Non-intrusive load monitoring for appliance status determination using feed-forward neural network

Energy efficiency regulations and initiatives have been implemented as part of proactive actions to address the energy crisis that has arisen due to the increasing demand and depletion of resources. A load monitoring system is used to provide real-time data for appropriate feedbacks towards electricity savings. It can also be used to evaluate the effectiveness of the implementation of an energy management scheme. However, monitoring all individual appliances by installing an energy meter for each appliance will incur high installation and maintenance costs. Therefore, this work aims to determine the status of individual appliances from an aggregated measurement using non-intrusive load monitoring (NILM) based on a feed-forward neural network. The establishment of a NILM model has for main processes, including, data acquisition, pre-processing, training and performance evaluation. In the pre-processing, a new approach using threshold is introduced to identify the status of appliances based on their power consumption readings. The performance of the proposed approach is then evaluated and compared with the traditional logistic regression technique in terms of accuracy. The results show that the NILM using a feed-forward neural network outperformed the traditional logistic regression by 5.78%. Moreover, the proposed approach with threshold helped to improve the accuracy further by 19.1% as compared to the same learning algorithm without considering the threshold. Consequently, the overall performance is improved by almost 25% as compared to the logistic regression as presented in the previous work. Hence, it clearly shows that the status of individual appliances can be determined from measurements at the main meter using NILM based on a feed-forward neural network with high accuracy.

Automated summary

This study aims to determine the status of individual appliances from aggregated measurements using non-intrusive load monitoring (NILM) based on a feed-forward neural network. A new approach using a threshold to identify the status of appliances based on their power consumption readings is introduced. The results show that the NILM using a feed-forward neural network outperformed the traditional logistic regression by 5.78% in terms of accuracy.