In order to meet the requirements of real-time online fault diagnosis of marine machinery and equipment, the support vector regression machine (SVRM) is studied in depth to solve the end effect problem of empirical mode decomposition (EMD). An EMD extremum fast continuation algorithm is proposed. The generation mechanism and influence of point effect are studied, and the advantages and limitations of typical end effect processing methods are analyzed. Then the basic principle of SVRM prediction is described, and the method of setting the extension length and sample number based on the signal extreme value scale is proposed. Finally, a SVRM-based extreme value prediction extension method is proposed by taking the signal extreme value point value and time value as samples. The simulation results show that the method can significantly improve the decomposition accuracy and operation efficiency of EMD, and provide support for the application of EMD technology in real-time monitoring and intelligent diagnosis of warship equipment.