Abstract: The reliability of transformer overheating fault identification is the basis for guaranteeing the service life of the transformer and the safe operation of the power system.Therefore, a method for identifying high temperature and overheating faults of transformers in power distribution rooms based on infrared temperature measurement technology is proposed in this paper. The base layer obtains the transformer temperature data through an infrared thermometer,uses BP neural network to correct the temperature measurement results,generates a thermal image according to the principle of infrared radiation and transmits it to the middle layer;the middle layer receives the thermal image and stores it in the image database,and the image is compared with the image in the prior knowledge base to detect whether the image is abnormal;the abnormal detection result is transmitted to the service layer,which uses the AlexNet convolutional neural network to classify and identify the thermal fault of the transformer,and displays the fault location through the client display screen. The test results show that the temperature difference correction effect of this method is good. When the temperature standard deviation is the largest,the maximum deviation between the test value of the transformer and the actual value is 0.98°C,which is less than 1°C. which can minimize the difference between the temperature measurement result and the actual temperature result,accurately identify thermal faults such as abnormal heat dissipation,contact heating in the outlet bushing,and clearly show the location of the transformer fault. The fault identification accuracy is as high as 94.6%,much higher than the traditional method,and the system has certain application value.