Substation Equipment Thermal Fault Diagnosis Model Based on ResNet and Improved Bayesian Optimization

At present, infrared detection technology is widely used to diagnose the thermal fault of substation equipment, and it is needed to use computers to intelligently handle the progressively increasing infrared images. A model based on improved Bayesian optimization to diagnose thermal fault is proposed in this paper. Firstly, in order to utilize the effective network structure of image classification, ResNet10 and ResNet14 with much smaller parameter size are created based on CNN and residual block. Secondly, Bayesian optimization based on Gauss process and expected improvement is adopted to select hyperparameter, and the constrains of validation accuracy and parameter size are set to improve the performance of the algorithm. Experiment result shows that the improved Bayesian optimization can converge faster and obtain a smaller converge value. Compared with Alexnet, VGG-16, ResNet18, the proposed model has a highest accuracy and smallest parameter size, which can diagnose the thermal fault more effectively.