Depth Load Identification Method of a Power Fingerprint Based on the Transferability of the V-I Trajectory

The traditional load identification method is not sufficiently accurate because of the high computational complexity and small load data of specific customer. A depth identification method for a power fingerprint based on the transferability of the V-I trajectory is proposed. First, the method uses the bilinear interpolation technique to solve the discontinuity of grayscale V-I trajectory images. The current, power, and phase numerical feature information are embedded in the grayscale V-I trajectory image by color encoding to generate a color V-I trajectory image that contains rich electrical features. Then, the color V-I trajectory images were used as input data for the ResNet34 neural network after model transfer. The residual structure in ResNet neural networks improves the ability of the model to extract load features without introducing additional parameters or increasing the complexity of the model. Finally, the ResNet34 neural network was pre-trained using the ImageNet dataset. The pre-trained ResNet34 neural network model was transferred to the new user appliance load identification by replacing the last fully connected layer with a new fully connected layer. Test results based on the PLAID dataset were used to verify the validity of the method