Research On Fault Warning Method Of Wind Turbine Based On Real-time Monitoring Data Mining

Wind energy is one of the fastest-growing renewable energy sources in recent years.Wind turbines will gradually enter a period of high failures as the operation time increases and serious mechanical failures will consume a large amount of operation and maintenance resources.Wind turbine fault warning research could warn the potential faults of the wind turbine,avoid serious mechanical failures,and provide the necessary time for the operation and maintenance screw to arrange spare parts and maintenance conduction.Based on the SCADA data,three kinds of wind turbine fault warning strategy,which are "single unit failure warning strategy","group units failure warning strategy" and "combination unit information fusion failure warning strategy",were studied.Then three kinds of failure warning strategy based on "single unit failure threshold","group units health similarity" and"multi-time scale warning information fusion",were designed separately.The main research of this paper are:1.Aiming at the problem that the original SCADA data dimension is high and contains a lot of "high wind speed and low power" type data,a wind turbine fault warning strategy based on "single unit" fault threshold is proposed.Using locality preserving projections algorithm,the core features of SCADA data are extracted while dimension reduction is achieved.With the kernel extreme learning machine algorithm which can be quickly modeled,the wind turbine prediction model is established;and based on the distribution of the prediction residual,the failure warning threshold of the wind turbine is determined,and the early warning information of online monitoring could provide the key maintenance suggestion for the operation and maintenance screw.2.Aiming at the problem that the incipient fault signal is easily overwhelmed by the change of the sensor signal generated by the surrounding meteorological factors,a fault warning strategy based on the group multi-dimensional feature similarity is proposed.The wind turbines are divided into multiple similar group,according to the similarity of the operating states.The multi-dimensional similarity between similar wind turbines is used to analyze the change of similarity between two wind turbines.Due to device degradation,the SCADA data space distribution of target wind turbine would migrate,also the similar state of the two wind turbines would change.This method can locate the target wind turbine with potential fault.3.Aiming at the problem that the multivariate time series metric in the proposed multi-dimensional similarity fault warning strategy is blunt and the warning results are not intuitive,a dynamic time warping algorithm based on hesitant fuzzy sets is proposed.Compared with the original algorithm,the proposed algorithm runs faster and gets better accuracy.It is a generalized dynamic time warping algorithm.The original dynamic time warping algorithm is the reduced case of the proposed algorithm.The proposed group multi-dimensional similarity fault warning strategy is used to verify the effectiveness of the proposed algorithm,and the results are more readable than the original fault warning strategy.4.The missing whole items of data is common in the SCADA data,the key warning information may be missed.The multiple fault warning information of wind turbines is inconsistent,and the reliability of the early warning results is low.To tackle the problems,a fault warning strategy based on data reconstruction and multi-time scale fault warning information fusion is proposed.The whole wind farm SCADA data was used to reconstruct the missing data of the target wind turbine,and the wind turbine deterioration degree was established to analyze the various stages of the deterioration process for wind turbine.To enhance the credibility of the early warning results,the warning strategy based on DS theory of evidence with single-group warning information fusion was studied.The fault warning fusion strategy could fuse multi-time scale warning information into fault warning information of any time scale and multiple conflict evidence could be handled.At the end of the paper,the conclusions of the thesis are given,and the suggestions for further research are prospected.