A Novel Ensemble Fault Diagnosis Model for Main Circulation Pumps of Converter Valves in VSC-HVDC Transmission Systems
Abstract
:1. Introduction
- (i)
- This study has conducted a comprehensive validation of different fault diagnosis models for the main circulation pump in the specific application scenario. The findings reveal that while LSTM performs the best among individual models, the ensemble comprising MLP, CNN, and LSTM surpasses both the individual models and other ensembles in terms of diagnostic performance. This highlights the strength of utilizing a combined approach for improved fault diagnosis outcomes.
- (ii)
- The contribution of this article also lies in the development of a proposed model that effectively harnesses the strengths of different AI models, thereby overcoming the limitations associated with existing fault diagnosis algorithms when applied to complex operating conditions and limited fault data scenarios in main circulation pumps. Specifically, a novel training approach called the coach-members method is introduced for ensemble models, enhancing the compatibility of diverse base learners across various working conditions and resulting in an ensemble fault diagnosis model that achieves superior performance. This innovative approach demonstrates the capability to overcome challenges associated with fault diagnosis in demanding operational environments.
2. Overview of Fault Diagnosis for the Main Circulation Pump
2.1. A Brief Introduction to the Main Circulation Pump
- (i)
- Bearing lubrication failure: usually caused by insufficient lubrication or the use of low-quality lubricants, leading to increased friction and heat generation in the bearings. This can result in premature failure of the bearings and potential damage to the shaft.
- (ii)
- Shaft center deviation: usually caused by improper assembly or installation, resulting in misalignment between the shaft and bearings. The consequences of shaft center deviation include increased vibration, wear and tear on the bearings, and potential damage to the impeller.
- (iii)
- Water leakage: due to wear and tear on the pump housing or impeller or inadequate sealing of the inlet or outlet. The consequences of water leakage include reduced coolant flow, increased pressure in the cooling system, and potential damage to other components in the system.
2.2. The Mathematical Model for Main Circulation Pump Fault Diagnosis
- (i)
- Accuracy () is used to measure the overall correctness of the fault diagnosis model, which is defined as follows:
- (ii)
- Precision () aims to determine the proportion of correctly diagnosed positive samples, which can be calculated using Equation (8):
- (iii)
- Recall rate () aims to determine the proportion of actual positive samples that are correctly predicted as positive, and it can be calculated using the following equation:
- (iv)
- F1 score () is a metric that reflects both accuracy and recall. It is defined as follows:
3. Model Construction
3.1. Base Learners for Fault Diagnosis
- (i)
- MLP: a feedforward neural network model that consists of multiple layers of interconnected nodes and can be trained using backpropagation. It is known for its ability to learn complex non-linear relationships between inputs and outputs, making it suitable for capturing patterns in fault data. The advantages of MLP in fault diagnosis include its flexibility in handling different types of data, ability to learn complex patterns, and potential for accurate classification and prediction. However, MLP may have limitations in handling noisy data, overfitting, and sensitivity to hyperparameter tuning.
- (ii)
- CNN: a type of deep neural network that is designed for image or data in the form of an array. It uses convolutional layers to automatically learn spatial hierarchies of features from data, making it highly suitable for fault diagnosis tasks where the fault data are presented as an array. CNN has been shown to achieve state-of-the-art performance in many fault diagnosis tasks, making it a popular choice in the field.
- (iii)
- LSTM: a type of recurrent neural network (RNN) that is designed to overcome the vanishing gradient problem in traditional RNNs, making it more effective in capturing long-term dependencies in sequential data. LSTM introduces specialized memory cells with gating mechanisms that can control the flow of information, allowing for the effective handling of sequences with varying time scales. LSTM has gained popularity in fault diagnosis tasks that involve time series data due to its ability to capture complex temporal dependencies, handle variable-length sequences, and mitigate the vanishing gradient problem.
3.2. The DRL-Based Weighting Model
3.3. The Coach-Members Method: Training of the Ensemble Fault Diagnosis Model
- (i)
- Normalize the sampled vibration signal data by mapping it into . The normalization formula is in shown in Equation (15):
- (ii)
- Divide the sample dataset into a training set, validation set and test set by the ratio of 3:1:1.
- (iii)
- For each base learner, obtain candidate models with different combinations of hyperparameters using manual experience and grid search methods.
- (iv)
- Perform the first-stage rough training for all candidate base learners on the training set.
- (v)
- Evaluate the trained candidate base learners on the validation set. From each type of candidate base learner, select the optimal one producing the lowest error, respectively, so as to form the combination of roughly trained base learners.
