distributed-resource-optimization¶

Distributed optimization algorithms for flexible resource coordination

Carrier-agnostic implementations of ADMM, COHDA, and averaging consensus — run in a single asyncio process or across a network via mango-agents.

Get started

Tutorials

GitHub

pip install distributed-resource-optimization

Algorithms¶

ADMM Sharing

Coordinate flexible resources so their aggregate output matches a target vector. Solves a convex QP locally at each agent; a central coordinator handles the global update.

ADMM

ADMM Consensus

Drive all agents to agree on a shared solution vector while satisfying individual box and coupling constraints.

ADMM

COHDA

Fully distributed heuristic for combinatorial schedule selection. Agents gossip solution candidates until the system configuration stabilises.

COHDA

Averaging Consensus

Distributed averaging of a parameter vector, with optional gradient corrections for economic dispatch and price-signal coordination.

Averaging Consensus

Quick look¶

ADMM Sharing

import asyncio
from distributed_resource_optimization import (
    create_admm_flex_actor_one_to_many,
    create_sharing_target_distance_admm_coordinator,
    create_admm_sharing_data, create_admm_start,
    start_coordinated_optimization,
)

async def main():
    flex1 = create_admm_flex_actor_one_to_many(10, [0.1,  0.5, -1.0])
    flex2 = create_admm_flex_actor_one_to_many(15, [0.1,  0.5, -1.0])
    flex3 = create_admm_flex_actor_one_to_many(10, [-1.0, 0.0,  1.0])
    coordinator = create_sharing_target_distance_admm_coordinator()
    start = create_admm_start(create_admm_sharing_data([-4, 0, 6], [5, 1, 1]))
    await start_coordinated_optimization([flex1, flex2, flex3], coordinator, start)
    print(flex1.x, flex2.x, flex3.x)

asyncio.run(main())

COHDA

import asyncio
from distributed_resource_optimization import (
    create_cohda_participant,
    create_cohda_start_message,
    start_distributed_optimization,
)

async def main():
    actor1 = create_cohda_participant(1, [[0.0, 1, 2], [1, 2, 3]])
    actor2 = create_cohda_participant(2, [[0.0, 1, 2], [1, 2, 3]])
    start = create_cohda_start_message([1.2, 2.0, 3.0])
    await start_distributed_optimization([actor1, actor2], start)
    print(actor1.memory.solution_candidate.schedules.sum(axis=0))

asyncio.run(main())