Optimizing power flows using harmony search with machine learning

1. A method for optimizing power flows using a harmony search, comprising:

decoupling electrical phases in a multi-phase power generation system into individual phase agents in a multi-phase power flow model for separately controlling at least one of phase variables or parameters;

ranking and selecting one or more harmony segments from harmony memory based on a utility value determined for each of the decoupled phases;

performing the harmony search with gradient descent learning to move the selected harmony segments to a better local neighborhood;

determining a new utility value for each of the selected segments based on historical performance, wherein the harmony memory is iteratively updated with the selected segments if one or more of the new utility values are higher than a utility value of a worst harmony segment stored in the harmony memory, and

dispatching power load based on the harmony search to minimize overall power distribution system costs.

2. The method recited in claim 1 , wherein the multi-phase power flow model is a three-phase power flow model.

3. The method recited in claim 2 , wherein the phase agents independently perform three harmony searches in parallel, the three harmony searches corresponding to the each of the decoupled phases.

4. The method recited in claim 1 , further comprising:

morphing the harmony segments into new harmony segments using the gradient descent learning during the harmony search; and

updating the harmony memory with a new utility value for the morphed harmony segments, the new utility value being determined based on historical performance.

5. The method recited in claim 1 , wherein the phase agents manage phase-related constraint violations and partial system cost increments.

6. The method recited in claim 1 , wherein the power flows are unbalanced power flows.

7. The method recited in claim 1 , further comprising improvising a new harmony, wherein the new harmony is improvised by at least one of selecting from original sets in harmony memory, moving to a new neighborhood, or random generation.

8. The method as recited in claim 1 , wherein at least one of the variables or parameters is randomly generated.

9. A system for optimizing power flows using a harmony search, comprising:

a power conditioning device for connecting to a power generation system to sense and decouple electrical phases in a multi-phase power flow model into corresponding individual phase agents for separately controlling at least one of phase variables or parameters; and

a processor configured to:

rank and select one or more harmony segments from harmony memory based on a utility value determined for each of the decoupled phases;

perform the harmony search with gradient descent learning to move the selected harmony segments to a better local neighborhood;

determine a new utility value for each of the selected segments based on historical performance, wherein the harmony memory is iteratively updated if one or more of the new utility values are higher than a utility value of a worst harmony segment stored in the harmony memory, and

dispatch power load based on the harmony search to minimize overall power distribution system costs.

10. The system recited in claim 9 , wherein the multi-phase power flow model is a three-phase power flow model.

11. The system recited in claim 10 , wherein the phase agents independently perform three harmony searches in parallel, the three harmony searches corresponding to the each of the decoupled phases.

12. The system recited in claim 9 , further comprising:

a harmony improviser configured to morph the harmony segments into new harmony segments using the gradient descent learning during the harmony search; and

updating the harmony memory with a new utility value for the morphed harmony segments, the new utility value being determined based on current and historical performance.

13. The system recited in claim 9 , wherein the phase agents manage phase-related constraint violations and partial system cost increments.

14. The system recited in claim 9 , wherein the power flows are unbalanced power flows.

15. The system recited in claim 9 , further comprising an improviser configured to an improvise a new harmony, wherein the new harmony is improvised by at least one of selecting from original sets in harmony memory, moving to a new neighborhood, or random generation.

16. The system recited in claim 9 , wherein at least one of the variables or parameters is randomly generated.

17. A non-transitory computer-readable storage medium including a computer-readable program, wherein the computer-readable program when executed on a computer causes the computer to perform the steps of:

decoupling electrical phases in a multi-phase power generation system into individual phase agents in a multi-phase power flow model for separately controlling at least one of phase variables or parameters;

ranking and selecting one or more harmony segments from harmony memory based on a utility value determined for each of the decoupled phases;

performing a harmony search with gradient descent learning to move the selected harmony segments to a better local neighborhood;

determining a new utility value for each of the selected segments based on historical performance, wherein the harmony memory is iteratively updated with the selected segments based on the new utility values until a stopping criteria is reached, and

dispatching power load based on the harmony search to minimize overall power distribution system costs.

18. The non-transitory computer-readable storage medium of claim 17 , wherein the multi-phase power flow is a three-phase power flow, and wherein the phase agents independently perform three harmony searches in parallel, the three harmony searches corresponding to each of the decoupled phases.