New parallel hybrid PHCNN-GRU deep learning model for multi-output NILM disaggregation

The Non-Intrusive Load Monitoring (NILM) technique has emerged as an efficient technique for conserving power and enhancing energy efficiency in residential buildings. This paper introduces a NILM disaggregation framework based on the multi-target regression approach, which is particularly suitable for real-time energy disaggregation. For this purpose, this work proposes a new Parallel Hybrid CNN-GRU (PHCNN-GRU) deep learning model for NILM disaggregation tasks. This technique takes advantage of the ability of Convolutional Neural Networks (CNN) to efficiently process spatial data and the excellent capability of Gated Recurrent Units (GRU) to process complex time-series data, due to their ability to retain memory of prior inputs. The proposed model has been tested and evaluated using two low-frequency benchmark databases: the UK-DALE database and the AMPds database. The experimental results demonstrate the effectiveness of the proposed model for energy disaggregation. Specifically, when using the UK-DALE database, the proposed model achieves an overall F1-score of 86.86% and an estimation accuracy of 87.16%. Moreover, when utilizing the AMPds database, the proposed model achieves an overall F1-score of 94.21% and an estimation accuracy of 94.13%. Furthermore, to better assess the performance of the proposed model, a noise signal was added to the input data. The obtained results indicate the effectiveness and robustness of the proposed model, even in the presence of noise.