Smart homes have significantly evolved and become more accessible to users with the
advancement of wireless technology and IoT (Internet of Things) technology in recent times.
Initially, home devices could only be controlled through a central interface, but later, they
could be managed via portable devices and an increasing number of interfaces throughout the
house. With the further development of technology, we have aimed to make smart homes
even more automated, not only in terms of comfort but also in energy efficiency and security.
To achieve this, we developed various algorithms for predicting event sequences, which are
one of the key factors in reaching this goal.
In this thesis, we reviewed some of the already known solutions in this area. We first focused
on simpler algorithms for predicting event sequences, such as SPEED and Active LeZi, which
are based on LZW compression algorithms and Markov chains. Then, we explored more
recent and complex methods that rely on deep learning. We presented the field of deep
learning and provided examples of the use of LSTM neural networks and the combination of
CNN-LSTM neural networks in a smart home.
In the second part of the thesis, we also tested a deep learning model on real data from a smart
home. This model was based on a GPT neural network. We found that the model has high
accuracy in predicting the next event; however, this accuracy did not change with the
modification of its parameters. Only the training time of the neural network increased. This
could be due to the simplicity of the data set, as devices in the home typically have a binary
nature, or due to the good performance of our network, as GPT models have proven to be very
effective in this field.
|
