Battery drain analysis and prediction for wireless devices
Technologies for wireless device battery analysis and prediction are disclosed. A described technology includes collecting wireless device information including battery drain information relating to a battery of a wireless device; determining, based on the collected wireless device information, a plurality of prediction outcomes corresponding respectively to a plurality of different battery usage prediction models; determining accuracies respectively for the battery usage prediction models; generating weight values respectively for the battery usage prediction models based on the determined accuracies; and generating a battery prediction result for the wireless device based on the prediction outcomes and the generated weight values.
1. A method implemented by data processing apparatus, the method comprising:
collecting wireless device information including battery drain information relating to a battery of a wireless device;
determining, based on the collected wireless device information, a plurality of prediction outcomes corresponding respectively to a plurality of different battery usage prediction models;
determining accuracies respectively for the battery usage prediction models;
generating weight values respectively for the battery usage prediction models based on the determined accuracies; and
generating a battery prediction result for the wireless device based on the prediction outcomes and the generated weight values,
wherein the battery usage prediction models includes a static model that is configured to extrapolate a battery usage profile for a new application from data collected for one or more other applications that have been previously measured for battery usage, wherein the battery usage profile is based on an extrapolation of one or more similarities between the new application and the one or more other applications.
2. A method implemented by data processing apparatus, the method comprising:
collecting wireless device information including battery drain information relating to a battery of a wireless device;
determining, based on the collected wireless device information, a plurality of prediction outcomes corresponding respectively to a plurality of different battery usage prediction models;
determining accuracies respectively for the battery usage prediction models;
generating weight values respectively for the battery usage prediction models based on the determined accuracies; and
generating a battery prediction result for the wireless device based on the prediction outcomes and the generated weight values,
wherein collecting the wireless device information comprises extracting one or more attributes of a new application, and wherein determining the prediction outcomes comprises:
retrieving from a database, one or more battery usage characteristics for the one or more extracted attributes of the new application;
extrapolating a battery usage profile for the new application based on the one or more retrieved battery usage characteristics for the one or more extracted attributes; and
producing, within a static model of the battery usage prediction models, a prediction outcome based on the extrapolated battery usage profile.
3. The method of claim 2 , wherein determining the prediction outcomes comprises:
producing, within a dynamic model of the battery usage prediction models, a prediction outcome based on the battery drain information, wherein the dynamic model is responsive to changes in the battery drain information that are observed via two or more measurements of the battery.
4. The method of claim 3 , wherein generating the weight values comprises:
setting, for a first duration, a weight value for the static model higher than weight values for the one or more dynamic models in response to a detection of the new application executing on the wireless device; and
setting, for a second duration, the weight value for the static model based on a relative performance of the static model with respect to the one or more dynamic models, the second duration being subsequent to the first duration.
5. The method of claim 1 , comprising:
determining an estimated time to a depleted state of the battery based on the battery prediction result.
6. The method of claim 1 , comprising:
sending a battery measurement sample to a server, the battery measurement sample including one or more process identifiers of processes executing on the wireless device; and
receiving from the server battery prediction information corresponding to the one or more process identifiers, wherein determining the prediction outcomes comprises using the battery prediction information.
7. A system comprising:
a storage device configured to store wireless device information including battery drain information relating to a battery of a wireless device; and
a processor communicatively coupled with the storage device, configured to (i) determine, based on the wireless device information, a plurality of prediction outcomes corresponding respectively to a plurality of different battery usage prediction models, (ii) determine accuracies respectively for the battery usage prediction models, (iii) generate weight values respectively for the battery usage prediction models based on the accuracies, and (iv) generate a battery prediction result for the wireless device based on the prediction outcomes and the weight values,
wherein the battery usage prediction models includes a static model that is configured to extrapolate a battery usage profile for a new application from data collected for one or more other applications that have been previously measured for battery usage, wherein the battery usage profile is based on an extrapolation of one or more similarities between the new application and the one or more other applications.
8. A system comprising:
a storage device configured to store wireless device information including battery drain information relating to a battery of a wireless device; and
a processor communicatively coupled with the storage device, configured to (i) determine, based on the wireless device information, a plurality of prediction outcomes corresponding respectively to a plurality of different battery usage prediction models, (ii) determine accuracies respectively for the battery usage prediction models, (iii) generate weight values respectively for the battery usage prediction models based on the accuracies, and (iv) generate a battery prediction result for the wireless device based on the prediction outcomes and the weight values,
wherein the wireless device information comprises one or more attributes of a new application, and wherein the processor is configured to
retrieve from a database, one or more battery usage characteristics for the one or more extracted attributes of the new application,
extrapolate a battery usage profile for the new application based on the one or more retrieved battery usage characteristics for the one or more extracted attributes, and
produce, within a static model of the battery usage prediction models, a prediction outcome based on the extrapolated battery usage profile.
9. The system of claim 8 , wherein the processor is configured to produce, within a dynamic model of the battery usage prediction models, a prediction outcome based on the battery drain information, wherein the dynamic model is responsive to changes in the battery drain information that are observed via two or more measurements of the battery.
10. The system of claim 9 , wherein the processor is configured to
set, for a first duration, a weight value for the static model higher than weight values for the one or more dynamic models in response to a detection of the new application executing on the wireless device, and
set, for a second duration, the weight value for the static model based on a relative performance of the static model with respect to the one or more dynamic models, the second duration being subsequent to the first duration.
11. The system of claim 7 , wherein the processor is configured to determine an estimated time to a depleted state of the battery based on the battery prediction result.