Řízení IoT nodu pomocí ANFIS

Abstract

The use of batteries in IoT nodes causes several problems, such as the need of replacing them, which increases the maintenance of the nodes, which is not desirable, especially for deviced located in poorly accessible places. In addition, batteries often contain heavy metals and reducing the usage can lead to the reduction of environmental burden on the planet. That is why, there is an effort to use energy harvesting methods allowing energy to be harvested at the location (making replacing the battery with a supercapacitor possible). This powering method is unstable and it is causing fluctuations of energy. More complex algorithms are often needed to effectively manage this energy fluctuations.

This bachelor thesis experiments with the use of ANFIS network to manage an energy independent IoT node. For bringing ANFIS its functionality it is required to have training data, which can be very challenging to ensure. The goal of this work is to determine the minimum size of training data for a two-input and three-input network so that the network can learn enough to manage the node. The structure of the training data leads to reducing only the number of rows because the number of columns is fixed. The results show that for a 2-input network needs to be provided at least 600 rows and for 3-input network it is 7000 rows.

The results of this work can lead to a reduction of the complex searching for training data. As a result, better training data can be received which will lead to better results of IoT node control, which could help in designing control algorithms for managing energy independent IoT nodes.