Abstract: The technology of supervised machine learning brings a revolutionary breakthrough to photovoltaic（PV）fault diagnosis,but it requires a large amount of labeled data for model training. A method based on semi-supervised extreme learning machine（SSELM）algorithm for PV fault diagnosis is proposed in this study which uses only a small number of labeled data. First,the variation of the I-V curve of PV modules under different fault conditions was analyzed,and feature parameters were extracted. Then,a method of feature parameter normalization was proposed to realize the conversion and normalization of feature parameters under different working conditions. Finally,SSELM fault diagnosis method was described. The advantage of the proposed method is that only a small amount of labeled simulation data is requred and a large number of unlabeled measured data can be used to build a PV fault diagnosis model,which greatly reduces the labor and time costs of data collection,and has high recognition accuracy. Simulation and real experiment verify that the proposed fault diagnosis method can effectively identify short-circuit,shade and abnormal aging faults of PV modules.