- (vi)
- Similarly, obtain candidate DRL-based weighting models with different combinations of hyperparameters. Obtain the same number of candidate ensemble fault diagnosis models by coordinating each candidate DRL-based weighting model with the roughly trained base learner combination, respectively.
- (vii)
- Perform the second-stage fine training for all candidate ensemble fault diagnosis models on the training set.
- (viii)
- Evaluate the trained candidate ensemble fault diagnosis models on the validation set and select the optimal one producing the lowest error.
- (ix)
- Test the optimal ensemble fault diagnosis model on the test set. Denormalize and analyze the output with the performance metrics, as shown in (6) to (9).
- (i)
- When members join the team, they should already have a good foundation in terms of their abilities and skills;
- (ii)
- The coach is able to optimize the team’s cooperation strategies in real time based on the situation on the field and the actual capabilities of each member;
- (iii)
- Members are able to optimize their collaboration with each other in real time based on the strategies provided by the coach, allowing for efficient communication and synergy among team members.
- (i)
- Before being incorporated as a base learner in the ensemble fault diagnosis model, each fault diagnosis model must have its hyperparameters and parameters fine-tuned to perform the fault diagnosis task independently and with satisfactory performance.
- (ii)
- The DRL-based weighting model should continuously optimize its parameters based on diagnostic errors. This allows it to determine the weight of each base learner more appropriately based on the operating status of the main circulation pump and the output of each base learner. As a result, each base learner can leverage its strengths and avoid weaknesses in diagnostic tasks under different working conditions.
- (iii)
- All base learners should continuously and simultaneously optimize their respective parameters based on diagnostic errors, so that when the DRL-based weighting model provides specific weight combinations, they can improve their collaboration with other base learners and enhance the accuracy of fault diagnosis.
4. Case Study
4.1. Data Source and Division
- (i)
- During routine maintenance, the converter valve was shut down nine times from 2018 to 2020, resulting in the main circulation pump’s vibration signal data being divided into roughly 10 sections by the nine breakpoints.
- (ii)
- Each data point is labeled by the operation and maintenance personnel of the converter station with a certain type of fault. To reduce the information extraction burden of the model, independent data sequences should only contain data points with the same fault type.
- (iii)
- The selection of the length of each independent data sequence, i.e., in (1), requires rigorous calculation. The principle of calculation is to minimize the complexity of training while ensuring the integrity of the temporal information contained in each data sequence. Trappenberg et al. [30] proposed a method for calculating the optimal q based on the principles mentioned above, and in this study, is set to 2380 using this method.
4.2. Case 1: Evaluation for Single Base Learners
4.3. Case 2: Evaluation for Ensemble Learning Using Different Combinations of Base Learners
4.4. Case 3: Evaluation for Ensemble Learning Using Different Weight Determination Method
5. Conclusions
Author Contributions
Funding
Institutional Review Board Statement
Informed Consent Statement
Data Availability Statement
Acknowledgments
Conflicts of Interest
Appendix A
Model | Name of Hyperparameter | Range | Optimal Setting |
---|---|---|---|
