In response to the limited data feature extraction capability of a single neural network and data feature loss in the serial connection of multiple neural networks in fault diagnosis, a parallel network structure model comprising Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks is proposed, called CNN-LSTM parallel network model. The model’s key is to simultaneously extract features from spatial and temporal dimensions of Supervisory Control and Data Acquisition (SCADA) data and make state judgments. Additionally, a partial ensemble learning meta-model was established to Identify fault types that are difficult to distinguish due to small differences in data. The CNN-LSTM parallel network model is employed for fault detection of the wind turbine using SCADA data. It‘s verified that the fault detection accuracy using the CNN-LSTM parallel network model is up to 99.60%, which is higher than the fault detection accuracy using single neural network models CNN and LSTM, as well as the CNN-LSTM serial connection model. Besides, the model outperforms the other models in terms of evaluation metrics, such as Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Mean Absolute Error (MAE).

Peng Xin, Xun Zhang, Chenglei Yuan, Chaoran Li. Fault Diagnosis of Wind Turbine Based on CNN-LSTM Parallel Network Model. Academic Journal of Engineering and Technology Science (2023) Vol. 6, Issue 12: 115-121. https://doi.org/10.25236/AJETS.2023.061217.

[1] LI Songfeng. Development Status and Development Trend of China's Wind Power Equipment Industry [J]. Value Engineering, 2019, 38(33): 37-38.

[2] Wang J, Ma Y, Zhang L, et al. Deep learning for smart manufacturing: Methods and applications[J]. Journal of Manufacturing Systems, 2018, 48.

[3] PENG Cheng ,LI Fengjuan,JIANG Jinyuan. Summary of fault diagnosis and prediction methods based on deep learning [J]. Modern Electronics Technique, 2022, 45(03): 111-120. 

[4] WEN Chenglin, LÜ Feiya. Review on Deep Learning Based Fault Diagnosis [J]. Journal of Electronics & Information Technology, 2020, 42(01): 234-248. 

[5] Jie T, Yilun L, Dalian Y. Bearing Fault Diagnosis Based on Deep Belief Network and Multisensor Information Fusion[J]. Shock and Vibration, 2016, 2016(Pt.6).

[6] LIANG Tao ,HUANG Yue. The Design of SCADA System for Wind Power Plants [J]. Hydropower and New Energy, 2023, 37(01): 1-4+10.

[7] Ren Yan. Research on anomaly identification method of wind turbine generator based on convolution neural network(CNN) [D]. Northeastern University, 2018.

[8] CAO Zhengzhi, YE Chunming. Application of Improved CNN-LSTM Model in Fault Diagnosis of Rolling Bearings [J]. Computer Systems & Applications, 2021, 30(03): 126-133.

[9] Zhou B, Khosla A, Lapedriza À, et al. Learning Deep Features for Discriminative Localization[J]. CoRR, 2015, abs/1512.04150.

[10] GUO Xiaojing, YIN Yuxuan, YUN Yujing. Aeroengine Life Prediction Method Based on Improved LSTM [J]. Machine Tool & Hydraulics, 2022, 50(20): 185-193.

[11] Jinming Huang,Meng Du. Build a used car price prediction model based on Stacking ensemble learning [J]. CHINA SCIENCE AND TECHNOLOGY INFORMATION,2023(14):88-89.

[12] Wang Tingshao, Ji Tianyao, Jiang Yuzi, Wang Jin, et al. Fault diagnosis method for wind turbines based on de-noise auto-encoder and one-dimensional convolution network [J]. Electrical Measurement & Instrumentation, 2023, 60(01): 87-93+173.

[13] CHEN Haipeng, ZHOU Yuehao, WANG Chenlu, WANG Junqi, HAN Hao, et al. Economic Analysis of System Spinning Reserve Based on Improved CNN-LSTM Short Term Wind Power Prediction [J]. High Voltage Engineering, 2022, 48(02): 439-448.

[14] XIN Peng, YANG Kaixun, WEN Xiaoqiang. Research on Wind Turbine Fault Diagnosis Method based on Improved SE-CNN [J].Journal of Jilin Institute of Chemical Technology,2023,40(01):34-40.

[15] LU Tongchao, MA Lele, FEI Ziyan, et al. Fault Identification of Aviation Engine State Perception Network Based on GMM－CNN－LSTM [J]. Industrial Control Computer, 2021, 34(10): 15-17.