Addressing the issue of electrical fires caused by DC series arc fault in long-serving photovoltaic systems due to line aging and poor contact, this paper proposes a fault detection method that combines multi-feature fusion fault criteria with machine learning algorithms. Initially, an arc generator is designed and integrated into the photovoltaic system to produce DC arc fault waveforms, with arc data collected using sensors and oscilloscopes. Subsequently, arc features are extracted from various domains and feature selection is performed using the random forest algorithm to eliminate redundant features, forming feature criteria. Finally, arc fault detection is carried out using a support vector machine optimized by particle swarm optimization. Experimental results show that this method can accurately identify DC series arc fault in photovoltaic systems and has a certain degree of anti-interference capability.