With the rapid development of renewable and energy power electronics

the DC system has been widely used in various fields.However

the DC arc fault caused by insulation damage and loose joint is the main factor threatening the normal operation of the DC system.In order to solve the problem that DC arc fault is difficult to detect due to the absence of zero crossing and shoulder phenomenon

the Gaussian process model is introduced into the arc fault research

and the series DC arc fault tests under different loads are carried out.Feature vectors are extracted from current signal including the mean of difference of current and the sum of harmonic energy at first.Then Gauss process classification is applied to arc fault training and prediction classification.Finally test results show that presented detection method can identify DC arc fault accurately under different load condition.