Towards Scalable Quantum Unit Commitment in Power Systems: Encoding and Complexity Reduction Techniques

AI summary

Overview

Research area: Quantum computing for power systems operations — specifically quantum optimisation applied to the Unit Commitment (UC) scheduling problem, combining a qubit-reduction encoding (Pauli Correlation Encoding, PCE) with a hybrid quantum-classical post-processing method (Sample-based Quantum Diagonalisation, SQD).

Technical level: Advanced. The paper is written for readers comfortable with QUBO/Ising formulations, adiabatic quantum evolution, and Hamiltonian encoding.

Scope (one sentence): The paper proposes and evaluates an end-to-end quantum UC pipeline — binary UC formulation, QUBO/Ising mapping, optional PCE compression, adiabatic evolution in Qibo, optional SQD refinement, and decoding back to the original UC space — to test whether qubit reduction and post-processing can preserve feasible, low-cost schedules as the number of generators grows.

What This Paper Is About

Unit Commitment is the problem of deciding which power generators to switch on or off so that demand is met at minimum cost. Quantum optimisation is attractive for this because the on/off decisions are naturally binary, but the number of qubits required grows quickly with the number of generators. The paper asks whether two add-on techniques — PCE, which compresses many logical variables into fewer qubits, and SQD, which classically refines the measured results — can make quantum UC more scalable without destroying the ability to recover feasible and near-optimal schedules.

Key Contributions

1. A complete QUBO/Ising formulation of UC for adiabatic quantum evolution. The UC problem is cast as a binary commitment model, converted to QUBO via a penalty term, normalised (cost normalisation $\alpha^C=\sum_g c_g$; power normalisation $\alpha^P=\gcd({P_g}, D)$), and mapped to an Ising Hamiltonian with a dimension-aware penalty $\lambda=\lambda_{\min}(1+1/N)$ using multiplier $\eta=0.5$. This is implemented in the Qibo framework.

2. PCE as a qubit-reduction preprocessing stage for UC. Each logical spin is replaced by a product of reduced Pauli-$Z$ operators over a support set, reducing the search space from $2^N$ to at most $2^m$ with $m<N$. Support sets are chosen by a demand-shell-guided search that weights feasible configurations by cost via $w(\mathbf{u}) \propto \exp[-\lambda_{\text{cost}}C(\mathbf{u})]$, and correlation orders are controlled by a parameter $k$ (the paper gives $k=(1,2)$ as an example).

3. SQD as a hybrid post-processing stage. A subspace is built only from measured bitstrings, capped at $K_{\text{target}}=\rho_{\text{SQD}}d$ and $M=\min(K_{\text{target}},U)$, then an auxiliary Hamiltonian mixing transverse-field and problem terms is projected onto that subspace and diagonalised classically. Because a Pauli-$X$ flips one qubit, sampled states couple only when they differ in exactly one bit.

4. A four-way comparative framework. The baseline adiabatic approach, PCE-only, SQD-only, and combined PCE+SQD are evaluated under identical criteria in the original UC space (feasible probability mass, optimal/best-known probability mass, relative qubit reduction $R_q$, relative optimality gap, and separated PCE/adiabatic/SQD runtimes). The authors state that, to their knowledge, this is the first framework combining PCE preprocessing and SQD post-processing for power-systems UC.

Main Findings

• PCE reduces the effective encoded problem size: the abstract reports that PCE reduces the effective size of the encoded problem, with the reduction bounded by the mapping from $2^N$ to at most $2^m$.
• SQD concentrates probability on useful solutions: SQD is reported to improve the concentration of probability on feasible and optimal solutions.
• The combination performs best overall: the abstract states that applying PCE and SQD together provides the best overall behaviour in several scalable case studies, achieving more reliable solutions at lower effective computational cost than the unreduced formulation.
• Qubit reduction is not free: the paper explicitly notes that the interaction order can increase during reduction — a quadratic Ising model can become an effective higher-order (HUBO) formulation on fewer variables — and that only configurations representable by the chosen encoding can be recovered after decoding.
• SQD is bounded by what was sampled: SQD "cannot introduce configurations outside the sampled subspace," so its quality depends on the adiabatic-and-measurement stage generating representative states.
• Numerical results are not available in the retrieved content: the paper text provided is truncated before the results section, so specific values for qubit counts per case study, feasible/optimal probabilities, optimality gaps, and runtimes ($T_{\text{PCE}}$, $T_{\text{AE}}$, $T_{\text{SQD}}$, $T_{\text{tot}}$) are not reported here. The abstract refers only to "several scalable case studies" without naming instances, generator counts, or datasets.

Methodology in Plain English

The authors build a pipeline with a fixed backbone and two optional add-ons, so four variants can be compared fairly.

1. Formulate. The full UC problem with generation costs, start-up costs, demand balance, and min/max output limits is simplified: ramp costs are ignored, so each time period can be solved independently and the time index drops out. What remains is: choose on/off states $u_g$ to minimise $\sum_g c_g u_g$ subject to $\sum_g P_g u_g = D$.

2. Convert to a quantum-friendly cost function. The equality constraint becomes a squared penalty added to the objective (QUBO), with coefficients normalised so that different cost and power scales behave comparably. The penalty weight is set automatically from an instance-dependent minimum, plus a small $1/N$ margin. Binary variables are then flipped into $\pm1$ spins and promoted to Pauli-$Z$ operators, giving an Ising Hamiltonian with local fields, pairwise couplings, and a constant offset.

