The impact of energy demand prediction on the automation of smart buildings management

Data mining and machine learning methods aimed at predicting the electrical energy demand of air conditioning systems are described with reference to a real-time control of these systems in smart buildings. The case study presented is applicable to higher-complexity systems, which may include several prosumer nodes, i.e. nodes that have to generate and implement strategies of energy supply, trade, consumption and storage. These nodes are connected to a smart grid and remotely supervised by a Distributed System Operator (DSO) using distributed control and monitoring systems. Accurate, continuously-recorded local weather data are used to predict energy demand in order to make decisions aimed at both reducing energy consumption and assuring pre-established comfort levels. Energy savings can be estimated by observing a buildings energy performance under the action of different meteorological agents. The proposed, real-time estimation architecture aims at exploiting a buildings intrinsic thermal inertia in order to anticipate energy demand whenever more convenient contextual conditions are met. The suggested on-line management system was validated through a laboratory experimental test site, and results are reported and discussed. © 2015, UK Simulation Society. All rights reserved.