Recently

high-voltage direct current（HVDC） cable projects have been developing quickly. However

studies on online detection and pattern recognition of cable and accessories still remain in primary stage. On the basis of cross-linked polyethylene（XLPE） cable

we designed four kinds of partial discharge（PD） defect models

such as air gap inside XLPE

scratch on the surface of XLPE

creepage on the outer semi-conducting layer

and corona inside XLPE on high voltage side. The Sparse Representation Classification technique based on compressed sensing was applied to pattern recognition of PD under DC voltage. The discharge repetition rate plots were drawn as the classification samples. All the training samples composed of an over-complete dictionary and sparse representations of test samples projecting on this dictionary were used to classify based on 1-norm minimization. Three 1-norm minimization algorithms were adopted under different dimensions

including Homotopy algorithm

non-negative least squares（NNLS）

and orthogonal matching pursuit（OMP）. The results reveal that the recognition rates of the three methods are approximate under lower dimension（10×10 or 15×15）. The Homotopy algorithm outperforms the others in recognition rate with the increase of dimension

and achieves 92.31% as the maximum recognition rate under 20×20 dimension. NNLS takes second place while consuming more time. Comprehensive comparison shows that Homotopy algorithm has obvious advantages in recognition accuracy and arithmetic speed

and 20×20 dimension can meet requirements.