Identifikation und Priorisierung minderleistender Windenergieanlagen und abnormen Betriebsverhaltens

Continuously decreasing levelized cost of energy (LCOE) for wind power led the global wind industry to extensive growth in the past decades. Subsidies are rapidly declining or not necessary anymore, e.g., bids of 0 ct/kWh in offshore wind power tenders. In this cost-driven and highly competitive environment, operation and maintenance(O&M) costs account for about 25 % to 35 % of the LCOE. O&M strategies aim at high energy yields and low O&M costs at the same time. Predicting downtimes and detecting energy losses as well as degradation is, therefore, of increasing importance for operational managers. Since operational managers often monitor large numbers of wind turbines (WTs), they depend on a toolset providing them with highly condensed information to identify and prioritize low performing WTs or schedule preventive maintenance measures. Examples for frequently used tools are key performance indicators (KPIs) to monitor the long-term performance of WTs or condition monitoring systems (CMS) to monitor specific components. Many research activities focus on methods to detect abnormal operational behavior (anomalies) of WTs without a need for additional measurements/sensors. Power curves are a frequently used tool to assess the performance of WTs. The power curve health value (HV) used in this work is supposed to detect power curve anomalies since small deviations in the power curve are not easy to identify. It evaluates deviations in the linear region of power curves by performing a principal component analysis. To calculate the HV, the standard deviation in direction of the second principal component of a reference data set is compared to the standard deviation of a combined data set consisting of the reference data and data of the evaluated period. This work examines the applicability of this HV for different purposes as well as its sensitivities and provides a modified HV approach to make it more robust and suitable for various data sets. The modified HV was tested based on ENGIE's open data windfarm and data of on- and offshore WTs from the WInD-Pool. It proved to detect anomalies in the linear region of the power curve in a reliable and sensitive manner and was also eligible to detect long term power curve degradation. Also, about 7% of all corrective maintenance measures were preceded by high HVs with a median alarm horizon of three days. Overall, the HV proved to be a promising tool for various applications.