Artificial intelligence methods based on Dissolved Gas Analysis (DGA) commonly face challenges related to input feature dimensions and parameters. In response to this challenge, a transformer fault diagnosis approach based on fuzzy rough set theory and Improved Particle Swarm Optimization-Support Vector Machine (IPSO-SVM) is proposed. Initially, an initial feature set is constructed by incorporating all gas dissolved ratios in transformer oil. Subsequently, attribute reduction of the feature set is carried out using fuzzy rough set theory. The standard Particle Swarm Optimization (PSO) algorithm has a limitation: it only adjusts the inertia weight, neglecting the learning factor. To overcome this, a new algorithm called IPSO is introduced. IPSO integrates asymmetrical learning factors and is designed specifically to optimize the parameters of the Support Vector Machine. Finally, the efficacy of the proposed method is demonstrated by utilizing DGA instance data and comparing it with the IEC Ratio Method and SVM optimized by other algorithms, indicating an enhancement in accuracy.