Online Prediction of Battery Lifetime for Embedded and Mobile Devices

[Introduction] [People] [Documents]

Introduction

In this project, we use statistical method to make online prediction of battery lifetime for the embedded and mobile devices. Our approach is based on the following observation: two battery discharge curve has linear relationship under constant load (see the following graph).

Battery Discharge Curve Relationship

This linear property enables us to use one constant load discharge curve as the "base curve" and transform any other discharge curves into a piece linear forms.
The next graph shows a real discharge curve under some variable load and its transformed counterpart.
real load curve transformation
 Then using the linear regression, we can make prediction of remaining battery lifetime based on history curve. The following graph shows the method.
discharge curve transformation
When compared to APM, which is a power management module widely used in Linux system, our method is far more accurate in terms of battery lifetime prediction. For constant loads, we can achieve <5% error (<50% for APM). For variable loads, we can achieve around 20% error (around 60% for APM).
I'm thinking about writing a prediction module for iPAQ to replace the battery lifetime prediction in APM...

People

Developer: Ye Wen , wenye at cs.ucsb.edu
Advisor: Rich Wolski , rich at cs.ucsb.edu

Documents

  1. Ye Wen, Rich Wolski, and Chandra Krintz, Online Prediction of Battery Lifetime for Embedded and Mobile Devices. Special Issue on Embedded Systems: Springer-Verlag Heidelberg Lecture Notes in Computer Science, V3164/2004, Dec 2004
  2. Ye Wen, Rich Wolski, and Chandra Krintz, History-based, Online, Battery Lifetime Prediction for Embedded and Mobile Devices. Workshop on Power-Aware Computer Systems (PACS), April 2003