Dual-channel Non-intrusive Load Recognition Based on the Inception-BO-CNN

Aiming at the problem of non-intrusive load recognition, traditional recognition models find it difficult to effectively identify loads with similar voltage-current trajectories, and the training parameters are too many, which take up more computational resources, a dual-channel load recognition method based on Inception network combined with Bayesian optimized convolutional neural network (BO-CNN) is proposed. The method first fuses the voltage-current trajectory feature matrix with the binary converted numerical feature matrix to obtain the hybrid feature matrix and then extracts the hybrid feature matrix information simultaneously using the proposed dual-channel network model to achieve effective identification of different categories of electrical load. Simulation verification is carried out through the public dataset PLAID, which proves that the proposed method can effectively identify different types of loads; comparative analyses show that the proposed method can effectively improve the identification efficiency while saving the computation time, and it has a certain degree of anti-noise ability.