Weight-Optimized Adaptive Boosting Detection for Decentralized Wind Farms

To address the challenges of grid-connected fault detection in decentralized wind power systems, this paper proposes a fault detection method based on Adaptive Boosting (AdaBoost) with input weight optimization. First, considering that decentralized wind power integration introduces high-frequency harmonic noise which may obscure critical features, this paper combines the dynamic estimation advantages of Kalman filtering with the multi-band feature extraction capability of wavelet transform. Specifically, Kalman filtering is used to remove noise from data, and then wavelet transform decomposes the data into multiple frequency bands to extract features from different bands, enhancing the identifiability of fault characteristics. Subsequently, the processed data is fed into the input weight-optimized AdaBoost algorithm to avoid its focus on noise. By dynamically adjusting feature weights, the algorithm enhances the classifier's responsiveness to key features. Additionally, Bayesian probability is used to calculate real-time probabilities of various fault types, enabling precise identification of system faults. Finally, experiments verify the effectiveness and reliability of the proposed method.