Abstract: Aiming at distinguishing false data and abnormal state points in power plant parameters, a cleaning method based on association rule, density-based spatial clustering of applications with noise（DBSCAN） algorithm and improved Gauss kernel relevance vector machine（RVM） is proposed. Firstly, association rules are introduced to analyze the association among parameters and find out the combination of parameters with strong association. Secondly, the DBSCAN algorithm is used to detect the abnormal point preliminarily, and the cleaning procedure combined with the associated parameters is proposed to distinguish the false data and the system abnormal state points. Finally, RVM is used to clean the false data, and the time cost is reduced by improving the Gaussian kernel space sample point form. Test results show that the cleaning method based on parameter correlation can effectively improve the accuracy and timeliness of cleaning.