A Novel Easy-to-construct Power Model for Embedded and Mobile Systems

Oussama Djedidi (1, 2, 3) , Mohand Djeziri (4) , Nacer M'Sirdi , Aziz Naamane (4)

1 SASV - Systèmes Automatisés à Structure Variable
2 LIS - Laboratoire d'Informatique et Systèmes
3 AMU - Aix Marseille Université
4 LSIS - Laboratoire des Sciences de l'Information et des Systèmes

This paper features a novel modeling scheme for power consumption in embedded and mobile devices. The model hereafter presented is built thought data fitting techniques using a NARX nonlinear neural net. It showcases the advantages of using a nonlinear model to estimate power consumption over the widely used linear regression models, where The NARX neural net is simpler, easier to implement, and more importantly more suitable as power changes are not always linear. Finally, experimental results validate the model with one with an accuracy of 97.1% on a smartphone.

Keywords: Data Fitting; Embedded Systems; Modeling; NARX; Neural Nets; Power Consumption; Smartphone

Domains: Engineering Sciences [physics]; Automatic; Cognitive science; Computer science; Physics [physics]; Physics [physics]; Data Analysis, Statistics and Probability [physics.data-an]

Oussama Djedidi, Mohand Djeziri, Nacer M'Sirdi, Aziz Naamane. A Novel Easy-to-construct Power Model for Embedded and Mobile Systems: Using Recursive Neural Nets to Estimate Power Consumption of ARM-based Embedded Systems and Mobile Devices. 15th International Conference on Informatics in Control, Automation and Robotics, Jul 2018, Porto, Portugal. ⟨10.5220/0006915805410545⟩. ⟨hal-01856579⟩