MLP | Number of layers | 1–4 | 3 |
Number of neurons in each layer | 2–256 | (64,32,8) | |
RF | Number of trees | 2–200 | 18 |
Maximum depth of trees | 2–20 | 5 | |
Minimum sample size of split nodes | 2–20 | 4 | |
Minimum sample size of leaf nodes | 2–20 | 4 | |
CNN | Number of layers | 1–4 | 2 |
Number of channels | 2–100 | (42,82) | |
LSTM | Number of layers | 1–4 | 2 |
Number of neurons in each layer | 2–256 | (21,14) | |
GRU | Number of layers | 1–4 | 2 |
Number of neurons in each layer | 2–256 | (20,10) |
Index | Model Combination | Accuracy (%) | Precision (%) | Recall (%) | F1 (%) |
---|---|---|---|---|---|
1 | {1,2} | 87.50 | 82.50 | 71.74 | 76.74 |
2 | {1,3} | 88.75 | 85.00 | 73.91 | 79.07 |
3 | {1,4} | 90.31 | 86.25 | 77.53 | 81.66 |
4 | {1,5} | 90.63 | 87.50 | 77.78 | 82.35 |
5 | {2,3} | 89.38 | 83.75 | 76.14 | 79.76 |
6 | {2,4} | 90.31 | 86.25 | 77.53 | 81.66 |
7 | {2,5} | 91.25 | 86.25 | 80.23 | 83.13 |
8 | {3,4} | 92.50 | 90.00 | 81.82 | 85.71 |
9 | {3,5} | 91.25 | 88.75 | 78.89 | 83.53 |
10 | {4,5} | 92.81 | 90.00 | 82.76 | 86.23 |
11 | {1,2,3} | 93.13 | 90.00 | 83.72 | 86.75 |
12 | {1,2,4} | 93.75 | 91.25 | 84.88 | 87.95 |
13 | {1,2,5} | 93.13 | 88.75 | 84.52 | 86.59 |
14 | {1,3,4} | 95.00 | 95.00 | 86.36 | 90.48 |
15 | {1,3,5} | 93.44 | 91.25 | 83.91 | 87.43 |
16 | {1,4,5} | 93.13 | 90.00 | 83.72 | 86.75 |
17 | {2,3,4} | 93.75 | 88.75 | 86.59 | 87.65 |
18 | {2,3,5} | 93.13 | 88.75 | 84.52 | 86.59 |
19 | {2,4,5} | 92.81 | 88.75 | 83.53 | 86.06 |
20 | {3,4,5} | 93.44 | 88.75 | 85.54 | 87.12 |
21 | {1,2,3,4} | 94.06 | 92.50 | 85.06 | 88.62 |
22 | {1,2,3,5} | 93.44 | 91.25 | 83.91 | 87.43 |
23 | {1,2,4,5} | 94.38 | 93.75 | 85.23 | 89.29 |
24 | {1,3,4,5} | 93.75 | 92.50 | 84.09 | 88.10 |
25 | {2,3,4,5} | 94.38 | 91.25 | 86.90 | 89.02 |
26 | {1,2,3,4,5} | 94.38 | 93.75 | 85.23 | 89.29 |
References
- Barnes, M.; Van Hertem, D.; Teeuwsen, S.P.; Callavik, M. HVDC Systems in Smart Grids. Proc. IEEE 2017, 105, 2082–2098. [Google Scholar] [CrossRef]
- Lescale, V.F. Modern HVDC: State of the art and development trends. In Proceedings of the 1998 International Conference on Power System Technology (POWERCON 98), Beijing, China, 18–21 August 1998; pp. 446–450. [Google Scholar]
- Kwak, J.S.; Kim, C.K.; Koh, B.E. Cooling system of Haenam-Jeju HVDC system. In Proceedings of the IEEE International Symposium on Industrial Electronics (ISIE), Pusan, Republic of Korea, 12–16 June 2001; pp. 1000–1005. [Google Scholar]
- Holweg, J.; Lips, H.P.; Tu, B.Q.; Uder, M.; Peng, B.S.; Zhang, Y.G. Modem HVDC thyristor valves for China’s electric power system. In Proceedings of the International Conference on Power System Technology, Kunming, China, 13–17 October 2002; pp. 520–527. [Google Scholar]
- Liu, Z.L.; Li, M.G.; Zhu, Z.F.; Xiao, L.H.; Nie, C.H.; Tang, Z.C. Health State Identification Method of Nuclear Power Main Circulating Pump Based on EEMD and OQGA-SVM. Electronics 2023, 12, 410. [Google Scholar] [CrossRef]
- Mansouri, S.A.; Ahmarinejad, A.; Javadi, M.S.; Heidari, R.; Catalao, J.P.S. Improved double-surface sliding mode observer for flux and speed estimation of induction motors. IET Electr. Power Appl. 2020, 14, 1002–1010. [Google Scholar] [CrossRef]
- Pecinka, L.; Stulik, P.; Zeman, V. Dynamic Response of the wwer 1000/320 reactor on The Pressure Pulsations Generated by Main Circulation Pumps. In Proceedings of the Pressure Vessels and Piping Conference of the American-Society-of-Mechanical-Engineers, Prague, Czech Republic, 26–30 July 2009; pp. 41–44. [Google Scholar]
- Vasyliv, K.M.; Mazurenko, L.I. A mathematical model of the electrical engineering complex for drive of main circulation pumps of nuclear reactor vver-1000 of nuclear power plants. Electr. Eng. Electromec. 2019, 6, 12–20. [Google Scholar] [CrossRef]
