: 10.1016/j.isatra.2022.09.041. DOI: https://doi.org/10.1016/j.isatra.2022.09.041
Downloads
Published
Issue
Section
License
Copyright (c) 2026 Raed AL-zaidi
This work is licensed under a Creative Commons Attribution 4.0 International License.
