Neural Fourier Energy Disaggregation.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Energy Systems

Background:

• Deploying energy disaggregation models in real-world applications faces challenges due to the high computational cost and storage requirements of deep neural networks.
• Existing deep learning models are often too resource-intensive for edge devices or cost-prohibitive for server-based deployment.
• Optimizing neural network size and computational efficiency without compromising performance is a significant hurdle.

Purpose of the Study:

• To propose a novel neural network architecture for energy disaggregation that minimizes learning parameters, reduces model size, and accelerates inference time.
• To achieve comparable or superior performance to existing state-of-the-art energy disaggregation models.
• To enable efficient real-world deployment of energy disaggregation technology.

Main Methods:

• Development of a novel neural network architecture incorporating a parameter-free Fourier transformation.
• Comparative performance evaluation against two established baseline energy disaggregation models.
• Analysis of model size, computational cost, and inference speed.

Main Results:

• The proposed architecture demonstrates a reduction in learning parameters and model size.
• Fast inference times were achieved without a trade-off in performance.
• The novel model performed on par with strong baseline models in energy disaggregation tasks.

Conclusions:

• The developed neural network architecture offers an efficient solution for energy disaggregation.
• The integration of Fourier transformation is key to achieving reduced computational complexity and size.
• This approach facilitates the practical, widespread deployment of energy disaggregation models.