基于深度置信网络的风力发电机故障诊断方法
|
摘要
为了避免严重的生产运行事故,同时降低设备运行维护成本,提高风力发电机的可靠性,本文提出一种基于深度置信网络(deep belief network,DBN)的新型风力发电机故障诊断(fault diag-nosis and isolation,FDI)方法。本文首先通过DBN网络构建了故障诊断模型,然后在风力发电机的基准模型中进行故障诊断仿真测试,并把该完全数据驱动型的故障诊断效果,与传统的基于模型的诊断方法和数据驱动型诊断方法的效果作对比。此外,在仿真中也采用高斯噪声来模拟风力发电机实际运行环境中的噪声,从而解决了实际使用中网络易受噪声干扰的问题,并进一步对基于DBN的故障诊断方法进行鲁棒性测试。仿真结果表明基于DBN的数据驱动型FDI方法对风力发电机的故障有着更好的诊断效果,同时在有噪声干扰的环境下也保持着较为稳定的诊断效果。
In order to improve the reliability of wind turbines,avoid serious accidents and reduce operation and maintenance costs,a fault diagnosis and isolation( FDI) method for wind turbines using deep belief network( DBN) is proposed. The DBN employed no knowledge of physical model but historical data without any selection. The proposed method was evaluated in a wind turbine benchmark model,in comparison with model-based algorithms and conventional data-driven methods. Besides,considering the disturbance in real application,extensive evaluation was taken to analyze the robustness of proposed method which applied Gaussian noise to simulate real noise. The simulation results show that the data-driven FDI method based on DBN for wind turbines achieves the highest accuracy,and it keeps stable diagnostic performance in the strong disturbance of noise.
In order to improve the reliability of wind turbines,avoid serious accidents and reduce operation and maintenance costs,a fault diagnosis and isolation( FDI) method for wind turbines using deep belief network( DBN) is proposed. The DBN employed no knowledge of physical model but historical data without any selection. The proposed method was evaluated in a wind turbine benchmark model,in comparison with model-based algorithms and conventional data-driven methods. Besides,considering the disturbance in real application,extensive evaluation was taken to analyze the robustness of proposed method which applied Gaussian noise to simulate real noise. The simulation results show that the data-driven FDI method based on DBN for wind turbines achieves the highest accuracy,and it keeps stable diagnostic performance in the strong disturbance of noise.
引文
[1] ODGAARD P F,STOUSTRUP J,KINNAERT M. Fault-tolerantcontrol of wind turbines:a benchmark model[J]. IEEE Transac-tions on Control Systems Technology,2013,42(4):1168.
[2]马宏忠,张正东,时维俊,等.基于转子瞬时功率谱的双馈风力发电机定子绕组故障诊断[J].电力系统自动化,2014,38(14):30.MA Hongzhong,ZHANG Zhengdong,SHI Weijun,et al. Doubly-fed induction generator stator fault diagnosis based on rotor instan-taneous power spectrum[J]. Automation of Electric Power Sys-tems,2014,38(14):30.
[3]苗锐,陈国初,李月,等.基于随机集含糊证据的风力发电机故障诊断方法[J].电力系统自动化,2012,36(7):22.MIAO Rui,CHEN Guochu,LI Yue,et al. A wind turbine faultdiagnosis method based on vague evidence of random set[J]. Auto-mation of Electric Power Systems,2012,36(7):22.
[4] DEY S,PISU P,AYALEW B. A comparative study of three faultdiagnosis schemes for wind turbines[J]. IEEE Transactions onControl Systems Technology,2015,23(5):1853.
[5] OZDEMIR A A,SEILER P,BALAS G J. Wind turbine fault de-tection using counter-based residual thresholding[J]. IFAC Pro-ceedings Volumes,2011,44(1):8289.
[6] BESSA I V D,PALHARES R M,FILHO J E C. Data-driven faultdetection and isolation scheme for a wind turbine benchmark[J].Renewable Energy,2016,87:634.
