Research on Machine Learning and Statistical Analysis for Crowdsourced Battery Big Data of Mobile Devices

• Main Authors: Tsai Sheng-Wen, 蔡盛文
• Other Authors: Lee Chung-Hong
• Format: Others
• Language: zh-TW
• Published: 2016
• Online Access: http://ndltd.ncl.edu.tw/handle/y33h9d

碩士 === 國立高雄應用科技大學 === 電機工程系博碩士班 === 105 === Lithium-ion battery has advantages of high capacity, low internal resistance, and low self-discharge rate. In recent years, the technique of battery manufacturing only had a little breakthrough, because lithium-ion battery requires highly strict and complex manufacturing techniques. To this end, the manufacturers of batteries and mobile devices have to commit substantial resources to battery data analysis, thereby achieving the best performance of power consumption for the overall system. In this study, we collected the data of mobile devices by crowdsourcing, and established a model for battery analysis. We used multiple regression to analyze correlation between different variables and filtered the specified datasets as training data for experimenting with support vector machines classifiers. We analyzed the correlation among the state of charge, voltage, temperature, CPU usage, and charging mode. Further, we used the support vector regression to verify the results of multiple regression analysis. Thus, this study completed an impact analysis between power consumption and different factors, and successfully characterized the discharge time from the experiment results. Our results provide a reference which can help users understand power consumption of mobile devices in their context of use. The experiment results show the method is clearly feasible.