Scholarlink

A Non-Intrusive Load Monitoring (NILM) is simple, low-cost and easy to implement as compared to Intrusive Load Monitoring (ILM). There has been a great interest in devising a universal NILM method for the purpose of energy disaggregation – identify the operation and its duration for any appliance from the aggregated electrical demand profile data. As a result, significant number of research articles are published in different conference proceedings and peer reviewed journals. In this research, a NILM has been developed to classify home appliances based on their power usage patterns employing the Reference Energy Disaggregation Dataset (REDD). In this regard, binary labelling is applied to represent clear status of various appliances and windowing technique is utilized to cleans data by removing outliers and power spikes. Additionally, data is balanced by employing various sample size limits, which reduce spurious transient data or non-consequential small data classes. Considering total power consumption as a feature, a simple feedforward neural network (FNN) with a hidden layer is trained for these purposes. The proposed methodology has been implemented on both REDD House 1 and House 3. Evaluation metrics such as accuracy, precision, recall, F1-score, demonstrate high performance using this method with varying levels of success. Further, model accuracy and loss plots illustrate the model’s performance across training epochs, confirming the effectiveness of the proposed NILM technique.

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

Feedforward Neural Network Assisted NILM Method Applied on Different Households and Appliance Groups with Selectable Balanced Data

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Leila Kamyabi,
  • Shafiqur Rahman Tito,
  • Snjezana Soltic,
  • Pieter Nieuwoudt,
  • Tek Tjing Lie

摘要

A Non-Intrusive Load Monitoring (NILM) is simple, low-cost and easy to implement as compared to Intrusive Load Monitoring (ILM). There has been a great interest in devising a universal NILM method for the purpose of energy disaggregation – identify the operation and its duration for any appliance from the aggregated electrical demand profile data. As a result, significant number of research articles are published in different conference proceedings and peer reviewed journals. In this research, a NILM has been developed to classify home appliances based on their power usage patterns employing the Reference Energy Disaggregation Dataset (REDD). In this regard, binary labelling is applied to represent clear status of various appliances and windowing technique is utilized to cleans data by removing outliers and power spikes. Additionally, data is balanced by employing various sample size limits, which reduce spurious transient data or non-consequential small data classes. Considering total power consumption as a feature, a simple feedforward neural network (FNN) with a hidden layer is trained for these purposes. The proposed methodology has been implemented on both REDD House 1 and House 3. Evaluation metrics such as accuracy, precision, recall, F1-score, demonstrate high performance using this method with varying levels of success. Further, model accuracy and loss plots illustrate the model’s performance across training epochs, confirming the effectiveness of the proposed NILM technique.