The partial discharge detection method is an important means to evaluate the operating status of electrical equipment. However

the research on the pattern recognition of partial discharge of direct current（DC） cross linked polyethylene（XLPE） cables is still in the early stage. According to the common insulation defects and partial discharge characteristics of DC XLPE cable

four kinds of DC cable defect models were designed and an experimental platform was set up to collect the partial discharge signals according to the common insulation defects and partial discharge characteristics of XLPE cables. Moreover

a DC cable pattern recognition method of time resolved partial discharge based on Adaptive Moment Estimation（ADAM）-optimized Deep Belief Networks（DBN） was studied. The recognition results of deep belief network model and the method based on time-frequency feature were compared by experiments. The applicability of recognition methods to various types of defects and the influence of training sample capacity on recognition models were analyzed. The results show that the recognition method based on the improved deep belief network can be adopted to deeply explore the effective characteristics of partial discharge pulses. The diagnosis accuracy of proposed method

which possesses a good recognition result for 4 types of defects

is higher than that of the support vector machine（SVM） and back propagation（BP） neural network classification method based on time-frequency features.The recognition accuracy rate is greatly improved with the increase of the training sample capacity.