A new method which is multiscale local-maximum sample entropy(MLSE)is put forward to measure the complexity of time series in this paper.It can suppress the noise and interference of vibration signals and improve the precision of sample entropy on each time scale compared with multi-scale entropy.The MLSE of hydraulic pump under different status are treated as feature vectors.Artificial bee colony(ABC)algorithm is selected to optimize the structural parameters of support vector machine(SVM).The optimized SVM is applied to fault mode identification of hydraulic pump.The fault identification rate is higher compared with SVM and optimized SVM,and it is verified by the test data of hydraulic pump.