Non-Intrusive Load Monitoring Using Denoising Autoencoder Neural Networks

One of the most critical parts of energy management is energy monitoring. As a result, it is necessary to monitor a facility's power consumption before implementing technical measures to reduce it. Without effective system monitoring, the functioning state of appliances cannot be determined with precision. By employing energy-saving techniques including using fewer energy-consuming devices, timing appliance usage properly, and minimizing energy-wasting activities, load monitoring primarily serves to aid in energy conservation. In this paper, we present a Non-Intrusive Load monitoring system using autoencoders to estimate the appliance wise energy consumption of a household and a simple plug and play type smart measuring device with a web application and mobile application to visualize the equipment wise energy consumption. The Autoencoder model was trained for energy disaggregation using 75% of refrigerator dataset from house 1 and tested using 25% of the refrigerator dataset from house 1 and unseen data from house 2 of refrigerator. It can be clearly observed that energy can be disaggregated from the metrics accurately using the proposed Autoencoder model for known data from house 1 and fairly enough for unseen data from house 2. We have obtained a dataset for a local household selecting refrigerator, microwave, and air conditioner as target devices. Then we have used 40% of local dataset for training separate autoencoder model for each targe device and, 20% for validation and 40% of dataset for testing the developed autoencoder models. We evaluated the performance of autoencoder models for local dataset and observed promising results.