Power Quality Disturbance Classification Based on Compressed Sensing and Deep Convolution Neural Networks

By analyzing the recovery and reconstruction process of various power quality single disturbances and composite disturbance signals, we proposed a set of acquisition methods suitable for power quality disturbance (PQD) signals. The proposed acquisition method is applied to the compression sensing (CS) technology for data compression, the demand for the acquisition device memory is reduced, and the transmission rate is increased. An end-to-end intelligent classification framework is designed, which can directly classify the collected data without any time-consuming data pre-processing operations. The model is designed with noise adaptation module, which can cope with the error of compressed sensing recovery and has also showed good classification performance in noise data. Simultaneously, the model applies a lot of easy-to-implement techniques, which makes the trained model have better generalization ability and classification effect. The proposed method is verified by both simulation and measured data. The method showed superior performance compared to the existing disturbance identification methods based on the classification results.