A Novel Non-Intrusive Load Decomposition Approach Based on TCN with RFAM Method

With the growing diversification and operational flexibility of load-side resources, improving electricity consumption efficiency through precise load management has become increasingly critical. While non-intrusive load decomposition (NILD) offers an effective solution for household energy monitoring, existing techniques face critical limitations, including inadequate spatial feature extraction and insufficient attention to complex operational patterns. This study presents an enhanced NILD framework that integrates a Receptive-Field Attention Mechanism (RFAM) within an improved Temporal Convolutional Network architecture to address these challenges. The proposed RFAM-TCN synergistically optimizes spatial feature perception through adaptive weight allocation across convolutional layers while enhancing multi-scale temporal pattern recognition. Validated on the AMPds dataset, the model demonstrates superior decomposition accuracy compared to conventional approaches, achieving a 6.47<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup> improvement in regression performance over BiLSTM-based solutions.