Abstract: Aiming at the low accuracy of dissolved gas analysis (DGA) in diagnosing transformer faults, this paper proposes a transformer fault diagnosis based on an improved Dingo optimization algorithm (IDOA) optimized deep hybrid kernel extreme learning machine (DHKELM). Firstly, kernel principal component analysis (KPCA) is used to reduce the dimension of gas data and extract effective feature quantities. Secondly, the polynomial kernel function and Gaussian kernel function are weighted to construct a new hybrid kernel function, and an auto encoder is introduced. The extreme learning machine is improved, and the DHKELM model is established. The superiority of the proposed model is verified by comparing it with other machine learning models. Integrating the reverse learning, Cauchy variation, and differential evolution algorithms into the dingo optimization algorithm and testing the IDOA performance by using two typical test functions demonstrates that IDOA is more stable and optimal. The key parameters of DHKELM are optimized by IDOA, and the IDOA-DHKELM transformer fault diagnosis model is established. Finally, the feature quantity extracted by KPCA is used as the input set, the model is simulated and analyzed, and the algorithm model of DHKELM is optimized by comparing with other optimization algorithms. The results show that IDOA-DHKELM has higher transformer fault diagnosis accuracy compared to other models.