Abstract: Based on the multivariate state estimation, a new condition monitoring methods for the generator set is presented to reduce the maintenance costs of power generation equipment. A new and improved matrix in memory is given. It can achieve a good coverage of the normal operation status of the milling system. When there is a large difference between the estimated value and the actual measurement value, the existence of the fault is indicated. Sliding window residual statistical methods is used to analyze the residuals. When the residual trend is beyond the set threshold value, the system generates the warning. The real-time data of power plant is used to verify the method. The simulation results show that this method can achieve early failure warning effectively.