- Mansouri, S.A.; Ahmarinejad, A.; Nematbakhsh, E.; Javadi, M.S.; Jordehi, A.R.; Catalao, J.P.S. Energy management in microgrids including smart homes: A multi-objective approach. Sustain. Cities Soc. 2021, 69, 102852. [Google Scholar] [CrossRef]
- Mansouri, S.A.; Nematbakhsh, E.; Jordehi, A.R.; Tostado-Veliz, M.; Jurado, F.; Leonowicz, Z. A Risk-Based Bi-Level Bidding System to Manage Day-Ahead Electricity Market and Scheduling of Interconnected Microgrids in the presence of Smart Homes. In Proceedings of the IEEE International Conference on Environment and Electrical Engineering (EEEIC)/IEEE Industrial and Commercial Power Systems Europe (I and CPS Europe) Conference, Prague, Czech Republic, 28 June–1 July 2022. [Google Scholar]
- Khrustalev, V.A.; Bashlykov, D.O. Application of Frequency Regulation of the Main Circulation Pump for Increasing the Nominal Power of VVER. Atom. Energy 2016, 120, 112–118. [Google Scholar] [CrossRef]
- Mansouri, S.A.; Jordehi, A.R.; Marzband, M.; Tostado-Veliz, M.; Jurado, F.; Aguado, J.A. An IoT-enabled hierarchical decentralized framework for multi-energy microgrids market management in the presence of smart prosumers using a deep learning-based forecaster. Appl. Energy 2023, 333, 120560. [Google Scholar] [CrossRef]
- Zhou, S.H.; Qin, L.; Sun, H.; Peng, B.; Ruan, J.J.; Wang, J.; Tang, X.; Wang, X.L.; Liu, K.P. TransFNN: A Novel Overtemperature Prediction Method for HVDC Converter Valves Based on an Improved Transformer and the F-NN Algorithm. Sensors 2023, 23, 4110. [Google Scholar] [CrossRef] [PubMed]
- Waqar, T.; Demetgul, M. Thermal analysis MLP neural network based fault diagnosis on worm gears. Measurement 2016, 86, 56–66. [Google Scholar] [CrossRef]
- Xia, F.; Zhang, H.; Peng, D.G.; Li, H.; Su, Y.K. Turbine Fault Diagnosis Based on Fuzzy Theory and SVM. In Proceedings of the International Conference on Artificial Intelligence and Computational Intelligence, Shanghai, China, 7–8 November 2009; p. 668. [Google Scholar]
- Yao, Q.; Wang, J.; Yang, L.; Su, H.X.; Zhang, G.G. A Fault Diagnosis Method of Engine Rotor Based on Random Forests. In Proceedings of the IEEE International Conference on Prognostics and Health Management (ICPHM), Carleton University, Ottawa, ON, Canada, 20–22 June 2016. [Google Scholar]
- Sun, G.; Shen, P.; Zhao, Y.; Zhu, H.; Ding, X. Intelligent recognition of power grid monitoring alarm event combining knowledge base and deep learning. Electr. Power Autom. Equip. 2020, 40, 40–47. [Google Scholar]
- Zou, P.; Hou, B.C.; Jiang, L.; Zhang, Z.J. Bearing Fault Diagnosis Method Based on EEMD and LSTM. Int. J. Comput. Commun. Control 2020, 15, 14. [Google Scholar] [CrossRef]
- Yuan, J.; Tian, Y. An Intelligent Fault Diagnosis Method Using GRU Neural Network towards Sequential Data in Dynamic Processes. Processes 2019, 7, 152. [Google Scholar] [CrossRef]
- Vaish, R.; Dwivedi, U.D.; Tewari, S.; Tripathi, S.M. Machine learning applications in power system fault diagnosis: Research advancements and perspectives. Eng. Appl. Artif. Intell. 2021, 106, 33. [Google Scholar] [CrossRef]
- Webb, G.I.; Zheng, Z.J. Multistrategy ensemble learning: Reducing error by combining ensemble learning techniques. IEEE Trans. Knowl. Data Eng. 2004, 16, 980–991. [Google Scholar] [CrossRef]
- Rojarath, A.; Songpan, W.; Pong-inwong, C. Improved Ensemble learning for Classification Techniques Based on Majority Voting. In Proceedings of the 7th IEEE International Conference on Software Engineering and Service Science (ICSESS), China Hall Sci & Technol, Beijing, China, 26–28 August 2016; pp. 107–110. [Google Scholar]
