Electrical grid control and optimization

1. A method performed by one or more data processing apparatus, the method comprising:

determining a plurality of simulated load curves, comprising, for each device in a population of devices:

determining, for each control from a set of multiple feasible controls for the device, a respective simulated load curve that simulates a load generated by the device based on application of the control to the device;

obtaining a plurality of weight values that comprises, for each of the plurality of simulated load curves, a respective weight value corresponding to the simulated load curve;

adjusting the plurality of weight values during a plurality of optimization iterations using an optimization technique to optimize a loss function that measures:

a sparsity of the plurality of weight values, and

an error between: (i) an aggregate load curve that is defined by combining the simulated load curves in accordance with the plurality of weight values, and (ii) a target load curve that defines a target for the aggregate load curve;

determining, for each device in the population of devices, a respective control to be applied to the device from the set of multiple feasible controls for the device based on the plurality of weight values after a final optimization iteration, comprising, for each device:

determining that the control to be applied to the device is a control corresponding to a simulated load curve having a highest weight value from among all the simulated load curves for the device; and

transmitting instructions to cause the respective control be applied to at least one device from the population of devices.

2. The method of claim 1 , wherein adjusting the plurality of weight values during the plurality of optimization iterations using the optimization technique to optimize the loss function comprises:

optimizing the loss function subject to a constraint that, for each device from the population of devices, a norm of the weight values corresponding to simulated load curves for the device has a predefined value.

3. The method of claim 2 , wherein the norm is a Euclidean norm and the predefined value of the Euclidean norm is one.

4. The method of claim 1 , wherein adjusting the plurality of weight values during the plurality of optimization iterations using the optimization technique to optimize the loss function comprises:

optimizing the loss function subject to a constraint that each weight value be non-negative.

5. The method of claim 1 , wherein the loss function comprises a sum of a first term that measures the sparsity of the plurality of weight values and a second term that measures the error between: (i) the aggregate load curve that is defined by combining the simulated load curves in accordance with the plurality of weight values, and (ii) the target load curve that defines the target for the aggregate load curve.

6. The method of claim 1 , wherein the aggregate load curve is defined by a linear combination of the simulated load curves, wherein each simulated load curve is scaled in accordance with the weight value corresponding to the simulated load curve.

7. The method of claim 1 , wherein for each device and for each control from the set of multiple feasible controls for the device, determining a respective simulated load curve that simulates load generated by the device based on application of the control to the device comprises:

generating a plurality of possible load curves that each simulate load generated by the device based on application of the control to the device; and

determining the simulated load curve by combining the plurality of possible load curves.

8. The method of claim 1 , wherein the optimization technique is an alternating direction method of multipliers (ADMM) optimization technique.

9. A system comprising:

one or more computers; and

one or more storage devices communicatively coupled to the one or more computers, wherein the one or more storage devices store instructions that, when executed by the one or more computers, cause the one or more computers to perform operations comprising:

determining a plurality of simulated load curves, comprising, for each device in a population of devices:

determining, for each control from a set of multiple feasible controls for the device, a respective simulated load curve that simulates a load generated by the device based on application of the control to the device;

obtaining a plurality of weight values that comprises, for each of the plurality of simulated load curves, a respective weight value corresponding to the simulated load curve;

adjusting the plurality of weight values during a plurality of optimization iterations using an optimization technique to optimize a loss function that measures:

a sparsity of the plurality of weight values, and

an error between: (i) an aggregate load curve that is defined by combining the simulated load curves in accordance with the plurality of weight values, and (ii) a target load curve that defines a target for the aggregate load curve;

determining, for each device in the population of devices, a respective control to be applied to the device from the set of multiple feasible controls for the device based on the plurality of weight values after a final optimization iteration, comprising, for each device:

determining that the control to be applied to the device is a control corresponding to a simulated load curve having a highest weight value from among all the simulated load curves for the device; and

transmitting instructions to cause the respective control be applied to at least one device from the population of devices.

10. The system of claim 9 , wherein adjusting the plurality of weight values during the plurality of optimization iterations using the optimization technique to optimize the loss function comprises:

optimizing the loss function subject to a constraint that, for each device from the population of devices, a norm of the weight values corresponding to simulated load curves for the device has a predefined value.

11. The system of claim 10 , wherein the norm is a Euclidean norm and the predefined value of the Euclidean norm is one.

12. The system of claim 9 , wherein adjusting the plurality of weight values during the plurality of optimization iterations using the optimization technique to optimize the loss function comprises:

optimizing the loss function subject to a constraint that each weight value be non-negative.

13. The system of claim 9 , wherein the loss function comprises a sum of a first term that measures the sparsity of the plurality of weight values and a second term that measures the error between: (i) the aggregate load curve that is defined by combining the simulated load curves in accordance with the plurality of weight values, and (ii) the target load curve that defines the target for the aggregate load curve.

14. The system of claim 9 , wherein the aggregate load curve is defined by a linear combination of the simulated load curves, wherein each simulated load curve is scaled in accordance with the weight value corresponding to the simulated load curve.

15. The system of claim 9 , wherein for each device and for each control from the set of multiple feasible controls for the device, determining a respective simulated load curve that simulates load generated by the device based on application of the control to the device comprises:

generating a plurality of possible load curves that each simulate load generated by the device based on application of the control to the device; and

determining the simulated load curve by combining the plurality of possible load curves.

16. The system of claim 9 , wherein the optimization technique is an alternating direction method of multipliers (ADMM) optimization technique.

17. One or more non-transitory computer storage media storing instructions that when executed by one or more computers cause the one or more computers to perform operations comprising:

determining a plurality of simulated load curves, comprising, for each device in a population of devices:

determining, for each control from a set of multiple feasible controls for the device, a respective simulated load curve that simulates a load generated by the device based on application of the control to the device;

obtaining a plurality of weight values that comprises, for each of the plurality of simulated load curves, a respective weight value corresponding to the simulated load curve;

adjusting the plurality of weight values during a plurality of optimization iterations using an optimization technique to optimize a loss function that measures:

a sparsity of the plurality of weight values, and

an error between: (i) an aggregate load curve that is defined by combining the simulated load curves in accordance with the plurality of weight values, and (ii) a target load curve that defines a target for the aggregate load curve;

determining, for each device in the population of devices, a respective control to be applied to the device from the set of multiple feasible controls for the device based on the plurality of weight values after a final optimization iteration, comprising, for each device:

determining that the control to be applied to the device is a control corresponding to a simulated load curve having a highest weight value from among all the simulated load curves for the device; and

transmitting instructions to cause the respective control be applied to at least one device from the population of devices.

18. The non-transitory computer storage media of claim 17 , wherein adjusting the plurality of weight values during the plurality of optimization iterations using the optimization technique to optimize the loss function comprises:

optimizing the loss function subject to a constraint that, for each device from the population of devices, a norm of the weight values corresponding to simulated load curves for the device has a predefined value.

19. The non-transitory computer storage media of claim 18 , wherein the norm is a Euclidean norm and the predefined value of the Euclidean norm is one.

20. The non-transitory computer storage media of claim 17 , wherein adjusting the plurality of weight values during the plurality of optimization iterations using the optimization technique to optimize the loss function comprises:

optimizing the loss function subject to a constraint that each weight value be non-negative.