Sampling control in environmental monitoring systems using recurrent neural networks
Obiora Sam Ezeora, Jana Heckenbergerová, Petr Musı́lek, James Rodway · 2016
Minimization of energy consumption of environmental monitoring systems is important to ensure their extended operational lifetime and low maintenance costs. One possible way to conserve energy is the use of low-frequency analog-to-digital conversion devices and associated data sampling techniques. In this paper, a new approach to lowering sampling frequency using model-based data imputation is proposed and discussed. Recorded data is used to develop models based on time-delay recurrent neural network. While sampling at low frequencies, the models are used to predict future and missing values. The models are updated when predicted values differ significantly from the actual measurements. The proposed approach is demonstrated using actual measurements sampled at different frequencies. The results are thoroughly analyzed from the perspective of approximation error and energy savings.