Recurrent Neural Networks with Long Short-Term Memory Application to Forecast Internet of Things Gateway Energy Consumption

The paper treats an issue of the neural networks application for solving various energy supply difficulties having become more urgent with the years to come. The neural networks of various kinds can be employed to analyze and forecast energy consumption. The recurrent neural networks with long short-term memory, an efficient neural networks variation, are appropriate for the energy supply analysis in conditions of the energy consumption parameters scatter. The feature of the neural networks discussed is the feedback availability. The paper considers the recurrent neural networks with long short-term memory application to forecast the internet of things gateway energy consumption, the telephone with limited power resources being used for the purpose. The sensors of distance, temperature and humidity and air pressure have been used for the sensor nodes. The input parameters for the recurrent neural network learning are the values of the output overall energy consumption and the traffic transferred at various time poll of the sensors. The history of the recurrent neural network learning measures 1000 epochs. The range of the neural networks operation error after learning was measured within 8,53%. The figure is considered to be a good result for the mean energy consumption cost per a sensor to be 253 mA.