Abstract: To address the issues of incomplete feature extraction, insufficient accuracy, and lack of timeliness in traditional wind farm fault warning methods—caused by model complexity limitations or improper parameter selection—a remote monitoring fault warning method for large-scale wind farm renewable energy centralized control wind turbine generator units was proposed. First, by deploying a high-precision sensor network and hybrid transmission architecture, real-time and complete collection and storage of environmental and equipment status data were achieved, providing a data foundation for warnings. A comparable feature set incorporating aggregated and transformed characteristics such as power consistency and wind energy utilization coefficient was constructed, enabling horizontal comparison of operational states among units. Based on Mahalanobis distance, a single-unit health baseline model was established to quantify deviations between unit status and health benchmarks, achieving early anomaly detection. A fault warning mechanism for turbine groups was designed based on dynamic thresholds and hierarchical responses, setting adaptive thresholds according to the statistical properties of Mahalanobis distance and initiating a hierarchical warning process based on fault severity. Experimental results demonstrate that this method significantly outperforms traditional approaches in both warning lead time and comprehensive warning scores (average F1-Score reaching 0.91), effectively enabling early fault warnings and reducing unplanned downtime risks.