ZEMCH 2015 - International Conference Proceedings | Page 243

International Conference | Bari - Lecce, Italy Session IV MODELLING OCCUPANT ACTIVITY PATTERNS FOR ENERGY SAVING IN BUIDINGS USING MACHINELEARNING APPROACHES Jose Luis Gomez Ortega1 & Liangxiu Han1 1School Of Computing, Mathematics & Digital Technology, Manchester Metropolitan University, UK, [email protected] 2School Of Computing, Mathematics & Digital Technology, Manchester Metropolitan University, UK, [email protected] Abstract Since occupants play an important role in the energy consumption in buildings, it is crucial to develop accurate models of occupant behaviour patterns for the automation of building management systems. With the advancement in information and communication technology (ICT), machine-learning technologies have been used for accurately modelling and learning occupant behaviour patterns based on environment sensing. In this work, we have developed a pattern recognition model based on non-linear multiclass Support Vector Machine classification approach, to accurately detect occupant behaviour patterns by modelling human activities of daily living (ADL). To evaluate the proposed model, we have selected multiple public datasets collected by different teams from diverse ambient sensors (i.e. motion, contact or pressure sensors). Two different data pre-processing techniques (i.e. Timeslice approach (TA) and chunk data (CDA)) have been developed for facilitating the construction of the models. Furthermore, we have also compared our model with other machine learning techniques (i.e. Hidden-Markov Model (HMM) and k-Nearest Neighbours (KNN)). The experimental results have shown that the proposed SVM-based model outperforms the other methods in terms of accuracy in three different scenarios for Dataset1 using the TA method and for the second dataset evaluated, combinin g CDA data pre-processing and SVM approaches, the accuracy is higher than other methods such as HMM and KNN. Keywords machine learning, SVM, mathematical modelling, energy efficiency, sensors. 241