Neural-network state-of-charge estimation

1. A method for monitoring battery operation, the method comprising:

receiving by a recurrent neural network (RNN), encoding characteristic behavior of a battery at multiple different ambient temperatures, a time series of values for one or more battery attributes for the battery, the one or more battery attributes comprising at least a current ambient temperature, wherein the RNN is trained based on backward-pass through the RNN of gradients of a loss function, representing the disparity between an actual state-of-charge value for the battery and a predicted value, computed for training set values corresponding to at least some of the multiple different ambient temperatures provided to the RNN during a respective forward-pass corresponding to the backward pass; and

determining by the RNN encoding the characteristic behavior of the battery at the multiple different ambient temperatures, based on the received time series comprising the current ambient temperature, a state estimate for the battery.

2. The method of claim 1 wherein the battery comprises a rechargeable battery, and wherein receiving the time series of values of the attributes includes receiving a time series spanning multiple charging and discharging cycles.

3. The method of claim 1 , wherein receiving time series of the values the one or more battery attributes comprises receiving values of one or more of a battery voltage, a battery current, and a battery temperature.

4. The method of claim 1 wherein the battery comprises a Lithium-ion (Li-ion) battery.

5. The method of claim 1 wherein the state estimate for the battery comprises a time series of values of state estimates corresponding to the time series of the values of the one or more battery attributes.

6. The method of claim 1 wherein the state estimate comprises a state-of-charge (SOC) estimate.

7. The method of claim 1 , wherein the state estimate for the battery comprises data representative of one or more of battery capacity, and battery resistance.

8. The method of claim 1 , further comprising controlling operation of at least one of the battery or a system powered by the battery based on the determined state estimate for the battery.

9. The method of claim 1 , further comprising periodically configuring the RNN according to one or more of age of the battery, and a long-term condition of the battery.

10. The method of claim 1 wherein the recurrent neural network includes a memory cell configured to be selectively updated according to inputs to the recurrent neural network.

11. The method of claim 10 wherein the recurrent neural network comprises a Long Short-Term Memory cell.

12. A system for monitoring battery operation comprising:

at least one sensor for measuring one or more battery attributes of at least one battery; and

a recurrent neural network (RNN) system, encoding characteristic behavior of a battery at multiple different ambient temperatures, configured according to stored numerical parameters to:

receive a time series of values for one or more battery attributes for the battery from the at least one sensor, the one or more battery attributes comprising at least a current ambient temperature, wherein the RNN system is trained based on backward-pass through the RNN system of gradients of a loss function, representing the disparity between an actual state-of-charge value for the battery and a predicted value, computed for training set values corresponding to at least some of the multiple different ambient temperatures provided to the RNN system during a respective forward-pass corresponding to the backward pass, and

determine, based on the received time series comprising the current ambient temperature, a state estimate for the battery;

wherein the RNN system comprises a memory cell configured to be selectively updated according to inputs to the recurrent neural network.

13. The system of claim 12 wherein the one or more battery attributes further comprise:

a battery voltage;

a battery current; and

a battery temperature.

14. The system of claim 12 wherein the at least one battery comprises a Lithium-ion (Li-ion) battery.

15. The system of claim 12 wherein the state estimate for the at least one battery comprises a state-of-charge (SOC) estimate.

16. A non-transitory computer readable media programmed with instructions, executed by a processor, which performs the functions of:

receiving by a recurrent neural network (RNN), encoding characteristic behavior of a battery at multiple different ambient temperatures, a time series of values for one or more battery attributes for the battery, the one or more battery attributes comprising at least a current ambient temperature, wherein the RNN is trained based on backward-pass through the RNN of gradients of a loss function, representing the disparity between an actual state-of-charge value for the battery and a predicted value, computed for training set values corresponding to at least some of the multiple different ambient temperatures provided to the RNN during a respective forward-pass corresponding to the backward pass; and

determining by the RNN encoding the characteristic behavior of the battery at the multiple different ambient temperatures, based on the received time series comprising the current ambient temperature, a state of charge estimate for the battery.