Modeling of smartphones’ power using neural networks

Abstract

In the work presented in this paper, we use data collected from mobile users over several weeks to develop a neural network-based prediction model for the power consumed by a smartphone. Battery life is critical to the designers of smartphones, and being able to assess scenarios of power consumption, and hence energy usage is of great value. The models developed attempt to correlate power consumption to users’ behavior by using power-related data collected from smartphones with the help of specially designed logging tool or application. Experiences gained while developing the model regarding the selection of input parameters to the model, the identification of the most suitable NN (neural network) structure, and the training methodology applied are all described in this paper. To the best of our knowledge, this is the first attempt where NN is used as a vehicle to model smartphones’ power, and the results obtained demonstrate that NNs models can provide reasonably accurate estimates, and therefore, further investigation of their use in this modeling problem is justified.