Predicting domestic energy consumption using inferencing at the edge

Abstract

This paper presents a novel system design approach which uses the concepts of Edge Computing and Machine Learning (ML) to predict the domestic power consumption. The device computes and displays the predicted power consumption values with its dollar equivalent amount at the user end. It also relays the expected power consumption values to the utility company on the backend. If the utility company has prior knowledge about the demand requirement, then the company can arrange and supply the deficit power onto the power grid. Thereby maintaining the supply and demand balance and hence preventing a potential blackout scenario. There is a strong correlation between the changing weather and the power consumption pattern. Using this idea as the basis we have modeled device to predict power consumption based on the changing weather and it uses machine learning. The device uses Global Positioning System (GPS) to get the real-time location data, Application Programming interface (API) to get the real-time location specific weather data and 16×2 matrix Liquid Crystal Display (LCD) to display the computed values. We have used San Diego State University's volta server to replicate the utility company server.