The conventional power system is undergoing a gradual transition into a novel power system, with new energy as the primary constituent. This transformation is characterised by a distinct ‘double high’ phenomenon, signifying a high proportion of renewable energy and advanced power electronic equipment. Consequently, the system power factor exhibits significant fluctuations, often reaching as low as 0.1 under high-speed rail, charging pile in light load or working conditions. However, the existing power meters are unable to accurately measure power factors between 0.01 and 0.25, leading to metrological inaccuracy in this range. Consequently, there is a significant need to conduct research on the classification of various typical scenarios under low power factor, and to master the classification and load characteristics of these scenarios. Given the complexity of adaptive feature extraction in load scene classification and the stringent requirements for high stability and accuracy of the classification results, this paper proposes a load scene classification method based on the combination of convolutional neural network (CNN) and KAN with CNN-KAN network. The method first exploits the automatic feature extraction capability of CNN to extract rich spatial features from load data to enhance the stability and training efficiency of the model. Subsequently, it exploits the advantages of KAN in key information capture or task-specific processing to further mine the key features in the data. Compared with traditional methods, this CNN-KAN network model has higher accuracy and better stability in load scene classification tasks.