A Data-Driven Approach to Fault Prognosis for Wind Turbines
Conventional machine learning algorithms suffer demerits such as sluggish training speed and low accuracy when dealing with wind turbine modeling.In view of this,the present work made an attempt on establishing the normal behavior model of wind turbine based on TPE-LightGBM algorithm,and using it as the basis of fault early warning scheme.First combined with the fan operation principle and XGBoost algorithm,the feature selection before modeling is completed,and the normal behavior models of wind turbine performance and gearbox are established by using the SCADA historical operation data after abnormal data processing.The deviation between the output of the normal behavior model and the actual value is taken as the early warning index,and the sliding window model is introduced to smooth the early warning index as the threshold index.Finally an experimental verification carried out by using the SCADA historical fault data indicates that the proposed early warning scheme can achieve alarm prior to SCADA system.
NETL
NSTL
万方数据
