A self-learning energy management system for a smart-grid-ready residential building

Abstract

Based on research and scientific advances in sensor and network technologies, machine learning, and standard statistical methods, a development and a deployment of energy management systems could reduce the cost of electricity in residential buildings. This paper describes two implementations of an energy management system. The objective of the algorithm is to reduce the energy consumption of a residential building maintaining the thermal comfort. The first prototype used a rule-based control flow and reduced the baseline consumption by 25%, whereas the smart version of energy management system reached almost 50% minimisation of consumption by predicting future changes in the house temperature via a tree based machine learning models generated in R language. This Smart Controller with these predictions and energy cost calculations makes decision to either turn on or off the heating system of the house. To test and evaluate the system, both energy management systems run a virtual building simulation environment such as EnergyPlus through its interface controller BCVTB and RESTful API service that controls the building simulation software and stores obtained results to its database.