YOLO-ResNet Automatic Recognition Algorithm for Abnormal Thermal State of Transformer Insulation

In order to capture subtle changes in insulation thermal state in real-time and accurately, the YOLO ResNet automatic recognition algorithm for the abnormal insulation thermal state of transformers is proposed. Thermal state features are extracted based on temperature change rate, and a matrix is constructed to determine the boundary of abnormal areas. Besides, we denoise and restore abnormal thermal state features. A ResNet recognition model is built. When the value of the abnormality degree index is different from the preset threshold, the stability of the insulation thermal state of the transformer is judged, and the weight of the abnormality degree index is calculated and extracted for time series matching. The experimental results show that the designed algorithm can accurately identify three fault samples and capture extreme abnormal operating conditions with temperature peaks exceeding $24^{\circ} \mathrm{C}$.