Occupancy prediction and its applications in smart homes

The intelligence built into the smart homes through the large deployment of sensing, computing and communication elements along with pattern recognition, prediction and adaptive control algorithms introduces new opportunities for better resource management, security, automation, and health monitoring solutions in future smart homes. In particular, activity and demand prediction of the smart home residents are key enablers of many effective smart home services such as energy management and home automation. In this thesis, we have adopted a sequential prediction approach for occupancy and movement prediction of residents in the smart home based on the Active LeZi algorithm. The Active LeZi algorithm is founded on an information theoretic approach and a data compression algorithm, which uses an order-k Markov model. We have specifically evaluated the effects of the order of the memory in the Markov model on the prediction accuracy. Our proposed algorithm extends the Active LeZi algorithm and improves its performance by limiting the memory of the Markov chain model, considering interrupted patterns and capturing the temporal information of the patterns in the model. Moreover, we have built a small-scale smart home testbed using motion detector sensors and a central microcontroller using Arduino technology to collect movement data streams using a small mobile robot. We have evaluated the performance of the proposed algorithm based on the simulated data and collected data from the testbed and have observed improvement in the prediction accuracy compared to the original Active LeZi algorithm. Finally, we have reviewed smart home applications that will benefit from such prediction models.