Abstract: Personalized and differentiated evaluation of transformers based on the technologies such as big data and data mining is an important means to improve the level of intelligent operation and maintenance of equipment. In the differentiated evaluation of transformers, the factors that affect the characteristics of dissolved gas in the oil rely on the subjective perception of the researcher and lack of objectivity, resulting in the inability to fully express the personalized and differentiated characteristics of the transformers and affecting the accuracy of the transformer evaluation. Therefore, a quantitative analysis method for the influence factors of the dissolved gas characteristics in power transformer oil was proposed in this paper. Considering the multiple dimensions and large volume of dissolved gas data, a fast clustering method based on CLARAS-Mahalanobis was proposed to mine the cluster centers under each influencing factor classification. Considering the centroid deviation of small samples data caused by abnormal and noisy data, the center point data set was selected based on the clustering center, and the average Hausdorff distance between the center point data sets was used as the quantitative indicator of the difference between transformers, thereby achieving quantification analysis. The actual case verification results show that the quantitative analysis method proposed in this paper can mine the influencing factors of dissolved gas characteristics, reflect the difference between transformers to the greatest extent, and realize accurate differential warning. Using the proposed method to obtain the best influencing factors, the accuracy of early warning of transformers can reach 98.4%.