Semi-Federated Learning Based on Autoencoder for Non-Intrusive Load Monitoring

Abstract

To addresses the challenges of data scarcity and weak computational capabilities of edge devices in the practical application of non-intrusive load monitoring (NILM) of power systems, we propose a semi-federated learning approach using an autoencoder network based on Mel-scale frequency cepstral coefficients (MFCC), referred to as semi-FedNILM. We introduce a MFCC-based autoencoder network to ensure the completeness of hidden layer feature information and incorporate a semi-federated learning framework into the network structure, such that only the training process of the autoencoder is performed on the clients. The encoded and encrypted data is then sent to the cloud server for training the global NILM model, significantly reducing the computational burden on edge clients while preserving user privacy. Experimental results demonstrate that the proposed Semi-FedNILM algorithm achieves better performance on homogenous and heterogeneous datasets of a practical task.