Redundant Anticipatory Control to Avoid Wind Turbine Failures Based on SCADA Data

摘要

Due to the characteristics of SCADA data, it is difficult, if not impossible, to use data-driven approach to predict wind turbine failures based on SCADA data. In this paper, we take a different tact. Resorting to unsupervised anomaly detection, we could detect suspicious turbine states, which might contain potential approaching failures. After that, a rule based decision maker could decide which redundant anticipatory operations, such as feathering, could eliminate the failures beforehand. Even if there is no failure actually, the redundant operations will not harm the equipment. To build such a system, we just use SCADA data, and do not need to modify the existing equipment. This paper presents the requirements and mechanism of the proposed system, as well as shows the feasibility using a case study about blade breakage.