An Integrated Method Based on Unequal Spanning Segmentation Clustering for Anomaly Detection in Wind Power Short-Term Prediction

Aiming at the problem of model accuracy degradation due to data anomalies in wind power prediction, the paper proposes a three-stage anomaly handling framework that integrates unequal span segmentation clustering, anomalous segment detection and dynamic correction. By improving the Non-dominated Sorting Genetic Algorithm II (NSGA-II) and introducing the Bayesian optimized self-evaluating variational operator, the optimal segmentation point of the wind speed time series data is dynamically located, and a multimodal category-based anomaly detection mechanism is constructed by combining the Fuzzy C-Means clustering and adaptive boxplot diagram, and a bi-directional Dynamic Time Warping (DTW) matching strategy for the historical data is proposed for the correction of anomalous segments. Experiments based on the SCADA system of wind farms in Turkey show that the method reduces the normalized root mean square error (NRMSE) of power prediction from 0.0845 to 0.0496, and the normalized mean absolute error (NMAE) from 0.0618 to 0.0396. The proposed method, which constructs an anomalous segment processing paradigm, provides an effective technical path for data cleaning in wind power prediction.