Effective Feature Selection and Deep Learning-Based Classification for Non-Intrusive Load Monitoring

Abstract

The approach detects the load consumption of every device from the aggregated home energy consumption. Most current solutions utilize machine learning (ML) methods resulting in models with many parameters and a high computational load. This article develops an effective feature selection and deep learning-based classification for NILM, which enhances the overall classification accuracy for appliance classification. At first, the data is collected, and the pre-processing stage is performed. Then, the event detection process is performed using the mean absolute deviation-sliding window approach. Then, feature extraction is performed using the Huang–Hilbert transform and time-domain descriptors methods. Then, the most significant features are selected using the new efficient optimization algorithm named Wildebeest herd optimization. Finally, the classification of the individual household appliances is performed using the hybrid wavelet convolutional sparrow search algorithm. The proposed work is simulated in MATLAB using the iAWE dataset. Various evaluation metrics have been computed for the proposed work and are compared to the current methods. Simulation results proved that the proposed method provides better outcomes for appliance classification than other approaches. The proposed model improves the appliance classification accuracy by 3.36%, 3.47%, 8.92%, and 6.38%, respectively, compared to CNN, DNN, KNN, and SVM.

Keywords:

• deep learning
• optimization
• feature selection
• event detection
• appliance classification

Data Availability Statement

Data sharing not applicable to this article.

Conflict of Interest

Authors have no conflict of interest to declare.

Additional information

Funding

No funding is provided for the preparation of manuscript.

Notes on contributors

Mamoon Elahi Barbhuyan

Mamoon Elahi Barbhuyan received the B.E. in Electronics and Communication Engineering from VTU, Belgaum, India in 2005 and received M.Tech degree in Communication Engineering from the VIT University, Vellore, India in 2007. He currently is Research Scholar at ASTU, Guwahati, Assam. His current research interests include the NILM, ML, wireless communication, embedded systems.

Pradyut Kumar Goswami

Pradyut Kumar Goswami, a Post Doctorate from Tokyo Metropolitan University, Japan, Dr PK Goswami did his PhD from Imperial College, University of London, UK and M. Tech in Electrical Engineering from IIT Madras. Dr PK Goswami has been a recipient of several international academic fellowships including JSPS fellowship for post doctoral research in Tokyo Metropolitan University, Japan; Quality Improvement Programme Scholarship from the Government of India and State Overseas Scholarship for PhD at Imperial College, London from the Government of Assam. Formerly, Dr Goswami has been Principal, Assam Engineering College. Ex-Vice Chancellor ASTU, Guwahati. His research interests include electric power components and systems.