With the development of computer and Internet, it is easy to get related information. For the catalytic cracking unit of oil refining industry, although there is a flood of information recorded data, the changes in yield of the product caused by the changes of production of raw materials and run-time still need to rely on the experience and learn from a running instance, so Data Mining has a unique advantage in these areas. In this paper, based on increasing the yield of oil products, Data Mining technology is researched to improve operation efficiency in oil refining device applications. It abandoned the traditional way of thinking named "Assumption- Modeling- Solving- Forecasting-Verification" which based on Lumped Kinetics. It gains the most valuable experience directly from the operation of a large number of records in the past, and guides the new operating parameters to optimal operation using these experiences to increase oil refining plant run more efficiently. Each instance of a record will run automatically to be a new one in the next state of the data warehouse, so it does not require any theoretical assumptions and models to solve and will improve the prediction accuracy in the continuous learning cycle. In this paper, the example is used to prove the yield of a refinery with an annual processing capacity of 20 million tons increased from the original 71.28% to 78.13% with Data Mining methods. It not only enhances the operating efficiency of oil refining equipment, but also produces significant economic benefits. In addition, it leads the Data Mining from a commercial into the engineering field, and brings new development opportunities to traditional oil refining industry.