Sequence-to-point learning with neural networks for nonintrusive load monitoring (1612.09106v3)

Abstract: Energy disaggregation (a.k.a nonintrusive load monitoring, NILM), a single-channel blind source separation problem, aims to decompose the mains which records the whole house electricity consumption into appliance-wise readings. This problem is difficult because it is inherently unidentifiable. Recent approaches have shown that the identifiability problem could be reduced by introducing domain knowledge into the model. Deep neural networks have been shown to be a promising approach for these problems, but sliding windows are necessary to handle the long sequences which arise in signal processing problems, which raises issues about how to combine predictions from different sliding windows. In this paper, we propose sequence-to-point learning, where the input is a window of the mains and the output is a single point of the target appliance. We use convolutional neural networks to train the model. Interestingly, we systematically show that the convolutional neural networks can inherently learn the signatures of the target appliances, which are automatically added into the model to reduce the identifiability problem. We applied the proposed neural network approaches to real-world household energy data, and show that the methods achieve state-of-the-art performance, improving two standard error measures by 84% and 92%.

• The paper introduces a sequence-to-point method that leverages CNNs to predict the power midpoint of appliances, significantly reducing prediction complexity.
• It demonstrates a robust performance by achieving an 84% reduction in MAE and a 92% reduction in SAE on benchmark UK-DALE and REDD datasets.
• The study paves the way for enhanced energy management systems and suggests potential applications in other blind source separation problems.

Sequence-to-Point Learning with Neural Networks for Non-Intrusive Load Monitoring

This paper addresses the critical problem of energy disaggregation in the context of Non-Intrusive Load Monitoring (NILM), presenting a novel approach known as sequence-to-point (seq2point) learning that leverages convolutional neural networks (CNNs). The aim of energy disaggregation is to decompose whole-house electricity consumption into appliance-specific energy usage using a single-channel signal. Given the unidentifiable nature of this blind source separation problem, the seq2point method offers a significant advancement over traditional methods, including factorial hidden Markov models and prior sequence-to-sequence (seq2seq) learning approaches.

Core Methodology

The seq2point approach diverges from the seq2seq paradigm by predicting a single point—the midpoint—of the target appliance’s power usage rather than an entire sequence. This methodological pivot is designed to reduce the complexity of the prediction task, focusing the model's representational capacity on critical features. The output at the midpoint benefits from full access to local input information, improving the handling of long time-series data challenges, often hindered by sliding window techniques in seq2seq models.

Numerical Evidence and Analysis

The authors demonstrate that seq2point learning achieves an impressive reduction in two standard error metrics—mean absolute error (MAE) and signal aggregate error (SAE), improving them by 84% and 92%, respectively. This robust performance is validated on well-known NILM datasets, UK-DALE and REDD, showcasing the practical applicability and generalization capability of the seq2point method beyond training household setups.

An analytical component comparing seq2seq and seq2point reveals that the latter provides a tighter approximation to the target posterior distribution. This is substantiated by mathematical modeling, which indicates that seq2point minimizes KL-divergence more effectively, ensuring a more accurate posterior approximation.

Implications and Future Directions

From a theoretical perspective, the seq2point architecture signifies a meaningful step towards tackling the identifiability issue by enabling the network to automatically learn appliance-specific features, like on/off states and power levels. This capacity to autonomously extract meaningful features from raw data shifts the paradigm away from manually engineered features, increasing the model's adaptability and potential for diverse application domains.

Practically, the improved accuracy and robustness of the seq2point approach in predicting appliance energy usage holds promise for household energy management systems aiming to inform users for energy conservation. Additionally, as visualized within the paper, the learned feature maps indicate that feature abstraction aligns closely with human-engineered signatures, reinforcing model reliability.

Looking forward, the exploration of similar seq2point frameworks in domains like audio and speech signal processing in single-channel blind source separation problems appears promising. The technique's versatility in handling long sequence data via focused point predictions could address analogous challenges in these fields.

In conclusion, the seq2point learning framework enriches the methodological landscape of NILM by advancing both the performance and interpretability of neural network applications in energy disaggregation, with implications for energy efficiency improvements and broader BSS applications.