Abstract: Accurate diagnosis and localisation of the faults in the PV arrays help improve the reliability of the PV power generation systems. In order to solve the problems that the existing diagnostic methods rely too much on a large number of labeled samples and cannot take into account the fault type diagnosis, the fault location and the costs at the same time, and combining the multi-sensor method with a semi-supervised learning algorithm, a semi-supervised learning algorithm (LP-ET) integrating the Label Propagation (LP) with the Extra-Trees (ET) is built.In order to overcome the less engineering fault samples and the lack of the fault labels, a PV array fault simulation model is built to obtain the samples.The LP algorithm is introduced to achieve the full labelling of the samples in the original fault sample set based on a small number of the labelled fault samples containing the fault type and location information.Then, the ET model is applied to continuously build a large number of decision trees to form an extreme random tree.A majority voting mechanism is used (Bagging) to obtain the fault type and location results. Experimental results showthat theproposed LP-ET model realizes high precisiondiagnoses under the short circuit, the open circuit, the degradation and the shading faults in the case of a large proportion of unlabeledsample data sets.It takes both the single component and multi-component faults into consideration, effectively solving the problem of PV array fault diagnosis and location.