Abstract: In order to accurately identify the fault of the disconnector, determine the fault type, and ensure the stable operation of the power grid, an autonomous-cognition deep temporal clustering representation model(AC-DTCR) is proposed to diagnose the fault of the disconnector. In the case of a small amount of data, and the class label information is not available, time series clustering is a very good unsupervised learning technology, and the AC-DTCR model integrates time reconstruction and K-means targets. In order to improve the ability of the encoder, a false sample generation strategy and auxiliary classification task are proposed to improve the cluster structure and obtain a cluster-specific time representation. According to the motor current data obtained from the fault simulation experiment of high voltage disconnector, the AC-DTCR model is divided into four parts to train the experimental data. The results show that the model has good classification performance. Compared with the traditional classification model and time series clustering model, it has higher accuracy and can be applied to the field of power equipment fault diagnosis.