The sampling mode of an intelligent substation in a reconstruction phase is complex, and it is difficult for the relay protection device to detect slight abnormalities in the sampling circuit, resulting in a serious lag in the exposure time of hidden dangers in the circuit. To address these issues, the sampling mode and secondary equipment configuration of an intelligent substation during the reconstruction period are analyzed, and a relay protection sampling circuit anomaly detection method based on homologous waveform data comparison is proposed. First, a language model and cosine similarity algorithm of bidirectional encoder representations from transformers (BERT) are used for channel matching of homologous waveform data. Then a resampling technique and Manhattan distance are used to achieve waveform sampling frequency unification and time domain alignment. Lastly, an improved algorithm based on dynamic time warping (DTW) is proposed, and an anomaly criterion of the sampling circuit is set in combination with the offset of the sampling point. Case studies demonstrate that this method can successfully match waveform data homologous channels, achieve waveform consistency alignment. Compared to the traditional DTW algorithm, the improved DTW algorithm exhibits higher sensitivity and accuracy in identifying abnormal states. The abnormal state of the relay protection sampling circuit can be reliably detected from the abnormal criterion, ensuring the safe and reliable operation of the intelligent substation.