With the increase in electrical energy demand, power systems are exposed to various faults, which lead to an increase in electrical and mechanical stresses in transformers and raise the probability of their failure. Accurate diagnosis of the type of fault in the transformer is crucial for maintaining the safety of power systems. Since transformer faults are complex and hidden, simple and crude methods have difficulty in correctly diagnosing the fault. In this work, a new method of transformer fault detection using a combined deep learning system and genetic algorithm has been proposed. In this modeling, an example of the most used transformers in Ilam city (315 kVA) is considered. Also, with the help of preliminary studies and investigations, a list of faults related to this type of transformer has been prepared. Simulation and coding of the most suitable algorithm for modeling (deep neural network) have been done in the MATLAB R2022a environment. In this modeling, the weights of the neural network are optimized using a genetic algorithm to minimize error and maximize accuracy compared to similar algorithms.