Microgrid Optimisation Platform
An interactive simulation of classical and quantum solvers for the Unit Commitment & Economic Dispatch problem — deciding which generators to turn on and how much power each should produce to meet demand at minimum cost. The platform models the full grid across a 9-layer hierarchy — solvers currently operate on L1 Supply; remaining layers are quantum solver targets.
How the solvers connect
Start with Gaussian Elimination to fit cost curves from raw generator data. Those curves feed into the dispatch solvers. Greedy gives you an instant answer; Branch & Bound gives you the provably optimal one. Simulated Annealing scales to larger fleets where B&B becomes too slow.
Compare Solvers
Run all classical and quantum solvers on the same problem — see cost, runtime, and gap side by side.
Quantum Solvers
Quantum annealing on QUBO-formulated microgrid siting and operational decisions using D-Wave Advantage.
Gate-based quantum optimization using QAOA and VQE on IBM's superconducting qubit systems via Qiskit.
Continuous-variable quantum optimization via QCI's Dirac-3 entropy quantum computing system for non-binary microgrid variables.
Classical Solvers
TC(q) = a + bq + cq² + dq³
Fits a cubic total-cost curve to observed (power, cost) data points by solving a 4×4 linear system via row reduction. Produces the MC and AC curves used by every downstream solver.
Foundation — all solvers inherit cost structure from here.
sort by cᵢ, commit until Σpᵢ = D
Ranks generators by variable cost and commits the cheapest first until demand is met. O(n log n), instant, but ignores start-up costs and may miss the global optimum.
Fast baseline — useful when speed matters more than optimality.
min Σ(cᵢpᵢ + SᵢUᵢ) s.t. Uᵢ ∈ {0,1}
Enumerates binary commitment combinations using a search tree. LP relaxation at each node gives a lower bound; branches that can't beat the incumbent are pruned. Returns proven optimal within the node cap.
Gold standard — compare all other solvers against this.
accept ΔC < 0 or with P = e^(−ΔC/T)
Jointly optimises commitment bits and output levels. Each iteration either flips a unit on/off or perturbs an output. Bad moves are accepted probabilistically to escape local minima, with temperature cooling over time.
Scalable heuristic — handles larger fleets where B&B is too slow.