Aiming at the issue that transformer differential protection devices are easily making a fault operation by inrush current

an identification technique of inrush current based on EMD-SVD-KELM and parameter optimization is presented. Firstly

with the help of empirical mode decomposition（EMD） and singular value decomposition（SVD）

inrush current and fault current signals are preprocessed to extract feature quantities for identification. These feature quantities can be as learning inputs of the following extreme learning machine with kernel（KELM）. Then

parameter C and γ have a greater impact on learning machine performance

so average accuracy rate of multiple models resulting from average training samples can be fitness function to provide evaluation criteria for the parameter optimization of KELM. EMTDC is used to generate the training samples and testing samples. Simultaneously

KELM is trained and tested by various optimization algorithms. Finally experimental results show that comparing with grid search（GS）

genetic algorithm（GA） and particle swarm optimization（PSO）

improved particle swarm optimization（IPSO）is better to quickly search optimal parameters. Meanwhile the trained IPSO-KELM model can properly identify inrush current and fault current

which verify correctness and feasibility of the proposed method in this paper.