[7] YIN S,WANG G,KARIMI H R. Data-driven design of robust faultdetection system for wind turbines[J]. Mechatronics,2014,24(4):298.
[8] SIMANI S,FARSONI S,CASTALDI P. Fault diagnosis of a windturbine benchmark via identified fuzzy models[J]. IEEE Transac-tions on Industrial Electronics,2015,62(6):3775.
[9] HINTON G E. Deep belief networks[J]. Scholarpedia,2009,4(6):5947.
[10] HINTON G E,OSINDERO S,TEH Y W. A fast learning algo-rithm for deep belief nets[J]. Neural Computation,2006,18(7):1527.
[11] BENGIO Y,LAMBLIN P,DAN P,et al. Greedy layer-wise train-ing of deep networks[C]//International Conference on NeuralInformation Processing Systems. MIT Press,2006:153.
[12] MOHAMED A,DAHL G,HINTON G. Deep belief networks forphone recognition[C]//Nips workshop on deep learning forspeech recognition and related applications. 2009,1(9):39.
[13]郭潇逍,李程,梅俏竹.深度学习在游戏中的应用[J].自动化学报,2016,(05):676.GUO Xiaoxiao,LI Cheng,MEI Qiaochu. Deep learning appliedto games[J]. Acta Automatica Sinica,2016,(05):676.
[14]王宪保,李洁,姚明海,等.基于深度学习的太阳能电池片表面缺陷检测方法[J].模式识别与人工智能,2014,27(6):517.WANG Xianbao,LI Jie,YAO Minghai,et al. Solar cells surfacedefects detection based on deep learning[J]. Pattern Recognitionand Artificial Intelligence,2014,27(6):517.
[15] LING Z H,DENG L,YU D. Modeling spectral envelopes usingrestricted boltzmann machines and deep belief networks for statis-tical parametric speech synthesis[J]. IEEE Transactions on Au-dio Speech&Language Processing,2013,21(10):2129.
[16] O'CONNOR P,NEIL D,LIU S C,et al. Real-time classificationand sensor fusion with a spiking deep belief network[J]. Fron-tiers in Neuroscience,2013,7:178.
[17] ZHANG X,ZHANG Q,ZHAO S,et al. Fault detection and isola-tion of the wind turbine benchmark:an estimation-based ap-proach[C]//Fault Detection and Isolation of the Wind TurbineBenchmark:An Estimation-Based Approach. 2011:8295.
[18] CHEN W,DING S X,HAGHANI A,et al. Observer-basedFDI schemes for wind turbine benchmark[J]. IFAC ProceedingsVolumes,2011,44(1):707.
[19] SVRD C,NYBERG M. Automated design of an FDI system forthe wind turbine benchmark[J]. Journal of Control Science andEngineering,2012,(2012-1-18),2012(1):8307.
[20] ODGAARD P F,STOUSTRUP J,KINNAERT M. Fault-tolerantcontrol of wind turbines:a benchmark model[J]. IEEE Transac-tions on Control Systems Technology,2013,42(4):1168.
[21] NAIR V,HINTON G E. Rectified linear units improve restrictedboltzmann machines[C]//International Conference on Interna-tional Conference on Machine Learning. Omnipress,2010:807.
[22] HINTON G E. Training products of experts by minimizing contras-tive divergence[J]. Neural Computation,2014,14(8):1771.
[2]马宏忠,张正东,时维俊,等.基于转子瞬时功率谱的双馈风力发电机定子绕组故障诊断[J].电力系统自动化,2014,38(14):30.MA Hongzhong,ZHANG Zhengdong,SHI Weijun,et al. Doubly-fed induction generator stator fault diagnosis based on rotor instan-taneous power spectrum[J]. Automation of Electric Power Sys-tems,2014,38(14):30.
[3]苗锐,陈国初,李月,等.基于随机集含糊证据的风力发电机故障诊断方法[J].电力系统自动化,2012,36(7):22.MIAO Rui,CHEN Guochu,LI Yue,et al. A wind turbine faultdiagnosis method based on vague evidence of random set[J]. Auto-mation of Electric Power Systems,2012,36(7):22.