- Atallah, R.; Al-Mousa, A. Heart Disease Detection Using Machine Learning Majority Voting Ensemble Method. In Proceedings of the 2nd International Conference on New Trends in Computing Sciences (ICTCS), Amman, Jordan, 9–11 October 2019; pp. 335–340. [Google Scholar]
- Tang, Y.; Dai, Q.; Yang, M.Y.; Du, T.Y.; Chen, L.F. Software defect prediction ensemble learning algorithm based on adaptive variable sparrow search algorithm. Int. J. Mach. Learn. Cybern. 2023, 21, 1967–1987. [Google Scholar] [CrossRef]
- Ekstrom, A.; Eklund, L. HVDC Thyristor Valve Development. IEEE Trans. Power Electron. 1987, 2, 177–185. [Google Scholar] [CrossRef]
- Chen, W.Y.; Wang, M.; Yu, T.; Destech Publicat, I. Reliability Evaluation of the Converter Valve Cooling System in HVDC Transmission System Based on Fault Tree Analysis. In Proceedings of the 4th International Conference on Mechanical Automation and Materials Engineering (ICMAME), Wuhan, China, 23–24 May 2015; pp. 614–619. [Google Scholar]
- Ji, Z.; Kang, C.Q.; Chen, Q.X.; Xia, Q.; Jiang, C.M.; Chen, Z.X.; Xin, J.B. Low-Carbon Power System Dispatch Incorporating Carbon Capture Power Plants. IEEE Trans. Power Syst. 2013, 28, 4615–4623. [Google Scholar] [CrossRef]
- Luo, Z.; Zhang, Z.Y.; Yan, X.; Qin, J.H.; Zhu, Z.D.; Wang, H.; Gao, Z.Y. Dissolved Gas Analysis of Insulating Oil in Electric Power Transformers: A Case Study Using SDAE-LSTM. Math. Probl. Eng. 2020, 2020, 1–10. [Google Scholar] [CrossRef]
- Zhou, S.; Qin, L.; Ruan, J.; Wang, J.; Liu, H.; Tang, X.; Wang, X.; Liu, K. An AI-Based Power Reserve Control Strategy for Photovoltaic Power Generation Systems Participating in Frequency Regulation of Microgrids. Electronics 2023, 12, 2075. [Google Scholar] [CrossRef]
- Trappenberg, T.; Ouyang, J.; Back, A. Input variable selection: Mutual information and linear mixing measures. IEEE Trans. Knowl. Data Eng. 2006, 18, 37–46. [Google Scholar] [CrossRef]
Model | Accuracy (%) | Precision (%) | Recall (%) | F1 (%) | |
---|---|---|---|---|---|
Index | Name | ||||
1 | MLP | 84.06 | 75.00 | 65.93 | 70.18 |
2 | RF | 86.56 | 81.25 | 69.89 | 75.14 |
3 | CNN | 89.69 | 86.25 | 75.82 | 80.70 |
4 | LSTM | 90.94 | 86.25 | 79.31 | 82.63 |
5 | GRU | 90.63 | 87.50 | 77.78 | 82.35 |
Method | Accuracy (%) | Precision (%) | Recall (%) | F1 (%) |
---|---|---|---|---|
Average | 91.56 | 86.25 | 81.18 | 83.64 |
Voting | 91.88 | 87.50 | 81.40 | 84.34 |
ISA | 93.44 | 92.50 | 83.15 | 87.57 |
DRL-1 | 92.19 | 90.00 | 80.90 | 85.21 |
DRL-2 | 95.00 | 95.00 | 86.36 | 90.48 |
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2023 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).
Share and Cite
Zhou, S.; Qin, L.; Yang, Y.; Wei, Z.; Wang, J.; Wang, J.; Ruan, J.; Tang, X.; Wang, X.; Liu, K. A Novel Ensemble Fault Diagnosis Model for Main Circulation Pumps of Converter Valves in VSC-HVDC Transmission Systems. Sensors 2023, 23, 5082. https://doi.org/10.3390/s23115082
Zhou S, Qin L, Yang Y, Wei Z, Wang J, Wang J, Ruan J, Tang X, Wang X, Liu K. A Novel Ensemble Fault Diagnosis Model for Main Circulation Pumps of Converter Valves in VSC-HVDC Transmission Systems. Sensors. 2023; 23(11):5082. https://doi.org/10.3390/s23115082
Chicago/Turabian StyleZhou, Sihan, Liang Qin, Yong Yang, Zheng Wei, Jialong Wang, Jing Wang, Jiangjun Ruan, Xu Tang, Xiaole Wang, and Kaipei Liu. 2023. "A Novel Ensemble Fault Diagnosis Model for Main Circulation Pumps of Converter Valves in VSC-HVDC Transmission Systems" Sensors 23, no. 11: 5082. https://doi.org/10.3390/s23115082
APA StyleZhou, S., Qin, L., Yang, Y., Wei, Z., Wang, J., Wang, J., Ruan, J., Tang, X., Wang, X., & Liu, K. (2023). A Novel Ensemble Fault Diagnosis Model for Main Circulation Pumps of Converter Valves in VSC-HVDC Transmission Systems. Sensors, 23(11), 5082. https://doi.org/10.3390/s23115082