Research On Energy Consumption Prediction Based On Android APP

Under the increasing mobile phone users,as the design of mobile phone hardware bottleneck,software energy consumption problem to be reckoned with,if the energy consumption for software,can carry on the energy consumption to the unknown APP application software hierarchy,can avoid the users unknowingly download a large amount of energy waste caused by high energy consumption software application situation.At present,a static and dynamic energy consumption detection method is proposed.In the aspect of dynamic detection,an intuitive data interaction and energy consumption measurement are provided.However,due to its professional detection equipment,there is a large learning cost for most mobile end users.There is a great deal of inconvenience in terms of mobility and portability.In the aspect of static detection,the energy consumption value and data transfer between the software layer and the hardware layer are modeled.This method has high reusability,simplicity,operability and rapidity.However,due to the large scale of energy consumption testing required for data information acquisition,the threshold threshold is defined,which leads to the current many static tests are unable to test and count the energy consumption of a large number of mobile applications.Training on the model leads to an increase in errors.Aiming at some problems in the above two methods,this paper proposes a new method for static testing,combined with the static test method using the deep learning model,through the training model,can learn more new features,mainly included in the method,conversion on the APK dex,XML files,generated by the grayscale of bytecode form,based on the deep learning algorithm to extract software grayscale characteristics of energy consumption.An automatic detection architecture MonkeyEnergyTest is used to automatically install the test data in sequence.At the same time,energy consumption data can be collected and then the integrated unit time and overall time energy consumption value can be output uniformly.As far as possible to avoid the possibility of artificial duplication of labor and statistical error birth.In subsequent studies,static energy consumption detection and dynamic energy consumption detection methods were compared.In the static method,by comparing different classical convolutional neural networks such as CNN,VGG and ResNet,theResNet model with deformable convolutional layer was fused,and the highest accuracy rate was 58.41%.In the dynamic comparison experiment,the Eroft plug-in with Dex decomcompiler was used.Compared with Eroft dynamic detection,the accuracy of the model is slightly higher than that of the dynamic detection method,which is 63%.The ResNet fusion deformable convolutional layer model is adopted in the model construction,which has a strong feature extraction effect for the byte code grayscale with difficult texture identification,thus providing a theoretical basis for accurate energy consumption prediction.At the same time,the method is simple,convenient and portable,which provides a new idea for the research of energy consumption prediction.