| Wind power generation is a clean,pollution-free,and resource-rich renewable energy generation technology that has developed rapidly worldwide with the depletion of traditional fossil energy in recent years.However,in order to obtain abundant wind energy resources,the actual wind farms are usually located in high altitudes on land or maritime areas,which makes the working environment of wind turbines extremely harsh.In particular,wind turbine blades are prone to icing fault when they are exposed to the external environment of low temperature and high humidity for a long time.The blade icing fault will interfere with the normal operation of the wind turbine,thereby reducing its power generation efficiency.In addition,the continuous accumulation of blade ice can even further lead to serious consequences of blade fracture and the subsequent collapse of wind turbines.Therefore,it is particularly important to detect and actively eliminate icing on wind turbine blades in a timely manner.At present,supervisory control and data acquisition(SCADA)systems have been generally installed in wind power station,and massive data reflecting the operation information of wind turbines are recorded and saved in real time.How to use the data-driven method to mine fault information from massive monitoring data and improve existing machine learning algorithms to adapt to different data scenarios has become a research hotspot in the current field of condition monitoring of new energy power equipment.In this paper,a series of improved extreme learning machine algorithms for wind turbine blade icing fault detection are proposed to solve the two practical problems that need to be solved urgently: data imbalance and the lack of label samples in wind turbine monitoring data.And a wind turbine fault detection system based on cloud-edge collaborative computation is designed.The following main research results have been achieved:(1)A novel adaptive weighted kernel extreme learning machine(AWKELM)algorithm is proposed for the data imbalance problem in wind turbine monitoring data,and it is applied to wind turbine blade icing fault detection.The support vector data description(SVDD)method is used to construct an adaptive weighting strategy to improve the existing weighted kernel extreme learning machine(WKELM)algorithm.Through a benchmark numerical simulation case and real-world monitoring data of two wind turbines for case analysis,it is demonstrated that the newly proposed AWKELM algorithm has superior unbalanced data processing capabilities.Compared with the conventional WKELM algorithm and some existing advanced data-driven wind turbine blade icing fault detection models,it has higher fault detection accuracy.(2)A novel ellipsoidal semi-supervised extreme learning machine(ESS-ELM)algorithm is proposed for the lack of label samples problem in wind turbine monitoring data,and it is applied to wind turbine blade icing fault detection.Fisher criterion is used to construct an ellipsoidal nearest neighbor graph(ENNG)calculation strategy to improve existing semisupervised extreme learning machine(SS-ELM)algorithm.The performance is verified by the benchmark industrial fault detection platform and real-world monitoring data of two wind turbines,it is demonstrated that the newly proposed ESS-ELM algorithm can effectively overcome the lack of labeled samples problem.Compared with the conventional SS-ELM algorithm,it has better detection performance,and it can achieve high-precision detection of icing faults with fewer label samples.(3)The existing wind turbine fault detection system is usually built using a cloud platform,which has obvious drawbacks of slow fault response speed and low online monitoring accuracy.In order to solve these problems,a wind turbine fault detection system based on cloud-edge collaborative computing is proposed,and the human-computer interaction interface of the system is built using MATLAB/GUI.An adaptive extreme learning machine-autoencoder(AELM-AE)algorithm with model adaptive online update capability is designed to realize the effective collaboration between cloud platform and edge computing module,and verify its performance on a benchmark industrial fault detection platform.Finally,the system function verification is carried out through the real-world monitoring data of wind turbine.The results show that the proposed system can effectively realize the visual analysis of wind turbine monitoring data and the online monitoring of operating status,and has the significant advantages of fast fault response speed and high online monitoring accuracy. |
