A large amount of data in field is necessary to effectively recognize the diagnosis of grounding grid

accompanied by heavy workload and influenced by interference. To further reduce the test workload in field and improve the accuracy of fault diagnosis of grounding grid

the combination of the virtual sample generation（VSG）technology and probabilistic neural network（PNN）are proposed in the paper. According to the measured results of magnetic field distribution of grounding grid in normal and typical fault conditions

the non-linear VSG technology is proposed to construct a large number of virtual samples by group discovery technology with hyper spherical characteristic equation. The probabilistic neural network（PNN）is trained to classify the measured results with the effective diagnosis of typical faults of grounding grid. The magnetic field distribution of ground grid of a 35 k V substation are measured in normal and typical fault conditions. The diagnosis results of typical faults have shown that the proposed nonlinear VSG technology is capable of holding the features of original samples with better adaptability and effectiveness. Moreover

the fault recognition rate of grounding grid based on PNN is clearly improved. The results could provide the important reference for the fault diagnosis of grounding grid.