[4] DEY S,PISU P,AYALEW B. A comparative study of three faultdiagnosis schemes for wind turbines[J]. IEEE Transactions onControl Systems Technology,2015,23(5):1853.
[5] OZDEMIR A A,SEILER P,BALAS G J. Wind turbine fault de-tection using counter-based residual thresholding[J]. IFAC Pro-ceedings Volumes,2011,44(1):8289.
[6] BESSA I V D,PALHARES R M,FILHO J E C. Data-driven faultdetection and isolation scheme for a wind turbine benchmark[J].Renewable Energy,2016,87:634.
[7] YIN S,WANG G,KARIMI H R. Data-driven design of robust faultdetection system for wind turbines[J]. Mechatronics,2014,24(4):298.
[8] SIMANI S,FARSONI S,CASTALDI P. Fault diagnosis of a windturbine benchmark via identified fuzzy models[J]. IEEE Transac-tions on Industrial Electronics,2015,62(6):3775.
[9] HINTON G E. Deep belief networks[J]. Scholarpedia,2009,4(6):5947.
[10] HINTON G E,OSINDERO S,TEH Y W. A fast learning algo-rithm for deep belief nets[J]. Neural Computation,2006,18(7):1527.
[11] BENGIO Y,LAMBLIN P,DAN P,et al. Greedy layer-wise train-ing of deep networks[C]//International Conference on NeuralInformation Processing Systems. MIT Press,2006:153.
[12] MOHAMED A,DAHL G,HINTON G. Deep belief networks forphone recognition[C]//Nips workshop on deep learning forspeech recognition and related applications. 2009,1(9):39.
[13]郭潇逍,李程,梅俏竹.深度学习在游戏中的应用[J].自动化学报,2016,(05):676.GUO Xiaoxiao,LI Cheng,MEI Qiaochu. Deep learning appliedto games[J]. Acta Automatica Sinica,2016,(05):676.
[14]王宪保,李洁,姚明海,等.基于深度学习的太阳能电池片表面缺陷检测方法[J].模式识别与人工智能,2014,27(6):517.WANG Xianbao,LI Jie,YAO Minghai,et al. Solar cells surfacedefects detection based on deep learning[J]. Pattern Recognitionand Artificial Intelligence,2014,27(6):517.
[15] LING Z H,DENG L,YU D. Modeling spectral envelopes usingrestricted boltzmann machines and deep belief networks for statis-tical parametric speech synthesis[J]. IEEE Transactions on Au-dio Speech&Language Processing,2013,21(10):2129.
[16] O'CONNOR P,NEIL D,LIU S C,et al. Real-time classificationand sensor fusion with a spiking deep belief network[J]. Fron-tiers in Neuroscience,2013,7:178.
[17] ZHANG X,ZHANG Q,ZHAO S,et al. Fault detection and isola-tion of the wind turbine benchmark:an estimation-based ap-proach[C]//Fault Detection and Isolation of the Wind TurbineBenchmark:An Estimation-Based Approach. 2011:8295.
[18] CHEN W,DING S X,HAGHANI A,et al. Observer-basedFDI schemes for wind turbine benchmark[J]. IFAC ProceedingsVolumes,2011,44(1):707.
[19] SVRD C,NYBERG M. Automated design of an FDI system forthe wind turbine benchmark[J]. Journal of Control Science andEngineering,2012,(2012-1-18),2012(1):8307.
[20] ODGAARD P F,STOUSTRUP J,KINNAERT M. Fault-tolerantcontrol of wind turbines:a benchmark model[J]. IEEE Transac-tions on Control Systems Technology,2013,42(4):1168.
[21] NAIR V,HINTON G E. Rectified linear units improve restrictedboltzmann machines[C]//International Conference on Interna-tional Conference on Machine Learning. Omnipress,2010:807.
[22] HINTON G E. Training products of experts by minimizing contras-tive divergence[J]. Neural Computation,2014,14(8):1771.