Competition in the crude oil refining industry, an industry that emits carbon, is intensifying due to the global trend of carbon reduction. Automating refining process control is critical to maintaining a competitive edge. Among the control automation technologies, the predictive model control method is the most effective, but it has a disadvantage in that it is difficult to identify the control target system. In this paper, the performance improvement method of the machine learning technique for system identification is presented and the performance improvement is verified by applying it to a commercial refinery.