Estimating Software Energy Consumption with Machine Learning Approach by Software Performance Feature

With the growing scale of the application and ability to compute, more and more people pay attention to software energy consumption. There is a huge potential for controlling energy consumption during the application's development phase. Although the energy consumption of software can be obtained by tools or models, the problem of how energy consumption is consumed is still not explained. To solve this problem, a model of energy consumption characterized by performance events is established with using the method of ridge regression machine learning, which can explain the origin of energy consumption, and the error rate is only 6.8%. Our model is based on performance events from perf tool and is independent of the application scenario. Using this model, it does not require programmers to measure and train their own applications, it can also decrypt the causes of energy consumption.