Aiming at the problem that valve internal leakage data is highly hidden and traditional detection means are often difficult to be effective, a data-driven valve internal leakage fault diagnosis method is proposed. Data-driven technology is used to construct a valve actual operating flow characteristic curve model by polynomial fitting historical valve opening and relative flow data and quantify the degree of valve internal leakage by combining expert knowledge. A internal leakage diagnosis method based on MV/PV data is innovatively proposed, and related application software is developed. It is practically applied in a continuous reforming unit of a refinery, realizing real-time monitoring and graded alarm of valve internal leakage.