Abstract: The paper is to establish a model of the energy consumption of mobile phones. The energy consumption of mobile phones is divided into four different parts, namely, system energy consumption, application energy consumption, periodic jump and white noise. Based on the model established, the energy consumption estimation algorithm for applications is proposed to analyze the energy consumption of the functions included in the application, and to locate the code segment which consumes energy abnormally. The model for the energy consumption mainly consists of the energy consumption evaluation of the operation system, the removal of noise arisen caused by the periodic jump, and the evaluation of functional average energy consumption. The denoising model for the noise caused by the periodic jump is based on DBSCAN algorithm, and the hyper parameters of the denoising model are selected by scanning hyper parameters combined with the theory of machine learning. Based on the results of the model above and verification of varied mobile phones combined with trace files, the average error is only 5.58% when evaluating the energy consumption of mobile phones, which means the model has good generalization ability in the calculation of energy consumption of functions and code segments.

Key words: DBSCAN, machine learning, energy consumption of applications, energy consumption of functions, energy consumption of code segments