Systems and Methods for Battery Health Prediction with On-Device Learning

An electronic device comprising:

• one or more displays;
• one or more non-transitory computer-readable storage media including instructions;
• and
• one or more processors coupled to the storage media, the one or more processors configured to execute the instructions to:
• access, from a remote server, a general battery-heath model corresponding to the mobile electronic device, wherein the general battery-health model is a first deep neural network trained to calculate a state-of-health value for a battery of the mobile electronic device based on historic battery output capacity, historic battery output voltage, and historic battery output current associated with the mobile electronic device;
• access real-time battery charging data from the mobile electronic device for a plurality of battery charging modes, wherein the real-time battery charging data comprises battery input voltage and battery input current during each of the plurality of battery charging modes;
• identify a plurality of data subgroups of battery charging data based on the real-time battery charging data, wherein each data subgroup is associated with one of the battery charging modes;
• train, using the general battery-health model as a starting model, a first customized battery-health model in a cascaded workflow using each of the plurality of data subgroups of battery charging data for the plurality of battery charging modes, wherein the first customized battery-health model is a second deep neural network trained to calculate state-of-health values associated with the mobile electronic device for the plurality of battery charging modes, respectively;
• and
• determine, using the first customized battery-health model, a plurality of first state-of-health values associated with the mobile electronic device for the plurality of battery charging modes, respectively, based on real-time battery output capacity, real-time battery output voltage, and real-time battery output current.

A computer-readable non-transitory storage media comprising instructions executable by a processor to:

• access, from a remote server, a general battery-heath model corresponding to a mobile electronic device, wherein the general battery-health model is a first deep neural network trained to calculate a state-of-health value for a battery of the mobile electronic device based on historic battery output capacity, historic battery output voltage, and historic battery output current associated with the mobile electronic device;
• access real-time battery charging data from the mobile electronic device for a plurality of battery charging modes, wherein the real-time battery charging data comprises battery input voltage and battery input current during each of the plurality of battery charging modes;
• identify a plurality of data subgroups of battery charging data based on the real-time battery charging data, wherein each data subgroup is associated with one of the battery charging modes;
• train, using the general battery-health model as a starting model, a first customized battery-health model in a cascaded workflow using each of the plurality of data subgroups of battery charging data for the plurality of battery charging modes, wherein the first customized battery-health model is a second deep neural network trained to calculate state-of-health values associated with the mobile electronic device for the plurality of battery charging modes, respectively;
• and
• determine, using the first customized battery-health model, a plurality of first state-of-health values associated with the mobile electronic device for the plurality of battery charging modes, respectively, based on real-time battery output capacity, real-time battery output voltage, and real-time battery output current.