3. Optionally compress (PCE). Instead of giving every generator its own qubit, several generators can be encoded as products of a smaller set of qubits. The support sets are not hand-picked: because demand is known in advance, the algorithm looks for configurations that balance demand ("demand shell") and prefers encodings that retain the cheap ones, using an exponential cost weighting.

4. Evolve adiabatically. Start in an equal superposition (the ground state of a transverse-field driver $-\sum_q X_q$) and slowly interpolate to the problem Hamiltonian, $H(t)=(1-t/T)H_{\text{easy}}+(t/T)H_{\text{problem}}$, using Qibo with identical evolution settings across variants so the encoding effect is isolated. Then measure repeatedly to get bitstring frequencies.

5. Optionally refine (SQD). Keep the most frequent measured bitstrings, forming a subspace of size $M$. Build an auxiliary Hamiltonian that mixes a transverse field with the problem Hamiltonian via a parameter $\beta$, project it onto that subspace (diagonal entries are problem energies, off-diagonal entries are one-bit-flip couplings weighted by $1-\beta$), and classically diagonalise the resulting $M\times M$ matrix. The ground eigenvector gives refined amplitudes over the sampled states — no new quantum evolution is run.

6. Decode and score in the original space. Reduced or original bitstrings are mapped back to UC commitment schedules, checked for demand balance and cost, and scored by probability mass on feasible schedules, probability mass on optimal/best-known schedules, qubit reduction, and optimality gap.

Why This Matters

Quantum optimisation papers in power systems often demonstrate a formulation without asking whether the reduced or sampled object still contains the schedules an operator actually cares about. This paper makes that question central: it evaluates everything after decoding back into the original UC space, and it separates the cost of encoding, evolution, and post-processing so the trade-offs are visible.

Real-world applications:

• Day-ahead market clearing and generation scheduling, where on/off decisions across a fleet must meet forecast demand at minimum cost.
• Unit commitment in systems with high renewable penetration, where commitment decisions are revisited frequently and fast, approximate re-optimisation has value.
• Microgrid and energy-community dispatch, where small fleets of assets make near-term quantum or hybrid hardware plausible.
• Transmission or distribution planning studies that embed commitment sub-problems inside larger combinatorial searches.

Industry relevance: the work targets the practical bottleneck for utilities and system operators — qubit counts and simulation cost — and offers two levers (fewer qubits, cheaper post-processing) that do not require new hardware. The framing around runtime accounting and optimality gap is the kind of evidence an operator would need before considering a hybrid solver in a production workflow.

Future Directions

• Report and stress-test scalability limits: the results section is not available in the retrieved content, so the quantitative crossover point at which PCE compression stops preserving good schedules — as a function of generator count and compression level $m$ — remains the key open question.
• Extend beyond the single-period, ramp-free simplification: restoring the time index $t$, start-up indicators $y_{g,t}$, ramp costs, and minimum up/down times would make the formulation match operational practice, but would multiply the qubit count and interact non-trivially with PCE.
• Systematically tune the encoding and SQD knobs: correlation orders $k$, the cost-prioritisation weight $\lambda_{\text{cost}}$, the retained fraction $\rho_{\text{SQD}}$, and the mixing parameter $\beta$ are all design choices whose sensitivity is not established in the available text.
• Test on real quantum and annealing hardware: the study is simulator-based (Qibo, with compute acknowledged on BSC's Quantum-Blue and Quantum-Red and CESGA's Finisterrae-QMIO-Qulacs), so noise, connectivity, and sampling overhead on physical devices are untested.
• Investigate the higher-order blow-up from PCE: since substituting Pauli strings into a quadratic Ising model can produce higher-order terms, methods for keeping the reduced Hamiltonian implementable are a natural follow-on.

Target Audience

Quantum computing researchers working on combinatorial optimisation and Hamiltonian encoding; power systems researchers interested in whether quantum methods can handle operational scheduling; and hybrid algorithm developers interested in combining compression with classical post-processing. Readers need familiarity with QUBO/Ising mappings and adiabatic evolution to follow the methodology, though the framing of the trade-off between qubit reduction and solution recoverability is accessible to a broader optimisation audience.

Authors’ abstract

This paper develops a Quantum optimisation Unit Commitment (UC) framework that benefits from Pauli Correlation Encoding (PCE) as a qubit reduction preprocessing technique and Sample-based Quantum Diagonalisation (SQD) as a hybrid post-processing refinement method. The UC problem is formulated in binary form, then transformed into a QUBO problem and mapped onto an Ising representation suitable for adiabatic quantum evolution within the Qibo framework. This study analyses how PCE and SQD techniques affect qubit scalability, computational effort, feasibility, and probability of recovering the optimal solution as the number of generators increases. The results show that PCE reduces the effective size of the encoded problem, SQD improves the concentration of probability on feasible and optimal solutions, and their combined application provides the best overall behaviour in several scalable case studies, achieving more reliable solutions with lower effective computational cost than the unreduced formulation. The repository is available open source in Github (link is not yet available until work is published. However, if requested by the reviewers, we can make it open for a scific time window, if requested).