||The ETI commissioned the HEMS & ICT Market project to undertake an in depth study and assessment of HEMS along with what data, processes and controls andpotential additional services enabled via a linked ICT system. The project delivers key insights and findings in terms of potential future offerings and capabilities of these products along with market assessment information. The aim of the project was to characterise the existing market for HEMS and ICT systems and to quantify the market/commercial opportunities for future HEMS and ICT propositions for both consumer and business.
The High Frequency Appliance Disaggregation Analysis (HFADA) project builds upon work undertaken in the Smart Systems and Heat (SSH) programme delivered by the Energy Systems Catapult for the ETI, to refine intelligence and gain detailed smart home energy data. The project analysed in depth datafrom five homes that trialed the SSH programme’s Home Energy Management System (HEMS) to identify which appliances are present within a building and when they are in operation. The main goal of the HFADA project was to detect human behaviour patterns in order to forecast the home energy needs of people in the future. In particular the project delivered a detailed set of data mining algorithms to help identify patterns of building occupancy and energy use within domestic homes from water, gas and electricity data.
In this deliverable Bournemouth Uni presents an original approach for mining utility data usage patterns relying on a novel Deep Hierarchical Dynamic model which consists of three modules, a Deep Belief network (DBN), a hierarchical mixture model which is based on Latent Dirichlet Allocation (LDA) and a Dynamic Bayesian Network based on Hidden Markov Model (HMM), called DBN-LDA-HMM. This architecture aims at extracting topics from data while taking into account the temporal structure of the data to model the inter-topic sequential dependency. While the mathematical details of the proposed algorithm are described elsewhere, a full empirical evaluation of this pattern mining algorithm using the ETI data is discussed, highlighting its performance on various mining tasks.