Abstract: This paper proposes an edge computing-based abnormal temperature detection method for transformer iron cores. It constructs a fractional-order thermal circuit model to describe iron core temperature changes, and designs an abnormal detection process integrating multi-dimensional features and dynamic thresholds. An experimental platform is built, and temperature rise tests and fault repetition experiments are conducted. Results show the method has a temperature abnormal detection accuracy of 98.5%, a false alarm rate of 2.3%, and fast response.