An intelligent impact load identification and localization method based on autonomic feature extraction and anomaly detection

•An intelligent impact load identification and localization method is proposed and verified on a thin-walled cylinder.•The method autonomically extracts key features from dynamic response and trains an ensemble learning model GBDT.•Abnormal impact samples are sifted statistically through iterative training and studied to further improve identification effect.•A retrained model after excluding abnormal interference performs semi-supervised anomaly detection on impact conditions.•The method is proved superior to other load identification methods.