Scholarlink

<p>Non-intrusive load monitoring (NILM) analyzes overall household energy consumption to estimate the usage of individual appliances. Numerous models have been proposed for this task; however, while neural networks effectively capture complex patterns, they remain computationally intensive. This study presents an optimized Feedforward Neural Network for NILM, incorporating oversampling and feature amplification to improve performance. The model is evaluated using seven metrics on the RAE, REFIT, and REDD datasets under unseen conditions, with additional REFIT experiments conducted both with and without noise. Compared to the benchmark approach, Denoising Auto-Encoder, our model improves the F1 score by <b>22.7%</b> and reduces the Normalized Error in Assigned Power (NEP) by <b>0</b>.<b>68</b>. On the REFIT dataset, it achieves an F1 score of <b>33%</b> and NEP of <b>1</b>.<b>87</b> under noise, which improves to <b>64%</b> and <b>0</b>.<b>97</b> after denoising. The code for this study is available on our GitHub repository at the following link: <a href="https://github.com/Axiid-7/NILM">https://github.com/Axiid-7/NILM</a>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Feed forward neural network for non-intrusive load monitoring

  • 期刊论文

  • 发表日期：

    2026年2月19日

查看全文
  • Muhammad Asad,
  • Manar Amayri,
  • Nizar Bouguila

摘要

Non-intrusive load monitoring (NILM) analyzes overall household energy consumption to estimate the usage of individual appliances. Numerous models have been proposed for this task; however, while neural networks effectively capture complex patterns, they remain computationally intensive. This study presents an optimized Feedforward Neural Network for NILM, incorporating oversampling and feature amplification to improve performance. The model is evaluated using seven metrics on the RAE, REFIT, and REDD datasets under unseen conditions, with additional REFIT experiments conducted both with and without noise. Compared to the benchmark approach, Denoising Auto-Encoder, our model improves the F1 score by 22.7% and reduces the Normalized Error in Assigned Power (NEP) by 0.68. On the REFIT dataset, it achieves an F1 score of 33% and NEP of 1.87 under noise, which improves to 64% and 0.97 after denoising. The code for this study is available on our GitHub repository at the following link: https://github.com/Axiid-7/NILM.