Cooperative and Interactive Learning to estimate human behaviours for energy applications

A new supervised learning approach is proposed to classify different instances related to human behaviour in a office context. It is based on a combination of Interactive and Cooperative Learning relying only on timed data from a set of sensors and feedback from occupants to collect the training data. The privacy of occupants is granted by avoiding the use of cameras and it can be easily adapted to different nonresidential and residential contexts, many profiles of users and different types of data. A new method is introduced to correct inconsistent labels, which assesses the errors and requests new information to the same context (updates). The methodology was tested for both occupancy and activity recognition, settling that using non-numerical labels leads to a more subjective environment but still reliable. Further, two different interaction criteria were analysed: density and spread rate. It was confirmed that the optimal sensors/features selected depend not only on the inhabitants characteristics but also on the interaction methodology applied. This approach can be used to promote sustainable behavioural changes by sending suggestions to the occupants and to study the impact of those suggestions (flexibility). It can be applied to energy management systems improving energy efficiency and performing active demand-side management. (c) 2021 Elsevier B.V. All rights reserved.