Abnormal State Identification Method for Wind Turbines Based on Improved DBSCAN and Random Forests

As a crucial component of wind power generation systems, wind turbines must operate safely to prevent sudden failures. This requires effective identification of abnormal operational states. In this study, we propose a novel approach for categorizing abnormal states based on operational data from wind turbines, leveraging an improved density-based spatial clustering of applications with noise (DBSCAN) algorithm in conjunction with random forests. We establish a wind turbine performance model as a benchmark, employing a DBSCAN–bidirectional long short-term memory network (DBSCAN-BiLSTM) framework for anomaly detection. The random forest algorithm is then applied to accurately identify abnormal data points. Furthermore, we implement real-time adjustments to operational state thresholds based on identified anomalies, facilitating precise classification of abnormal operational data. A case study is presented, detailing steps including abnormal data cleaning, performance model construction, and abnormal state classification to validate our approach. Our results demonstrate that the DBSCAN-BiLSTM method significantly reduces the root mean square error (RMSE) by 41.5% and 49.7% compared to traditional LSTM and convolutional neural network (CNN) algorithms, respectively. This research provides an effective solution for identifying abnormal states in supervisory control and data acquisition (SCADA) data of wind turbines, which is vital for fault detection and maintenance in the wind power industry.