Accurate prognosis of wind turbine generator failures is essential for reducing operation and maintenance costs in wind farms. Existing methods rely on expensive, purpose-built condition monitoring systems to conduct diagnosis and prognosis of wind turbine generator failures. In this paper, we present a prognosis method to predict the remaining useful life (RUL) of generators, which requires no additional hardware support beyond widely adopted SCADA system. This work first introduces a notion, Anomaly Operation Index (AOI), to quantitatively measure wind turbine performance degradation in runtime. It then presents a data-driven wind turbine anomaly detection method and a time series analysis method to predict the wind turbine generator RUL. Experimental study on real-world wind farm data demonstrates that the proposed methods can achieve accurate prediction of wind turbine generator RUL and provide sufficient lead time for scheduling maintenance and repair.