Search

The use of IoT technology can help reduce the energy use of the built environment and increase occupant’s well-being, but growing privacy concerns may slow down its adoption. This project introduces an algorithm called Compact Auto-Encoder (CAE) that anonymises time-series data by taking into account data minimisation and self determinations principles outlined by the General Data Protection Regulation (GDPR), data utility loss and privacy gains. The CAE algorithm is composed of two neural networks, (1) an encoder responsible for creating a latent vector (z) and (2) a decoder that recovers relevant time-series information. During the prediction stage, edge devices host the encoder from the CAE algorithm, i.e. close to where data is being generated, whereas the data-recipient will host its corresponding decoder. Two training scenarios of the CAE algorithms are presented: unsupervised and supervised. The algorithm achieves 28% fewer utility losses in terms of Root Mean Squared Error (RMSE) and higher detection losses when transforming electricity data in the unsupervised scenario. In the second case, supervised anonymisation of CO2 data achieved 43% fewer utility losses and a 5% drop in its correlation with motion detection measurements, a sensitive measurement in the context of this work. The presented solution is considered an enabler of technologies that aim energy savings in the built environment by protecting user’s privacy when sharing the data that is needed to make more accurate analyses., Dissertation work done as a part of the Master of Science degree in Data Science, Technology and Innovation. Submitted to the School of Informatics, University of Edinburgh.

“Methodology for determination of virtual sensor and meters for flexibility” is a deliverable of the iBECOME H2020 project focused on showcasing the work that has been achieved towards identifying and calculating appropriate flexibility metrics, required to inform implicit and explicit demand response services.&nbsp