In order to analyze the problem of abnormal parameters in the process of using the cloud server, the process of parameter data acquisition, data cleaning, and effective analysis of the cloud server is introduced. Aiming at the problems that the conditional FP-tree construction process is too redundant and the larger amount of data causes lower processing efficiency in the existing FP-Growth algorithm, an improved FP-Growth algorithm is proposed. It introduces the array tagging strategy, and each FP-tree node retains only pointers to the parent node. It does not need to generate a conditional FP-tree during the mining process, thus reducing time and space consumption. Experimental results show that the improved FP-Growth parallel algorithm can effectively improve the correlation analysis efficiency of abnormal data of cloud platform virtual machines, and is also suitable for data mining of large-scale data sets.