Detection of abnormal temperature increases for dry-type transformers based on working condition recognition and WKN-Mixer

In the realm of maintaining secure and stable distribution networks, the detection of abnormal temperature rises in dry-type transformers plays a pivotal role. Conventional approaches relying on thermal-model based anomaly detection, however, struggle to accommodate the intricate and fluctuating working conditions prevalent in these transformers. To address this issue, a comprehensive framework for detecting abnormal temperature rises in dry-type transformers is proposed, which is based on the accurate recognition of working conditions. Firstly, the Soft-DTW approach is employed to conduct curve clustering on the working condition parameters of dry-type transformers. Through gauging the similarity of time-series data, it effectively captures the working condition features embedded in the intricate monitoring data, thereby enhancing the precision of working condition recognition. Subsequently, a WaveletKernelNet-Mixer (WKN-Mixer) is devised to precisely forecast the three-phase winding temperatures. WaveletKernelNet is seamlessly integrated into the input layer of the Mixer, enabling the network focus on features in the time-frequency domain. Finally, case studies verified that the proposed method reduced MAE by 31.9% after recognizing working conditions. Under normal operation, the WKN-Mixer achieved the lowest FPR and an average F1-score 7% higher than its peers at a 3.0% anomalous level.